Reviews in Mineralogy - Geochemistry Volume 72 - Diffusion in Minerals - Melts

Reviews in Mineralogy & Geochemistry
Vol. 72 pp. 1-4, 2010 1

Copyright © Mineralogical Society of America
Diffusion in Minerals and Melts: Introduction

Youxue Zhang
Department of Geological Sciences
The University of Michigan
Ann Arbor, Michigan, 48109-1005, U.S.A.
youxue@umich.edu
Daniele J. Cherniak
Department of Earth & Environmental Sciences
Rensselaer Polytechnic Institute
Troy, New York, 12180, U.S.A.
chernd@rpi.edu
INTRODUCTION: RATIONALE FOR THIS VOLUME

Because diffusion plays a critical role in numerous geological processes, petrologists
and geochemists (as well as other geologists and geophysicists) often apply diffusion data
and models in a range of problems, including interpretation of the age of rocks and thermal
histories, conditions for formation and retention of chemical compositional and isotopic zoning
in minerals, controls on bubble sizes in volcanic rocks, and processes influencing volcanic
eruptions. A major challenge in the many applications of diffusion data is for researchers to
find relevant and reliable data. For example, diffusivities determined in different labs may differ
by orders of magnitude. Sometimes the differences are a result of limitations not recognized in
certain diffusion studies due to the materials or methodologies used. For example, diffusivities
determined through bulk analyses are often orders of magnitude greater than those obtained
from directly measured diffusion profiles; the former are often affected by cracks, extended
defects and/or other additional diffusion paths whose influence may not be recognized without
direct profiling. Differences in depth resolution of analytical techniques may also contribute to
discrepancies among measured diffusivities, as can the occurrence of non-diffusional processes
(e.g., convection, crystal dissolution or surface reaction) that may compromise or complicate
diffusion experiments and interpretations of results. Sometimes the discrepancies among datasets
may be due to subtle variations in experimental conditions (such as differing oxygen fugacities,
pressures, or variations in H2O content of minerals and melts used in respective experimental
studies). Experts in the field may be able to understand and evaluate these differences, but
those unfamiliar with the field, and even some experimental practitioners and experienced
users of diffusion data, may have difficulty discerning and interpreting dissagreements among
diffusion findings. For those who want to investigate diffusion through experiments, it is critical
to understand the advantages and limitations of various experimental approaches and analytical
methods in order to optimize future studies, and to obtain a clear sense of the “state of the art”
to put their own findings in perspective with earlier work.
Two early books were important landmarks in diffusion studies in geology. One was a
special publication by Carnegie Institution of Washington edited by Hofmann et al. (1974)
titled “Geochemical Transport and Kinetics.” The other was a Reviews of Mineralogy volume
edited by Lasaga and Kirkpatrick (1981) titled “Kinetics of Geochemical Processes.” Various
recent tomes are available on diffusion theory in metallurgy, chemical engineering, materials
1529-6466/10/0072-0001$05.00 DOI: 10.2138/rmg.2010.72.1
2 Zhang & Cherniak
science, and geology (e.g., Kirkaldy and Young 1987; Shewmon 1989; Cussler 1997; Lasaga
1998; Glicksman 2000; Balluffi et al. 2005; Mehrer 2007; Zhang 2008) and the mathematics
of solving diffusion problems (e.g., Carslaw and Jaeger 1959; Crank 1975). There have also
been summaries of geologically relevant diffusion data (e.g., Freer 1981; Brady 1995), review
articles and book chapters presenting diffusion data for specific mineral phases (e.g., Yund 1983;
Giletti 1994; Cherniak and Watson 2003) and for specific species in minerals and melts (e.g.,
Chakraborty 1995; Cole and Chakraborty 2001; Watson 1994) and applications of diffusion
in geology (e.g., Ganguly 1991; Watson and Baxter 2007; Chakraborty 2008). However, there
is no single resource that reviews and evaluates a comprehensive collection of diffusion data
for minerals and melts, and previously published summaries of geologically-relevant diffusion
data predate the period in which a large proportion of the existing reliable diffusion data have
been generated. This volume of Reviews in Mineralogy and Geochemistry attempts to fill this
void. The goal is to compile, compare, evaluate and assess diffusion data from the literature for
all elements in minerals and natural melts (including glasses). Summaries of these diffusion
data, as well as equations to calculate diffusivities, are provided in the chapters themselves
and/or in online supplements. Suggested or assessed equations to evaluate diffusivities under a
range of conditions can be found in the individual chapters. The aim of this volume is to help
students and practitioners to understand the basics of diffusion and applications to geological
problems, and to provide a reference for and guide to available experimental diffusion data
in minerals and natural melts. It is hoped that with this volume students and practitioners
will engage in the study of diffusion and the application of diffusion findings to geological
processes with greater interest, comprehension, insight, and appreciation.
SCOPE AND CONTENT OF THIS VOLUME

This volume begins with three general chapters. One chapter presents the basic theoretical
background of diffusion (Zhang 2010), including definitions and concepts encountered in later
chapters. This chapter is not meant to be comprehensive, as detailed, book-length treatments of
diffusion theory can be found in other sources. Some discussion of advanced topics of diffusion
theory and mechanisms can be found in individual chapters throughout the volume, including
models for diffusion in melts (Lesher 2010), multi-species diffusion (Zhang and Ni 2010),
multicomponent diffusion (Liang 2010; Ganguly 2010), and defect chemistry (Chakraborty
2010; Cherniak and Dimanov 2010; Van Orman and Crispin 2010). Diffusion data for minerals
and melts are most commonly obtained through experimental studies which require analyses
of the experimental products; these considerations are reflected in the topics of the next two
chapters. For readers who are interested in carrying out experimental research or understanding
experimental results and diffusion data, the second general chapter (Watson and Dohmen 2010)
covers experimental methods in diffusion studies, with focus on nontraditional and emerging
methods. Additional discussion of experimental methods in diffusion studies is provided in
Ganguly (2010) and Farver (2010). The third general chapter reviews a range of analytical
techniques applied in analyses of diffusion experiments (Cherniak et al. 2010). Experimental
methods and analytical techniques are also described in other chapters in the context of
discussion of specific diffusion studies.
The next five chapters are on diffusion in melts (including glasses), focusing on natural
melts relevant in geological systems. Zhang and Ni (2010) discuss the diffusion of H, C and O
in silicate melts, which involves multi-species diffusion, where one species (such as molecular
H2O) may contribute to the diffusion of two elements (such as H and O in this case). They also
assess the relative importance of various diffusing species, and extract oxygen diffusion data in
hydrous silicate melts from diffusion data for water. Behrens (2010) offers a thorough review
and evaluation of noble gas diffusion data for natural silicate melts and industrial glasses. Lesher
(2010) elaborates on the various diffusion models for self diffusion, tracer diffusion, isotopic
Introduction to Diffusion in Minerals and Melts 3
diffusion and trace element diffusion. Zhang et al. (2010) summarize available diffusion data
(focusing on effective binary diffusivities) of all elements in silicate melts. Liang (2010)
presents a systematic assessment of multicomponent diffusion studies for silicate melts.
The next eleven chapters review and evaluate diffusion data for minerals. Farver (2010)
reviews H and O diffusion data for a range of mineral phases and examines the effect of oxygen,
hydrogen and water fugacities on diffusion. Noble gas diffusion in minerals, notably diffusion
of the important radiogenic nuclides 40Ar and 4He for application in closure temperature
determinations and thermochronometry, is reviewed by Baxter (2010). Ganguly (2010) assesses
cation diffusion data in garnet, with discussion of multicomponent diffusion in garnet and its
geological applications. Chakraborty (2010) focuses on diffusion in (Fe,Mg)2SiO4 polymorphs
(olivine, wadsleyite and ringwoodite) with a discussion of the role of defects in diffusion and
the effects of pressure on diffusion in these phases. Diffusion of major and trace elements
in pyroxenes, amphibole, and mica is discussed by Cherniak and Dimanov (2010). Cherniak
(2010a) reviews diffusion data for feldspars, examining the effects of feldspar composition on
diffusion in this common crustal mineral. Cherniak (2010d) summarizes diffusion data for the
silicate phases quartz, melilite, silicate perovskite, and mullite. Van Orman and Crispin (2010)
discuss diffusion in oxide minerals including periclase, magnesium aluminate spinel, magnetite,
and rutile, and explore the intricacies of defect chemistry and its effects on diffusion in these
deceptively simple compounds. Cherniak (2010b) reviews diffusion in the accessory minerals
zircon, monazite, apatite, and xenotime, phases important in geochronologic studies. Diffusion
in other minerals, including carbonates, sulfide minerals, fluorite and diamond, is reviewed
by Cherniak (2010c). Brady and Cherniak (2010) take a broad overview of extant diffusion
data for minerals, examining possible relations among diffusivities for various mineral phases
and diffusants to assess trends and correlations that may be of value in developing or refining
predictive models and empirical relations.
The next two chapters discuss the specialized topics of grain-boundary diffusion and
computational methods for determining diffusion coefficients. Dohmen and Milke (2010)
present existing data for grain boundary diffusion in polycrystalline materials, discuss
theoretical underpinnings and the different types of grain-boundary diffusion regimes, and
outline mathematical treatments and experimental approaches for quantifying grain-boundary
diffusion. Computation of diffusion coefficients using ab initio methods and molecular
dynamics simulations are reviewed by De Koker and Stixrude (2010) with focus on recent
progress and what the future may bring for these rapidly-developing techniques.
The final chapter is devoted to geological applications of diffusion data (Mueller et al.
2010). The applications outlined include not only forward problems of applying diffusion
theory and data to infer rates and extents of diffusion-related processes, but also inverse
problems of thermochronology and geospeedometry.
REFERENCES
Balluffi RW, Allen SM, Carter WC, Kemper RA (2005) Kinetics of Materials. Wiley-Interscience, Hoboken, N.J
Baxter EF (2010) Diffusion of noble gases in minerals. Rev Mineral Geochem 72:509-557
Behrens H (2010) Noble gas diffusion in silicate glasses and melts. Rev Mineral Geochem 72:227-267
Brady JB (1995) Diffusion data for silicate minerals, glasses, and liquids. In: Mineral Physics and Crystallography,
A Handbook of Physical Constants, Reference Shelf 2. Ahrens TJ (ed) AGU, Washington, D.C., p 269-290
Brady JB, Cherniak DJ (2010) Diffusion in minerals: an overview of published experimental diffusion data. Rev
Mineral Geochem 72:899-920
Carslaw HS, Jaeger JC (1959) Conduction of Heat in Solids. Clarendon Press, Oxford
Chakraborty S (1995) Diffusion in silicate melts. Rev Mineral Geochem 32:411-504
Chakraborty S (2008) Diffusion in solid silicates; a tool to track timescales of processes comes of age. Ann Rev
Earth Planet Sci 36: 153-190
4 Zhang & Cherniak
Chakraborty S (2010) Diffusion coefficients in olivine, wadsleyite and ringwoodite. Rev Mineral Geochem
72:603-639
Cherniak DJ (2010a) Cation diffusion in feldspars. Rev Mineral Geochem 72:691-734
Cherniak DJ (2010b) Diffusion in accessory minerals: zircon, titanite, apatite, monazite and xenotime. Rev
Cherniak DJ (2010c) Diffusion in carbonates, fluorite, sulfide minerals, and diamond. Rev Mineral Geochem
72:871-897
Cherniak DJ (2010d) Diffusion in quartz, melilite, silicate perovskite, and mullite. Rev Mineral Geochem
72:735-756
Cherniak DJ, Dimanov A (2010) Diffusion in pyroxene, mica and amphibole. Rev Mineral Geochem 72:641-
690
Cherniak DJ, Hervig R, Koepke J, Zhang Y, Zhao D (2010) Analytical methods in diffusion studies. Rev Mineral
Geochem 72:107-169
Cherniak DJ, Watson EB (2003) Diffusion in zircon. Rev Mineral Geochem 53:113-143
Cole DR, Chakraborty S (2001) Rates and mechanisms of isotopic exchange. Rev Mineral Geochem 43:83–223
Crank J (1975) The Mathematics of Diffusion. Clarendon Press, Oxford
Cussler EL (1997) Diffusion: Mass Transfer in Fluid Systems. Cambridge Univ. Press, Cambridge, England
de Koker N, Stixrude L (2010) Theoretical computation of diffusion in minerals and melts. Rev Mineral
Geochem 72:971-996
Dohmen R, Milke R (2010) Diffusion in polycrystalline materials: grain boundaries, mathematical models,
and experimental data. Rev Mineral Geochem 72:921-970
Farver JR (2010) Oxygen and hydrogen diffusion in minerals. Rev Mineral Geochem 72:447-507
Freer R (1981) Diffusion in silicate minerals and glasses: a data digest and guide to the literature. Contrib
Mineral Petrol 76:440-454
Ganguly J (2010) Cation diffusion kinetics in aluminosilicate garnets and geological applications. Rev Mineral
Geochem 72:559-601
Ganguly J (ed) (1991) Diffusion, Atomic Ordering, and Mass Transport: Selected Topics in Geochemistry.
Advances in Physical Geochemistry, Vol. 8. Springer
Giletti BJ (1994) Isotopic equilibrium/disequilibrium and diffusion kinetics in feldspars. In: Feldspars and their
reactions. NATO Advanced Study Institutes Series. Series C: Mathematical and Physical Sciences. Parsons
I (ed) D. Reidel Publishing, Dordrecht-Boston, 421:351-382
Glicksman ME (2000) Diffusion in Solids: Field Theory, Solid-State Principles, and Applications. Wiley, New
York
Hofmann AW, Giletti BJ, Yoder HS, Yund RA (1974) Geochemical Transport and Kinetics. Carnegie Institution
of Washington Publ., Vol 634. Washington, DC
Kirkaldy JS, Young DJ (1987) Diffusion in the Condensed State. The Institute of Metals, London
Lasaga AC (1998) Kinetic Theory in the Earth Sciences. Princeton University Press, Princeton, NJ
Lasaga AC, Kirkpatrick RJ (eds) (1981) Kinetics of Geochemical Processes. Reviews in Mineralogy, Vol 8.
Mineralogical Society of America, Washington DC
Lesher CE (2010) Self-diffusion in silicate melts: theory, observations and applications to magmatic systems.
Rev Mineral Geochem 72:269-309
Liang Y (2010) Multicomponent diffusion in molten silicates: theory, experiments, and geological applications.
Mehrer H (2007) Diffusion in Solids: Fundamentals, Methods, Materials, Diffusion-Controlled Processes.
Springer, Berlin
Mueller T, Watson EB, Harrison TM (2010) Applications of diffusion data to high-temperature earth systems.
Shewmon PG (1989) Diffusion in Solids. Minerals. Metals & Materials Society, Warrendale, PA
Van Orman JA, Crispin KL (2010) Diffusion in oxides. Rev Mineral Geochem 72:757-825
Watson EB (1994) Diffusion in volatile-bearing magmas. Rev Mineral 30:371-411
Watson EB, Baxter EF (2007) Diffusion in solid-Earth systems. Earth Planet Sci Lett 253:307-327
Watson EB, Dohmen R (2010) Non-traditional and emerging methods for characterizing diffusion in minerals
and mineral aggregates. Rev Mineral Geochem 72:61-105
Yund RA (1983) Diffusion in feldspars. In: Feldspar Mineralogy, Short Course Notes 2. Ribbe P (ed)
Mineralogical Society of America, p 203-222
Zhang Y (2008) Geochemical Kinetics. Princeton University Press, Princeton, NJ
Zhang Y (2010) Diffusion in minerals and melts: theoretical background. Rev Mineral Geochem 72:5-59
Zhang Y, Ni H (2010) Diffusion of H, C, and O components in silicate melts. Rev Mineral Geochem 72:171-
225
Zhang Y, Ni H, Chen Y (2010) Diffusion data in silicate melts. Rev Mineral Geochem 72:311-408
Vol. 72 pp. 5-59, 2010 2
Diffusion in Minerals and Melts:

Theoretical Background
Youxue Zhang
The University of Michigan
Ann Arbor, Michigan, 48109-1005, U.S.A
youxue@umich.edu
INTRODUCTION
Diffusion is due to thermally activated atomic-scale random motion of particles (atoms,
ions and molecules) in minerals, glasses, melts, fluids, and gases (Fig. 1). The random motion
leads to a net flux when the concentration (more strictly speaking, the chemical potential) of
a component is not uniform. Even though diffusion is a microscopic process, it can lead to
macroscopic effects. For example, the initial phase of explosive volcanic eruptions (or more
commonly encountered champagne eruptions) is powered by bubble growth, which in turn is
controlled by diffusion that brings gas molecules into bubbles. This chapter provides a brief
review of the theory of diffusion in minerals and melts (including glasses). More complete
coverage of diffusion theory can be found in Crank (1975), Kirkaldy and Young (1987),
Shewmon (1989), Cussler (1997), Lasaga (1998), Glicksman (2000), Balluffi et al. (2005),
Mehrer (2007), and Zhang (2008).
In minerals, diffusive
Youxue Zhang (Ch 2) transport is the only
Page mechanism
1 for particles to move from one
location to another. For example, homogenization of a zoned crystal and loss of radiogenic

Figure 1. An example of random motion of particles. Initially (the left panel), all A particles (such as Fe2+
ions in garnet) represented by filled circles are in the lower side, and all B particles (such as Mg2+ ions in
garnet) Fig.
represented by open
1. An example circles
of random are in
motion the upperInitially
of particles. side. (the
Dueleftto panel),
random motion,
all A particlesthere
(suchisas aFenet fluxinof A
2+ ions
from the lower side to the upper side, and a net flux of B from the upper side to the lower side (the middle
and right panels).
garnet) As timebyincreases,
represented A are
filled circles andinBthewill eventually
lower become
side, and all randomly
B particles (such asand
Mg2+uniformly distributed in
ions in garnet)
the whole system. This situation for diffusion is often encountered in diffusion experiments and is referred
represented by open circles are in the upper side. Due to random motion, there is a net flux of A from the lower side
to as a diffusion couple.
to the upper side, and a net flux of B from the upper side to the lower side (the middle and right panels). As time
1529-6466/10/0072-0002$10.00 DOI: 10.2138/rmg.2010.72.2
increases, A and B will eventually become randomly and uniformly distributed in the whole system. This situation
for diffusion is often encountered in diffusion experiments and is referred to as a diffusion couple.
6 Zhang
nuclides (such as 40Ar from the decay of 40K) from a mineral are through diffusion. In silicate
melts, mass transport can be through either diffusion or flow (or convection). Only diffusion
is covered in this chapter. Even when convection is present, it is still necessary to understand
diffusion because in the boundary layer mass transport is through diffusion. Diffusion also plays
a role during crystal growth and dissolution in a melt, key processes in magma solidification
and evolution.
One of the most important geological applications of diffusion is the inverse problem, to
infer the details of thermal histories and factors such as closure temperature, apparent equilibrium
temperature, and cooling rates from diffusion properties (Zhang 2008). Thermochronology
and its application to the understanding of tectonic uplift and erosion rates, require a thorough
understanding of diffusion in minerals.
The mathematics of diffusion is complicated. An excellent reference book is by Crank
(1975), which provides analytical solutions to many diffusion problems. The mathematical
description of diffusion is similar to that of heat conduction. Hence, analytical solutions to heat
conduction problems (e.g., Carslaw and Jaeger, 1959) can also be applied to diffusion. Because
the mathematical treatment is in itself specialized and can be found in the aforementioned
treatises, in this chapter, I focus on concepts of diffusion relevant to geological and experimental
diffusion studies, rather than the mathematical solutions. Solutions for specific diffusion
problems will be given without derivations.
FUNDAMENTALS OF DIFFUSION
Basic concepts
The German physiologist Adolf Fick (1829-1901) investigated diffusive mass transport
and proposed the following phenomenological law that describes diffusion by analogy to
Fourier’s law of heat conduction
∂C
J = −D (1)
∂x
where J is the diffusive mass flux (a vector), D is the diffusion coefficient (also referred to
as the diffusivity), C is the concentration of the component under consideration (in mass per
unit volume, such as kg/m3, or number of atoms per m3, or mol/m3), x is distance, ∂C/∂x is
the concentration gradient (a vector), and the negative sign means that the direction of the
diffusive flux is opposite to the direction of the concentration gradient (i.e., diffusive flux
goes from high to low concentration, but the gradient is from low to high concentration).
Hence, when the concentration gradient is large (i.e., the concentration profile is steep), the
diffusive flux is also large. The unit of D is length2/time, such as m2/s, mm2/s, and µm2/s (1
m2/s = 106 mm2/s = 1012 µm2/s). The value of the diffusivity is an indication of the “rate”
of diffusion and, hence, is essential in quantifying diffusion. Diffusivities depend on several
factors, including temperature, pressure, composition, and physical state and structure of the
phase, and sometimes oxygen fugacity. Some general relations between diffusivities and other
parameters will be presented later in this chapter. Diffusivity values in various systems are the
main focus of this volume, of which this chapter is a part.
When diffusion is mentioned without special qualification, it refers to volume diffusion
occurring inside a phase due to thermally activated random motion (in contrast to grain-
boundary diffusion or eddy diffusion in natural waters). Typical values of diffusion coefficients
are (see Fig. 2 for diffusivity of a neutral gas species as a function of temperature; see also
Watson and Baxter 2007 for generalized diffusion behaviors in geological materials):
• In gas, D is large, about 10−5 m2/s in air at 300 K;
Youxue Zhang (Ch 2) Page 2
Theoretical Background of Diffusion in Minerals and Melts 7
Figure 2. Ar diffusion data in air (gas) (calculated using relations in Cussler 1997), water (liquid) (Wise
and Houghton 1966), basalt melt (Nowak et al. 2004), rhyolite melt (Behrens and Zhang 2001) and the
mineral Fig.
hornblende (Harrison
2. Ar diffusion data in1981).
air (gas) (calculated using relations in Cussler 1997), water (liquid) (Wise and Houghton
1966), basalt melt (Nowak et al. 2004), rhyolite melt (Behrens and Zhang 2001) and the mineral hornblende
• In aqueous
(Harrison 1981). solution, D is intermediate, about 10−9 m2/s in water at 300 K;
• In silicate melts, D is small, about 10−11 m2/s at 1600 K for divalent cations;
• In minerals, D is extremely small, about 10−17 m2/s at 1600 K for divalent cations.
Grain-boundary diffusion is diffusion along interphase interfaces, including mineral-
fluid interfaces (or surfaces) or mineral-mineral interfaces. Eddy (or turbulent) diffusion in
fluid phases is due to non-thermal random disturbances such as waves, fish swimming, boats
cruising, etc. Hence, eddy diffusion is fundamentally different from thermally activated volume
diffusion. Both grain-boundary diffusivities and eddy diffusivities are often several orders of
magnitude higher than the respective volume diffusivities listed above.
In Fick’s first law, the diffusive flux is related to the concentration gradient. In diffusion
studies, often we need to determine how a concentration profile would evolve with time given
the initial concentration distribution. For this purpose, we need an equation (referred to as the
diffusion equation) to describe how the concentration is related to space and time, such as
C(x,t) for the one-dimensional case. The one-dimensional diffusion equation often takes the
following form
∂C ∂ 2C
=D 2 (2)
∂t ∂x
where D is independent of C and x. Equation (2) is also referred to as Fick’s second law.
Below is a derivation of Equation (2) from Equation (1) and the mass balance condition.
Consider diffusion across a thin sheet with the left side at x and the right side at x+dx (thickness
of dx). Assume that the flux is one-dimensional along the x direction (Fig. 3). Then the total
mass variation in the volume defined by thickness dx and an arbitrary area S and equals the
flux into the sheet from the left side (x), JxS, minus the flux out of the sheet from the right side
(x+dx), Jx+dxS:
8 Zhang
Figure 3. Sketch of fluxes into and out of an

Jx Jx+dx element volume. The flux along the x-axis
C points to the right (the x-axis also points to
the right). The flux at x is Jx, and that at x+dx
is Jx+dx. The net flux into the small volume
is (Jx − Jx+dx), which causes the mass and
density in the volume to vary.
x x+dx
Fig. 3. Sketch of fluxes into and out of an element volume. The flux along the x-axis points to the right (the x-axis
also points to the right). The flux at x is Jx, and that at x+dx is Jx+dx. The net flux into the small volume is (Jx -
∂C ∂J ( x )
Sdx = J x S − J x + dx S = − x
Jx+dx), which causes the mass and density in the volume to vary.
Sdx (3)
∂t ∂x
where Jx (a scalar) is the flux along increasing x direction (the vector flux J = Jxi where i is the
unit vector along the x axis). Hence,
∂C ∂J ( x )
=− x ( 4)
∂t ∂x
Combining the above with Fick’s first law (Eqn. 1) leads to:
∂C ∂ ∂C
= (D ) (5)
∂t ∂x ∂x
If D is independent of C and x, the above is simplified to Equation (2).
In three dimensions, the diffusive flux for a component (Eqn. 1) takes the following form:
J = – D∇C (6)
the mass balance equation (Eqn. 4) becomes:
∂C
= −∇ ⋅ J ( 7)
∂t
and the diffusion equation (Eqn. 5) becomes:
∂C
= ∇ ⋅ ( D∇C ) (8)
∂t
where ∇ is the gradient operator when it is applied to a scalar C, and the divergent operator
when it is applied to a vector ∇C (i.e., ∇ turns a scalar to a vector and a vector to a scalar).
From Equation (2), it can be seen that if ∂2C/∂x2 = 0 at a position (e.g., point 1 in Fig. 4),
i.e., if C is locally a linear function of x (including the case of constant concentration), then
∂C/∂t = 0, meaning that the concentration at the position would not vary with time. If ∂2C/∂x2 >
0 at the position (point 2 in Fig. 4; concave up), then ∂C/∂t > 0, meaning that the concentration
at the position would increase with time. If ∂2C/∂x2 < 0 at the position (point 3 in Fig. 4;
concave down), then the concentration at the position would decrease with time.
Although we often talk about diffusion “rate”, and the rate is related to the diffusion coeffi-
cient, diffusion is a peculiar process in which there is no single diffusion “rate”. From solutions
Youxue Zhang (Ch 2)
Theoretical Background
Page 4
of Diffusion in Minerals and Melts 9
Figure 4. Concentration profile C versus x, and the 4. Concentration profile∂C2C/∂x

Fig.corresponding 2 and the corresponding ∂2C/∂x2 versus x (arbitrary units) to illustrate
versus x, versus x (arbitrary units) to
illustrate whether C increases, decreases or stays the same with time. At point ∂2C/∂x
1, with
whether C increases, decreases or stays the same
2
time. At = 0 1,and
point hence
∂2C/∂x 2 = 0 and hence ∂C/∂t = 0. At point 2
2 2 2 2
∂C/∂t = 0. At point 2 (concave up), ∂ C/∂x > 0 and hence ∂C/∂t > 0. At point 3 (concave down), ∂ C/∂x
(concave up), ∂2C/∂x2 > 0 and hence ∂C/∂t > 0. At point 3 (concave down), ∂2C/∂x2 < 0 and hence ∂C/∂t < 0.
< 0 and hence ∂C/∂t < 0.
of the diffusion equation, the diffusion distance is proportional not to duration, but to the square
root of duration; the relation is often written as
x ≈ Dt ( 9)
This distance may also be referred to as the mid-concentration distance, or half distance of
diffusion (Zhang 2008, p. 201-204), which will become clearer later. Defining the diffusion
“rate” as how rapidly the diffusion distance advances with time (dx/dt), then the “rate” equals
0.5(D/t)1/2, and is infinity at t = 0 and then decreases gradually with time.
Fig. 4. Concentration profile C versus x, and the corresponding ∂2C/∂x2 versus x (arbitrary units) to illustrate
The diffusivity increases rapidly with temperature, following the Arrhenius relation (Fig. 2),
whether C increases, decreases or stays the same with time. At point 1, ∂2C/∂x2 = 0 and hence ∂C/∂t = 0. At point 2
D = D0 e − E / RT
(concave up), ∂2C/∂x2 > 0 and hence ∂C/∂t > 0. At point 3 (concave down), ∂2C/∂x2 < 0 and hence ∂C/∂t < 0. (10)
where T is the absolute temperature in K, D0 is the pre-exponential factor and equals the value
of D at T = ∞, E is the activation energy and is a positive number, and R is the universal gas
constant.
The pressure dependence of diffusivities can be either positive or negative. The following
equation is often used to describe both the temperature and pressure dependence of diffusivity
− ( E + P ∆V ) / RT
D = D0 e (11)
where ∆V is referred to as the activation volume, which can be either positive (leading to a
decrease of D with increasing P) or negative (leading to an increase of D with increasing P).
Negative ∆V is not rare. From Equation (11), the activation energy depends on pressure (when
∆V ≠ 0). Similarly, the activation volume ∆V may also depend on temperature, which would
change the form of the above equation (see later discussion).
Microscopic view of diffusion
Microscopically and statistically, diffusion can be quantified using random walk of
particles (atoms, ions, or molecules). Consider, for example, diffusion of Mg2+ (counter-
balanced by Fe2+ in the opposite direction) in garnet along any direction, labeled as the x
direction. (A cubic crystal is used here so that the effect of diffusional anisotropy does not have
to be considered.) Consider two adjacent lattice planes (left and right) at distance l apart. If the
jumping distance of Mg2+ is l and the frequency of Mg2+ ions jumping away from the original
position is f, then the number of Mg2+ ions jumping from left to right is ½nLfdt and that from
right to left is ½nRfdt, where nL and nR are the number of Mg2+ ions per unit area on the left and
10 Zhang
right planes. The factor ½ in the expressions is due to the fact that the ions in each plane can
jump to both sides, but we consider only one direction. The jumping frequency f is assumed
to be the same from left to right or from right to left, i.e., random walk is assumed. Therefore,
the net flux from the left plane to the right plane is
1
J= ( nL − n R ) f (12)
2
Since nL = lCL and nR = lCR where CL and CR are the concentrations of Mg2+ on the left and
right planes, then
1
J = l (CL − CR ) f (13)
2
Because CL – CR = –l∂C/∂x, we have
1 ∂C
J = − l2 f (14)
2 ∂x
Comparing this with Fick’s law (Eqn. 1), we have
1
D = l2 f (15)
2
Thus, microscopically, in one-dimensional diffusion, the diffusion coefficient may be
interpreted as one-half of the jumping distance squared times the overall jumping frequency.
Since l is of the order 3×10−10 m (the interatomic distance in a lattice), the jumping frequency
can be roughly estimated from D. For D ≈ 10−17 m2/s, as in a typical mineral at high temperature,
the jumping frequency is 2D/l2 ~ 220 per second. Because ion jumping requires a site to
accept the ion, the jumping frequency in minerals depends on the concentration of vacancies
and other defects. Hence, high defect concentrations lead to high diffusivities. In melts, the
jumping frequency is much higher (about 108 per second), depending on the flexibility of the
liquid structure, and may also be related to viscosity.
The above analysis can be carried forward to the full statistical treatment of random
walk using either theoretical analysis (Gamow 1961) or computer simulations (Kleinhans and
Friedrich 2007). If initially a large number (trillions) of particles were at a single position
(defined as x = 0), after more than 100 jumping steps, the distribution can be approximated
well by a continuous function. For one-dimensional diffusion, the concentration of particles at
x (or the probability of finding a particle at x) follows the Gaussian distribution:
M 2
C ( x, t ) = e − x /( 4 Dt ) (16)
( 4 πDt )1/ 2
where M is the total number of particles, all of which were initially at x = 0. This diffusion
problem is known as diffusion from an instantaneous plane source.
Various kinds of diffusion
There are many kinds of diffusion encountered in nature and experimental studies. The
definitions may differ somewhat in the literature, making it less straightforward to deal with
the terms. Below is a summary of the various kinds of diffusion described in most of the
geological literature. Because diffusion involves a diffusing species in a diffusion medium,
it can be classified based on either the diffusion medium or the diffusing species. When
considering the diffusion medium, thermally activated diffusion may be classified as volume
diffusion and grain-boundary diffusion. Volume diffusion is diffusion in the interior of a phase;
an example is the diffusion of Mg and Fe2+ in a garnet crystal, leading to the homogenization of
a garnet crystal initially zoned in Fe2+ and Mg (Ganguly 2010, this volume). Volume diffusion
is what is typically referred to when we simply say “diffusion” without further qualifiers. In
volume diffusion, the diffusion medium can be either isotropic or anisotropic. In an isotropic
diffusion medium, diffusion properties do not depend on direction. Both melts (and glasses)
and isometric minerals are isotropic diffusion media, but non-isometric minerals are in general
anisotropic diffusion media (although in some cases, the dependence of diffusivities on the
direction is weak). Anisotropic diffusion will be treated later in this chapter.
Grain-boundary diffusion is diffusion along interphase interfaces, including mineral-fluid
interfaces (or surfaces), interfaces between the same minerals, and those between different
minerals. Because many bonds are not satisfied for atoms on the interface, there are generally
very high concentrations of defects, leading to very high grain-boundary diffusivities
compared with volume diffusivities. For example, at 1473 K, the grain-boundary diffusivity
of Si at forsterite-forsterite boundaries is about 9 orders of magnitude greater than the volume
diffusivity of Si in forsterite (Farver and Yund 2000). Grain-boundary diffusion will be the
subject of a chapter in this volume (Dohmen and Milke 2010, this volume).
Considering differences in the diffusing species, diffusion can be classified as self
diffusion, tracer diffusion, or chemical diffusion that can be further distinguished as trace
element diffusion, binary diffusion, multispecies diffusion, multicomponent diffusion, and
effective binary diffusion. Below is a discussion of these terms; first the definition used in this
work is shown, then alternative definitions are also mentioned.
Self diffusion. There is no chemical potential gradient in the system in terms of elemental
composition but there is difference in the isotopic ratios (or chemical potential gradients are
present only in isotopes) (Lasaga 1998; Zhang 2008). The diffusion is monitored through
difference in isotopic fractions. For example, in a diffusion couple made of basalt melt, one
side may have a high 44Ca/SiCa ratio and the other side has normal Ca isotope ratios, but
the elemental composition of the melts in both sides of the couple is uniform (e.g., in the
experiments of LaTourrette et al. 1996; LaTourrette and Wasserburg 1997). In Figure 1,
one may view the solid circles as 44Ca-enriched Ca, and the open circles as normal Ca, and
the matrix is the haplobasalt melt. Because there are no chemical (or elemental) gradients,
the diffusivity, which often depends on chemical composition of the system, is assumed to
be constant. This works well for self diffusion without exceptions. Small differences, ≤1%
relative, in the diffusivities of different isotopes have been measured by, e.g., Richter et
al. 2008. Such small differences are important in understanding isotopic fractionation but
negligible in quantifying the self diffusion coefficient of an element.
Other definitions of self diffusion. Some authors consider self diffusion to be the
diffusion of the exact same species (not even with isotopic differences) (e.g., Lesher 2010,
this volume; Mungall, personal communication). Such self-diffusivities cannot be measured,
however. Others may use self diffusivity to mean the binary diffusivity in the hypothetical
ideal mixing case (e.g., Lesher 1994; Ganguly 2002, p 275), which was referred to as intrinsic
diffusivity by Zhang (1993). Still others may use self diffusivity to mean the diagonal
diffusivity in a multicomponent diffusion matrix (e.g., De Koker and Stixrude 2010, this
volume); multicomponent diffusion will be explained later. The diffusion described as self
diffusion in the preceding paragraph is sometimes referred to as isotopic exchange (e.g.,
Lesher, personal communication), or tracer diffusion (e.g., Ganguly 2002, p 275; Mungall,
personal communication).
Tracer diffusion. A tracer is introduced into the system with undefined low concentrations
(e.g., Fig. 5). The tracer is often a radioactive isotope, such as 134Cs used to study Cs diffusion
in albite melt (Jambon and Carron 1976), but can also be an otherwise detectable trace element
as long as there are no major concentration gradients. Some authors distinguish tracer diffusion
using a radioactive isotope versus trace element diffusion (Baker 1989). Zhang et al. (2007)
12 Zhang
(a) (b)
Figure 5. Thin-source diffusion. (a) Set up of thin-source diffusion. Initially there is no or low diffusant
in the inside of the cylinder (or disk), but one surface (in the drawing it is the top surface) has a thin layer
of the diffusant. “Thin” means much thinner than the diffusion distance. (b) The resulting diffusion profile
(C versus x where x is distance from the upper surface vertically downward) after the experiment for two
different times. Both the surface concentration and the length of the diffusion profile depend on time.
discussed the difference between 14C tracer diffusivities and CO2 trace element diffusivities
but the difference is likely due to limited spatial resolution in b-particle mapping (International
Commission on Radiation Units and Measurements 1984; Mungall 2002) when 14C tracer
diffusivities were determined. In tracer diffusion, the bulk composition of the system is roughly
uniform, with the only variation being the concentration of the tracer, and the dilution effect by
the tracer on other elemental concentrations. Hence, the diffusivity is assumed to be constant,
meaning that the diffusivity does not depend on the concentration of the component itself at
low concentrations. This is true for many components, but at least for H2O diffusion (using a
3
H tracer), it has been found that the H2O diffusivity depends on its own concentration even at
low concentration levels of hundreds or even tens of ppm (Drury and Roberts 1963). Hence,
there may be exceptions to the assumption that a tracer diffusion profile can be well described
by a constant diffusivity. Another complexity is that the deposited component containing the
diffusant (e.g., 134Cs is deposited as CsCl in the study of Jambon and Carron 1976) may have
very high solubility in the material, which would result in compositional variation, for example
in the case of Jambon and Carron (1976), from almost pure CsCl to the albite melt in a short
distance, meaning that there could be significant chemical potential gradients.
Other definitions of tracer diffusion. In the definition of some authors, tracer diffusion
would include self diffusion (or isotopic exchange) defined above (e.g., Ganguly 2010, this
volume). Other authors would include the condition that the trace element behaves as Henrian
(meaning constant activity coefficient) (e.g., Lesher 2010, this volume), which would mean
that diffusion of 3H2O would not be tracer diffusion because its chemical activity is not
proportional to its concentration (Drury and Roberts 1963).
Either self diffusion or tracer diffusion. When a radioactive tracer is introduced into a
system that contains stable isotopes of the tracer, the diffusion may be referred to as either self
diffusion or tracer diffusion. For example, 24Na tracer diffusion into an albite melt (Jambon
and Carron 1976) can also be said to be self diffusion.
Chemical diffusion. There is chemical potential gradient in major and minor components.
Among chemical diffusion, trace element diffusion, binary diffusion (also referred to as
interdiffusion or mutual diffusion), multispecies diffusion, multicomponent diffusion, and

effective binary diffusion may be distinguished.
Trace element diffusion. Some authors separate trace element diffusion from tracer
diffusion (e.g., Baker 1989), where a trace element means that the concentration is no more
than thousands of ppm). Baker (1989) discussed tracer versus trace element diffusion in which
trace elements diffusion occurs in the presence of concentration gradients of major oxides
such as SiO2 and MgO. If the concentration gradient is only in the trace element and the
concentration variation of other components is due to the dilution effect of the trace element,
the trace element diffusion is similar to tracer diffusion. If there are concentration gradients in
major components, trace element diffusion often show uphill diffusion and must be treated in
the framework of multicomponent diffusion. In this work, trace element diffusion is arbitrarily
limited to the diffusion of an element with concentrations of up to1 wt% (that is, it includes
minor elements) when the other concentration gradients are due to dilution by the diffusion
component, so that it is similar to tracer diffusion.
Binary diffusion (also referred to as interdiffusion, or mutual diffusion) refers to
diffusion in a binary system (such as MgO-SiO2 diffusion in MgO-SiO2 binary melts, or Fe-
Mg diffusive exchange in olivine). Self diffusion may be viewed as a special type of binary
diffusion (no chemical composition gradient), such as 18O-16O exchange in dry quartz. Tracer
diffusion may also be viewed as another special type of binary diffusion (the tracer exchanges
with the rest of the system). Furthermore, tracer diffusion is the limiting case of binary
diffusion. Consider Fe-Mg interdiffusion in olivine. As the system composition approaches
pure forsterite, Fe concentration becomes low, and hence the case approaches Fe tracer
diffusion in forsterite. As the system approaches pure fayalite, Mg concentration is low, and
the case approaches Mg tracer diffusion in fayalite.
Multispecies diffusion. When the diffusing component can be present in two or more
species, such as diffusion of H2O that is present as H2O molecules and hydroxyl groups
(Doremus, 1969, 1995; Zhang et al. 1991a,b), or diffusion of CO2 that may be in the form
of carbonate ions and CO2 molecules (Nowak et al. 2004), the diffusion is referred to as
multispecies diffusion. The understanding of multispecies diffusion is one of the contributions
made by geologists to the theory of diffusion (Zhang et al. 1991a,b; Zhang and Behrens 2000;
Behrens et al. 2007).
Multicomponent diffusion. If diffusive transport involves three or more components
in the system, the diffusion is referred to as multicomponent diffusion (e.g., Cussler 1976;
Lasaga 1979; Ghiorso 1987; Trial and Spera 1994; Kress and Ghiorso 1993, 1995; Liang et
al. 1997; Mungall et al. 1998). Natural melts and many minerals are multicomponent systems.
Because of the complexity in treating multicomponent diffusion, simple treatments, which
work well in some cases, but not in others, have often been applied. The most often applied
simple treatment is effective binary diffusion, discussed below.
Effective binary diffusion. When diffusion of a component in a multicomponent system
is treated as simply due to its own concentration gradient (equivalent to either (i) treating all
other components as one combined component, or (ii) ignoring the cross diffusivities in the
multicomponent diffusion matrix), the diffusion of the component is referred to as effective
binary diffusion (Cooper 1968). In the case of tracer or trace element diffusion without major
elemental concentration gradients, strictly speaking, if there is a chemical potential gradient
in the tracer component, there will also be chemical potential gradients in other components,
however small they may be. Therefore, tracer diffusion may be viewed as a special type of
effective binary diffusion. Later in this chapter, I will classify effective binary diffusion further
into the first kind of effective binary diffusion (FEBD) and second kind of effective binary
diffusion (SEBD).
14 Zhang
Equation (2) describes binary (as well as self and tracer) diffusion with a constant diffusivity
and along one direction (often in an isotropic system, but can also be an anisotropic system
along a principal axis of diffusion, as explained later). Measurable diffusion effects require at
least a binary system, because diffusion in a one-component system, even though theoretically
conceivable, does not lead to measurable or macroscopic consequences: the system is always
uniform in composition (100% of the component). Two measurably different components (e.g.,
18
O-enriched versus 16O-enriched materials, or Fe2+ and Mg2+ exchange) are a minimum for
diffusion to lead to detectable elemental or isotopic concentration profiles. Binary diffusion
is the simplest diffusion when the mathematics of diffusion is discussed (i.e., when diffusion
problems are solved). Many other types of diffusion problems, if mathematically more
complicated, are often transformed into a binary diffusion equation.
The distinction between isotropic versus anisotropic (different diffusivities along different
directions on an interphase interface) diffusion and among self diffusion, tracer diffusion, and
chemical diffusion can also be made for grain-boundary diffusion (Dohmen and Milke 2010,
this volume).
General mass conservation and various forms of the diffusion equation
Even though some simple forms of the diffusion equation have been presented earlier
(Eqns. 2, 5 and 8), to understand the general diffusion problem, it is necessary to examine
general mass conservation and diffusion. Mass conservation means that mass is conserved
except during nuclear reactions. Even during nuclear decay, mass loss from a radioactive
nuclide is still only < 0.03%. Hence, total mass is conserved for the whole system. On the
other hand, mass conservation for an element or nuclide must include sinks due to radioactive
decay and sources due to radiogenic growth. For each chemical species, reactions producing or
consuming it must be included in the mass balance equations.
For a closed system, mass conservation means that total mass in the system is constant. For
an open system, mass conservation means that the mass increase or decrease is quantitatively
due to mass flux into or out of the system. The differential form for total mass conservation in
a representative element volume is as follows:
∂ρ
= −∇ ⋅ J (17)
∂t
where r is density, J is total mass flux (whereas J in Eqn. 7 is the flux of a component), and ∇·J
is the divergence of J. The total mass flux J is related to the bulk flow and can be written as ru,
where u is the flow velocity of the bulk material. Hence, the above equation can also be written
in the following form:
∂ρ
= −∇ ⋅ (ρu ) (18)
∂t
which is the mass conservation equation, also known as the continuity equation in fluid
mechanics.
For a given species, the flux can be divided into convective (or advective) flux (i.e., flow)
and diffusive flux. As clarified by Richter et al. (1998), the distinction between diffusive flux
and convective flux is a matter of reference frame. In a barycentric (or mass-fixed) reference
frame, the convective velocity u is defined as:
N
u = ∑ wi u i (19)
i =1
where N is the number of components, wi is the mass fraction of component i, ui is the flow
velocity of i. (Similarly, volume fixed and molar fixed reference frames can be defined by
interpreting wi in Eqn. 19 to be volume fraction of i or mole fraction of i.) The diffusive flux
of any component i, or more generally, any species k, relative to the motion of the local center
of mass, can be written as follows:
J k = ρwk ( u k − u ) = Ck ( u k − u ) (20)
The conservation equation for a species depends on whether other species can react to
form or consume the species under consideration. Without reaction terms, the mass balance
equation is Equation (7). In the presence of reaction terms, the conservation for species k can
be written as:
∂Ck m dξ
= −∇ ⋅ J k + ∑ ν kj j (21)
∂t j =1 dt
where Ck is the mole concentration of k (such as mol/m3) because reaction terms are expressed
in mole concentrations, dxj/dt is the net chemical reaction rate (i.e., rate of forward reaction
minus rate of backward reaction) of reaction j, nkj is the stoichiometric coefficient of species
k in reaction j, and m is the total number of reactions, including not only the independent but
also the dependent reactions. The value of nkj is positive when component k is a product and
negative when component k is a reactant. Positive reaction terms are also called sources, and
negative reaction terms are also called sinks. Specific examples of the reaction terms can be
found below. For a binary system without reaction terms, combining the above equation with
Fick’s law in three dimensions, Jk = −D∇Ck, leads to Equation (8).
One famous example with a reaction term is mass conservation of radiogenic 40Ar in a
mineral (such as hornblende). Because 40K decays to 40Ar at a rate of le40K = le40K0e−lt, where
40
K0 is the initial concentration of 40K, l is the overall decay constant of 40K, and le is the
branch decay constant of 40K to 40Ar, the concentration of 40Ar can be expressed as
∂ 40 Ar ∂ 2 ( 40 Ar)
=D + λ e 40 K (22)
∂t ∂x 2
where 40Ar and 40K are atomic concentrations (such as mol/m3).

Another example is OH groups in a silicate melt. OH and molecular H2O (H2Om) can
convert to each other through the following reaction (Stolper 1982a,b):
H 2Om (melt ) + O(melt )  2OH(melt ) (23)
where “melt” indicates the melt phase. Assuming the above reaction is elementary, meaning
it is accomplished by a single step (which may not be correct), with forward and backward
reaction rate coefficients of kf and kb, respectively, the concentration variation of OH with time
may be expressed as
∂COH ∂ 2COH
=D + 2 kf CH2 Om CO − 2 kbCOH
2
(24)
∂t ∂x 2
where CH2Om, COH, and CO are mole concentrations (mol/m3) of H2Om, OH, and anhydrous
oxygen, and the factor 2 is the stoichiometric coefficient in Reaction 23. Experimental studies
show that the diffusive term above is negligible, and the OH concentration change is almost
entirely due to Reaction (23) (Zhang et al. 1991a).
One way to avoid dealing with chemical reactions in the diffusion equation is to use
components whose concentrations are independent of chemical reactions. One choice is to use
elemental concentrations (such as H concentration). For H2O diffusion, the convention is to use
total H2O concentration (H2Ot where CH2Ot = COH/2 + CH2Om). Then the diffusion equation would
not include reaction terms, but will include diffusive fluxes of different species, as follows:
16 Zhang
∂CH2 Ot 1 ∂  ∂CH2 Om  1 ∂  ∂C 
= −∇ ⋅ J H2 Om − ∇ ⋅ J OH =  DH2 Om +  DOH OH  (25)
∂t 2 ∂x  ∂x  2 ∂x  ∂x 
Hence, in treating the diffusion of a multispecies component, two approaches may be used.
In the first approach, the diffusion equation of a non-conservative species is used, with only
one diffusion term but with extra reaction terms. In the second approach, the diffusion for the
total component is considered, which contains diffusive contributions from different species
but does not contain the reaction terms. In the literature, the second approach is often used
(e.g., Zhang et al. 1991a,b; Zhang and Behrens 2000; Behrens et al. 2007). The most often-
encountered multispecies diffusion problems in geology are H2O, CO2, and oxygen diffusion in
silicate melts, which will be reviewed by Zhang and Ni (2010, this volume). Similar diffusion
problems that have been examined to some degree include diffusion of multivalent elements,
such as Fe (Fe2+ and Fe3+), S (S2−, S4+ in SO2, and S6+ in SO42−), Eu (Eu2+ and Eu3+), Sn (Sn2+
and Sn4+), etc. (e.g., Behrens et al. 1990; Behrens and Hahn 2009).
There are other variants of the diffusion equation. The concentration in the fundamental
diffusion equation is mole per unit volume (mol/m3), especially when there are reactions
because the stoichiometric coefficients are in terms of moles. If there are no reaction terms,
the concentration unit can also be mass per unit volume (such as kg/m3). Often concentrations
are measured as mass fractions (or wt%), or mole fractions. If the mass density of the diffusion
medium of a binary system is roughly constant, i.e., r = C1+C2 in kg/m3, is constant, then
∂w
= ∇ ⋅ ( D∇w ) (26)
∂t
where w is mass fraction C/r (or wt%) of either component. One rough example is diffusion
in silicate melts. If the bulk molar density of a binary system, i.e., r = C1+C2 in mol/m3, is
constant, then
∂X
= ∇ ⋅ ( D∇X ) (27)
∂t
where X is mole fraction of either component. One rough example is Fe2+-Mg2+ exchange in
olivine. In reality, neither molar density nor mass density is perfectly constant in a system,
but if the variation is small (e.g., < 10%), the approximations are often made in literature for
simplicity and the errors from such approximations are small. The choice of the equations is
based on convenience instead of rigorousness. In minerals, mole fractions are usually used. In
melts, mass fractions are often used. When there is large variation in density (e.g., > 10%, such
as from a mineral to a melt), concentrations in mol/m3 or kg/m3 should be used.
Equation (8) is the general equation for diffusion in a binary system without reaction
terms or multiple species. If D is constant, Equation (8) becomes
∂C
= D∇ 2C (28)
∂t
If diffusion is one-dimensional, Equation (8) becomes Equation (5). If diffusion is one-
dimensional and D is independent of C and x, then Equation (8) becomes Equation (2).
All the above equations are for binary systems and isotropic diffusion media. Diffusion in
multicomponent systems or anisotropic diffusion media is more complex and will be discussed
in later sections (and chapters). Diffusion equations in three dimensions in isotropic media are
discussed below.
Diffusion in three dimensions (isotropic media)

In general, three-dimensional diffusion is much more complicated unless the boundary
shape is simple (such as spherical surfaces) and there is high symmetry (such as spherical
symmetry). The forms of the diffusion equations are summarized below (for details, see Crank
1975; Carslaw and Jaeger 1959).
The three-dimensional diffusion equation takes the following form in Cartesian
coordinates:
∂C ∂  ∂C  ∂  ∂C  ∂  ∂C 
= D + D + D (29)
∂t ∂x  ∂x  ∂y  ∂y  ∂z  ∂z 
If D is constant, then the above becomes
∂C  ∂2 C ∂2 C ∂2 C 
= D 2 + 2 + 2  (30)
∂t  ∂x ∂y ∂z 
In cylindrical coordinates, defining x = r cosq, and y = r sinq, where r is the planar radial
coordinate, then the diffusion equation becomes:
∂C 1 ∂  ∂C  1 ∂  ∂C  ∂  ∂C 
= Dr + D + D (31)
∂t r ∂r  ∂r  r 2 ∂θ  ∂θ  ∂z  ∂z 
If (i) concentration is uniform along z (i.e., only two dimensional radial diffusion is considered),
(ii) there is rotational symmetry (i.e., C is independent of q), and (iii) D is constant, then the
above equation becomes:
∂C 1 ∂  ∂C   ∂ 2C 1 ∂C 
=D r = D  2 +  (32)
∂t r ∂r  ∂r   ∂r r ∂r 
In spherical coordinates, defining x = r sinq cosf, y = r sinq sinf, and z = r cosq, where r
is the three-dimensional radial coordinate, then the diffusion equation becomes:
∂C 1  ∂  2 ∂C  1 ∂  ∂C  1 ∂  ∂C  
=   Dr  +  D sin θ  + 2 D  (33)
∂t r 2  ∂r  ∂r  sin θ ∂θ  ∂θ  sin θ ∂φ  ∂φ  
If there is spherical symmetry (meaning C is independent of q and f) and D is constant, the
above equation is simplified to:
∂C 1 ∂  ∂C   ∂ 2C 2 ∂C 
= D 2  r2 = D  2 +  (34)
∂t r ∂r  ∂r   ∂r r ∂r 
Comparing the last terms in Equations (32) and (34), the difference between cylindrical
and spherical diffusion is only the factor of 1 or 2 in front of (1/r)∂C/(∂r), but this seemingly
trivial difference leads to completely different solutions. Equation (34) can also be written as
(for r > 0):
∂(rC ) ∂ 2 (rC )
=D (35)
∂t ∂r 2
Defining u = rC, then the above equation becomes:
∂u ∂ 2u
=D 2 (36)
∂t ∂r
18 Zhang
which has the same form of the basic equation for one-dimensional diffusion (Eqn. 2). That is,
three-dimensional diffusion in the case of spherical symmetry can be simplified to one-dimen-
sional diffusion. However, two-dimensional diffusion cannot be simplified in a similar way.
SOLUTIONS TO BINARY AND ISOTROPIC DIFFUSION PROBLEMS

This section presents solutions to binary diffusion problems in isotropic media, which are
often encountered in experimental studies and in natural systems. Solving a diffusion problem
requires knowledge of initial and boundary conditions. Experimental studies are often designed
so that the analytical solutions to extract diffusivities are simple. Geological diffusion problems
in nature are often much more complicated. The solutions below are for relatively simple
diffusion problems, and are presented without derivation, but the experimental or geological
aspects to which the solution can be applied will be explained.
For constant D, two methods are commonly applied to solve diffusion problems. One is
the Boltzmann transformation method, which is widely applied to diffusion in infinite or semi-
infinite media. The second method is separation of variables, which is applied to boundary value
problems (finite diffusion media). In addition, Laplace and other integral transforms, Green’s
function methods, and numerical methods can all be employed to solve diffusion equations.
These and other methods are covered in Carslaw and Jaeger (1959), Crank (1975), Lasaga
(1998), Glicksman (2000), and Zhang (2008). When D is not constant, analytical solutions are
often not available, and numerical solution is necessary.
Thin-source diffusion
This diffusion problem belongs to the class of problems referred to as “instantaneous
source” diffusion in infinite or semi-infinite space in which the source can be a plane (one
dimensional diffusion), a line (two dimensional), or a point (three dimensional). Mathematically,
this class of solutions is also useful in deriving other solutions. Experimentally, this is the basis
of the thin-source diffusion method, also called thin-film method, which was widely used in the
past to determine diffusivities (e.g., Jambon and Carron 1976; Hofmann and Magaritz 1977;
Behrens 1992). Thin-source diffusion means diffusion proceeds from the surface of a material
(a plane source) into the interior of a material, but with the diffusion distance much smaller than
the extent of the material (Fig. 5) so that the medium can be treated as semi-infinite.
In the jargon of diffusion mathematics, the thin-source diffusion problem is diffusion in
a semi-infinite space with the initial condition that all of the diffusant is at a single location
of x = 0; and C = 0 at x > 0. There is no additional flux from either side of the sample, which
means that ∂C/∂x = 0 at both ends (the end with the thin film is x = 0, and the other end is x =
∞ if diffusion has not reached this end). This mathematical problem is similar to that of random
walk in one dimension (Eqn. 16), except that in the thin source problem, diffusion goes in only
one direction instead of both directions. Hence, the resulting concentration profile (i.e., the
solution to this diffusion problem) is two times that in Equation (16):
M 2 2
C ( x, t ) = e − x /( 4 Dt ) = C0e − x /( 4 Dt ) (37)
( πDt )1/ 2
where x is distance measured from the surface on which the tracer was applied, C is the con-
centration of the diffusant (e.g., measured by counting the number of decays in the case of a
radiotracer), M is the initial mass of the diffusant in the thin film per applied area, C0 is the con-
centration of the diffusant at the surface (x = 0), which decreases by half as time is quadrupled.
Defining the mid-concentration distance (x1/2) as the distance at which C = C0/2, then
x1/2 = 1.6651(Dt)1/2 (38)
The above is similar to the general form of Equation (9). If the thin film thickness is < 0.1x1/2,
then the solution (Eqn. 37) applies well. Otherwise, the solution may not be accurate. If the
“thin” film thickness is > 0.2x1/2, the source is not thin any more, and the problem should
be treated as extended source diffusion or finite-medium diffusion (e.g., Zhang 2008). If the
“thin” film thickness is > 2x1/2, then the tracer diffusion is almost equivalent to a diffusion
couple (discussed in a later section), with one half being the “thin” film, and the other half
the diffusion medium of interest. In this case, the tracer diffusion becomes chemical diffusion
across two very different compositions (effective binary diffusion in a diffusion couple).
When a radiotracer is used as the diffusing species, the integrated concentrations are
often measured using the residual activity method (e.g., Jambon and Carron 1976; Behrens
1992). After the experiment, the radioactive nuclide on the surface is washed away, and the
radioactivity in the whole sample is measured. Then a thin layer of the sample (e.g., 0.005
mm) is polished off, and the total residual radioactivity of the remaining sample is measured.
And another layer is polished off, and the residual activity measured, and so on. Hence, every
measurement is total radioactivity from x to ∞ where x starts at zero (the first measurement)
and gradually increases. Hence, the solution is the integration of Equation (37):
∞ ∞
x
A( x, t ) = ∫ C ( x, t )dx = C0 ∫ e − x
2
/( 4 Dt )
dx = A0 erfc (39)
x x 4 Dt
where A is defined as the measured residual radioactivity, and erfc is the complementary error
function. To the uninitiated, the shapes of the two profiles (Eqns. 37 and 39) may appear
similar, but there are important differences between the two profiles. For example, the slope is
zero at x = 0 for Equation (37), but the slope is the steepest at x = 0 for Equation (39).
Comments about fitting data
When analytical data are fit by Equation (37) or (39), one may choose to carry out
nonlinear fit using the equations directly. In the past, this was difficult because one would have
to write a software program to do so (e.g., Press et al. 1992). More recently, nonlinear fitting
has become easier because many commercially available programs can carry out such fitting.
Another approach is to linearize the relations and do a linear fitting; an advantage is that it is
simple and visually easy to verify such relations. Hence, many authors have used linearized
fitting. Equation (37) is linearized as follows:
x2
ln C = ln C0 − ( 40)
4 Dt
A plot of lnC versus x2 would be a straight line and D can be found from the slope. Equation
(39) is linearized as follows:
x  A
= erfc −1   (41)
4 Dt  A0 
where erfc−1 means the inverse of the complementary error function. A plot of erfc−1(A/A0)
versus x would be a straight line and D can also be found from the slope.
If analytical errors are much smaller than every measured concentration (e.g., 1% relative
precision for all measured concentrations), linearized fitting will work well. However, the
relative uncertainty of measurements at low concentrations is often large. Therefore, the
error in lnC (which is the relative error for C) increases as x increases in Equation (40), and
error in erfc−1(A/A0) also increases as x increases. One must be careful either to do an error-
weighted fitting, or only use data with high relative precision (e.g., Fig. 3-29b in Zhang 2008).
Otherwise, the fit might be dominated by data with large errors and D from the fitting would
20 Zhang
not be accurate. Hence, nonlinear fitting has the advantage of handling errors much better (the
data with small concentrations and consequently large errors are not emphasized in nonlinear
fitting) and is the preferred method, especially since nonlinear fitting programs are now more
readily available. The above comments about fitting data also apply to fitting other kinds of
diffusion profiles discussed below.
Sorption or desorption
Sorption or desorption of gases into or from a mineral occurs often in nature. For example,
loss of radiogenic Ar and He (important for thermochronology) as well as other volatiles from
minerals can be considered desorption. Sorption of water into minerals and glasses occurs in
nature and can change the properties of the mineral and glasses. In diffusion studies, sorption
and desorption experiments are often undertaken to obtain effective binary diffusivities of
volatile components in melts and minerals (e.g., Dingwell and Scarfe 1985). The method has
also been applied to determine 18O diffusivities in melts and minerals under hydrous conditions
(e.g., Giletti et al. 1978).
In desorption experiments, a mineral or glass initially containing volatiles is heated in
a gas medium that is devoid of the volatile component of interest. The surface condition is
hence a zero concentration (or some low equilibrium concentration). In sorption experiments, a
mineral or glass initially free (or almost free) of the volatile component of interest is heated in
a gas or fluid containing the component of interest in the diffusion study. The surface boundary
condition is a fixed concentration of the volatile component.
Mathematically, the two problems (sorption and desorption) are similar, with the only
difference being the initial and surface concentrations. This diffusion problem is known as the
half-space diffusion problem with constant initial and surface concentrations. If the diffusivity
D is constant, and diffusion from one surface has not reached the center of the sample (hence a
semi-infinite medium), the resulting diffusion profile is as follows:
x
C = Cs + (Ci − Cs ) erf (42)
4 Dt
where erf is the error function, Ci is the initial concentration of the volatile component in the
sample, and Cs is the surface concentration. Figure 6 shows a diffusion profile during sorption
of Ar into a rhyolite melt.
For desorption experiments, if the surface concentration is zero, the solution becomes:
x
C = Ci erf ( 43)
4 Dt
For sorption experiments, if the initial concentration is zero, the solution becomes:
x
C = Cs erfc ( 44)
4 Dt
If concentration profiles can be measured, the above equations can be used to fit data and D can
be obtained. The mid-concentration distance for sorption and desorption is:
x1/2 = 0.9539(Dt)1/2 (45)
In sorption or desorption experiments, the concentration of the volatile component often
only changes by hundreds or thousands of ppm, meaning the concentration gradients of major
components are small. Hence, the diffusivity is often constant across the profile and the above
solutions can be applied. For some diffusant such as H2O in glass or minerals, even when the
concentration is low (thousands of ppm, even down to tens of ppm), the concentration profiles
Figure 6. Diffusion profile for Ar sorption into a rhyolite melt (experiment RhyAr4-0
at 1375 K and 0.5 GPa of Ar pressure; Behrens and Zhang (2001).
Fig. 6. Diffusion profile for Ar sorption into a rhyolite melt (experiment RhyAr4-0 at 1375 K and 0.5 GPa of Ar
pressure; Behrens and Zhang 2001).

cannot be fit by the above equations (e.g., Drury and Roberts 1963; Delaney and Karsten 1981;
Zhang et al. 1991a; Wang et al. 1996), signifying that D must depend on the concentration itself.
After sorption or desorption experiments, sometimes the concentration profiles cannot
be measured, but only the total mass gain or loss as a function of time is measured. If D is
constant, the total mass gain or loss from both surfaces of a parallel plate (if loss from other
surfaces is negligible) can be described by the following equation (Crank 1975):
M t 4 Dt  ∞
nL 
M∞
=
πL


1 + 2 π ∑
n =1
( −1)n ierfc 
2 Dt 
( 46)
where Mt and M∞ are the amount of the volatile component of interest entering (or exiting)
the plate of thickness L at time t and time ∞, and ierfc is the integrated complementary error
function. For small times (more specifically, when Mt/M∞ ≤ 0.6), diffusion has not reached the
center yet and the above equation can be simplified as (Crank 1975):
Mt 4 D
≈ t ( 47)
M∞ πL
That is, a plot of Mt versus t1/2 is a straight line. If D depends on concentration, the linearity
between Mt and t1/2 still holds, but the diffusivity derived from such data is an average
diffusivity, and depends on whether sorption data are averaged (from which one obtains the
diffusion-in diffusivity, Din), or desorption data are averaged (from which one obtains the
diffusion-out diffusivity, Dout). Din and Dout can be different, depending on how D depends on
concentration.
In some experiments, one single sphere, or more often, many spheres of roughly equal
radius a, are investigated for mass gain or loss to obtain diffusivities. The equation to describe
such results is (Crank 1975):
Mt Dt  ∞
na  Dt
=6 1 + 2 π ∑ ierfc −3 2 ( 48)
M∞ πa  n =1 Dt  a
22 Zhang
where Mt and M∞ are the amount of diffusant (for example, 18O in the case of oxygen diffusion
studied using an 18O tracer) entering (or exiting) the sphere of radius a at time t and time ∞.
The above equation converges rapidly for small times. Furthermore, if Mt/M∞ ≤ 0.9, the above
equation can be simplified as (Zhang 2008, p 291):
Mt 6 Dt Dt
≈ −3 2 ( 49)
M∞ π a a
In the literature (e.g., Muehlenbachs and Kushiro 1974), the following equation is also used to
fit experimental data for spheres, which converges rapidly at large diffusion times (Crank 1975):
Mt 6 ∞ 1  n2 π2 Dt 
= 1 − 2 ∑ 2 exp  −  (50)
M∞ π n =1 n  a2 
The sorption problem and the thin-source diffusion problem are similar in that in both
cases diffusion starts from a surface, but they are different in that in the sorption problem, the
surface concentration is constant, whereas in the thin-source diffusion problem, a fixed amount
of diffusant is applied on a surface so that the surface concentration decreases with time. The
solutions to the two problems are different. In fact, the solution to the sorption problem (Eqn.
44) is similar to the integration (Eqn. 39) of the thin-source problem.
Diffusion couple or triple
The diffusion couple problem is also often encountered in nature and in experimental
studies conducted to obtain self diffusivities (e.g., LaTourrette et al. 1996), interdiffusivities
(e.g., Freda and Baker 1998), effective binary diffusivities (e.g., Koyaguchi 1989; Zhang and
Behrens 2000), and multicomponent diffusion matrix (Kress and Ghiorso 1995; Liang et al.
1996). In this method, two cylinders (each is called a half) of the same phase but different
composition (for self diffusion studies the difference is only in the isotopes of the element(s)
of interest) are joined together at flat and sometimes polished and annealed surfaces (Fig. 7).
For studying diffusion in melts, the two halves are oriented vertically so that the interface is
horizontal to minimize convection. Assuming diffusion has not reached either end yet, the
diffusion problem is one-dimensional diffusion in infinite space. If the diffusivity is constant,
the concentration at the interface is the simple (not weighted) average concentration of the two
halves. One may view each half as behaving as a sorption or desorption problem with constant
concentration at the interface, so the solution would be an error function.
Define the vertical direction as z, z = 0 at the interface, and z > 0 in the upper half. The
combined solution of both halves is:
CU + CL CU − CL z
C= + erf (51)
2 2 4 Dt
where the CU and CL are the initial concentrations in the upper and lower halves. Measured
concentration profiles can be fit to the above equation to obtain D. In such fitting, CU and CL
can often be obtained from measured concentrations at the two ends (each can be obtained
by averaging many points) and can be fixed in the fitting. Hence, there is essentially only one
unknown parameter, D, to be obtained from the fitting. However, often the interface position is
not known accurately, although it may be roughly estimated. Hence, the fitting often takes the
following form:
CU + CL CU − CL z − z0
C= + erf (52)
2 2 4 Dt
where z0 (the position of the Matano interface, defined by mass balance so that the diffusant
loss from one side is equal to the diffusant gain on the other side; see Eqn. 54 below) is also
Figure 7. The diffusion couple setup and the resulting concentration profiles. On the left is a drawing of the
diffusion couple configuration with high concentration of the component of interest in the upper half, and
low concentration in the lower half (see also Fig. 1). The evolution of the concentration with time is shown
on the right for three different times (arbitrary unit).
a fitting parameter and allowed to vary to optimize the fitting. The value of z0 does not have
much meaning; it only indicates how well one estimated the interface position before the
fitting.
The definition of the mid-concentration distance takes some thought for a diffusion couple.
If it were defined as the mid-concentration between the two halves, then it would not move
at all, inconsistent with diffusive flux into a medium. The adopted definition is to consider
diffusion in each half as having a constant surface concentration. Then the mid-concentration
distance is the same as in sorption or desorption experiments, with x1/2 = 0.9539(Dt)1/2.
Some authors use diffusion triples (e.g., Behrens and Hahn 2009), which are essentially
two diffusion couples sharing one common half, in one experiment. In a diffusion triple, three
glass or mineral cylinders are stacked together as upper, middle and lower thirds, making two
diffusion couples.
In nature, diffusion between two layers of a crystal differing in elemental or isotopic
compositions may be viewed as a diffusion couple, as can diffusion between two layers
of melts (though it is difficult to avoid convection in natural systems). For the complete
homogenization of a diffusion couple, the initial concentration evolution is similar to Equation
(51), but the concentration evolution after the diffusant has reached at least one end of the
material depends on the boundary conditions at the ends (e.g., whether the ends are kept at
constant concentration or there is no flux from the outside) as well as the dimensions of the
initial two layers.
Diffusive crystal dissolution
Crystal dissolution and growth are common in magma chambers. Diffusive crystal
dissolution has been applied to obtain chemical diffusivities and to treat multicomponent
diffusion (Harrison and Watson 1983; Zhang et al. 1989; Liang 1999). Crystal dissolution
rather than crystal growth is adopted in diffusion studies because crystal dissolution can be
controlled well; for crystal growth experiments, where new crystals form cannot be well-
controlled. The modifier “diffusive” is also important: it means that convection needs to be
avoided to study diffusion.
24 Zhang
In the design of diffusive crystal dissolution, a gem-quality crystal disk and a glass cylinder
are joined vertically with a horizontal interface to minimize convection (Fig. 8). If the melt due
to the dissolution of the crystal has a higher density than the ambient (or initial) melt, the crystal
is placed at the bottom; otherwise the crystal is placed on the top to minimize convection. Thus
mass transport is entirely controlled by diffusion. At a fixed high temperature, the dissolution
of the crystal often rapidly establishes a constant melt composition at the interface (Zhang et
al. 1989; Chen and Zhang 2008, 2009), and diffusion carries the flux into the melt interior. The
diffusion is often complicated due to (i) multicomponent effects and (ii) major compositional
variation in the melt.
Figure 8. Setup of an olivine dissolution

experiment. Because the dissolution of
olivine produces a melt (interface melt) with
greater density than the initial melt, olivine
is placed at the bottom of a melt to minimize
convection in the melt. In this case, the
olivine crystal is larger in diameter than the
melt so that the edge of olivine is preserved
for the accurate determination of the olivine
dissolution distance (Chen and Zhang 2008).
For the dissolution of low-solubility minerals such as zircon, the concentration gradients
in major oxides are often negligible, and the diffusivity of the main mineral component is
roughly constant along a profile. The solution to this diffusion problem is (Zhang et al. 1989):
( x − L)
erfc
C = Ci + ( C s − Ci ) 4 Dt (53)
(− L )
erfc
4 Dt
where Ci is the initial concentration of the main mineral component (such as ZrO2) in the melt,
Cs is the concentration of the component in the interface melt, which is a fitting parameter,
and L is the melt growth distance, which is often negligible for dissolution of low-solubility
minerals (which are also slowly dissolving minerals) such as zircon.
For the dissolution of high-solubility minerals such as pyroxenes and olivine, the
concentration gradients in major oxides are significant and the above equation does not
work well for most components because of the multicomponent effects. However, for the
major mineral component (the component whose concentration in the mineral is much higher
than that in the melt, such as MgO during olivine dissolution), it is often possible to treat
its diffusion as effective binary diffusion. In such cases, Equation (53) may be applied to fit
the data to estimate the effective binary diffusivity. Furthermore, for high-solubility minerals
(which are also rapidly dissolving minerals), the melt growth distance L must be determined
independently (often from the mineral dissolution distance multiplied by the ratio of the
mineral density over the melt density) to apply Equation (53) to fit data.
In earlier experimental studies of crystal dissolution, convection was often present (e.g.,
Brearley and Scarfe 1986), but was either not considered or incorrectly treated (see Zhang et
al. 1989 for more discussion). Hence, the extracted diffusivities based on crystal dissolution
experiments in these studies were often incorrect.
Theoretically, there is also a short diffusion profile in the crystal, which is too short to
be measured. Furthermore, the dissolution of the crystal shortens the diffusion profile in the
crystal (Zhang 2008, p 378-389).
Variable diffusivity along a profile
In some diffusion experiments, the diffusivity may vary along a concentration profile.
This can happen in at least two scenarios. One is when the major element composition changes
significantly along a diffusion profile, such as in the case of Fe-Mg interdiffusion in olivine
(Chakraborty 2010, this volume), in which diffusion has a strong compositional dependence.
The other is in the case of components such as H2O, where the diffusivity varies with its
own concentration due to the effects of speciation even when the compositional variation of
major components is negligible. To solve the diffusion equation with concentration-dependent
diffusivity, numerical methods are necessary (e.g., Crank 1975; Press et al. 1994), which often
is only slightly more difficult than working on complicated analytical solutions to a diffusion
problem. In experimental studies, however, the interest is in obtaining the diffusivities from
the measured concentration profiles, which is an inverse problem.
There are two methods to extract diffusion coefficients if the diffusivity varies along a
concentration profile. In one method, the functional form of the variation of the diffusivity
with concentration is known, even though some parameters in the function are not known.
For example, the diffusivity might be proportional to the concentration: D = aC, where a is
the value of D when C = 1. Or the diffusivity may be linear in C: D = aC+b. Or the diffusivity
might be an exponential function of concentration: D = b exp(aC) (i.e. lnD is linear in C),
where b is the value of D when C = 0. If the functional form is known but not the parameters
a and b, the diffusion equation can be solved for given values of a and b, and the solution is
compared with the experimental profile. By adjusting a and b to fit the concentration profile,
the parameters can be found, so that the way in which D varies with C can be determined. The
fitting can be complicated but specific programs have been written to accomplish this task
(e.g., Zhang et al. 1991a; Zhang and Behrens 2000; Ni and Zhang 2008).
If the functional form of the dependence of D on C is not known and cannot be guessed,
then Boltzmann-Matano method, based on an application of the Boltzmann analysis by Matano
(1933), can be applied to obtain diffusivities at every point along a profile. This method is
most often applied to diffusion couples. In the original method, it is necessary to first find the
Matano interface between the two halves of the diffusion couple (which may or may not be
the physically marked initial interface between the two halves), so that x defined relative to the
Matano interface (i.e., x = 0 at the Matano interface) satisfies:
CU
∫CL
xdC = 0 (54)
where CL and CU are the concentrations at the two ends, x < 0 in the lower half of the couple, and
x > 0 in the upper half of the couple. After obtaining the Matano interface, then the diffusivity
at any x = x0 (which also means at a C corresponding to x0) can be found (Crank 1975):
CU
Dat C ( x0 ) =
∫C ( x0 )
xdC
(55)
2t ( dC / dx ) x = x0
where t is the experimental duration. The key in minimizing the errors in extracting D using
the above expression is to obtain accurate integrals and slopes, which requires smooth
concentration profiles. Often the experimental data are smoothed objectively, either manually
26 Zhang
or by some kind of piecewise fitting (because it is not known what function can fit the whole
profile). Furthermore, D values obtained using the above method near the two ends often
have large errors. If the method is applied carefully, the general trend of D versus C is often
acceptable, but small undulations may be artifacts of inaccurate slopes and integrations.
A trivial variation of Equation (55) is
C ( x0 )
−∫ xdC
CL
Dat C ( x0 ) = (56)
2t (dC / dx ) x = x0
A modified approach based on the Boltzmann analysis is provided by Sauer and Freise
(1962). The advantage of this method is that there is no need to find the Matano interface.
Define
y=
(C − CL ) (57)
(CU − CL )
which may be referred to as the normalized concentration. D can be found as follows:
1  y| ∞ (1 − y)dx + (1 − y | ) x0 ydx 
2t (dy / dx ) x = x0  0 ∫x0
Dat C ( x0 ) = x0 ∫−∞ (58)
x

Again, the key in obtaining reliable D is to obtain accurate integrals and slopes, which requires
smooth concentration profiles.
The Boltzmann analysis can also be adapted for use in studies of diffusive crystal
dissolution in order to extract diffusivities. The equation is (Zhang 2008):
Ci
1
Dat C ( x0 ) =
2t (dC / dx ) x = x0 ∫
C ( x0 )
( x − L )dC (59)
where the upper limit of the integration Ci is the initial concentration in the melt, x is the
distance from the crystal-melt interface, x0 is the position at which the diffusivity is obtained,
and L is the melt growth distance.
Homogenization of a crystal with oscillatory zoning
In nature, a crystal (such as plagioclase) may be oscillatorily zoned. Idealize the initial
oscillatory zones as follows: the concentration in the zones can be described by a sine or cosine
function (which also implies constant width of every zone):
 2πx 
CAn t =0
= a + bsin   (60)
 p 
where a is the average An content in a plagioclase crystal, b is the peak amplitude (or half of
peak-to-peak amplitude), and p is the width of each zone (or period of the oscillation), e.g.,
from one maximum to the neighboring maximum in Figure 9. As diffusion proceeds in a
closed system (nothing entering or leaving the system), the concentration profile would evolve
as·
2
L2  2πx 
CAn = a + be −4π Dt sin   (61)
 p 
where D is the diffusivity of the coupled cation exchange Ca+Al ↔ Na+Si, which changes the
concentration of the An component in plagioclase. That is, both the average An content and the
period of the zoning stay the same, but the compositional amplitude of the zoning decreases
Figure 9. Homogenization of an oscillatorily zoned

Fig. 9. Homogenization crystal
of an with
oscillatorily time.
zoned crystal with time.
exponentially with time (Fig. 9). If the initial oscillatory zoning is periodic but has sharp
boundaries, the solution would be an infinite series of sine or cosine functions.
One dimensional diffusional exchange between two phases at constant temperature
Often, two minerals in contact have common components that may be exchanged. For
example, garnet and olivine both contain Fe2+ and Mg2+, and the two cations can exchange
through diffusion (Fig. 10). Garnet and spinel may exchange Mn-Fe2+-Mg in divalent sites,
and Al-Cr-Fe3+ in the trivalent sites. The following results are from Zhang (2008, p 426-430).
Assume (i) each phase is initially uniform in composition, (ii) the exchange is between only
two components (binary diffusion), (iii) the contact interface between the two minerals is flat
(planar), (iv) either the mineral is isotropic or diffusion in an anisotropic mineral is along a
Fig. principal axis
9. Homogenization of an of diffusion
oscillatorily (see
zoned crystal withbelow
time. on diffusion in anisotropic medium), (v) diffusion has
not proceeded to the center of either mineral yet, and (vi) D in each mineral phase is constant.
Then, the problem is one dimensional and has analytical solutions. Furthermore, assume that
there is instantaneous equilibrium at the contact between the surfaces of two minerals and that
the equilibrium condition is described by a constant exchange coefficient KD (which depends
on temperature):
KD =
(X B
2 X1B )
x =+0
(62)
(X A
2 X A
)
1 x =−0
where X means mole fractions, superscripts A and B are the two mineral phases (e.g., A =
olivine and B = garnet), subscripts 1 and 2 are the two components (e.g., 1 = Mg and 2 = Fe),
X1A is the mole fraction of component 1 in mineral A, the interface is at x = 0, mineral B is on
the side of x > 0, mineral A is on the side of x < 0, “x = +0” means the surface of mineral B (x
approaches zero (interface) from x > 0 side), and “x = –0” means the surface of mineral A (x
approaches zero (interface) from x < 0 side).
The solution for the concentration evolution as a function of time is:
x
(
X1A = X1Ai + X1A, −0 − X1Ai erfc ) (63a )
4D A t
x
B
(
X1B = X1Bi + X1,+0 − X1Bi erfc ) (63b)
4DB t
where X1Ai and X1Bi are the initial mole fractions of component 1 in minerals A and B, X1,A− 0 and
28 Zhang
Olivine Garnet
Figure 10. Fe-Mg exchange between olivine and garnet. The figure on the left shows the geometry of the
diffusional exchange, with the horizontal direction being along c-axis of olivine. The figure on the right
shows the calculated diffusion profile in the two phases using Equations (63a) and (63b). KD = (Fe/Mg)Gt/
(Fe/Mg)Ol = 1.7.
B
X1,+0 are the mole fractions of component 1 at the interfaces of minerals A and B, DA and DB
are interdiffusivities between components 1 and 2 in minerals A and B. The mole faction of
component 2 in each mineral can be found by stoichiometry (e.g., the sum of mole fractions of
1 and 2 in every mineral is 1). Given initial conditions X1Ai and X1Bi, and diffusivities, there are
still two unknowns ( X1,A− 0 and X1,+0
B
) in the above two equations, which can be solved from two
equations: Equation (62) (surface equilibrium) and the following (mass balance):
(X A
1, − 0 ) (
− X1Ai ρA D A = X1,+0
B
)
− X1Bi ρB D B (64)
where rA and rB are the molar densities of components 1 and 2 in minerals A and B. For
example, ρolivine
Fe+Mg = 43.48 mol/L and ρ Fe+Mg = 25.64 mol/L if there are no other divalent cations.
garnet
Calculated profiles are shown in Figure 10b.

Spinodal decomposition
Spinodal decomposition is the spontaneous decomposition of a single phase to two phases.
For example, alkali feldspar at high temperature can be a single phase. As the temperature
becomes lower, it may spontaneously separate into two phases, albite and orthoclase. The
intergrowth of the two phases is called perthite. The separation of a single uniform phase
into two phases of similar structure is called spinodal decomposition. It is accomplished by
diffusion and thermal fluctuation. In the process, diffusion may transport elements from low
concentration to high concentration (referred to as uphill diffusion), opposite to the transport
direction during normal diffusion.
Spinodal decomposition in a binary system illustrates that diffusion is not simply
responding to concentration differences to homogenize the system, but is a response to
the chemical potential (or chemical activity) difference. Diffusion reduces the Gibbs free
energy of the system. In ideal or close to ideal binary mixtures, the entropy portion of the
Gibbs free energy dominates the total Gibbs free energy of mixing. The chemical potential
of a component increases as the concentration of the component increases. Hence, diffusion
homogenizes the system, which minimizes the Gibbs free energy of the system. In highly
non-ideal binary mixtures, the chemical potential (or activity) may decrease as concentration
increases when the enthalpy part of the Gibbs free energy dominates the total mixing energy.
Then, diffusion would still be downhill in terms of the chemical potential (or activity) gradient,
but can be uphill in terms of concentration gradient. Hence, Fick’s law is an approximation of
the following more accurate diffusion law in a binary system (Zhang 1993):
D
J=− ∇a (65)
γ
where a and g are the chemical activity and activity coefficient of either of the two components
with a = gC, and D is the “intrinsic” binary diffusivity, different from the normal binary
diffusivity D. When g is constant, the above equation reduces to Equation (6). Comparing
Equations (65) and (6), the “intrinsic” effective binary diffusivity and the normal effective
binary diffusivy are related as:
 d ln γ 
D = D 1 +  (66)
 d ln C 
The “intrinsic” binary diffusivity D is always positive, but D calculated from the above equation
is negative in the compositional range when the phase is unstable, leading to uphill diffusion.
In binary systems, uphill diffusion occurs rarely, only in cases when the phase is
unstable. On the other hand, in multicomponent systems, uphill diffusion often occurs even
when the phase is stable. Furthermore, in binary systems, diffusion is always downhill of the
chemical potential gradient of any given component, even in uphill diffusion. However, in
multicomponent systems, the diffusive flux may also be uphill against the chemical potential
gradient.
Diffusive loss of radiogenic nuclides and closure temperature
Diffusive loss of radiogenic nuclides is probably the most common application of
diffusion in geology, with critical implications for geochronology. One example is the loss of
40
Ar (produced by the decay of 40K) from minerals. Another is the loss of 4He (produced by the
decay of 238U, 235U and 232Th, and other nuclides) from minerals. 40Ar is used in the following
examples for clarity. The decay of 40K is a branched decay, with 10.48% to 40Ar and the rest
to 40Ca. Because 40Ar as a noble gas is not readily incorporated in any mineral structure, it is
especially prone to diffusive loss from a mineral, and the process is similar to desorption.
The diffusive loss occurs at high temperature and during cooling, in which the diffusivity
depends on time since the temperature varies with time. At high temperature, the diffusivity
is high, leading to extensive 40Ar loss and little 40Ar accumulation in the mineral. As the
temperature decreases, the diffusivity becomes lower, until it becomes sufficiently low
that there is little 40Ar loss, and essentially all radiogenic 40Ar accumulates. A numerical
example is as follows. For hornblende, Ar diffusivity depends on temperature as D = 2.4×10−6
exp[−268000/(RT)] = exp(−12.94 − 32257/T) where R is 8.3144 J·mol−1·K−1, T is in K, D is in
m2/s (Harrison 1981). Consider the average hornblende crystal radius to be 0.5 mm. A plutonic
rock typically cools at 10 K/Myr (e.g., Harrison and McDougall 1980), or 1 K per 100,000
years. First, consider diffusion at a typical magmatic temperature of 1200 K, at which the Ar
diffusivity is 5.1×10−18 m2/s. In 100,000 years, the diffusion distance estimated from x1/2 =
(Dt)1/2 would be 4 mm, much larger than the radius of hornblende, meaning that essentially
all radiogenic 40Ar would be lost from the hornblende. Next, consider diffusion at 300 K. The
extrapolated Ar diffusivity is 5×10−53 m2/s (which may not be accurate because of the large
down-temperature extrapolation from the experimental data range of 773-1173 K, but can
nonetheless used for an order of magnitude estimate). In 4.5×109 years (the age of the Earth;
Patterson 1956), the diffusion distance estimated from x1/2 = (Dt)1/2 is 3×10−18 m, much less
than the atomic distance of 10−10 m. That is, Ar atoms would not have moved even one atomic
layer by staying at 300 K for the whole history of the Earth, meaning all radiogenic Ar would
30 Zhang
have been kept inside the hornblende. As the rock cools down from 1200 K to 300 K, the
hornblende would become closed to Ar loss at some intermediate temperature, which is called
the closure temperature (Fig. 11).
A rigorous definition of the closure temperature is as follows. For a given cooling history
(Fig 11a), the accumulation of 40Ar in a given mineral is a continuous function. Initially, there
is little accumulation because of fast diffusion of Ar at high temperatures. Gradually, as the
temperature becomes lower, there is some accumulation, but not complete retention of all
radiogenic 40Ar. As the rock cools down further, new radiogenic 40Ar is completely retained
in the mineral. The accumulation of 40Ar in the mineral as a function of time is shown as the
thin solid curve in Figure 11b. After a geochemist collects the rock, the thermal history is
not known and the age of the rock is calculated based on the present-day 40Ar and 40K ratio
assuming the rock was a closed system from the beginning (the thick dashed line in Fig. 11b).
The calculated age is called the closure age. All geochronology methods give the closure age.
As can be seen from the difference of the thin solid curve and the thick dashed line in Figure
11b, the difference between the real age and the closure age in this case is about 5 Myr. The
temperature corresponding to 2)the closure age is defined
Youxue Zhang (Ch Pageto
11be the closure temperature, which is
about 850 K in this case. In this case, the thermal history is assumed and forward calculation is
Figure 11. Closure temperature

Fig. 11. (Tc) and(Tclosure
Closure temperature time (closure age, t ). The solid curve in (a) shows the
c) and closure time (closure age,ctc). The solid curve in (a) shows the temperature
temperature history of the mineral. The solid curve in (b) corresponds to the accumulation of 40Ar in the
mineral with time. Theofdashed
history curve
the mineral. insolid
The (b) curve
showsin the assumed 40toArtheaccumulation
(b) corresponds accumulation ofin40the
Ar inage
the calculation,
mineral with time. The
with the calculated age at tc, which corresponds to the time when the mineral temperature was Tc.
dashed curve in (b) shows the assumed 40Ar accumulation in the age calculation, with the calculated age at tc, which
corresponds to the time when the mineral temperature was Tc.

carried out so that the closure temperature can be readily determined. When rocks are studied,
the thermal history is not known and is to be determined from geochronology and other studies,
which requires the estimation of the closure temperature given the diffusion properties.
To treat 40Ar retention, it is necessary to consider diffusive loss during continuous cooling.
Solving the diffusion equation in which the diffusivity depends on time is relatively easy in
most cases by simply replacing Dt in the solutions by ∫Ddt where the integration is from t = 0
to the time of interest. Hence, the mid-concentration diffusion distance is roughly:
t
x1/ 2 ≈ ∫ Ddt
0
(67)
Now consider a mineral in a continuously cooled rock. The cooling rate is often high
initially at high temperature and then slower at lower temperature. One function to approximate
the temperature history is the so-called asymptotic cooling (Ganguly 1982; Zhang 1994):
T0
T= (68)
1 + t / τ1
where T is temperature, T0 is the initial temperature (at t = 0), and t1 is the time for the rock
to cool from T0 to T0/2. This function is often used because it leads to simple expressions in
treating diffusion during cooling. The initial cooling rate (cooling rate at T0) q is
T0
q= (69)
τ1
The diffusivity as a function of temperature is given by the Arrhenius equation (Eqn. 10), and
can be expressed as a function of time:
D = D0 e − E /( RT ) = D0 e − E (1+ t / τ1 )/( RT0 ) = DT0 e − t / τ2 (70)
where DT0 = D0e−E/(RT0) is the initial D at T = T0, and t2 = t1(RT0/E) is the characteristic time for
D to decrease from the initial value to 1/e of the initial value.
Combining Equations (70) and (67), the diffusion distance during cooling can be found as:
∞
x1/ 2 = ∫0
DT0 e − t / τ2 dt = DT0 τ2 (71)
The upper integration limit is set to be infinity because (i) it makes the final result simple, and
(ii) the upper limit can be treated as ∞ because diffusion at room temperature and slightly above
room temperature is negligible over the whole history of the Earth for Ar and most other species.
At the closure temperature, the mineral roughly becomes closed, implying that the
mineral size must be significantly larger than the diffusion distance at and below the closure
temperature Tc (i.e., treating the initial temperature to be Tc in Eqn. 71)
a 2 >> Dc τ2 (72)
where Dc is the diffusivity at Tc, a is the half thickness for a platy mineral, radius for a cylinder
and sphere. To quantify, the above condition is written as:
GD0e − E /( RTc ) τ1RTc GD0e − E /( RTc ) RTc 2
a = GDc τ2 =
2
= (73)
E qE
where q is the cooling rate when the temperature of the rock was Tc. G is greater than 1 and
the exact value depends on the shape of the mineral and must be determined by solving the
diffusion equation. Rearranging the above equation leads to Dodson’s famous equation for
closure temperature (Dodson 1973, 1979):
32 Zhang
E  GD0Tc2 
= ln  2  (74)
RTc  a qE / R 
The value of G has been determined by Dodson (1973) to be 55 for spheres (where a is the
radius), 27 for infinitely long cylinders (a is the radius) and 8.65 for plates (a is the half-
thickness).
The closure temperature of a mineral can be estimated using the above equation given the
diffusion parameters (D0 and E), the shape and radius of the mineral grains, and the cooling rate.
The dependence of the closure temperature on the cooling rate makes it necessary to estimate
the cooling rate before inferring the closure temperature, making it more complex to infer
thermal histories. For slowly cooled plutonic rocks, the closure temperature and closure age
can be obtained using different minerals and different isotopic systems, hence a temperature-
time history (thermal history) may be obtained. For rapidly cooled rocks, the application of the
closure temperature concept can confirm that the age obtained from a radiogenic system is the
true rock formation age. Below are two examples.
Consider a hornblende grain (a = 0.2 mm) in a rapidly cooled volcanic rhyolite lava
(Lewis-Kenedi et al. 2005). The estimated cooling rate is 100 K/year. Using D = exp(−12.94
− 32257/T) and Equation (74) with iteration, the closure temperature is 1304 K. That is, if the
initial temperature is below 1304 K, there would be essentially no Ar loss. Because a typical
eruption temperature for rhyolite lava is about 1200 K or lower, little Ar would be lost from
hornblende at this cooling rate and the Ar-Ar age is the eruption age. Such high cooling rates
cannot be inferred from geochronology because small age differences cannot be determined
from isotopic dating (but cooling rates of rhyolite lavas can be estimated by other methods,
e.g., Zhang 1994; Zhang et al. 2000; Xu and Zhang 2002).
Now consider a hornblende grain (a = 0.3 mm) in a slowly cooled granitoid (Harrison
and McDougall 1980). Start by assuming a cooling rate of 10 K/Myr. Using D = exp(−12.94
− 32257/T) and Equation (74) with iteration, the closure temperature is 825 K. Hence, the
Ar-Ar age of hornblende (about 110 Ma) corresponds to a temperature of about 825 K. The
zircon age of this rock is about 116 Ma. The Pb closure temperature in 60-µm zircon is about
1200 K (Cherniak and Watson 2000), higher than the placement temperature of about 1050 K
(Harrison and McDougall 1980). Hence, the U-Pb age reflects the intrusion age, and the rock
cooled from 1050 K to 825 K in about 6 Myr, about a factor of four times the assumed cooling
rate of 10 K/Myr. More closure temperature and closure age data can be obtained from other
dating systems to further constrain the cooling history and cooling rate. If necessary, better
constrained cooling rates can be used to recalculate the closure temperatures to obtain more
accurate temperature versus time histories.
The above results are for whole mineral grains when the initial temperature is sufficiently
high to have full equilibration of the diffusant in the mineral grain. Dodson (1979) has
extended the theory to individual points inside a mineral, in anticipation that age profiles may
be measured in many different points in a mineral grain. Ganguly and Tirone (1999) extended
the theory to cases when the initial temperature is not high enough to achieve full equilibration
of the diffusant in the mineral grain at the onset of cooling.
DIFFUSION IN ANISOTROPIC MEDIA

The above solutions are all for isotropic diffusion media, such as melts, glasses and
isometric minerals. In such media, the diffusivity is a scalar. Most minerals are not isometric,
and hence most diffusion problems geologists encounter are anisotropic diffusion although
diffusional anisotropy is not necessarily pronounced in many cases. Such diffusion is
complicated and almost never treated with full rigor. Rather, simplifications are made so that
the problems can be handled with reasonable effort as well as reasonable usefulness. In this
section, diffusion in anisotropic media is reviewed.
Diffusional anisotropy means that diffusivity depends on the crystallographic directions
in a mineral. For example, the diffusivity of 18O in quartz under hydrothermal conditions
along the c-axis is about two orders of magnitude greater than that along a direction in the
plane perpendicular to the c-axis (Giletti and Yund 1984). In mathematical terms, diffusivity
is a tensor, more specifically, a second-rank symmetric tensor representable by a 3×3 matrix.
Only when the diffusion medium has high internal structural symmetry, would the matrix
be simplified. For melts, glasses and isometric minerals, the diffusivities along all directions
are the same, and the tensor can be simplified to a single number, a scalar. Generally, the
diffusivity tensor is denoted as D, and its matrix form is:
 D11 D12 D13   D11 D12 D13 
   
D =  D21 D22 D23  =  D12 D22 D23  (75)
   
 D31
 D32 D33   D13 D23 D33 
where underlined terms are tensors or elements in a tensor, rather than elements in a “normal”
matrix, such as the diffusivity matrix for multicomponent diffusion (to be discussed next).
The second equal sign in the above equation utilizes the property that the diffusivity tensor is
symmetric. The diffusivity tensor in different crystallographic symmetries and the calculation
of diffusivities along a given direction (that is, measured along the direction of the concentration
gradient) are summarized in Appendix 1.
Fick’s first law of diffusion (Eqn. 1) for an anisotropic medium is written as:
J = − D∇C (76)
where J is the flux vector, and ∇C is the concentration gradient (a vector). In matrix form, the
above equation can be written as:
 Jx   D11 D12 D13   ∂C / ∂x 

    
 J y  = −  D12 D22 D23   ∂C / ∂y  (77)
    
J
  z
 D13
 D23 D33   ∂C / ∂z 
Based on mass balance (Eqn. 7), ∂C/∂t can be written as:

∂C ∂J ∂J ∂J
= −∇⋅ J = − x − y − z (78)
∂t ∂x ∂y ∂z
Therefore, the full diffusion equation for a binary system becomes
∂C ∂  ∂C ∂C ∂C  ∂  ∂C ∂C ∂C 
=  D11 + D12 + D13  +  D12 + D22 + D23  (79)
∂t ∂x  ∂x ∂y ∂z  ∂y  ∂x ∂y ∂z 
∂ ∂C ∂C ∂C 
+  D13 + D23 + D33 
∂z  ∂x ∂y ∂z 
If all the tensor components are constant, then
∂C ∂ 2C ∂ 2C ∂ 2C ∂ 2C ∂ 2C ∂ 2C
= D11 2 + D22 2 + D33 2 + 2 D12 + 2 D13 + 2 D23 (80)
∂t ∂x ∂y ∂z ∂x∂y ∂x∂z ∂y∂z
34 Zhang
The above equation can be simplified using coordinate transformation to become:

∂C ∂ 2C ∂ 2C ∂ 2C
= Dα 2 + Dβ 2 + Dγ 2 (81)
∂t ∂α ∂β ∂γ
where a, b, and g are the principal axes of diffusion, and Da, Db and Dg are the principal
diffusivities (diffusivities along the principal axes). For the above equation to be applicable,
the boundary conditions must also be transformable. The fluxes along a principal axis of
diffusion can be written as:
J α = − Dα ∂C / ∂α (82a )
J β = − Dβ∂C / ∂β (82 b)
J γ = − Dγ ∂C / ∂γ (82c)
One-dimensional diffusion equation along each principal axis of diffusion can be written
as:
∂C ∂ 2C
= Dα 2 (83a )
∂t ∂α
∂C ∂ 2C
= Dβ 2 (83b)
∂t ∂β
∂C ∂ 2C
= Dγ 2 (83c)
∂t ∂γ
Diffusion along each principal axes is hence unaffected by diffusion in other directions.
In melts, glasses and isometric minerals, any direction can be taken as the principal diffusion
direction. In non-isometric minerals, the crystallographic axes can be taken as the principal
diffusion axes if the symmetry is at least orthorhombic. In addition, in hexagonal, tetragonal,
and trigonal minerals, diffusion along any direction in the plane perpendicular to the c-axis can
be taken as a principal axis and can be treated simply as one-dimensional diffusion. Diffusion
along a direction that is not a principal axis cannot be treated as one-dimensional diffusion,
because diffusion along other directions would also contribute to the flux along the direction
of consideration; i.e., the diffusive flux J is not parallel to −∇C.
The three-dimensional diffusion equation (Eqn. 81) can be simplified further with the
following axis transformation:
α′ = α / Dα (84a )
β′ = β / Dβ (84 b)
γ′ = γ / Dγ (84c)
The transformed diffusion equation becomes

∂C ∂ 2C ∂ 2C ∂ 2C
= + + (85)
∂t ∂α′2 ∂β′2 ∂γ′2
which is identical to the diffusion equation in isotropic media with D = 1. Hence, after convo-
luted transformations, the complicated diffusion equation in anisotropic media with a constant
diffusivity tensor (Eqn. 80) is simplified to the diffusion equation in isotropic media. Hence,
in theory, analytical solutions for diffusion in anisotropic media can be obtained using these
transformations. However, the three-dimensional initial and boundary conditions must also be
transformed, which is not always easy. Furthermore, the transformations may have drastically
changed the shape of a crystal, from

the physical shape in natural coor-
dinates to the effective shape in the
transformed coordinates. The effec-
tive shape might be highly unintui-
tive. Figure 12 gives an example.
Diffusion in anisotropic media
is often treated with major simpli- Figure 12. Comparison of the physical
fications. If total loss or gain of the shape and effective shape in terms of
diffusant is of interest, as in the dif- diffusion. The upper diagram shows the
physical shape of mica (plane sheet), and
fusive loss of radiogenic nuclides in the shape on the right hand side is the
understanding closure temperature shape after transformation to lengthen the
and closure age, the shape of the vertical axis by a factor of 100 (because the
mineral grains (necessary for deter- difference in oxygen diffusivities is 4 orders
mining the shape factor) is the ef- of magnitude). Hence, diffusion along the
z direction can be ignored, and the mica
fective shape. For example, Fortier can be treated as a cylinder in terms of 18O
and Giletti (1991) showed that under diffusion.
hydrothermal conditions the 18O dif-
fusivity in mica parallel to the c-axis
is about 4 orders of magnitude slower
than diffusion in the plane perpendicular to the c-axis (even though mica is monoclinic, it is
approximately hexagonal in terms of many of its properties). Hence, the physical shape of mica
is platy, but the effective shape of mica in terms of 18O diffusion under hydrothermal conditions
is an “infinitely” long cylinder, meaning that bulk mass loss or gain is through diffusion in the
plane perpendicular to the c-axis (Fig. 12).
MULTICOMPONENT DIFFUSION
Diffusion in multicomponent systems (having three or more components) is also
complicated. In-depth treatment of multicomponent diffusion will be covered in other chapters
(Liang 2010; Ganguly 2010) in this volume. The general aspects and some simple treatments
are covered here.
Fick’s law (Eqn. 1) is for binary systems only. When three of more components (such
as A, B, C and D) are present, experiments show that the diffusion of a given component A
depends not only on the concentration gradient of A, but also on the concentration gradients
of B and C. Consider a melt with N components. Because the summation of concentrations
(such as mole fractions or mass fractions) of all components is 100%, there are only N−1
independent components in an N-component system. (For example, in a binary system, only
the concentration of one component is independent.) Let n = N−1. If the N-component system
is a stable phase (i.e., no spinodal decomposition), the diffusive flux of components can be
written as (De Groot and Mazur 1962):
J1 = – D11∇C1 – D12∇C2 ... – D1n∇Cn (86a )
J 2 = – D21∇C1 – D22∇C2 ... – D2 n∇Cn (86 b)

J n = – Dn1∇C1 – Dn 2∇C2 ... – Dnn∇Cn (86 n )
where Dij’s are diffusion coefficients for component i due to concentration gradients of
component j. Dii’s are referred to as the main or on-diagonal diffusivities, and Dij’s when i≠j
are referred to as the off-diagonal or cross diffusivities. In matrix notation, the above becomes:
36 Zhang
 J1   D11 D12 ... D1n   ∇C1 

    
 J 2  = −  D21 D22 ... D2 n   ∇C2 
(87)
 ...   ... ... ... ...   ... 
    
 Jn   Dn1 Dn 2 ... Dnn   ∇Cn 
The diffusivity matrix is not to be confused with the diffusivity tensor; they are different in
at least two aspects: (i) the meanings are different, one refers to diffusion in a multicomponent
system but isotropic medium, and the other refers to binary diffusion in an anisotropic medium;
(ii) the diffusivity tensor in anisotropic system is always represented by a 3×3 symmetric matrix,
but the diffusivity matrix (always a square matrix) is n by n where n ≥ 2 and is nonsymmetrical.
For diffusion in a multicomponent system in an anisotropic medium, the full rigorous description
would require a diffusivity matrix in which every element is a tensor. Such a diffusion problem
has not yet been solved. Even if the mathematical complexity can be overcome in the foreseeable
future, it would be impossible to obtain all the coefficients in the tensor matrix experimentally.
Because multicomponent and anisotropic minerals are common (e.g., mica, hornblende,
pyroxenes, etc), diffusion in nature is truly complicated. Hence, simplifications are usually
made out of necessity because requiring theoretical rigor would simply mean getting nowhere
in understanding geological problems which otherwise can be approximately quantified. Even
for diffusionally isotropic natural silicate melts with typically four or more major components
and numerous minor and trace components, no reliable diffusivity matrix has been obtained.
In the near future, it may be possible to extract diffusivity matrices of the major components.
However, unless theories can be developed to calculate the diffusivity matrix for trace elements
in natural melts, the diffusivity matrix approach is unlikely workable for trace elements because
it is impractical to experimentally obtain diffusivity matrices involving both major and trace
elements (about 80 components in total). Hence, in the foreseeable future, the simple effective
binary approach, or some modified simple approach, will be necessary.
In this section, the various approaches and their advantages and disadvantages are briefly
outlined.
Effective binary approach, FEBD and SEBD
The effective binary diffusion approach (Cooper 1968) is the most widely used simple
treatment of diffusion in a multicomponent system. In this approach, the diffusion of a
component in a multicomponent system is treated as diffusion in a binary system, of which
one is the component of interest and the other is all the other components combined. That is,
the flux of component i in one dimension is simply expressed as (Eqn. 6):
Di ∂Ci
Ji = − (88)
∂x
The diffusion equation is hence Equation (5). Compared to the multicomponent diffusivity
matrix approach (one dimensional form of Eqn. 87) in which Ji = −SDij∂Cj/∂x, it can be seen
that
n ∂C j / ∂x n ∂C j
Di = ∑ Dij = Dii + ∑ Dij (89)
j =1 ∂Ci / ∂x j =1, j ≠ i ∂Ci
Hence, the effective binary diffusivity depends on the on-diagonal and off-diagonal diffusivities
involving the component of interest, as well as the concentration gradients of all components.
With the effective binary approach, the solutions for binary diffusion are used for diffu-
sion of a component in a multicomponent system and to extract effective binary diffusivities.
Almost all chemical diffusion data in minerals and melts are obtained using this approach.
These chemical diffusivities are called effective binary diffusivities (EBD, which can also
mean effective binary diffusion). The approach
only works when there is no uphill diffusion
(e.g., Fig. 13); otherwise, the effective binary diffusivity would change from positive to nega-
tive in a single profile.
Figure 13. Uphill diffusion in a diffusion couple experiment (Van Der Laan et al. 1994).
Fig. 13. Uphill diffusion in a diffusion couple experiment (Van Der Laan et al. 1994).
Because the EBD approach does not work for uphill diffusion profiles, authors typically
ignore uphill diffusion profiles in data treatment, often simply marking them as “uphill”.
In binary systems, the presence or absence of uphill diffusion can be predicted from
thermodynamics: uphill diffusion would occur when the phase is not stable and undergoes
spinodal decomposition. However, in multicomponent systems, whether uphill diffusion would
occur for a given element (such as Ca or Ce) in a diffusion couple or during crystal dissolution
(Zhang et al. 1989) cannot be predicted yet.
Even in the absence of uphill diffusion, the EBD approach is still prone to many limitations.
The most serious one is that effective binary diffusivity of an element may depend not only on
the major oxide concentrations, but also on their gradients (Liang 2010, this volume). This may
sound strange but can be understood in the following example. The SiO2 diffusivity in CaO-Al2O3-
SiO2 melts along constant CaO concentration (the CaO concentration gradient is nearly zero,
and the Al2O3 concentration gradient roughly compensates the SiO2 concentration gradient) is
essentially the SiO2-Al2O3 interdiffusivity, but is essentially the SiO2-CaO interdiffusivity along
a constant Al2O3 concentration. It is hence understandable that the two are different even at the
same bulk composition, as shown by Liang et al. (1996). Along other compositional directions,
even though the diffusivity cannot be simply viewed as an interdiffusivity, the effective binary
diffusivity is expected to also depend on other compositional gradients. Because compositional
gradients can change as diffusion proceeds, especially in a finite system, the EBD can also
change as diffusion proceeds (Liang 2010, this volume).
Even knowing that it has limitations, EBD approach is still the most widely used approach
in treating diffusion in multicomponent systems because other approaches are too complicated.
In the foreseeable future, the EBD approach will still likely be the method of choice for trace
element diffusion even if the diffusivity matrix for major components could be obtained.
Hence, it is necessary to understand under what conditions the EBD approach is more reliable
than others. Cooper (1968) summarized that the EBD approach is meaningful when (i) the
concentration gradients of all components are in the same direction; and (ii) either a steady
state exists or the diffusion media is infinite or semi-infinite. However, there could still be
uphill diffusion even when the two conditions are satisfied, and uphill diffusion is difficult to
treat using the EBD approach. Below, some specific situations are discussed:
38 Zhang
(a) The EBD approach works well if the base composition of a diffusion system is the
same, but there is one relatively minor component (or a few minor to trace components)
up to several percent that diffuses in or out, such as in cases of sorption or desorption,
hydration or dehydration, bubble growth, or a diffusion couple between two rhyolite
melts with only small difference in Sr concentration. That is, the compositional
difference in the system is due to the dilution effect of the presence (or absence) of this
component under consideration. If the component diffuses more rapidly than other
components, the effective binary diffusion approach works even better. Diffusivities
from such experiments can be applied accurately to similar situations when the base
compositions are the same.
(b) In a multicomponent system, if the initial compositional difference is only between
two components such as Na2O and K2O, the diffusion of each of the two components
can be treated as effective binary diffusion. In fact, this case is similar to interdiffusion
between two components, and hence the two components should have the same inter-
diffusivity. Diffusion of other components in the system may not be treated as EBD.
(c) Even if there are major concentration gradients in multiple components, as long as
they are consistent in the direction and relative magnitude, the EBD approach can be
applied to the component with the largest concentration gradient. Examples include
dissolution of a specific mineral in a specific melt (such as olivine dissolution in a
basalt melt, or quartz dissolution in an andesite melt). In this case, the component with
the largest concentration gradient (such as MgO during olivine dissolution in a basalt
melt, or SiO2 during quartz dissolution in an andesite melt, referred to as the principal
equilibrium determining component by Zhang et al. 1989) can be treated as EBD.
Such effective binary diffusivities can be applied to the identical situations in nature
under similar boundary conditions such as semi-infinite diffusion media, meaning
MgO EBD extracted from olivine dissolution experiments in basalt melt can be
applied to predict MgO diffusion in nature during olivine dissolution in a basalt melt,
but it may not be applicable to MgO diffusion during clinopyroxene dissolution in a
similar melt, or olivine dissolution in a different melt. Diffusion of other components
may or may not be treated as EBD.
(d) In diffusion couple studies when two different melts are placed together, effective
binary diffusivities may be extracted for components showing large concentration
differences between the two halves. These EBD values may be applied to diffusion
couples with similar compositions in the two halves. However, if the concentration
gradient is switched for some major component (e.g., in one diffusion couple, Al2O3
concentration is 13 wt% in the basalt half and 17 wt% in the andesite half, whereas
in the other diffusion couple, Al2O3 is 17 wt% in the basalt half and 13 wt% in the
andesite half), EBD from one case may not be applied to the other. Liang et al. (1996)
(their Fig. 5) showed an example of this.
As can be seen from the above, effective binary diffusivities are a large category and cover
many different situations. Some EBD values are more reliable than others. For easy reference,
I propose three types of effective binary diffusivities based on their consistency and reliability:
(1) Interdiffusivity or interdiffusion (ID). Binary diffusion is interdiffusion. In multicom-
ponent systems, if the concentration gradients are primarily in two components, and
concentrations of all other components are roughly uniform (case (b) above), then it
can be referred to as multicomponent interdiffusion. In this case, the diffusivities of
the two components are roughly the same, and can be treated well and consistently
using the effective binary approach. The interdiffusivity would depend on the bulk
composition but not on the concentration gradients of other components (these are es-
sentially zero). Diffusion of other components may not be treated well by the effective
binary approach (e.g., often there can be uphill diffusion for other components).
(2) FEBD (first type of effective binary diffusion, or first type of effective binary
diffusivity). FEBD corresponds to the diffusion situation of case (a) above. Because
this situation is often encountered in experiments and nature (especially sorption/
desorption, bubble growth, and explosive volcanic eruptions), and because of the
high degree of consistency of this type of effective binary diffusivity, it is considered
to be a special type of EBD, called the first type of effective binary diffusion, with
the acronym FEBD. In principle, when the concentration of the component of interest
becomes low enough, FEBD approaches the tracer diffusivity (or trace element
diffusivity in the absence of major concentration gradients).
(3) SEBD (second type of effective binary diffusion, or second type of effective binary
diffusivity): All other types of effective binary diffusivities are less reliable and are
grouped as SEBD, even though some may be more consistent than others. SEBD
values can be applied to systems very similar to experimental systems in terms of
bulk composition as well as the direction and size of concentration differences. In
studies of diffusion in multicomponent systems, often all monotonic concentration
profiles (no uphill diffusion) are treated using the effective binary diffusion approach,
assuming a constant SEBD. The fits may not be perfect (e.g., the SEBD of MgO during
olivine dissolution in an andesite melt seems to increase with increasing SiO2 content;
Fig. 14). In such a case, one is tempted to make efforts to determine how the SEBD
varies along the profile and associate the variation with SiO2 or other concentration
changes. However, this may not be correct, because the SEBD variation might be due
Youxue Zhang (Ch 2)to concentration gradient
Page 14 variations, rather than the compositional variations. Hence,
in treating SEBD profiles, the simplest approach is to fit the profile with a constant
SEBD and ignore the small misfits because of the complexity of SEBD.
Figure 14. MgO diffusion

profile in the melt during ol-
ivine dissolution in andesite
melt (Zhang et al. 1989). The
fit curve, assuming constant
SEBD, does not match the data
well, e.g., at the region near x
= 1 mm. The slower decrease
of the concentration implies
higher diffusivity in this region
even though the SiO2 concen-
tration here is high (about 57
wt%) compared to SiO2 near x
= 0 (about 52 wt%). This misfit
is most likely due to the depen-
dence of the SEBD of MgO on
concentration gradients rather
than on the bulk composition
itself.
Modified effective binary approach (activity-based effective binary approach)

Fig. 14. MgO diffusion profile in the melt during olivine dissolution in andesite melt (Zhang et al. 1989). The fit
The
curve, assuming modified
constant effective
SEBD, does not match thebinary
data well, approach (also
e.g., at the region near xcalled theslower
= 1 mm. The activity-based effective binary
approach) was proposed by Zhang (1993) and based on rough chemical
decrease of the concentration implies higher diffusivity in this region even though the SiO2 concentration here is activity estimation
in silicate melts. It is assumed that the diffusive flux of a component i is proportional to the
high (about 57 wt%) compared to SiO2 near x = 0 (about 52 wt%). This misfit is most likely due to the dependence
of the SEBD of MgO on concentration gradients rather than on the bulk composition itself.
40 Zhang
activity gradient of the component alone (Eqn. 65):

Di
Ji = − ∇ai (90)
γi
where ai and gi are the chemical activity and activity coefficient of component i with ai = giCi,
and Di is the “intrinsic” effective binary diffusivity, different from the normal effective binary
diffusivity. The above equation reduces to Equation (88) when gi is constant. The difference
between the effective binary approach and the modified effective binary approach is that the
concentration gradient is replaced by the activity gradient. From Equation (66) (or comparing
Eqns. 88 and 90), the “intrinsic” effective binary diffusivity and the normal effective binary
diffusivity are related as:
 d ln γ i 
Di = Di  1 +  (91)
 d ln Ci 
Zhang (1993) showed that the approach could fit and predict uphill diffusion profiles during
crystal dissolution experiments (Fig. 15), which cannot be treated by effective binary diffusion.
Lesher (1994) developed a similar model to treat uphill diffusion of trace elements in a
diffusion couple. Even though the model can handle uphill diffusion profiles and may also fit
monotonic diffusion profiles better, the approach has two disadvantages: (i) it is complicated,
and (ii) the activity model for silicate melts is uncertain. The applicability (or inapplicability)
of the approach needs to be explored further.
Figure 15. Fitting an uphill

diffusion profile of FeO in an
andesite melt during olivine
dissolution using the modified
effective binary diffusion
model (from Zhang 1993).
Diffusivity matrix approach

Fig. 15. Fitting an uphill diffusion profile of FeO in an andesite melt during olivine dissolution using the modified
The diffusivity matrix approach is the classical and rigorous method to describe
effective binary diffusion model (from Zhang 1993).
multicomponent diffusion (Eqns. 86 and 87; see in-depth discussion by Liang 2010, this volume).
This approach works well if the N-component mixture is not very non-ideal. The approach fails
when the mixture is unstable, leading to spinodal decomposition because some eigenvalues
would change from positive to negative, similar to the case of spinodal decomposition in a
binary system. When the system is highly non-ideal, the individual diffusivity values would
be highly variable with composition even if the mixture is stable. For ideal and nearly ideal
systems, the diffusion equation for an N-component system (n = N−1) is of the following form:
∂Ci n
= ∑ ∇ ( Dij ∇C j ) (92)
∂t j =1
For one-dimensional diffusion and constant Dij values, the diffusion equation becomes:
∂Ci n ∂ 2C j
= ∑ Dij (93)
∂t j =1 ∂x 2
Therefore, the concentration evolution of component i depends on the concentration gradients
of other components. Equation (93) contains coupled equations. Hence, it is necessary to solve
the concentration evolution of all n components simultaneously in the N-component system. In
the matrix form, the concentrations can be solved as follows.
The one-dimensional diffusion equation with constant diffusivity matrix can be written
as:
∂C ∂ 2C
=D 2 (94)
∂t ∂x
where C is the concentration vector, or the transpose of (C1, C2, ..., Cn), and D is the diffusivity
matrix. If the N-component mixture is stable, it can be shown that all n eigenvalues of the D
matrix are real and positive (e.g., De Groot and Mazur 1962), and D can be diagonalized:
D = TlT−1 (95)
where l is a diagonal matrix made of the eigenvalues of D, and T is a matrix made of the
eigenvectors of D. Replacing Equation (95) into Equation (94) leads to
∂u ∂2u
=l 2 (96)
∂t ∂x
where u = T−1C is the transformed composition vector. Because l is a diagonal matrix, the
above equation is equivalent to:
∂ui ∂ 2u
= l i 2i (97)
∂t ∂x
where i can be 1, 2, ..., n. Hence, in the transformed compositional space, the diffusion
equation for each ui depends only on its own concentration gradient with a real and positive
diffusivity of li. If the initial and boundary conditions can also be transformed, Equation (97)
is in the same form as binary diffusion and ui can be solved. After solving for every ui, the
final solution is
C = Tu (98)
When the diffusivity matrix is not constant (e.g., some of the elements in the matrix depend
on concentration, which is common), the above analytical solution is not possible, and the
multicomponent diffusion equation must be solved numerically (this complexity also exists
for the effective binary treatment).
In principle, the diffusivity matrix (i.e., diffusion of all elements in a diffusivity matrix)
can be obtained from experimental diffusion studies, similar to binary or effective binary
diffusivities, by fitting experimental concentration profiles using the diffusivities of elements
as fitting parameters. This has been done for some simple melts (e.g., Vignes and Sabotier
1969; Sugawara et al. 1977; Liang et al. 1996; Liang and Davis 2002). However, for natural
silicate melts, due to the large number of fitting parameters involved (e.g., the diffusivity matrix
of a 10-component system is made of 81 individual values), this is a daunting task. Strategies
42 Zhang
have been proposed (Trial and Spera 1994) and bold attempts have been made (Kress and
Ghiorso 1995; Mungall et al. 1998), but the diffusivity matrices have not been verified and
more follow-up work is necessary. More on multicomponent diffusion as well as empirical
models for multicomponent diffusion matrices can be found in Liang (2010, this volume).
Activity-based diffusivity matrix approach
Even the complicated treatment of multicomponent diffusion using diffusivity matrices
is not enough to treat phase separation in multicomponent systems. A more fundamental
approach is to use the activity-based diffusivity approach as in the binary system (Eqn. 65).
The diffusive flux is expressed as (Zhang 2008)
n Dij
J i = −∑ ∇aj (99)
j =1 γj
where Dij are the “intrinsic” diffusivities. The diffusion equation for component i is hence
∂Ci n D 
= −∇J i = ∑ ∇  ij ∇a j  (100)
∂t  γj 
j =1  
The advantage of this approach is that it can treat highly non-ideal mixtures as well as spinodal
decompositions smoothly without invoking negative diffusivities. The disadvantage of this
approach is its complexity. Even the concentration-based diffusivity matrix approach is
already too complicated to be carried out for melts and many minerals. When accurate activity
models become available, it may be possible to contemplate the activity-based diffusivity
matrix approach.
Origin of the cross-diffusivity terms
In the concentration-based diffusivity matrix, the cross diffusivities are significant, often
resulting in uphill diffusion of many components. Uphill diffusion is the most clear demonstra-
tion of the cross diffusivities, and the need to include the cross diffusion terms. Furthermore,
uphill diffusion is also the best example of the failure of the concentration-based effective bi-
nary approach. One fundamental question is what factors contribute to the cross-diffusivity
terms of Dij (i≠j).
The cross-diffusivities may arise from two sources, one is non-ideality of the multicomponent
mixture (thermodynamic effect), and the other is a purely kinetic effect. Some uphill diffusion
profiles in a stable phase may be attributed to non-ideal mixing, i.e., the thermodynamic effects,
and can be modeled using activity-based diffusivities (e.g., Zhang 1993). It is not clear how
much contribution to the cross terms is from kinetic effects, although there has been some
discussion (e.g., Liang et al. 1997). One rigorous way to separate the two effects is to consider
diffusion in a perfectly ideal mixing system of different isotopes. At least three isotopes (such as
28
Si, 29Si, and 30Si) are needed so as to make a multicomponent system, with each isotope being
a component. For example, diffusion in two chemically identical haplorhyolite melt halves, one
normal in silicon isotopes, and one enriched in 30Si, can be investigated. The fraction of 29Si
out of all Si can be kept constant in both halves. After diffusion experiments, silicon isotope
fractions are measured. Because the chemical composition is uniform and mixing between
the three silicon isotopes is expected to be perfectly ideal, the cross diffusivities would be
zero if cross diffusivities are entirely due to thermodynamic contributions. In such a case, the
29
Si fraction would stay as a constant across the whole profile because there was no initial
29
Si gradient and hence no 29Si flux. However, if there is a kinetic contribution to the cross
diffusivities, the cross terms would cause 29Si fluxes, leading to uphill diffusion profiles when
the fraction of 29Si is plotted against distance.
DIFFUSION COEFFICIENTS
Over the years, diffusivities in many systems have been determined experimentally. Many
authors have examined how the diffusivities vary with experimental conditions, and developed
numerous relations. Some of them have strong theoretical basis and are widely applicable, but
others are largely empirical and also uncertain. Experimentally determined values of diffusion
coefficients in minerals and melts are reviewed and summarized in later chapters of this
volume. In this section, various relations relating diffusion coefficients and other parameters
are reviewed.
Diffusivity values and activation energy depend on the phase in which diffusion occurs
(Watson and Baxter 2007). Figure 2 shows Ar diffusivities in gas, aqueous solutions, basalt
melt, rhyolite melt, and the mineral amphibole for comparison. The diffusivity decreases and
the activation energy increases from air to water to melt to mineral. From one species to
another in a liquid or melt, the diffusivity depends on the bond strength as well as the size of
the diffusing species. From one liquid to another, the diffusivity depends on both the diffusing
component as well as the liquid composition. Similarly, in minerals, from one species to
another, diffusion can depend on size and charge of the diffusing species, and the sites on
which they diffuse; from one mineral to another, diffusivities of a particular species may
depend on properties such as mineral composition, structure and bond strengths.
Temperature dependence of diffusivities; Arrhenius relation
The temperature dependence of diffusivities is well known and works well almost without
exceptions. This relation is the Arrhenius equation (Eqn. 10) D = D0e−E/(RT), meaning that logD
(or lnD) versus 1/T (or 1000/T) is a straight line with a negative slope, as shown in Figure 2
for Ar diffusivities in water, basalt melt, rhyolite melt and the mineral hornblende. (Fig. 17a in
a later section also shows some diffusion data in an Arrhenius plot.) The Arrhenius equation
can be derived from either the collision theory or the transition state theory (e.g., Lasaga
1998), in which the activation energy is identified to be the necessary enthalpy for forming the
activated complex. The preexponential factor D0 is proportional to T1/2 in the collision theory
and to T in the transition state theory. There were efforts in the early years to test how the
preexponential factor D0 depends on temperature (e.g., Perkins and Begeal 1971; Shelby and
Keeton 1974), but it requires high-quality data (e.g., with ≤ 10% relative error in D) in a large
temperature range (e.g., 400-1200 K), and the results are inconclusive: high-quality data over
a large temperature range may show a small curvature in lnD versus 1/T, indicating that either
D0 or E depends on temperature, but the exact relation is not well constrained. On the other
hand, for most diffusion data in melts and minerals in the geological literature, the uncertainty
in D is often of the order of 30% and the temperature range is not large enough, so that the
Arrhenius relation works well within uncertainty. In limited cases when the temperature range
is large from glass to melt, the Arrhenius equation fits the data well and it is rarely necessary to
invoke dependence of D0 on T, or E on T, or a discontinuity due to glass transition (e.g., Zhang
et al. 1991a; Zhang and Behrens 2000; Behrens and Zhang 2009). In minerals, although some
authors proposed discontinuities in lnD versus 1/T (e.g., Buening and Buseck 1973 for Fe-Mg
interdiffusion in olivine), later studies show no such discontinuity is present (e.g., Chakraborty
2010, this volume). In summary, the temperature dependence of all experimental diffusivity
data in the geological literature can be summarized well by the Arrhenius relation in the form
of D = D0e−E/(RT) with positive D0 and E.
Pressure dependence of diffusivities
The dependence of diffusivities on dry pressure (P) is more complicated than the
temperature dependence. (Hydrous pressure, on the other hand, has two effects: one is pressure,
and the other is the presence of water, which may accelerate diffusion.) In a small pressure
range, lnD is often linear in P. Such a relation is consistent with the transition state theory
44 Zhang
because the enthalpy is linear to pressure in liquid and solid phases, leading to Equation (11),
D = D0e−(E+P∆V)/RT, in which E is the activation energy at zero pressure, ∆V is the activation
volume (identified as the volume difference between the activated complex and the non-
activated state), and E+P∆V is the activation energy at pressure P. Even though the activation
energy for diffusion is always positive, the activation volume can be either positive (meaning
D decreases with increasing pressure, such as Ar diffusivity in rhyolite melt — Behrens and
Zhang 2001, see Fig. 16a; H2O diffusivity in rhyolite and dacite melts — Zhang and Behrens
2000, Ni and Zhang 2008, Ni et al. 2009; divalent cation diffusivities in garnet and in oxide
minerals — Ganguly 2010, this volume, Van Orman and Crispin 2010, this volume) or negative
(meaning D increases with increasing pressure, such as oxygen self diffusivity in dacite melt
below 4 GPa; Tinker and Lesher 2001, see Figure 16b; Mg self diffusion above 60 GPa, Van
Orman and Crispin 2010, this volume).
Whereas the activation energy E is a large positive value and hence small changes due to
temperature and pressure variations do not affect E noticeably, the activation volume ∆V as the
volume difference between the activated complex and the ground state is small and hence can
change significantly as pressure or temperature changes. Such changes produce at least two
effects (or complexities) when compared with the temperature dependence of D:
(1) The dependence of ∆V with temperature means more terms are needed in fitting
lnD as a function of T and P. For example, if ∆V is linear to T as ∆V = a + bT, then
Equation (11) becomes
E + P∆V E + aP + bPT a +bP
ln D = a0 − = a0 − = ( a0 + b0 P ) − 1 1 (101)
RT RT T
where a0 = lnD0, b0 = -b/R, a1 = E/R, and a1 = a/R. That is, the preexponential term
would depend on pressure even though this is not in the original equation derived
from the transition state theory (Eqn. 11). If ∆V is a more complicated function of T,
there may be more terms in fitting lnD versus T and P. When fitting diffusion data as a
function of T and P, usually Equation (11) is used. However, if the fitting quality is not
good enough, then Equation (101) or even more complicated function may be used.
(2) The activation volume ∆V can also change with pressure, even from positive to nega-
tive or vice versa as pressure increases (Fig. 16b). Therefore, at a given temperature,
in a large pressure range (several GPa), lnD versus P may not be linear: lnD may first
increase with P and then decrease with P, or first decrease with P and then increase
with P. Large changes in ∆V, especially from positive to negative or vice versa, may
be associated with structural changes in silicate melts.
In addition to dependence on total pressure, H2O pressure may have a disproportionally
large effect on diffusivity of some components in hydrothermal diffusion experiments. For
example, 18O diffusivities in nominally anhydrous felsic minerals are roughly proportional
(rather than exponential) to H2O pressure or fugacity (e.g., Yund and Anderson 1978; Farver
and Yund 1990), increasing by two orders of magnitude as H2O fugacity increases from
0.001 GPa to 0.2 GPa. This dependence on pressure is different from the dependence on
total pressure, and is related to H218O molecules entering a mineral and carrying 18O into
the mineral (Zhang et al. 1991b; McCornell 1995). The contribution by H2O diffusion to the
diffusion of 18O has been quantified (Zhang et al. 1991b) and is related to the product of H2O
concentration and diffusivity in the mineral. For some minerals (e.g., zircon), however, there
is little dependence of 18O diffusion on H2O pressure in hydrothermal experiments, although
there is a difference between oxygen diffusivities for dry and hydrothermal experiments (e.g.,
Watson and Cherniak 1997), probably due to the saturation of mobile H2O in the mineral.
For cations such as Na and Sr in feldspar, because H2O does not carry them, H2O diffusion
would not directly contribute to their diffusion (although H2O in a mineral may weaken the
Figure 16. Different pressure dependences of diffusivities. (a) The diffusivity decreases with pressure (pos-
itive activation volume); (b) The diffusivity first increases with pressure and then decreases with pressure.
structure, enhancing diffusion). For example, H2O pressure of 200 MPa does not change the
diffusivity of cations in feldspar compared to room pressure dry experiments (Behrens et al.
1990; Cherniak 2010, this volume).
In summary, the dependence of D on pressure is more complicated than on temperature,
and for many systems such dependence has not been investigated well. Because pressure can
vary widely from the surface to the deep Earth, understanding the pressure effect quantitatively
so that diffusivities can be predicted remains an important task for geochemists.
Diffusion in crystalline phases and defects
Crystals contain various defects where the periodicity of the lattice is locally disturbed.
In crystalline phases, diffusion often occurs through defects in the crystalline structure. For
example, a cation may jump into a nearby vacancy (one kind of point defect), leaving another
vacancy defect in its original position (into which another cation may jump), and then jump
into another vacancy as it comes nearby, and so on. Hence, the diffusivity is often proportional
to defect concentrations.
For a review of defect theory as well as its applications to minerals, readers are referred
to Schock (1985) and Lasaga (1998). Defects may be classified as point defects and extended
46 Zhang
defects. Point defects include vacancies (unoccupied sites that are normally occupied),
interstitials (atoms occupying normally unoccupied sites), and impurities (atoms occupying
sites that are normally occupied by other types of atoms, such as Al occupying an Si site). If a
defect is caused by impurities, it is called an extrinsic defect. Otherwise, it is called an intrinsic
defect. The number of defects in a material must satisfy the condition of charge neutrality. A
nearby pair of vacancy and interstitial is called a Frenkel defect. A stoichiometric proportion
of cation and anion vacancies is called a Schottky defect. Extended defects include line defects
(such as dislocations), plane defects, domain boundaries, grain boundaries, and bulk defects
or impurities (such as a melt inclusion). Point defects play the most important role in volume
diffusion, but extended defects may produce fast diffusion paths.
In treating defects, the Kröger-Vink notation is conventionally used. The format of the
p
notation takes the form M b where M is the species, including (1) elements such as Fe, Mg, Si,
O, etc, (2) vacancies (either V or v since V is also the element symbol for vanadium), (3)
electrons (e), and (4) electron holes (h). The subscript b indicates the lattice site that the species
M occupies, such as Na site in NaCl, Fe site in olivine, or an interstitial site (indicated as i).
The superscript p indicates the electronic charge relative to the site it occupies (so that if the
site is occupied by its normal occupant, the charge is zero, indicated by x), with positive
charges indicated by • (the number of • indicates the number of positive charges), and
negative charges indicated by ′.
At a given temperature, there is an equilibrium concentration of intrinsic defects. For
Schottky defects when cation to anion ratio is 1:1, such as MgO, the reaction can be written
as (in Kröger-Vink notation):
Ø  v′′Mg + v••O (102)
where Ø means nothing, v′′Mg means a vacant magnesium site with an effective charge (relative
to that of the occupied site) of −2, and v••O means a vacant oxygen site with an effective charge
of +2. If this kind of defects dominate, the equilibrium constant is
K1 = [v′′Mg ][vO•• ] = [v′′Mg ]2 = [v••O ]2 (103)
where brackets mean mole fractions. Hence, the equilibrium concentration of vacancy pairs is:
[v′′Mg ] = [v••O ] = K11/ 2 = e − ∆Gf1 /( 2 RT ) = e ∆Sf1 /( 2 R )e − ∆Hf1 /( 2 RT ) (104)
where ∆Gf1, ∆Sf1, and ∆Hf1 are the formation free energy, entropy and enthalpy for the pair of
vacancies.
For Frenkel defects, the reaction can be written as (using Mg in MgO as an example):
Mg xMg  v′′Mg + Mg••i (105)
where Mg xMg means a magnesium ion sitting on a magnesium lattice site, and Mg••i means an
interstitial magnesium ion. If this kind of defects dominate, the equilibrium constant is:
K 2 = [v′′Mg ][ Mg ••i ] = [v′′Mg ]2 = [ Mg ••i ]2 (106)
Hence, the expression for the equilibrium concentration of Frenkel defects is similar to the
case of Schottky defects:
[v′′Mg ] = [ Mg ••i ] = K 21/ 2 = e − ∆Gf2 /( 2 RT ) = e ∆Sf2 /( 2 R )e − ∆Hf2 /( 2 RT ) (107)
where ∆Hf2 is the formation energy for the pair (i.e., moving Mg from a lattice position to an
interstitial position, leaving behind a vacancy).
Because ions with higher valence are more strongly bonded, it is expected that ∆Hf increases
with valence. Other factors, such as oxygen fugacity, can also affect defect concentrations (see
below).
Diffusivities and oxygen fugacity
Variation in oxygen fugacity in a system results in at least two effects, one is to change the
oxidation state of multivalent elements, and the second is to change (often induce extra) defect
sites in the structure. The two effects are somewhat correlated at the atomic level, but the
resulting effects in terms of diffusivity are different. The first effect is important especially in
melts and minerals when the element of interest can be present in multiple valences, such as Fe
(Fe2+ and Fe3+), S (S2−, S4+ in SO2, and S6+ in SO42−), Eu (Eu2+ and Eu3+), Sn (Sn2+ and Sn4+),
etc (e.g., diffusion of Fe in feldspars, Behrens et al. 1990; diffusion of Eu in orthopyroxene,
Cherniak and Liang 2007; and diffusion of Eu in melts, Behrens and Hahn 2009). The
diffusivity of the lower valent species is often higher than the higher valent species because
of weaker bonding between the low valent ion and the rest of the structure. Hence, as oxygen
fugacity increases, the diffusivity of the element of interest often decreases. The diffusion of
the element can be quantified using multispecies diffusion equations following the approach
of Zhang et al. (1991b), e.g., for Fe diffusion:
DFe = X Fe2+ DFe2+ + X Fe3 + DFe3 + (108)
where XFe2+ = Fe2+/(Fe2++Fe3+).
The second effect has been observed and examined largely in mafic minerals, and can
be understood by examining the defect concentration in a mineral structure. For example,
increasing oxygen fugacity would oxidize a small fraction of Fe2+ to Fe3+, producing vacancy
defects (because two Fe3+ ions are equivalent to three Fe2+ in terms of charges). In wüstite, the
reaction in Kröger-Vink notation can be written as:
6 Fe xFe + O2 (gas)  4 Fe •Fe + 2v’’Fe + 2 FeO (109)
The equilibrium constant can be written as:
[ Fe •Fe ]4 [v′′Fe ]2 [ FeO]2
K= (110)
[ Fe xFe ]6 fO2
From stoichiometry in Reaction 109, [Fe •Fe] = 2[v′′Fe]. Hence, Equation (110) can be written as:
16[v′′Fe ]6 [ FeO]2
K= (111)
[ Fe xFe ]6 fO2
Because the activities of Fe xFe and FeO in wüstite are roughly constant, it can be seen that the
concentration of this type of vacancy is proportional to the 1/6 power of oxygen fugacity.
In other ferromagnesian minerals, the fO12/ 6 relation also holds (e.g., Lasaga 1998), which is
consistent with the increase of Fe-Mg interdiffusivity in olivine with oxygen fugacity (Dohmen
and Chakraborty 2007a,b). If other defects are present, their effect on diffusivity must also be
quantified to obtain the full effect of oxygen fugacity.
Compositional dependence of diffusivities
Diffusivity of a component depends on the major chemical composition, and occasionally
also on the minor element composition. For example, the effect of oxygen fugacity on diffusion
may be viewed as a compositional effect. The effects of H2O content (or H2O pressure) on
diffusivities in melts and minerals have been investigated: Minor amounts of H2O may affect
the diffusivity of some components in minerals significantly as in the case of the dependence
of 18O diffusivity in feldspars and quartz on H2O pressure (Farver and Yund 1990, 1991; Zhang
48 Zhang
et al. 1991b), as well as diffusivities in melts, especially silicic melts (Mungall and Dingwell
1997; Behrens et al. 2007). The effects of SiO2 and alkalies on diffusivities in melts have
also been investigated to some degree. In general, more effort is needed to understand the
compositional effects on diffusivities.
Limited data often indicate that D is an exponential function of some concentration (that
is, lnD is linear with respect to concentration). For example, Behrens and Zhang (2001) showed
that the Ar diffusivity in rhyolite melts increases exponentially with H2O content. Dohmen and
Chakraborty (2007a,b) showed that the Fe-Mg interdiffusivity in olivine increases exponentially
with the mole fraction of fayalite (lnD is linear to XFa). Other relations have also been found.
For example, for 22Na tracer diffusivity in plagioclase melts, lnD is not linear with respect to
XAn (Behrens 1992) but requires a second-order term (Zhang et al. 2010, this volume). For
Th and U diffusion in a haplorhyolite melt, lnD seems to be linear with respect to the square
root of the H2O concentration, rather than the H2O concentration itself (Mungall and Dingwell
1997; Zhang et al. 2010, this volume). Diffusivity of the hydrous component in orthopyroxenes
depends strongly on the composition of orthopyroxenes (Farver 2010, this volume).
In addition, the relation between diffusivity and concentration may be modified by the
role of speciation of the diffusing component. For example, molecular H2O diffusivity in
rhyolite and dacite melts conforms well with the exponential relation (Zhang and Behrens
2000; Liu et al., 2004; Ni and Zhang 2008; Ni et al. 2009), but the diffusivity of total H2O
(which includes both molecular H2O and hydroxyls) increases first proportionally with total
H2O content and then exponentially. Diffusivity of the hydrous component in pyrope shows
complicated behavior, due to both the speciation effect and the oxygen fugacity in pyrope
affecting the H2O/H2 ratio (Wang et al. 1996).
SiO2 is the dominant component in silicate melts, and its effect on diffusivities can be
complicated. Diffusivities of most components decrease as SiO2 content increases (and the
degree of polymerization increases), and often the relation between lnD and SiO2 concentration
is roughly linear (e.g., Lesher and Walker 1986; Behrens et al. 2004; Ni et al. 2009). However,
diffusivities of some components (such as He, Li, and Na) increase as SiO2 content increases
(Behrens 2010; Zhang et al. 2010).
Overall, compositional effects on diffusivities are complicated, especially in silicate melts,
but also in complicated minerals such as hornblende and pyroxenes. We have only scratched
the surface for such effects. One complexity is due to the numerous components in natural
melts and many minerals because we often can only examine the effect of one component at
a time. Another complication arises because there are no theoretical relations to guide studies
on the compositional effects although empirical models are available (Mungall 2002) in which
the diffusivities of non-alkalies are related to viscosity, but those of alkalies are not related to
viscosity. It is anticipated that in the future much more effort will be devoted to understanding
the compositional effects on diffusion in melts and minerals.
Relation between diffusivity, particle size, particle charge, and viscosity
There are famous relations between diffusivity, particle size and viscosity, all of which
only have limited applicability to diffusion in silicate melts. One such famous relation is based
on Einstein’s analyses of Brownian motion of neutral and spherical particles in a continuous
fluid medium (Einstein 1905), resulting in the following relation between diffusivity, viscosity,
particle radius and temperature:
kB T
D= (112)
6 πaη
where kB is the Boltzmann constant (1.3807×10−23 J/K), a is the radius of the particle, and h is
the viscosity of the fluid. The above equation is often referred to as the Einstein equation, or
Stokes-Einstein equation. There are other variations of this equation, such as the Sutherland
equation (Sutherland 1905) differing from the above equation by a factor of 1.5, and the
Glasstone equation (Glasstone et al. 1941), differing from the above equation by a factor of
3π. Another equally famous equation is the Eyring equation (Glasstone et al. 1941), relating
self diffusivity, viscosity, diffusive jump distance and temperature:
kB T
D= (113)
lη
where l is the diffusive jump distance. The jump distance is often estimated to be interatomic
distances or diameter of the diffusing species. If the diameter of the diffusing species is used
as the jump distance, then the Eyring equation becomes the same as the Glasstone equation,
and the Eyring diffusivity is 3p times the Einstein diffusivity. The assumptions employed in
deriving the Einstein equation and the Eyring equation are different: the Einstein equation is
derived for the diffusion of large neutral species, whereas the Eyring equation is derived for the
case when the diffusion of the species is also responsible for the viscous flow. Both equations
claim that D is inversely proportional to viscosity.
The Einstein equation seems to work well (within a factor of 2) for diffusion of relatively
large neutral molecules (e.g., heavier noble gases) in aqueous solutions, but does not work
at all (orders of magnitude difference) for the diffusion of any known neutral species in
silicate melts (e.g., Ni and Zhang 2008). The Eyring relation appears to work well for O (and
possibly Si) self diffusion in dry silicate melts (but see Lesher et al. 1996; Liang et al. 1996;
LaTourrette and Wasserburg 1997), but does not work at all (orders of magnitude difference)
for 18O diffusion in hydrous silicate melts (Behrens et al. 2007). Because most natural melts
are hydrous, the applicability of the Eyring relation to natural melts is very limited. For other
components, especially for rapidly diffusing components, the Eyring relation does not apply.
For example, when 3.5 wt% H2O is added to a dry rhyolite at 1000 K, the viscosity decreases
by 7 orders of magnitude (Zhang et al. 2003), but 22Na tracer diffusivity only increases by a
factor of about 2 (Watson 1981).
Not only is the diffusivity of most components in melts not inversely proportional to
melt viscosity, but the negative correlation between D and h is also violated sometimes. For
example, tracer diffusivities of Li and Na increase as viscosity increases from basalt to dacite
to andesite to rhyolite melts (Jambon and Semet 1978; Jambon 1982; Zhang et al. 2010).
There is also no universal relation between the size of the diffusing species and the
diffusivity, contrary to the Einstein equation. For example, for the alkali elements, the ionic
radius increases from Li+ to Na+ to K+ to Rb+ to Cs+, but the tracer diffusivity decreases from
Na to Li to K to Rb to Cs by orders of magnitude (e.g., Jambon 1982). On the other hand, for
alkaline earth elements, the ionic radius increases from Be2+ to Mg2+ to Ca2+ to Sr2+ to Ba2+,
but the tracer diffusivity decreases from Sr and Ba to Ca to Mg to Be (Mungall et al. 1999),
with smaller cations having greater diffusivity, completely opposite to the prediction of the
Einstein equation. Even though the opposite trend may be rationalized by the diffusing species
of a component being different from the cations, it does show that even for isovalent cations,
the size effect cannot be predicted a priori.
Diffusivities of ionic species in general decrease with increasing positive or negative
charge, from 0 valence (neutral atoms or molecules) to ±1, to ±2, etc., although the trend is not
perfect either. The relation is covered by Brady and Cherniak (2010, this volume) for minerals
and Zhang et al. (2010, this volume) for melts.
Mungall (2002) developed empirical relations to predict cation diffusivities in silicate
melts as a function of cation radius and compositional parameters for alkali elements, and
of melt viscosity, ionic strength, and compositional parameters for other elements. These are
50 Zhang
the best predictive models available currently although the uncertainty can still be orders of
magnitude (Behrens and Hahn 2009; Zhang et al. 2010, this volume).
Diffusivity and ionic porosity
Diffusion is driven by random motion of particles. If there is more “empty” space in a
mineral or melt, a given species should diffuse more rapidly, and vice versa. Based on this
concept, it has been proposed that diffusivity of a species depends on the ionic porosity (IP),
a measure of the “empty” space in a structure. Ionic porosity is defined as the unoccupied
volume divided by the total volume:
Vions
IP = 1 – (114)
Vtotal
where Vions is volume occupied by the ions. Below is an example to calculate IP of quartz. The
molar volume of quartz is 22.69×10−6 m3/mol. In one mole of quartz, there are two moles of
O2− and one mole of Si4+. Ionic radius of O2− is taken to be 1.38×10−10 m. Ionic radius of Si4+
in a tetrahedral site is 0.26×10−10 m (Shannon 1976). Hence, volume occupied by the ions is:
Vions = (6.02214×1023)(4π/3) × (2×1.383 + 0.263)×10−30 = 13.30×10−6 m3/mol
Therefore, IP = 0.414. Ionic porosity of some common minerals can be found in Zhang (2008,
p. 310). Fortier and Giletti (1989) applied this approach to relate 18O diffusivity in different
minerals under hydrothermal conditions. This issue is discussed further by Brady and Cherniak
(2010) in this volume.
Compensation “law”
The compensation “law” is an empirical “law” proposed by Winchell (1969). This “law”
states that the logarithm of the preexponential factor D0 is linear to the activation energy E for
the diffusion of different components in a single phase (Hofmann 1980; Hart 1981), or the
same diffusion species in different phases (Bejina and Jaoul, 1997; Zhao and Zheng 2007):
lnD0 = a + bE (115)
where a and b are two fitting parameters. From the above equation, one may derive the
following (Hart 1981; Lasaga 1998)
 E   E    1  
D = exp  lnD0 –  = exp  a + bE –  = exp a + b − E (116)
 RT   RT    RT  
Hence, in the context of the compensation “law”, a compensation temperature Tcomp can be
defined as Tcomp = 1/(bR). At this temperature, diffusivities of all components would be the
same, exp(a). In other words, when plotted in lnD vs. 1/T, all lines would intersect a common
point at the compensation temperature. Examination of experimental data shows that the “law”
has large uncertainties even for a single group of elements. Figure 17 shows that for tracer
diffusion of alkali elements in a rhyolite melt, the compensation “law” does not work well
(there is no single compensation temperature). In some other systems, the “law” works better,
but the uncertainty in predicting D is still about an order of magnitude.
Interdiffusivity and self diffusivity
There are two models relating the interdiffusivity DAB in a binary system between
components A and B, and the self diffusivities DA and DB in the same system with the same
composition. The models depend on whether diffusion is ionic or neutral atomic.
For ionic interdiffusion between isovalent ions (e.g., Na-K interdiffusion between albite
and orthoclase melts, or Fe-Mg interdiffusion in olivine), Helfferich and Plesset (1958) and
Barrer et al. (1963) derived the following relation:

Figure
Fig. 17. Diffusivities
17. Diffusivities ofrhyolite
of alkalis in alkalismelts.
in rhyolite melts.
Data sources: Li: Data
Jambonsources:
and Semet (1978) and Jambon (1982).
Li: Jambon and Semet (1978) and Jambon (1982).
Fig. 17. Diffusivities of alkalis in rhyolite melts. Data sources: Li: Jambon and Semet (1978) and Jambon (1982).
D AD B ( X A + X B )  d ln γ B 
DAB = 1 +  (117)
X AD A + X BD B  d ln X B 
where XA and XB are mole concentrations of A and B, and gB is the chemical activity coefficient
of component B. Note that DA and DB are those at the same composition at which DAB is
estimated (DA and DB may depend on composition). For ionic diffusion between ions with
different valences, there is a relation accounting for the valence difference (e.g., Zhang 2008)
but more assumptions on how charges are balanced must be made. Lasaga (1979) and Liang
et al. (1997) extended the ionic model to multicomponent ionic diffusion for the calculation
of the multicomponent diffusivity matrix. (Note that dlngA/dlnXA = dlngB/dlnXB from
thermodynamics, meaning that A and B can be exchanged in Eqn. 117.)
For neutral atomic interdiffusion (such as in alloys), the following equation has been
formulated (Darken 1948; Shewmon 1989; Kirkaldy and Young 1987)
X BD A + X AD B  d ln γ B 
DAB = 1 +  (118)
X A + X B  d ln X B 
52 Zhang
Equation (118) is often referred to as the Darken-Hartley-Crank equation. Cooper (1965),

Richter (1993), and Liang et al. (1997) extended the above model to multicomponent systems
for the calculation of the multicomponent diffusivity matrix.
By comparing Equation (66) with either Equation (117) or Equation (118), and recognizing
that binary diffusivity is the interdiffusivity, it can be seen that the intrinsic diffusivity in
Equation (66) is related to the self diffusivities of the two components, and the exact relation
depends on whether the diffusion is ionic or neutral atomic.
To illustrate the dependence of DAB on composition in ideal and nonideal systems, Figure
18a shows calculated DAB in an ideal solution (equivalent to a regular solution with W = 0
where W is the interaction parameter) and Figure 18b shows that in a regular solution with W/
(RT) = 2.4 using Equations (117) and (118). When W/(RT) > 2 for a regular solution, the binary
mixture would decompose into two phases in the compositional region centered at XA = XB =
0.5. It can be seen that
Youxue (1)(ChD2)
Zhang
AB values from Equations (117) and (118) can be very different
Page 18
although the two equations may look similar; and (2)Page

Youxue Zhang (Ch 2) DAB 18values depend strongly on how ideal
Figure 18. Interdiffusivity DAB as a function of composition for (a) ideal binary mixture, and (b) a binary
mixture that can be described by a regular solution model with W/(RT) = 2.4.
Fig. 18. Interdiffusivity DAB as a function of composition for (a) ideal binary mixture, and (b) a binary mixture that
Fig. 18. Interdiffusivity DAB as a function of composition for (a) ideal binary mixture, and (b) a binary mixture that
can be described by a regular solution model with W/(RT) = 2.4.
the mixture is. In nonideal solutions, DAB values are negative in the compositional region in
which the mixture would spontaneously decompose into two phases, as expected.
Before applying either of the above equations to a mineral or melt, it is necessary to
decide whether diffusion is ionic or through neutral species because the two seemingly similar
equations produce rather different interdiffusivities. For diffusion in minerals, it is relatively
easy to decide whether the diffusion is ionic or neutral. For example, Ca-Mg interdiffusion in
garnet is ionic, but Cu-Au interdiffusion in gold is through neutral atoms. For silicate melts,
however, it is more difficult to determine which of the two models to apply, or whether any
would apply. For example, Kress and Ghiorso (1995) used the diffusivity matrix they obtained
for a basalt melt to test the model of Richter (1993) and found that the model does not work.
CONCLUSIONS
Diffusion theory is well developed. Nonetheless, diffusion in minerals and melts are
complicated and difficult to describe mostly due to (i) multiple components in melts as well
as many minerals, (ii) diffusional anisotropy of many minerals, and (iii) highly non-ideal
mixing (including the miscibility gap) of some mixtures. For a given geologic problem, the
complexities may also be due to complex boundary conditions and unknown or complicated
initial conditions. Hence, in treating diffusion in minerals and melts, simplifications are
usually made. These simplifications have led to successful understanding of many geological
processes, although not complete understanding of the diffusion behavior of the whole system.
Some problems (such as crystal growth and dissolution; Liang 1999, 2000) do require more
rigorous treatment of diffusion.
On the other hand, there is much room to improve theories and models regarding diffusion
coefficients, especially how they should depend on the composition and structure of the system.
Such work would provide guidelines on future diffusion work. For example, the understanding
of the role of speciation in the diffusion of some components has allowed us not only to gain
insights into the diffusion mechanism but also to describe the diffusion behavior of these
components well so that practical problems such as bubble growth in melts can be addressed
(Proussevitch and Sahagian 1998; Liu et al. 2000; Zhang 2009).
ACkNOwLEDGMENTS
This research is supported by NSF grants EAR-0711050, EAR-0838127, and EAR-
1019440. I thank Yan Liang and Daniele Cherniak for insightful and constructive comments.
REFERENCES
Baker DR (1989) Tracer versus trace element diffusion: diffusional decoupling of Sr concentration from Sr
isotope composition. Geochim Cosmochim Acta 53:3015-3023
Balluffi RW, Allen SM, Carter WC, Kemper RA (2005) Kinetics of Materials. Wiley-Interscience, Hoboken,
N.J
Barrer RM, Bartholomew RF, Rees LVC (1963) Ion exchange in porous crystals. Part II. the relationship
between self- and exchange- diffusion coefficients. J Phys Chem Solids 24:309-317
Behrens H (1992) Na and Ca tracer diffusion in plagioclase glasses and supercooled melts. Chem Geol 96:267-
275
Behrens H, Hahn M (2009) Trace element diffusion and viscous flow in potassium-rich trachytic and phonolitic
melts. Chem Geol 259:63-77
Behrens H, Johannes W, Schmalzried H (1990) On the mechanisms of cation diffusion processes in ternary
feldspars. Phys Chem Miner 17:62-78
54 Zhang
Behrens H, Zhang Y (2001) Ar diffusion in hydrous silicic melts: implications for volatile diffusion mechanisms
and fractionation. Earth Planet Sci Lett 192:363-376
Behrens H, Zhang Y (2009) H2O diffusion in peralkaline to peraluminous rhyolitic melts. Contrib Mineral
Pertrol 157:765-780
Behrens H, Zhang Y, Leschik M, Miedenbeck M, Heide G, Frischat GH (2007) Molecular H2O as carrier for
oxygen diffusion in hydrous silicate melts. Earth Planet Sci Lett 254:69-76
Behrens H, Zhang Y, Xu Z (2004) H2O diffusion in dacitic and andesitic melts. Geochim Cosmochim Acta
68:5139-5150
Bejina F, Jaoul O (1997) Silicon diffusion in silicate minerals. Earth Planet Sci Lett 153:229-238
Blank JG (1993) An experimental investigation of the behavior of carbon dioxide in rhyolitic melt. California
Institute of Technology, Pasadena, p 210
Brady JB, Cherniak DJ (2010) Diffusion in minerals: an overview of published experimental diffusion data.
Brearley M, Scarfe CM (1986) Dissolution rates of upper mantle minerals in an alkali basalt melt at high
pressure: an experimental study and implication for ultramafic xenolith survival. J Petrol 27:1157-1182
Buening DK, Buseck PR (1973) Fe-Mg lattice diffusion in olivine. J Geophys Res 78:6852-6862
Carroll MR, Stolper EM (1991) Argon solubility and diffusion in silica glass: implications for the solution
behavior of molecular gases. Geochim Cosmochim Acta 55:211-225
Carslaw HS, Jaeger JC (1959) Conduction of Heat in Solids. Clarendon Press, Oxford
Chen Y, Zhang Y (2008) Olivine dissolution in basaltic melt. Geochim Cosmochim Acta 72:4756-4777
Chen Y, Zhang Y (2009) Clinopyroxene dissolution in basaltic melt. Geochim Cosmochim Acta 73:5730-5747
Cherniak DJ (2010) Cation diffusion in feldspars. Rev Mineral Geochem 72:691-734
Cherniak DJ, Liang Y (2007) Rare earth element diffusion in natural enstatite. Geochim Cosmochim Acta
71:1324-1340
Cherniak DJ, Watson EB (2000) Pb diffusion in zircon. Chem Geol 172:5-24
Cooper AR (1965) Model for multi-component diffusion. Phys Chem Glasses 6:55-61
Cooper AR (1968) The use and limitations of the concept of an effective binary diffusion coefficient for multi-
component diffusion. In: Mass Transport in Oxides, Vol 296. Wachman JB, Franklin AD (eds) Nat Bur
Stand Spec Publ, p 79-84
Crank J (1975) The Mathematics of Diffusion. Clarendon Press, Oxford
Cussler EL (1997) Diffusion: Mass Transfer in Fluid Systems. Cambridge Univ Press, Cambridge, England
Darken LS (1948) Diffusion mobility and their interrelation through free energy in binary metalic systems.
Trans AIME 175:184-201
DeGroot SR, Mazur P (1962) Non-Equilibrium Thermodynamics. Interscience, New York
Delaney JR, Karsten JL (1981) Ion microprobe studies of water in silicate melts: concentration-dependent
water diffusion in obsidian. Earth Planet Sci Lett 52:191-202
Dingwell DB, Scarfe CM (1985) Chemical diffusion of fluorine in melts in the system Na2O-Al2O2-SiO2. Earth
Planet Sci Lett 73:377-384
Dodson MH (1973) Closure temperature in cooling geochronological and petrological systems. Contrib.
Mineral. Petrol. 40:259-274
Dodson MH (1979) Theory of cooling ages. In: Lectures in Isotope Geology. Jager E, Hunziker JC (eds)
Springer-Verlag, New York, p 194-202
Dohmen R, Becker HW, Chakraborty S (2007a) Fe-Mg diffusion in olivine I: experimental determination
between 700 and 1200 °C as a function of composition, crystal orientation and oxygen fugacity. Phys
Chem Miner 34:389-407
Dohmen R, Chakraborty S (2007b) Fe-Mg diffusion in olivine II: point defect chemistry, change of diffusion
mechanisms and a model for calculation of diffusion coefficients in natural olivine. Phys Chem Miner
34:409-430
Dohmen R, Milke R (2010) Diffusion in polycrystalline materials: grain boundaries, mathematical models,
and experimental data. Rev Mineral Geochem 72:921-970 Doremus RH (1969) The diffusion of water
in fused silica. In: Reactivity of Solids. Mitchell JW, Devries RC, Roberts RW, Cannon P (eds) Wiley,
New York, p 667-673
Doremus RH (1995) Diffusion of water in silica glass. J Mater Res 10:2379-2389
Drury T, Roberts JP (1963) Diffusion in silica glass following reaction with tritiated water vapor. Phys Chem
Glasses 4:79-90
Einstein A (1905) The motion of small particles suspended in static liquids required by the molecular kinetic
theory of heat. Ann Phys 17:549-560
Farver JR, Yund RA (1990) The effect of hydrogen, oxygen, and water fugacity on oxygen diffusion in alkali
feldspar. Geochim Cosmochim Acta 54:2953-2964
Farver JR, Yund RA (1991) Oxygen diffusion in quartz: dependence on temperature and water fugacity. Chem
Geol 90:55-70
Farver JR, Yund RA (2000) Silicon diffusion in forsterite aggregates: implications for diffusion accommodated
creep. Geophys Res Lett 27:2337-2340
Fortier SM, Giletti BJ (1989) An empirical model for predicting diffusion coefficients in silicate minerals.
Science 245:1481-1484
Fortier SM, Giletti BJ (1991) Volume self-diffusion of oxygen in biotite, muscovite, and phlogopite micas.
Geochim Cosmochim Acta 55:1319-1330
Freda C, Baker DR (1998) Na-K interdiffusion in alkali feldspar melts. Geochim Cosmochim Acta 62:2997-
3007
Gamow G (1961) One, Two, Three ... Infinity: Facts & Speculations of Science. Viking, New York, p 340
Ganguly J (1982) Mg-Fe order-disorder in ferromagnesian silicates, II: thermodynamics, kinetics and
geological applications. In: Advances in Physical Geochemistry Vol 2. Saxena SK (ed) Springer-Verlag,
New York, p 58-99
Ganguly J (2010) Cation diffusion kinetics in aluminosilicate garnets and geological applications. Rev Mineral
Geochem 72:559-601
Ganguly J, Tirone M (1999) Diffusion closure temperature and age of a mineral with arbitrary extent of
diffusion: theoretical formulation and applications. Earth Planet Sci Lett 170:131-140
Ghiorso MS (1987) Chemical mass transfer in magmatic processes, III: crystal growth, chemical diffusion and
thermal diffusion in multicomponent silicate melts. Contrib Mineral Petrol 96:291-313
Giletti BJ, Yund RA (1984) Oxygen diffusion in quartz. J Geophys Res 89:4039-4046
Giletti BJ, Semet MP, Yund RA (1978) Studies in diffusion—III. Oxygen in feldspars: An ion microprobe
determination. Geochim Cosmochim Acta 42:45-57
Glasstone S, Laider KJ, Eyring H (1941) The Theory of Rate Processes. McGraw-Hill, New York
Glicksman ME (2000) Diffusion in Solids: Field Theory, Solid-State Principles, and Applications. Wiley, New
York
Gregg MC, Sanford TB, Winkel DP (2003) Reduced mixing from the breaking of internal waves in equatorial
waters. Nature 422:513-515
Harrison TM (1981) Diffusion of 40Ar in hornblende. Contrib Mineral Petrol 78:324-331
Harrison TM, McDougall I (1980) Investigations of an intrusive contact, northwest Nelson, New Zealand, I:
thermal, chronological and isotopic constraints. Geochim Cosmochim Acta 44:1985-2003
Harrison TM, Watson EB (1983) Kinetics of zircon dissolution and zirconium diffusion in granitic melts of
variable water content. Contrib Mineral Petrol 84:66-72
Harrison TM, Watson EB (1984) The behavior of apatite during crustal anatexis: equilibrium and kinetic
considerations. Geochim Cosmochim Acta 48:1467-1477
Hart SR (1981) Diffusion compensation in natural silicates. Geochim Cosmochim Acta 45:279-291
Helfferich F, Plesset MS (1958) Ion exchange kinetics. a nonlinear diffusion problem. J Chem Phys 28:418-425
Hofmann AW (1980) Diffusion in natural silicate melts: a critical review. In: Physics of Magmatic Processes.
Hargraves RB (ed) Princeton Univ Press, p 385-417
Hofmann AW, Magaritz M (1977) Diffusion of Ca, Sr, Ba, and Co in a basaltic melt: implications for the
geochemistry of the mantle. J Geophys Res 82:5432-5440
Jambon A (1982) Tracer diffusion in granitic melts: experimental results for Na, Rb, Cs, Ca, Sr, Ba, Ce, Eu to
1300°C and a model of calculation. J Geophys Res 87:10797-10810
Jambon A, Carron JP (1976) Diffusion of Na, K, Rb and Cs in glasses of albite and orthoclase composition.
Jambon A, Semet MP (1978) Lithium diffusion in silicate glasses of albite, orthoclase, and obsidian
compositions: an ion-microprobe determination. Earth Planet Sci Lett 37:445-450
Kirkaldy JS, Young DJ (1987) Diffusion in the Condensed State. The Institute of Metals, London
Kleinhans D, Friedrich R (2007) Continuous-time random walks: simulation of continuous trajectories. Phys
Rev E 76:061102
Koyaguchi T (1989) Chemical gradient at diffusive interfaces in magma chambers. Contrib Mineral Pertrol
103:143-152
Kress VC, Ghiorso MS (1993) Multicomponent diffusion in MgO-Al2O3-SiO2 and CaO-MgO-Al2O3-SiO2
melts. Geochim Cosmochim Acta 57:4453-4466
Kress VC, Ghiorso MS (1995) Multicomponent diffusion in basaltic melts. Geochim Cosmochim Acta 59:313-
324
Lasaga AC (1979) Multicomponent exchange and diffusion in silicates. Geochim Cosmochim Acta 43:455-469
Lasaga AC (1998) Kinetic Theory in the Earth Sciences. Princeton Univ Press, Princeton, NJ
LaTourrette T, Wasserburg GJ, Fahey AJ (1996) Self diffusion of Mg, Ca, Ba, Nd, Yb, Ti, Zr, and U in
haplobasaltic melt. Geochim Cosmochim Acta 60:1329-1340
Lesher CE (1994) Kinetics of Sr and Nd exchange in silicate liquids: theory, experiments, and applications to
uphill diffusion, isotopic equilibrium and irreversible mixing of magmas. J Geophys Res 99:9585-9604
Lesher CE, Walker D (1986) Solution properties of silicate liquids from thermal diffusion experiments.
56 Zhang
Lesher CE, Hervig RL, Tinker D (1996) Self diffusion of network formers (silicon and oxygen) in naturally
occurring basaltic liquid. Geochim Cosmochim Acta 60:405-413
Lewis-Kenedi CB, Lange RA, Hall CM, Delgado-Granados H (2005) The eruptive history of the Tequila
volcanic field, western Mexico: ages, volumes, and relative proportions of lava types. Bull Volcano
67:391-414
Liang Y (1999) Diffusive dissolution in ternary systems: analysis with applications to quartz and quartzite
dissolution in molten silicates. Geochim Cosmochim Acta 63:3983-3995
Liang Y (2000) Dissolution in molten silicates: effects of solid solution. Geochim Cosmochim Acta 64:1617-
1627
Liang Y, Davis AM (2002) Energetics of multicomponent diffusion in molten CaO-Al2O3-SiO2. Geochim
Cosmochim Acta 66:635-646
Liang Y, Richter FM, Chamberlin L (1997) Diffusion in silicate melts, III: empirical models for multicomponent
diffusion. Geochim Cosmochim Acta 61:5295-5312
Liang Y, Richter FM, Watson EB (1996) Diffusion in silicate melts, II: multicomponent diffusion in CaO-
Al2O3-SiO2 at 1500°C and 1 GPa. Geochim Cosmochim Acta 60:5021-5035
Liu Y, Zhang Y (2000) Bubble growth in rhyolitic melt. Earth Planet Sci Lett 181:251-264
Liu Y, Zhang Y, Behrens H (2004) H2O diffusion in dacitic melt. Chem Geol 209:327-340
MacKenzie JM, Canil D (2008) Volatile heavy metal mobility in silciate liquids: implications for volcanic
degassing and eruption prediction. Earth Planet Sci Lett 269:488-496
Matano C (1933) On the relation between the diffusion coefficient and concentrations of solid metals. Japan
J Phys 8:109-113
McConnell JDC (1995) The role of water in oxygen isotope exchange in quartz. Earth Planet Sci Lett 136:97-
107
Mehrer H (2007) Diffusion in Solids: Fundamentals, Methods, Materials, Diffusion-Controlled Processes.
Springer, Berlin
Moulson AJ, Roberts JP (1961) Water in silica glass. Trans Faraday Soc 57:1208-1216
Muehlenbachs K, Kushiro I (1974) Oxygen isotope exchange and equilibrium of silicates with CO2 or O2.
Carnegie Institution of Washington Yearbook:232-236
Mungall JE (2002) Empirical models relating viscosity and tracer diffusion in magmatic silicate melts.
Mungall JE, Dingwell DB (1997) Actinide diffusion in a haplogranitic melt: Effects of pressure, water content,
and pressure. Geochim Cosmochim Acta 61:2237-2246
Mungall JE, Dingwell DB, Chaussidon M (1999) Chemical diffusivities of 18 trace elements in granitoid
Mungall JE, Romano C, Dingwell DB (1998) Multicomponent diffusion in the molten system K2O-Na2O-
Al2O3-SiO2-H2O. Am Mineral 83:685-699
Ni H, Behrens H, Zhang Y (2009) Water diffusion in dacitic melt. Geochim Cosmochim Acta 73:3642-3655
Ni H, Zhang Y (2008) H2O diffusion models in rhyolitic melt with new high pressure data. Chem Geol 250:68-
78
Nowak M, Schreen D, Spickenbom K (2004) Argon and CO2 on the race track in silicate melts: a tool for the
development of a CO2 speciation and diffusion model. Geochim Cosmochim Acta 68:5127-5138
Nye JF (1985) Physical Properties of Crystals. Clarendon Press, Oxford
Patterson C (1956) Age of meteorites and the earth. Geochim Cosmochim Acta 10:230-237
Perkins WG, Begeal DR (1971) Diffusion and permeation of He, Ne, Ar, Kr, and D2 through silicon oxide thin
films. J Chem Phys 54:1683-1694
Press WH, Flannery BP, Teukolsky SA, Vetterling WT (1992) Numerical Recipes. Cambridge Univ Press,
Cambridge, England
Proussevitch AA, Sahagian DL (1998) Dynamics and energetics of bubble growth in magmas: analytical
formulation and numerical modeling. J Geophys Res 103:18223-18251
Rapp RP, Watson EB (1986) Monazite solubility and dissolution kinetics: implications for the thorium and
light rare earth chemistry of felsic magmas. Contrib Mineral Petrol 94:304-316
Richter FM (1993) A model for determining activity-composition relations using chemical diffusion in silicate
Richter F, Liang Y, Minarik WG (1998) Multicomponent diffusion and convection in molten MgO-Al2O3-SiO2.
Richter RM, Watson EB, Mendybaev RA, Teng FZ, Janney PE (2008) Magnesium isotope fractionation in
silicate melts by chemical and thermal diffusion. Geochim Cosmochim Acta 72:206-220
Roering JJ, Kirchner JW, Dietrich WE (1999) Evidence for nonlinear, diffusive sediment transport on hillslopes
and implications for landscape morphology. Water Resources Res 35:853-870
Sauer VF, Freise V (1962) Diffusion in binaren Gemischen mit Volumenanderung. Z Elektrochem Angew Phys
Chem 66:353-363
Schock RN (1985) Point Defects in Minerals. Geophysical Monograph, Geophys Union, Washington, DC
Shannon RD (1976) Revised effective ionic radii and systematic studies of interatomic distances in halides and
chalcogenides. Acta Cryst A32:751-767
Shelby JE, Keeton SC (1974) Temperature dependence of gas diffusion in glass. J Appl Phys 46:1458-1460
Shewmon PG (1989) Diffusion in Solids. Minerals, Metals & Materials Society, Warrendale, PA
Sierralta M, Nowak M, Keppler H (2002) The influence of bulk composition on the diffusivity of carbon
dioxide in Na aluminosilicate melts. Am Mineral 87:1710-1716
Stolper EM (1982a) Water in silicate glasses: an infrared spectroscopic study. Contrib Mineral Petrol 81:1-17
Stolper EM (1982b) The speciation of water in silicate melts. Geochim Cosmochim Acta 46:2609-2620
Sugawara H, Nagata K, Goto KS (1977) Interdiffusivities matrix of CaO-Al2O3-SiO2 melt at 1723 K to 1823
K. Metall Trans 8B:605-612
Sutherland W (1905) A dynamical theory of diffusion for non-electrolytes and the molecular mass of albumin.
Phil Mag 9:781-785
Tinker D, Lesher CE (2001) Self diffusion of Si and O in dacitic liquid at high pressures. Am Mineral 86:1-13
Trial AF, Spera FJ (1994) Measuring the multicomponent diffusion matrix: experimental design and data
analysis for silicate melts. Geochim Cosmochim Acta 58:3769-3783
Turcotte DL, Schubert G (1982) Geodynamics: Applications of Continuum Physics to Geological Problems.
Wiley, New York, NY
Van der Laan SR, Zhang Y, Kennedy A, Wylie PJ (1994) Comparison of element and isotope diffusion of K and
Ca in multicomponent silicate melts. Earth Planet Sci Lett 123:155-166
Vignes A, Sabatier JP (1969) Ternary diffusion in Fe-Co-Ni alloys. Trans AIME 245:1795-1802
Wang L, Zhang Y, Essene EJ (1996) Diffusion of the hydrous component in pyrope. Am Mineral 81:706-718
Watson EB (1981) Diffusion in magmas at depth in the Earth: the effects of pressure and dissolved H2O. Earth
Watson EB, Cherniak DJ (1997) Oxygen diffusion in zircon. Earth Planet Sci Lett 148:527-544
Watson EB, Baxter EF (2007) Diffusion in solid-Earth systems. Earth Planet Sci Lett 253:307-327
Winchell P (1969) The compensation law for diffusion in silicates. High Temp Sci 1:200-215
Wise DL, Houghton G (1966) The diffusion coefficients of ten slightly soluble gases in water at 10-60°C.
Chem Eng Sci 21:999-1010
Xu Z, Zhang Y (2002) Quench rates in water, air and liquid nitrogen, and inference of temperature in volcanic
eruption columns. Earth Planet Sci Lett 200:315-330
Yund RA, Anderson TF (1978) The effect of fluid pressure on oxygen isotope exchange between feldspar and
water. Geochim Cosmochim Acta 42:235-239
Zhang Y (1993) A modified effective binary diffusion model. J Geophys Res 98:11901-11920
Zhang Y (1994) Reaction kinetics, geospeedometry, and relaxation theory. Earth Planet Sci Lett 122:373-391
Zhang Y (2005) Global tectonic and climatic control of mean elevation of continents, and Phanerozoic sea level
change. Earth Planet Sci Lett 237:524-531
Zhang Y (2008) Geochemical Kinetics. Princeton Univ Press, Princeton, NJ
Zhang Y (2009) Degassing of lunar basalts. Geochim Cosmochim Acta 73:A1514
Zhang Y, Behrens H (2000) H2O diffusion in rhyolitic melts and glasses. Chem Geol 169:243-262
225
Zhang Y, Stolper EM, Wasserburg GJ (1991a) Diffusion of water in rhyolitic glasses. Geochim Cosmochim
Acta 55:441-456
Zhang Y, Stolper EM, Wasserburg GJ (1991b) Diffusion of a multi-species component and its role in the
diffusion of water and oxygen in silicates. Earth Planet Sci Lett 103:228-240
Zhang Y, Walker D, Lesher CE (1989) Diffusive crystal dissolution. Contrib Mineral Petrol 102:492-513
Zhang Y, Xu Z, Behrens H (2000) Hydrous species geospeedometer in rhyolite: improved calibration and
application. Geochim Cosmochim Acta 64:3347-3355
Zhang Y, Xu Z, Liu Y (2003) Viscosity of hydrous rhyolitic melts inferred from kinetic experiments, and a new
viscosity model. Am Mineral 88:1741-1752
Zhang Y, Xu Z, Zhu M, Wang H (2007) Silicate melt properties and volcanic eruptions. Rev Geophys
45:RG4004, doi:4010.1029/2006RG000216
Zhao ZF, Zheng YF (2007) Diffusion compensation for argon, hydrogen, lead, and strontium in minerals:
Empirical relatioinships to crystal chemistry. Am Mineral 92:289-308
58 Zhang
APPENDIx 1.
ExPRESSION OF DIFFUSION TENSOR IN CRYSTALS
wITH DIFFERENT SYMMETRY
This appendix gives explicitly the representation of the diffusion tensor in crystals with
different symmetry following similar analyses in Nye (1985). Note that the diffusion tensor is
a symmetric tensor based on the Onsager Reciprocal Principle (De Groot and Mazor 1962).
Hence, a maximum of 6 coefficients are needed to describe the diffusion tensor in crystals with
the lowest symmetry.
In glasses and crystals of cubic systems, the D tensor in any arbitrary orthogonal axes can
be written as a diagonal matrix with equal diagonal elements (or simply written as a scalar D):
D 0 0 
 
D= 0 D 0=D ( A1)
 0 0 D
 
The diffusivity along any direction is D.
In hexagonal, tetragonal and trigonal systems, the D tensor can be written as follows (2
independent coefficients) by choosing the z-axis along c direction, and x- and y-axes as any
two mutually orthogonal axes in the plane perpendicular to c:
 D⊥ c 0 0 
 
D= 0 D⊥ c 0  ( A2)
 0
 0 D||c 
The diffusivity along the direction of the concentration gradient (diffusion direction) is D(q) =
D^csin2q + D||c cos2q, where q is the angle between the diffusion direction and c.
In orthorhombic systems, the D tensor can be written as follows (3 independent
coefficients) by choosing the x-, y-, and z-axes along the crystallographic a, b, and c directions:
 Da 0 0 
 
D= 0 Db 0  ( A3)
 0
 0 Dc 
The diffusivity along any direction is D = Dacos2qa + Dbcos2qb + Dccos2qc, where qa, qb, and
qc are the angles between the diffusion direction and the respective crystallographic axes a, b
and c.
In monoclinic systems, the D tensor can be written as follows (4 independent coefficients)
by choosing the y-axis along b direction, and x- and z-axes to be two mutually perpendicular
directions in the a-c crystallographic plane:
 D11 0 D13 
 
D= 0 D22 0  ( A4)
 
 D13
 0 D33 
Triclinic systems (6 independent coefficients) (x-, y- and z-axes can be any three mutually
perpendicular directions):
 D11 D12 D13 

 
D =  D12 D22 D23  ( A5)
 
 D13
 D23 D33 
In monoclinic and triclinic systems, diffusion tensors (A4) and (A5) can be transformed
to the diagonal form:
 D1 0 0 
 
D= 0 D2 0  ( A6)
0
 0 D3 
which is similar to Equation (A3) but the axes are the principal diffusion axes 1, 2, 3 for
diffusion (not necessarily the crystallographic directions a, b and c), and D1, D2 and D3 are the
principal diffusivities along the principal axes. (In the monoclinic system, principal diffusion
axis 2 is the same as the crystallographic direction b, and D2 = Db.) The diffusivity along any
direction is D = D1cos2q1 + D2cos2q2 + D3cos2q3, where q1, q2, and q3 are the angles between
the diffusion direction and the respective principal diffusion axes 1, 2 and 3.
Vol. 72 pp. 171-225, 2010 5
Diffusion of H, C, and O Components

in Silicate Melts
Youxue Zhang
University of Michigan
Ann Arbor, Michigan 48109-1005, U.S.A.
youxue@umich.edu
Huaiwei Ni
Bayerisches Geoinstitut
University of Bayreuth
D-95440 Bayreuth, Germany
INTRODUCTION
Silicate melts are complicated in structure and composition. Hence, diffusion in a melt (in
this chapter, a melt includes both liquid and glass) can be very complicated due to the presence
of many components, each of which can be present in different species, as well as various kinds
of diffusion, such as self-diffusion, tracer diffusion, interdiffusion, effective binary diffusion and
multi-component diffusion. In this chapter, the diffusion of H, C and O components is reviewed.
H may be in the form of H2O molecules (hereafter referred to as H2Om), OH groups (hereafter
referred to as OH), H2 molecules (hereafter referred to as H2), and other species. H2Om and OH
will be collectively referred to as the H2O component, or the hydrous component. C may be
present as CO2 molecules (hereafter referred to as CO2,molec), CO32− groups (hereafter referred
to as CO32−), and other species (such as CO and CH4). CO2 molecules and CO32− groups will be
collectively referred to as total CO2 (CO2,total) or the CO2 component. O may be present in the
form of network oxygen (such as bridging oxygen, non-bridging oxygen, and free O2−), H2Om,
OH, CO2,molec, CO32−, O2 molecules (hereafter referred to as O2), etc. H2O and CO2 are the major
volatile components in melts and their diffusion plays a critical role in bubble growth (e.g.,
Proussevitch and Sahagian 1998; Gardner et al. 2000; Liu and Zhang 2000; Wang et al. 2009),
magma degassing (e.g., Bottinga and Javoy 1990; Navon and Lyakhovsky 1998; Proussevitch
and Sahagian 1998), magma fragmentation (Zhang 1999a), volcanic eruptions (e.g., Proussevitch
and Sahagian 1996), and welding of erupted volcanic particles (e.g., Sparks et al. 1999). Hence,
understanding H2O and CO2 diffusion is important in terms of volcanic hazard mitigation and
global volatile budgets. Diffusion of H2O may also play a role during diffusive and convective
growth and dissolution of hydrous minerals (e.g., Chen and Zhang 2008, 2009; Zhang 2008). In
glass industry, H2O diffusion is important to glass stability and strength (e.g., Doremus 1973). On
the other hand, oxygen is the most abundant element in silicate melts, and its diffusion controls
many reactions and transport properties of melts and is related to the melt structure. The diffusion
of H2O, CO2 and oxygen is assessed together because of the following: (1) The diffusion of these
three components is all affected by the presence of multiple species. Hence, the diffusion belongs
to the same category of multi-species diffusion. (2) To understand oxygen diffusion, it is often
necessary to understand the diffusion of the individual oxygen species including especially H2O
(as well as other oxygen-bearing species such as bridging oxygen, non-bridging oxygen, free
O2−, O2 molecules, OH, CO2 and CO32−). (3) As the major volatile components in silicate melts,
1529-6466/10/0072-0005$10.00 DOI: 10.2138/rmg.2010.72.5

172 Zhang & Ni
H2O diffusion and CO2 diffusion often must be considered together. For completeness of this
review, molecular H2 and O2 diffusion is also included. Below, we first review H2O (including
OH) and molecular H2 diffusion, then CO2, and finally oxygen diffusion (including molecular
O2 diffusion). Other possible species of H, C, and O include CO and CH4. However, in natural
silicate melts, no data are available for the diffusion of these species, and the concentrations of
these species are low, meaning that they are unlikely to play a major role in the transport of H, C
and O in natural silicate melts.
DIffUSION Of THe H2O COMpONeNT

A review covering H2O diffusion was published recently (Zhang et al. 2007). Nonetheless,
major progress has been made since then. As will be seen below, syntheses of experimental data
to formulate more general models in several silicate melts have been carried out since 2007.
Because H2O speciation is the key in understanding the complicated behavior of H2O diffusion,
we first discuss both the equilibrium and kinetic aspects of H2O speciation to prepare for the
discussion of H2O diffusion.
H2O speciation: equilibrium and kinetics
In natural silicate melts, H is present in the oxidized form as the hydrous component.
Dissolved H2O in silicate melts is present in at least two species: one is H2Om, and the other
is OH (Stolper 1982a,b). Total H2O content will be referred as H2Ot. Under highly reducing
conditions, there may also be molecular hydrogen (H2) that may transport the hydrogen
component. In natural and experimental melts, the presence of H2 has not been detected yet. H2
diffusion will be discussed in a later section.
H2Om and OH can interconvert through the following reaction in melts (Stolper 1982a,b):
H2Om(melt) + O(melt)  2OH(melt) (1)
where O means an anhydrous oxygen ion, ionic charge is ignored, and the phase is indicated
inside the parentheses. The two hydrous species, H2Om and OH, can be detected easily by near-
infrared (NIR) spectroscopy (Stolper 1982a; Cherniak et al. 2010, this volume). To convert IR
peak heights or areas to H2Om and OH concentrations requires independent calibration to
determine molar absorptivities. The equilibrium constant of the above reaction can be written
as:
[OH]2e
K= (2)
[H 2 Om ]e [O]e
where brackets mean activities approximated by mole fractions, and the subscript “e” means at
stable or metastable equilibrium. The mole fractions are calculated as:
Cw
[H 2 O t ] = 18.015 (3)
Cw (100 − Cw )
+
18.015 W
C1
[H 2Om ] = [H 2O t ]
Cw
=
[OH] 2([H 2O t ] − [H 2Om ])
[O] =
1 − [H 2Om ] − [OH]
where Cw is wt% of H2Ot (at 2 wt%, Cw = 2), C1 is wt% of H2Om, and W is the molar mass of the
dry melt on a single oxygen basis (Stolper 1982a; Zhang 1999b; see Table 1). The composition
Table 1. Chemical composition (wt%) of melts for H2O diffusion studies.
ID Composition SiO2 TiO2 Al2O3 feO MnO MgO CaO Na2O K2O p2O5 W Ref
1 Rhyolite 1 76.59 0.08 12.67 1.00 0.03 0.52 3.98 4.88 32.55 a-c
2 AOQ 76.14 13.53 4.65 5.68 32.60 d
3 Rhyolite 2 77.30 0.10 13.00 0.50 0.1 0.5 3.8 4.7 32.39 e
4 MAC 72.26 0.04 15.83 0.61 0.06 0.02 0.22 4.14 3.66 32.37 f
5 NSL 75.45 0.19 10.05 4.29 0.12 0 0.17 5.27 4.56 33.26 f
6 CBS rhyolite 76.38 0.20 10.32 4.37 5.11 4.66 33.19 g
7 Dacite 1 65.03 0.67 16.63 4.16 1.96 5.10 3.95 2.70 33.84 h-j
8 Dacite 2 67.54 0.77 15.74 4.28 0.12 1.43 4.40 3.58 2.15 33.49 i-k
9 Andesite 1 57.21 0.84 17.50 7.58 0.11 4.27 7.59 3.31 1.60 34.98 j
10 Andesite 2 62.54 0.70 16.74 5.55 0.10 2.97 6.48 3.20 1.69 34.13 k
11 Haploandesite 62.20 20.10 0.03 0.02 2.32 10.11 4.30 0.98 0.02 33.57 l
12 Trachyte 59.90 0.39 18.00 0.89 0.12 3.86 2.92 4.05 8.35 0.21 34.55 m
13 Basalt 1 50.60 1.88 13.90 12.5 0.23 6.56 11.4 2.64 0.17 0.21 36.59 n
14 Basalt 2 46.12 1.50 16.11 10.84 0.20 7.60 13.32 3.56 0.76 37.15 k
The compositions are listed on the anhydrous basis. W is weight of the dry melt per mole of oxygen (g/mol) assuming all Fe is ferrous. (Including ferric Fe would lower W values
Diffusion of H, C, O Components in Silicate Melts
slightly.) For simplicity in treatment, small difference in W between similar compositions is ignored in literature. For example, for compositions 1-4, W = 32.49; for compositions 5 and
6, W = 33.24 (Wang et al. 2009 used a slightly different value of 33.14 by considering Fe2O3), etc. MAC is a peraluminous rhyolite; NSL and CBS rhyolite are peralkaline rhyolites.
References: a. Shaw (1974), Delaney and Karsten (1982), Karsten et al. (1982), and Zhang et al. (1991a); b. Zhang and Behrens (2000); Behrens et al. (2007); c. Ni and Zhang (2008);
d. Nowak and Behrens (1997); e. Okumura and Nakashima (2004); f. Behrens and Zhang (2009); g. Wang et al. (2009); h. Liu et al. (2004b); i. Ni et al. (2009b); j. Behrens et al. (2004);
k. Okumura and Nakashima (2006); l. Ni et al. (2009a); m. Freda et al. (2003); n. Zhang and Stolper (1991).
173
174 Zhang & Ni
and W for some melts are listed in Table 1. The temperature dependence of K takes the following
form (e.g., Zhang et al. 1997a):
lnK = A + B/T (4)
where A and B are constants related to the standard state entropy and enthalpy of Reaction (1).
The equilibrium speciation of rhyolite, dacite and andesite melts has been investigated.
Before an understanding of equilibrium speciation, it is necessary to understand (i) the kinetics
on how long it would take to reach equilibrium (e.g., Zhang et al. 1991a, 1995, 1997b, 2000;
Withers et al. 1999; Liu et al. 2004a; Hui et al. 2008), and (ii) the kinetic effect on the species
concentrations during cooling (Dingwell and Webb 1990; Zhang 1994) so as to understand
whether species concentrations can be preserved upon quench. These have been reviewed in
Zhang (1999b), Zhang et al. (2007) and Hui et al. (2008) and only a brief summary is given
below. Speciation data include both quench data and in situ data. Both the quench data and in
situ data suffered from problems in the early stages as shown below. The experimental studies
of H2O speciation are probably the most debated experimental topics concerning H2O in silicate
melts, with two major issues (the quench effect on quench experiments, and the temperature
dependence of molar absorptivities on in situ IR analyses) summarized below.
Quench effect. The equilibrium speciation data obtained by the quench technique before
1990 (e.g., Silver and Stolper 1989; Stolper 1989; Silver et al. 1990) at temperatures above
~1000 K suffered from the quench effect. The data obtained at significantly lower temperatures
in these papers are generally acceptable. The time for Reaction (1) to reach equilibrium decreases
as temperature and H2Ot increase. When the experimental temperature is high (typically ~1000
K, but depending on H2Ot and cooling rate), it often takes less than a second (shorter at higher
H2Ot) for Reaction (1) to reach equilibrium. That is, there is continuous reaction as the sample
cools down. To facilitate discussion, we define quotient Q as:
[OH]2
Q= (5)
[H 2Om ][O]
The expression of Q is the same as that of K except that K is restricted to the case when equilib-
rium is reached. If Q > K, the reaction goes to the left to form H2Om. Otherwise, it goes to the
right to form OH. For a sample cooled to room temperature, the measured species concentra-
tion can be used to calculate Q, which is the apparent equilibrium constant Kae of this sample
(it is called “apparent equilibrium” since the sample did not exactly reach equilibrium at the
corresponding temperature). This apparent equilibrium constant corresponds to an apparent
equilibrium temperature (Tae) as follows,
B
lnQ= A + (6)
Tae
That is, the apparent equilibrium temperature is the temperature calculated from the quotient
Q assuming Q were the same as K even though equilibrium is not reached at any temperature
(comparing Eqns. 4 and 6). During cooling at various quench rates, the variation of Q and Tae as
a function of temperature is illustrated in Figure 1 (Zhang 1994, 2008; Zhang et al. 1997b, 2000).
Starting from a high temperature, if cooling rate is high, the resulting Tae after cooling
down is also high, and vice versa. Even at a high cooling rate, species concentrations may
change upon cooling. For a rhyolite melt containing 2.5 wt% H2Ot equilibrated at a temperature
of 1000 K and cooled down at 70 K/s, the resulting Tae is only about 800 K (Zhang et al.
2000), rather than the experimental equilibrium temperature of 1000 K. Dingwell and Webb
(1990) were the first to analyze this problem using the concept of cooling-rate-dependent glass
transition temperature, and assumed that the apparent equilibrium temperature for Reaction
(1) is similar to the glass transition temperature, which was verified later (e.g., Zhang et al.
Youxue Zhang and Ni (Ch 5) Page 1
Diffusion of H, C, O Components in Silicate Melts 175
figure 1. Schematic evolution of Q and Tae for rapid cooling and slow cooling of a hypothetical reaction.
Fig 1. Schematic evolution of Q and Tae for rapid cooling and slow cooling of a hypothetical reaction.
1997b, 2003; Behrens and Nowak 2003; Hui et al. 2008). Because the temperature was high
for many early experiments using the quench method, equilibrium H2O speciation was not
fully quenchable even with the fastest achievable quench rate. That is, the reported data did not
reflect true speciation at the experimental temperature; only Tae can be obtained. On the other
hand, when the experimental temperature is low enough (usually below 900 K, but depending
on H2Ot and quench rate), reaction during quench is negligible and Tae is roughly the same as
the experimental temperature. However, the experimental temperature cannot be too low (such
as < 600 K) either, otherwise it would require too long a duration to reach equilibrium. Based
on the above considerations, the appropriate temperature range for quench method is typically
650-900 K (which are referred to as intermediate temperatures).
Dependence of molar absorptivities on temperature. The first in situ IR measurements
indicated that the molar absorptivities depend on temperature (e.g., Keppler and Bagdassarov
1993). However, for some years afterwards, temperature effects were not considered in
interpreting in situ IR speciation data (Nowak and Behrens 1995; Shen and Keppler 1995;
Sowerby and Keppler 1999) and the results were inconsistent with speciation data quenched
from intermediate temperatures (e.g., Zhang et al. 1991a, 1995; Ihinger et al. 1999). To resolve
the inconsistency, two groups with different views collaborated (Zhang and Behrens 1998;
Withers et al. 1999). They demonstrated that (i) molar absorptivities vary with temperature
even below the glass transition, which led to an apparent change in species concentrations as
well as inaccuracy of K values in Nowak and Behrens (1995), Shen and Keppler (1995), and
Sowerby and Keppler (1999); and (ii) after accounting for the temperature effect of the molar
absorptivities, the speciation data from quench experiments from intermediate temperatures
and those from in situ experiments are consistent, as also demonstrated by other studies (Zhang
1999b; Withers and Behrens 1999; Nowak and Behrens 2001; Behrens and Nowak 2003).
176 Zhang & Ni
However, the two data sets cover different temperatures (with only small overlap) and seem to
indicate a slightly different standard state enthalpy change for Reaction (1).
Summary of speciation data. After two vigorous debates discussed above, one on the
quench effect, and the other on the in situ measurements, it is now known that the quench method
can be used to obtain reliable speciation data when the experimental temperature is low enough
(usually below ~900 K, but the exact temperature depends on H2Ot and quench rate, and must
be determined experimentally for a given system), and the in situ method can also be used to
obtain reliable speciation data when the temperature dependence of the molar absorptivities is
accounted for. In addition, the speciation constant as a function of temperature may be inferred
by estimation of the glass transition temperature from viscosity models and assuming the glass
transition temperature is the apparent equilibrium temperature (Dingwell and Webb 1990; Zhang
et al. 2003); referred to as the fictive temperature method hereafter. Subsequently, speciation in
the following melts have been investigated further: albite and some synthetic sodium-calcium
aluminosilicate melts (Ohlhorst et al. 2000), AOQ (Qz28Ab38Or34 where Qz means quartz, Ab
means albite and Or means orthoclase, and the values indicate wt%) melt (Nowak and Behrens
2001; Behrens and Nowak 2003), dacite melt (Liu et al. 2004a), andesite melt (Botcharnikov et
al. 2006), NS4 (Na2O·4SiO2, or Na2Si4O9) and NS6 (Na2O·6SiO2, or Na2Si6O13) melts (Behrens
and Yamashita 2008), high-pressure rhyolite melts (Hui et al. 2008), and haploandesite melt (Ni
et al. 2009a). The precision of the speciation data depends on the method (the 2σ precision in
lnK is often about 0.025 for the quench method, 0.1 for the in situ method, and 0.1 to 0.5 for the
fictive temperature method). The accuracy, on the other hand, depends on the accuracy of the
infrared calibration for species concentrations, which is not well characterized. For example,
even for rhyolite glass, which has been studied most extensively (Newman et al. 1986; Zhang et
al. 1997a; Withers and Behrens 1999), the possible variation of molar absorptivities with H2Ot
and quench rate is still not accurately known. Studies on hydrous speciation in Fe-rich melts such
as dacite melt have another difficulty: the high iron concentration leads to significant and broad
absorption bands that overlap with hydrous species bands in the NIR. What makes the problem
worse is the variation of the shape and intensity of the iron-related bands before and after heating.
Hence, many authors have opted to investigate iron-free analogs of natural silicate melts.
With the above preamble and precaution, some speciation data are shown and compared in
Figure 2. The pressure effect on speciation is examined by Hui et al. (2008); K may increase or
decrease with pressure, but the effect is not very large and not included in Figure 2. From Figure
2, several observations can be made: (i) K in sodium silicates (NS4 and NS6) is much greater
than in natural aluminosilicate melts and their analogs (as well as albite melt), supporting the
suggestion of formation of NaOH clusters (Kohn et al. 1989), especially in the absence of Al;
(ii) K in a natural rhyolite is similar to that in haplorhyolite (AOQ); the similarity also holds for
natural andesite and haploandesite; (iii) K increases with temperature (meaning Reaction (1)
is an endothermic process); (iv) K increases slightly from rhyolite to dacite to andesite melts,
which means that K values do not simply increase with Na2O because the Na2O (as well as the
K2O) content decreases from rhyolite to dacite to andesite melts. The speciation models (Zhang
et al. 1997a; Liu et al. 2004a; Ni et al. 2009a) that are often used in diffusion studies (Zhang and
Behrens 2000; Behrens et al. 2007; Ni and Zhang, 2008; Behrens and Zhang 2009; Wang et al.
2009; Ni et al. 2009a,b) are:
 3110 
=
K rhyolite exp  1.876 − (7a)
 T 
 2634 
=K dacite exp  1.49 − (7b)
 T 
 2453 
= exp  1.55 −
K haploandesite (7c)
 T 
figure 2. Speciation data (with 2σ errors) and regression lines in different melts at ≤ 0.8 GPa. For rhyolite
melts, the data are also limited to H2Ot ≤ 3 wt% because the given calibration (Zhang et al. 1997a) is best
applied to HFig. 2. Speciation data (with 2σ errors) and regression lines in different melts at ≤ 0.8 GPa. For rhyolite melts, the
2Ot ≤ 3 wt%. Error bars for lnK in rhyolite and haploandesite melts are smaller than the size
of the symbols. Data
data are sources:
also limited to H2rhyolite
Ot ≤ 3 wt%melt using
because the calibration
the given quench method: Zhang
(Zhang et al. 1997a) et al. applied
is best (1997a);to H2Ihinger
Ot ≤ 3 et al.
(1999); Hui et al. (2008); AOQ1: AOQ melt (Table 1) using the in situ method (Nowak and Behrens 2001);
AOQ2: AOQ wt%.melt
Errorusing
bars forthe
lnK fictive temperature
in rhyolite method
and haploandesite (Behrens
melts are smaller thanand Nowak
the size 2003); Data
of the symbols. dacite melt using
sources:
the quench method: Liu et al. (2004a); haploandesite melt using the quench method: Ni et al. (2009a);
rhyolite melt using the quench method: Zhang et al. (1997a); Ihinger et al. (1999); Hui et al. (2008); AOQ1: AOQ
andesite melt using the fictive temperature method: Botcharnikov et al. (2006); NS4 (Na2O·4SiO2) and NS6
(Na2O·6SiOmelt
2) melts
(Table using
1) usingthe fictive
the in temperature
situ method (Nowak and method: Behrens
Behrens 2001); AOQ2:andAOQYamashita (2008).
melt using the fictive temperature
method (Behrens and Nowak 2003); dacite melt using the quench method: Liu et al. (2004a); haploandesite melt
Other expressions of equilibrium

using the quench method: Ni et al. constants aremeltavailable
(2009a); andesite (e.g.,
using the fictive Ihinger
temperature et al.
method: 1999; etNowak
Botcharnikov al. and
Behrens 2001; Hui et al. 2008), but they have not been much used in diffusion
(2006); NS4 (Na2O·4SiO2) and NS6 (Na2O·6SiO2) melts using the fictive temperature method: Behrens and studies either
because they are new, or because they are complicated, or because of the need to maintain
internal consistency with other studies. Using different K values would impact the retrieval of
Yamashita (2008).
the diffusivities of individual H2O species but only negligibly the retrieval of DH2Ot because the
latter is essentially constrained by the length and shape of an H2Ot concentration profile.
More detailed studies of hydrous species reaction kinetics have been carried out through
isothermal experiments or controlled cooling rate experiments (Zhang et al. 1995, 1997b,
2000; Liu et al. 2004a; Hui et al. 2008). However, in treating H2O diffusion, for simplicity,
quasi-equilibrium is often assumed for the species reaction, and complicated kinetics are not
incorporated. Hence, the kinetics is not discussed in detail here.
The variations of H2Om and OH concentrations with H2Ot for a fixed equilibrium constant
K are shown in Figure 3. It can be seen that H2Om is not proportional to H2Ot (that is, the Xm/X
curve in Fig. 3 is not a horizontal line). Instead, at low H2Ot, H2Om is roughly proportional to
the square of H2Ot (in other words, H2Om/H2Ot ratio is proportional to H2Ot), meaning a more
rapid increase of H2Om concentration as H2Ot increases. Below about 0.2 wt% H2Ot, almost
all H2Ot is OH (which does not mean that OH would play a main role in H2O diffusion, as
will be clear later in this chapter). This behavior arises from the stoichiometric coefficient 2
for OH in Reaction (1), which means the square of OH concentration is proportional to H2Om
concentration, leading to the complicated relation between H2Om and H2Ot concentrations as
well as the complicated behavior of H2Ot diffusion.
178 Youxue Zhang and Ni (Ch 5) Zhang
Page&
3 Ni
figure 3. Species fractions, dXm/dX, Xm/(dXm/dX) and (Xm/X)/(dXm/dX) versus H2Ot at a fixed K of 0.5
where Xm = [H2O m]3.and
Fig. = [H2O
X fractions,
Species dXt]. XmX/X
m/dX, =m[H
m/(dX /dX)2O ]/[H
andm(X 2Ot]m/dX)
m/X)/(dX (fraction
versus H2of
Ot atmolecular Hwhere
a fixed K of 0.5 2O).XF
m =is fraction of
OH. DH2Ot and DH2Om are related through dXm/dX; D18O and DH2Ot are related through Xm/(dXm/dX); and 2H-
1 [H2Om] and X = [H2Ot]. Xm/X = [H2Om]/[H2Ot] (fraction of molecular H2O). F is fraction of OH. DH2Ot and
H self-diffusivity in hydrous melts and DH2Ot (chemical diffusivity) are related through (Xm/X)/(dXm/dX).
DH2Om are related through dXm/dX; D18O and DH2Ot are related through Xm/(dXm/dX); and 2H-1H self diffusivity in
H2O diffusion literature

hydrous melts and DH O are related through (Xm/X)/(dXm/dX).
2 t
Technological improvements and theoretical modeling were the keys in the development
of H2O diffusion studies. Most H2O diffusion experiments can be characterized as follows:
hydration of an originally homogeneous and almost water-free sample by exposure to water
vapor or fluid, dehydration of an initially homogeneous hydrous glass by exposure to dry N2
or Ar, and diffusion couple (two samples with different water contents juxtaposed against each
other). In the early years of H2O diffusion studies in silicate melts, diffusion profiles could not
be measured, so either the bulk H2O mass loss from dehydration or mass gain from hydration
was determined by weight change or mass spectrometry or infrared spectroscopy, from
which the diffusivity was inferred (e.g., Shaw 1974). Later, advancements in ion microprobe
analyses (Coles and Long 1974; Hofmann 1974) allowed the microanalytical determination of
H2O diffusion profiles (Delaney and Karsten 1981), but species concentrations could not be
determined. Afterwards, advancement in infrared spectroscopy (Stolper 1982a,b; Acocella et al.
1984) led to the measurement of concentration profiles of both H2Om and OH species (Zhang
et al. 1991a). Coupled speciation and diffusion studies (Zhang et al. 1991a) helped establish
the basic mechanism of H2O diffusion. Later studies, especially those at high temperatures,
used only H2Ot concentration profiles from micro-IR to model H2O diffusion because species
concentrations cannot be quenched from high temperatures.
In the glass and materials science literature, H2O diffusion studies began in the 1960’s. The
studies were limited to low H2O concentrations, with H2Ot typically no more than 0.1 wt% (e.g.,
Drury and Roberts 1963; Cockram et al. 1969; Burn and Roberts 1970; Lanford et al. 1979;
Houser et al. 1980; Tsong et al. 1980; Nogami and Tomozawa 1984a,b; and the recent review
by Shelby 2008). The average OH concentration (proportional to H2Ot) in a thin wafer upon
heat treatment is monitored by IR and the diffusivity is obtained from the diffusive mass loss
equation for a thin wafer (Crank 1975):
4 D
F= t (8)
πL
where F is the fractional mass loss (H2Ot loss divided by initial H2Ot), L is the thickness of the
wafer, t is time, and D is apparent diffusivity (more specifically, D is diffusion-out diffusivity
of H2Ot, denoted as Dout, see below). It is often observed that Dout is proportional to H2Ot, from
which it was suggested that the diffusing species is H2Om (because the H2Om concentration is
proportional to the square of the H2Ot concentration: a rigorous explanation can be found below
in Eqn. 11) (Doremus 1969, 1973; Ernsberger 1980; Smets and Lommen 1983; Nogami and
Tomozawa 1984a,b), but other explanations included the interdiffusion of hydronium H3O+ and
cations (e.g., Cockram et al. 1969; Doremus 1975; Lanford et al. 1979; Houser et al. 1980;
Tsong et al. 1980), or the depolymerization of the silicate network (e.g., Haller 1963; Roberts
and Roberts 1966; Delaney and Karsten 1981). These other explanations gradually become less
cited as later direct measurements of species concentration profiles and modeling showed that
assuming H2Om as the diffusing species can model the detailed diffusion profiles almost perfectly.
Shaw (1974) was the first to investigate H2O diffusion in a geologically relevant silicate
melt. He determined total mass gain due to hydration, and inferred that H2Ot diffusivity depends
strongly on H2Ot content. Jambon (1979) carried out dehydration experiments and measured
the mass loss from the sample. This study was later corrected by Jambon et al. (1992): because
of an error in the estimated initial H2Ot in the sample (0.38 wt% in Jambon 1979 versus the
correct 0.114 wt% in Jambon et al. 1992), the diffusion-out H2Ot diffusivity was corrected
upward by a factor of 11.
Delaney and Karsten (1981) and Karsten et al. (1982) were the first to measure H2O
concentration profiles using the ion microprobe after hydration experiments. The profiles
were fit by numerical solutions assuming some dependence of DH2Ot on H2Ot. Good fits were
obtained for DH2Ot = D0exp(bCw) where Cw is the H2Ot concentration and D0 and b are two
fitting parameters. Hence, it was assumed that DH2Ot depends exponentially on H2Ot. However,
the expression does not work at low H2Ot (such as below 0.5 wt%) because it implies that
DH2Ot approaches a constant as Cw approaches zero, but prior experimental data in the glass
science literature showed that DH2Ot is proportional to Cw even down to very low Cw. Lapham
et al. (1984) used a similar technique and compared diffusion of 1H2O and 2H2O (or D2O)
in rhyolite melts using ion microprobe measurements. Their data apparently showed that
the 1H2O diffusivity was 2 times the 2H2O diffusivity. However, a later publication from the
same laboratory (Stanton et al. 1985) retracted this claim and found no measurable difference
between 2H2O and 1H2O diffusivities.
Wasserburg (1988) analyzed the role of H2O speciation in H2O diffusion. Zhang et al.
(1991a) were the first to utilize FTIR to measure concentration profiles of both H2Om and OH
species in dehydration studies of natural obsidian, at 676-823 K, 0.1 MPa, and ≤ 1.7 wt%
H2Ot. By considering the interconversion reaction between H2Om and OH and the diffusion
of both H2Om and OH in concert, they concluded that OH diffusivity is negligible compared
to H2Om diffusivity (i.e., H2Om is the diffusing species), and H2Om diffusivity is roughly
independent of H2Ot in the samples studied. At low H2Ot, a constant H2Om diffusivity leads to
proportionality between the H2Ot diffusivity and H2Ot content, consistent with the glass science
literature. The much higher diffusivity of H2Om than OH is understandable because H2Om is
a neutral molecule hence its diffusion does not require breaking strong bonds, whereas OH
is bonded to other cations (Si4+, Al3+, Na+, etc) and its motion requires breaking strong bonds
in the silicate structure. In terms of particle size, H2Om is similar or slightly smaller than OH
(Shannon 1976; Zhang and Xu 1995). Zhang et al. (1991a) also distinguished H2Ot diffusivities
during diffusion-out (dehydration) versus diffusion-in (hydration) experiments (Moulson and
Roberts 1961). Because of the dependence of DH2Ot on H2Ot content, the diffusion-in diffusivity
is 1.78 times the diffusion-out diffusivity when DH2Ot is proportional to H2Ot under otherwise
identical conditions (see also Wang et al. 1996). Zhang and Stolper (1991) investigated H2O
diffusion in basalt melt at 1573-1773 K, 1.0 GPa, and ≤0.42 wt% H2Ot, and found that the H2Ot
180 Zhang & Ni
diffusivity is roughly proportional to H2Ot, consistent with H2Om being the diffusing species
and the H2Om diffusivity being independent of H2Ot. Jambon et al. (1992) re-interpreted their
earlier dehydration data and found that (i) the data are consistent with Zhang et al. (1991a) and
(ii) disequilibrium in Reaction (1) can significantly affect DH2Ot.
Nowak and Behrens (1997) carried out diffusion couple experiments and investigated H2O
diffusion in a synthetic rhyolite melt (AOQ) at 1073-1473 K, 0.05-0.5 GPa, and ≤ 9.0 wt%
H2Ot. They demonstrated that the H2Ot diffusivity at H2Ot > 3 wt% increases exponentially
with H2Ot, implying that the H2Om diffusivity cannot be constant but must increase with H2Ot
content at H2Ot > 3 wt%. They fit logDH2Ot as a polynomial function of H2Ot. Zhang and
Behrens (2000) studied H2O diffusion in rhyolite melts at 673-1473 K, 0.0001-0.81 GPa, and ≤
7.7 wt% H2Ot. They confirmed the observations of Nowak and Behrens (1997) and found that
the diffusion profiles can be accurately modeled by assuming H2Om is the diffusing species and
DH2Om increases exponentially with H2Ot, DH2Om = D0eaX, where X is mole fraction of H2Ot, D0
is the DH2Om value as X approaches zero, and a is a parameter characterizing how rapidly DH2Om
increases with H2Ot. The exponential increase means that at low H2Ot, DH2Om increases only
slowly with H2Ot and can be treated roughly as a constant (e.g., for a = 25 and from 0 to 1 wt%
H2Ot, eaX varies from 1 to 1.56, only slightly outside the experimental uncertainty of diffusivity
determinations at high temperatures), consistent with the observations of Zhang et al. (1991a).
The exponential increase is also consistent with the diffusivities of neutral molecular species
such as Ar (Behrens and Zhang 2001) and CO2 (Watson et al. 1982; Watson 1991).
Later experimental studies in general followed the framework established in Zhang et
al. (1991a), Zhang and Stolper (1991), and Zhang and Behrens (2000). Freda et al. (2003)
investigated H2O diffusion in a trachyte melt at 1334-1601 K, 1 GPa, and ≤ 2.0 wt% H2Ot.
Okumura and Nakashima (2004, 2006) developed the in situ FTIR measurements of H2O
loss due to dehydration to obtain diffusivities, and reported results on rhyolite melts at ≤ 4.1
wt% H2Ot and on dacite, andesite and basalt melts at ≤ 1.1 wt% H2Ot. Their method works
well for rapidly obtaining diffusion data but cannot resolve how DH2Ot depends on H2Ot. Liu
et al. (2004b) explored H2O diffusion in dacite melts by dehydration experiments at 824-910
K, ≤ 0.15 GPa, and ≤ 2.5 wt% H2Ot. Behrens et al. (2004) studied H2O diffusion in dacite
and andesite melts at 1458-1858 K, 0.5-1.5 GPa, and ≤ 6.3 wt% H2Ot; some of their data are
consistent with concentration-independent DH2Ot. Behrens et al. (2007) obtained some H2O
diffusion data in rhyolite melts in a study comparing oxygen and H2O diffusion. Ni and Zhang
(2008) further resolved the pressure effect on H2O diffusion in rhyolite melts and constructed a
general H2O diffusivity model over a large range of T, P and H2Ot. Behrens and Zhang (2009)
and Wang et al. (2009) quantified H2O diffusion in peralkaline rhyolite melts. Ni et al. (2009a,b)
examined H2O diffusion in dacite and haploandesite melts. All the melt compositions that have
been studied for H2O diffusion are listed in Table 1.
H2O diffusion, theory and data summary
Strictly speaking, silicate melts are multicomponent systems, and H2O diffusion should
fall into the category of multicomponent diffusion. However, the anhydrous melt composition
generally does not vary along a diffusion profile in the typical design of the experiments, and
addition of H2O mainly causes a dilution effect. Therefore, H2O diffusion in the literature is
treated as effective binary diffusion, or the first kind of effective binary diffusion defined by
Zhang (2010, this volume) and Zhang et al. (2010, this volume).
Because DH2Ot depends on H2Ot concentration, the general equation for one-dimensional
diffusion is as follows:
∂X ∂  ∂X 
=  DH2 Ot (9)
∂t ∂x  ∂x 
where t is time, x is distance, DH2Ot is the diffusivity (effective binary diffusivity) of H2Ot, and
X = [H2Ot] (see Eqn. 3a). The above diffusion equation can be solved given how DH2Ot varies
with H2Ot, and boundary and initial conditions. For hydration and dehydration experiments,
the equilibrium surface concentration and the initial concentration (usually uniform) are the
boundary and initial conditions, respectively. For diffusion couple experiments, no boundary
condition is needed and the initial concentrations in the two starting halves are the initial
condition. If DH2Ot were independent of H2Ot (or X), then the diffusion profile could be fit by an
error function solution. However, this is rarely the case: the variation of DH2Ot with H2Ot turns
out to be complicated.
Experimental results in the glass and geological literature show that when H2Ot is
sufficiently low (≤ 2 wt%, but depending on temperature), DH2Ot is proportional to H2Ot (or
proportional to X in Eqn. 9). Such an equation can be solved numerically (e.g., Crank 1975) and
the solution has been applied to fit diffusion profiles. If a profile can be fit within experimental
uncertainty (e.g., Fig. 4), then the proportionality relation is assumed to describe how DH2Ot
varies with H2Ot.
figure 4. Experimental H2Ot diffu-

sion profile (points) in a peralumi-
nous rhyolite melt and two fits. The
dashed curve is an error function fit
(assuming DH2Ot is independent of
H2Ot). The solid curve is a fit assum-
ing DH2Ot is proportional to H2Ot.
The solid curve agrees well with ex-
perimental data. From Behrens and
Zhang (2009).
Fig. 4. Experimental H2Ot diffusion profile (points) in a peraluminous rhyolite melt and two fits. The dashed curve
At high H2OtD(such
is an error function fit (assuming
as >3 wt%, especially at relatively low temperatures such as 800 K),
H O is independent of H2Ot). The solid curve is a fit assuming DH O is
2 t 2 t
the proportionality relation often does not work well (Nowak and Behrens 1997; Zhang and
proportional to H2Ot. The solid curve agrees well with experimental data. From Behrens and Zhang (2009).
Behrens 2000; Liu et al. 2004b; Ni and Zhang 2008; Ni et al. 2009a,b; Wang et al. 2009). One
example is given in Figure 5. Hence, more complicated relations are proposed to describe how
DH2Ot varies with H2Ot. Based on our knowledge of H2O speciation, the general and mechanistic
approach is to consider the diffusion of both species:
∂X ∂  ∂X 1 ∂X 
=  DH2 Om m + DOH OH  (10)
∂t ∂x  ∂x 2 ∂x 
where Xm is the mole fraction of H2Om, XOH is the mole fraction of OH, and DH2Om and DOH are
the diffusivities of H2Om and OH. The factor 1/2 is due to the fact that one mole of H2Om reacts
to form two moles of OH (Reaction 1).
Experimental data show DOH << DH2Om in rhyolite melt so that diffusive flux due to OH
diffusion can be ignored (Zhang et al. 1991a). (The small diffusivity of OH does not mean
that OH profile is flat, because interconversion between OH and H2Om can change OH
concentration.) Therefore, Equation (10) can be simplified as:
182 Zhang & Ni
figure 5. Experimental H2Ot dif-

fusion profile (points) in a peral-
kaline rhyolite melt and three fits.
The long-dashed (blue in online
version) curve is an error function
fit (assuming DH2Ot is indepen-
dent of H2Ot). The short-dashed
(red in online version) curve is a
fit assuming DH2Ot is proportional
to H2Ot. The solid curve is a fit as-
suming DH2Om = D0eaX (Eqn. 12).
The solid curve agrees well with
experimental data. From Wang et
al. (2009).
Fig. 5. Experimental H2Ot diffusion profile (points) in a peralkaline rhyolite melt and three fits. The long-dashed
∂X ∂  ∂X 
=  DH2 Om m is a fit
(11)
∂x  curve
(blue) curve is an error function fit (assuming DH2Ot is independent of H2Ot). The short-dashed (red) curve
2 t
∂t ∂x
assuming DH O is proportional to H2Ot. The solid curve is a fit assuming DH O = D0eaX (Eq. 12). The solid
2 m
The above equation means that DH2Ot = DH2OmdXm/dX if equilibrium for Reaction (1) is reached
agrees well with experimental data. From Wang et al. (2009).
at every point along a profile (dXm/dX for K = 0.5 is shown in Fig. 3). (Equilibrium for Reaction
(1) is necessary because without equilibrium the partial differential ∂Xm/∂X would depend on
time and cannot be simplified to dXm/dX.) At low H2Ot (e.g., below 2 wt% but depending on the
temperature), DH2Om is roughly independent of H2Ot, and dXm/dX is proportional to H2Ot (Fig.
3), leading to DH2Ot proportional to H2Ot. However, at high H2Ot, DH2Om must increase with H2Ot
(Nowak and Behrens 1997) and has been shown to increase exponentially with H2Ot (Zhang
and Behrens 2000; Liu et al. 2004b; Ni and Zhang 2008; Ni et al. 2009a,b; Wang et al. 2009):
DH2 Om = D0e aX (12)
where a is a fitting parameter that may vary with T and P. Numerous studies have found that
a increases with decreasing temperature, indicating a stronger dependence of DH2Om on H2Ot
at lower temperatures. This finding is not surprising since at low T the presence of H2O has a
larger impact on melt properties such as viscosity (e.g., Zhang et al. 2003), which is not simply
related to the formation of the number of OH but is also related to the larger effect of each
OH and H2Om on the melt viscosity at low temperatures. The above approach can fit all H2Ot
profiles within experimental uncertainty (as well as species concentration profiles if the species
concentrations can be preserved during quench), including those profiles that cannot be fit by
assuming DH2Ot is proportional to H2Ot (Fig. 5). Furthermore, experiments using samples with
different H2Ot generally yield the same set of parameters D0 and a, which further supports the
validity of Equation (12). The exponential dependence of the diffusivity of H2Om as well as
other neutral molecules (Watson 1991; Behrens and Zhang 2001) on H2Ot concentration might
be related to the dramatic change of melt structure, as evidenced by the significantly reduced
melt viscosity caused by the addition of H2O (e.g., Shaw 1974; Zhang et al. 2003).
If equilibrium for Reaction (1) is reached at every point along a profile, the general relation
between DH2Ot and DH2Om is (Wang et al. 2009):
dX m  (0.5 − X ) 
=
DH2 Ot DH2 O= DH2 Om 1 − [ X (1 − X )((4 / K ) − 1) + 0.25]1/ 2  (13)
m
dX  
Diffusion of H, C,Youxue
O Components
Zhang and Ni (Ch 5)
in Silicate Melts
Page 6
183
where K is the equilibrium constant

of Reaction (1) and depends on
temperature (to a lesser extent, on
pressure or H2Ot), and X and Xm are
mole fractions of H2Ot and H2Om
on a single oxygen basis. Hence,
if K is known or fixed, DH2Ot can
be calculated from DH2Om using the
above equation. Figure 6 illustrates
how DH2Om and DH2Ot vary with H2Ot
concentration at given K and a. At
low H2Ot, DH2Ot is approximately
proportional to H2Ot; while at high-
er H2Ot, the relation transitions to
exponential. For rhyolite, dacite and
haploandesite melts, K is given by figure 6. DH2Om and DH2Ot as function of H2Ot for a given
Equations (7a) to (7c). Diffusion set of K and a, calculated from Equations (12) and (13). The
Fig. 6. DH2Om and DH2Ot as function of H2Ot for a given set of K and a, calculated from Eqs. 12 and 13. The relation
data in various melts are summa- relation between DH2Ot and Hto2O
between DH O and H2Ot shifts from quasi-proportional t shifts from
quasi-exponential
quasi-propor-
as H2Ot increases. DH O is lower than
rized below. tional to quasi-exponential as H2Ot increases. DH2Ot is lower
2 t 2 t
DH O .
2 m than DH2Om.
If equilibrium for Reaction
(1) is not reached (at relatively low
temperatures and low H2Ot), one would have to consider the kinetics of Reaction (1) and couple
the kinetic equation with the diffusion equation 11 to solve the concentration profiles. Such
full treatment has not been carried out yet, but experimental data have been noted where these
conditions exist (Jambon et al. 1992; Zhang et al. 2007).
Metaluminous and peraluminous rhyolite melts. “Normal” metaluminous rhyolite melts
including haplorhyolite melts (compositions 1-3 in Table 1) have been studied most extensively
(Shaw 1974; Delaney and Karsten 1981; Karsten et al. 1982; Zhang et al. 1991a; Nowak and
Behrens 1997; Zhang and Behrens 2000; Okumura and Nakashima 2004; Behrens et al. 2007;
Ni and Zhang 2008). Figure 7 compares DH2Ot in rhyolite melts at 1 wt% H2Ot from various
laboratories. It can be seen that the original diffusivity data from different laboratories are
largely consistent at this H2Ot when cast in the same way. Furthermore, there is a significant
pressure effect on DH2Ot, especially at low temperature.
Ni and Zhang (2008) combined their new experimental data and literature data at 676-1900
K, 0-1.9 GPa, 0-8 wt% H2Ot (reflecting the coverage of conditions by the experimental data),
using K from Equation (7a), and obtained DH2Om (m2/s) as follows:
rhyolite  12939 + 3626 P − 75884 X 
DH2 Om =exp  −14.26 + 1.888P − 37.26 X −  (14)
 T 
where T is in K, P is in GPa, and X is mole fraction of H2Ot on a single oxygen basis. The
H2Ot diffusivity in peraluminous rhyolite melts (composition 4 in Table 1) at ≤ 0.5 GPa is
indistinguishable from that in “normal” rhyolite melts and can be described by the above
expression as well (Behrens and Zhang 2009). Hence, DH2Ot for metaluminous and peraluminous
rhyolite melts can be calculated from the above DH2Om using Equation (13) with K from Equation
(7a). The 2σ uncertainty in lnDH2Ot is about 0.5 for this and other melts discussed below.
The above equation implies that DH2Om increases rapidly with H2Ot at low temperatures
but slowly at higher temperatures. Therefore, at higher temperatures, DH2Ot is proportional to
H2Ot from zero to higher concentrations of H2Ot. Because the proportionality equation is easy
to apply and works well at low H2Ot (which means ≤ 1 wt% at 773 K and ≤ 3 wt% at 1473 K),
it is given below (Ni and Zhang 2008):
184 Zhang & Ni
figure 7. Comparison of experimental data on H2Ot diffusivities in “normal” rhyolite melts at 1 wt% H2Ot.
Data sources:Fig.
Data are from Zhang et al. (1991a), Zhang and Behrens (2000), and Ni and Zhang (2008),
7. Comparison of experimental data on H2Ot diffusivities in “normal” rhyolite melts at 1 wt% H2Ot. Data
unless otherwise indicated as: Nowak and Behrens (1997); Karsten et al. (1982); Shaw (1974).
sources: Data are from Zhang et al. (1991a), Zhang and Behrens (2000), and Ni and Zhang (2008), unless otherwise
indicated as: Nowak and Behrens (1997); Karsten et al. (1982); Shaw (1974).
rhyolite  9699 + 3626 P 
DH O = Cw exp  −18.10 + 1.888P −  (15)
2 t
 T 
where T is in K, P is in GPa, and Cw is wt% of H2Ot (Cw = 1 for 1 wt% H2Ot).
Peralkaline rhyolite melts. Behrens and Zhang (2009) and Wang et al. (2009) investigated
H2O diffusion in peralkaline (PA) rhyolite melts (compositions 5 and 6 in Table 1) covering
789-1516 K, 0-1.4 GPa, 0-4.6 wt% H2Ot. Behrens and Zhang (2009) investigated diffusion at
low H2Ot, and Wang et al. (2009) expanded the studied range of H2Ot and pressure. Even though
these studies are not numerous, the effort was coordinated and systematic so that the data are
sufficient to allow the construction of a general model for DH2Om (and hence DH2Ot). Figure 8
compares the H2Ot diffusivities in peralkaline rhyolite melts to metaluminous rhyolite melts.
On average, DH2Ot at 0.5 GPa and 1 wt% H2Ot in peralkaline rhyolite melt is ~2 times that in
metaluminous rhyolite melt. The effect of pressure on DH2Ot in peralkaline rhyolite melts is
smaller than that in metaluminous rhyolite melts. For example, lnDH2Ot decreases by ~1.6 as
pressure increases from 0.0001 to 0.5 GPa for metaluminous rhyolite melts; but by only ~0.3
for peralkaline rhyolite melts. At various P-T conditions, the difference between DH2Ot in per-
alkaline rhyolite melts and that in metaluminous rhyolite melts is often a factor of 2 or smaller.
Using results from both Behrens and Zhang (2009) and Wang et al. (2009), and K from
Equation (7a), Wang et al. (2009) constructed the following expression for DH2Om (m2/s):
PA rhyolite  13939 + 1230 P − 60559 X 
DH2 Om = exp  −12.79 − 27.87 X −  (16)
 T 
where T is in K, P is in GPa, and X is mole fraction of H2Ot on a single oxygen basis. DH2Ot
for peralkaline rhyolite melts can be calculated from the above DH2Om using Equation (13) with
K from Equation (7a). The resulting DH2Om and DH2Ot are slightly larger than that in “normal”
rhyolite melts, but no more than a factor of 2. By combining the data in both Behrens and Zhang
(2009) and Wang et al. (2009), a simple equation to calculate directly DH2Ot at low H2Ot (which
means ≤ 1 wt% at 773 K and ≤ 3 wt% at 1473 K) is as follows:
figure 8. Comparison between H2Ot diffusivities in peralkaline rhyolite melt at 0.5 GPa (filled circles) and
those in metaluminous (or “normal”) rhyolite melts at 0.0001 and 0.5 GPa (open circles and squares). Data
Fig. 8. Comparison between H2Ot diffusivities in peralkaline rhyolite melt at 0.5 GPa (filled circles in red) and
sources for peralkaline rhyolite: Behrens and Zhang (2009) and Wang et al. (2009). See Figure 7 for data
sources for metaluminous rhyolite.
those in metaluminous (or “normal”) rhyolite melts at 0.0001 and 0.5 GPa (black open circles and squares). Data
sources for peralkaline rhyolite: Behrens and Zhang (2009) and Wang et al. (2009). See Fig. 7 for data sources for
metaluminous rhyolite.
PA rhyolite  10870 + 1101P 
DH2 Ot = Cw exp  −16.55 −  (17)
 T 
where T is in K, P is in GPa, and Cw is wt% of H2Ot.
Dacite melts. Liu et al. (2004b), Behrens et al. (2004), Okumura and Nakashima (2006),
and Ni et al. (2009b) quantified H2O diffusion in dacite melts (compositions 7 and 8 in Table 1),
covering 786-1800 K, 0-1 GPa, 0-8 wt% H2Ot. Liu et al. (2004b) and Okumura and Nakashima
(2006) conducted dehydration experiments at intermediate temperature and low pressure;
Behrens et al. (2004) carried out diffusion couple experiments at high temperature and high
pressure; and Ni et al. (2009b) investigated H2O diffusion at intermediate temperature and high
pressure. All data have been combined to evaluate the effect of temperature, pressure and H2Ot
on H2O diffusion. Figure 9 shows some experimental diffusion data and compares these with
data in rhyolite melts. At 1 wt% H2Ot, DH2Ot in dacite melts is lower than that in rhyolite melts at
< 1470 K. Furthermore, for both rhyolite and dacite melts, DH2Ot decreases as pressure increases.
As pressure increases from 0.0001 to 0.5 GPa, lnDH2Ot decreases by ~1.6 for rhyolite melt; but
only by ~1.0 for dacite melts. That is, the pressure effect becomes smaller from rhyolite melt
to dacite melt.
Using results from Liu et al. (2004b), Behrens et al. (2004) and Ni et al. (2009b), and K
from Equation (7b), DH2Om at 786-1800 K, 0-1 GPa, and 0-8 wt% H2Ot (reflecting conditions
covered by the experimental data) can be expressed as (Ni et al. 2009b):
 19064 + 1477P − 108882 X 
DH2 Om dacite = exp  −9.42 − 62.38 X −  (18)
 T 
where T is in K, P is in GPa, and X is the mole fraction of H2Ot on a single oxygen basis. DH2Ot
for dacite melts can be calculated from the above DH2Om using Equation (13) with K from
Equation (7b). At ≤0.7 wt% H2Ot at 773 K, and ≤ 6.2 wt% H2Ot at 1673 K, DH2Ot is proportional
to H2Ot and can be expressed as follows:
186 Zhang & Ni
figure 9. Comparison of DH2Ot in dacite melts (solid circles, triangles and squares) and rhyolite melts
(open circles Fig.
and9.squares)
Comparisonatof1Dwt% H2Ot. melts
H2Ot in dacite
Data(colored
sourcessolidfor dacite
circles, melts:
triangles <0.15 and
and squares) GPa: Liumelts
rhyolite et al.(black
(2004b); 0.5
GPa and 1 GPa: Behrens et al. (2004) and Ni et al. (2009b). Data sources for rhyolite melt: see Figure 7.
The data of Okumura
open circles and Nakashima
and squares) at 1 wt%(2006) aresources
H2Ot. Data not shown
for dacitebecause
melts: <0.15there
GPa: is
Liuaetsmall systematic
al. (2004b); 0.5 GPa anddifference
between their1data and other data (Ni et al. 2009b).
GPa: Behrens et al. (2004) and Ni et al. (2009b). Data sources for rhyolite melt: see Fig. 7. The data of Okumura
and Nakashima (2006) are not shown because there is a small systematic difference between their data and other
data (Ni et al. 2009b).

dacite  15722 + 1466 P 
DH2 Ot = Cw exp  −13.32 −  (19)
 T 
where T is in K, P is in GPa, and Cw is wt% of H2Ot. Note that the proportionality relation holds
for a smaller H2Ot range (≤ 0.7 wt%) in dacite than in rhyolite at intermediate temperatures
such as 773 K, but for a larger H2Ot range (≤ 5 wt%) in dacite than in rhyolite (≤ 3 wt%) at high
temperatures such as 1473 K.
Andesite melts. There have been three studies (Behrens et al. 2004; Okumura and
Nakashima 2006; Ni et al. 2009a) on andesite melts (compositions 9-11 in Table 1). However,
the data are insufficient for the construction of a general model for DH2Om or DH2Ot. There is
also a significant compositional difference (e.g., 5 wt% difference in SiO2, 7 wt% difference in
FeO) between the three andesites used in these studies. H2O diffusion data are summarized in
Figure 10.
Examining Figure 10, the pressure dependence of DH2Ot at 1 wt% H2Ot in andesite melts is
small, at least at high temperature. For example, from 0.5 to 1.5 GPa at high temperature (tri-
anglar points), the variation of DH2Ot is within uncertainty. Furthermore, the difference in DH2Ot
between haploandesite melt and normal andesite melt is small. Hence, we ignore the pressure
dependence and compositional difference and fit all DH2Ot data at low H2Ot in three andesite
melts to obtain the following equation:
andesite  18340 
DH O = Cw exp  −11.80 − (20)
2 t
 T 
where T is in K, D is in m2/s, and Cw is wt% of H2Ot. The above proportionality relation is
applicable to ≤ 0.7 wt% H2Ot at 773 K and ≤ 3.5 wt% H2Ot at 1473 K. At temperatures ≥ 1573
K, there is ambiguity regarding how DH2Ot depends on H2Ot: whether DH2Ot is proportional to
H2Ot or independent of H2Ot (Behrens et al. 2004).
figure 10. Comparison of DH2Ot in andesite melts (solid circles, triangles, diamonds, and squares) and
dacite melts (open circles and squares) at 1 wt% H2Ot. Data
Fig. 10. Comparison of DH2Ot in andesite melts (colored
sources for andesite melts: Andesite1: Behrens
solid circles, triangles, diamonds, and squares) and dacite
et al. (2004); Andesite2: Okumura and Nakashima (2006); Haploandesite: Ni et al. (2009a). Data sources
for dacite melt:
meltssee Figure
(black 9. and squares) at 1 wt% H2Ot. Data sources for andesite melts: Andesite1: Behrens et al.
open circles
(2004); Andesite2: Okumura and Nakashima (2006); Haploandesite: Ni et al. (2009a). Data sources for dacite melt:
At intermediate
see Fig. 9. temperatures and high H2Ot, DH2Ot increases strongly with H2Ot and
calculation of DH2Ot is best through DH2Om. For haploandesite melts (composition 11 in Table 1)
at 743-873 K, 0.1 GPa, 0-2.5 wt% H2Ot, using K from Equation (7c), DH2Om can be expressed
as (Ni et al. 2009b):
haploandesite  18172 − 73136 X 
DH O = exp  −12.27 − 13.91X −  (21)
2 m
 T 
where T is in K and X is the mole fraction of H2Ot on a single oxygen basis. DH2Ot for haploandes-
ite melts (also for Andesite 2 in Table 1, judging from similarity in DH2Ot in the two melts in Fig.
10) can be calculated from the above DH2Om using Equation (13) with K from Equation (7c), but
only under the limited conditions of 743-873 K, 0.1 GPa, 0-2.5 wt% H2Ot.
For other silicate melts (such as basalt and trachyte), the equilibrium constant K for Reaction
(1) has not been characterized. Hence, no DH2Om values or equations have been reported in the
literature, and only DH2Ot values and expressions under limited conditions are obtained. These
are summarized below.
Basalt melts. For basalt melts (compositions 13 and 14 in Table 1), the investigations are
limited (Zhang and Stolper 1991; Okumura and Nakashima 2006; Persikov et al. 2010) and
the range of H2Ot covered in experiments is only ≤ 1.1 wt%. Assuming DH2Ot is proportional to
H2Ot and ignoring the pressure effect on the diffusivity (the validity of this assumption needs
confirmation), Zhang et al. (2007) summarized the data and expressed DH2Ot as follows at 673-
1773 K, ≤ 1 GPa and ≤ 1.1 wt% H2Ot:
basalt  19110 
DH O = Cw exp  −8.56 − (22)
2 t
 T 
where T is in K and Cw is wt% of H2Ot. The recent high temperature data on a haplobasalt
melt (62.2 wt% SiO2, 10.5 wt% Al2O3, 6.8 wt% MgO; 14.2 wt% CaO and 6.3 wt% Na2O) by
Persikov et al. (2010) are about 0.17 times those on a MORB melt (composition 13 in Table 1)
188 Zhang & Ni
by Zhang and Stolper (1991) when compared at 1 wt% H2Ot assuming. It is not clear whether
this difference is due to the significant compositional difference between the two melts, or due
to the different H2Ot concentration coverage, or due to experimental uncertainties.
Trachyte melts. For trachyte melt (composition 12 in Table 1), there has been only one
study (Freda et al. 2003). The equilibrium constant K for Reaction (1) for trachyte melts has not
been characterized. Assuming DH2Ot is proportional to H2Ot, Zhang et al. (2007) summarized
the data and expressed DH2Ot as follows at 1334-1601 K, 1 GPa and ≤ 2 wt% H2Ot:
trachyte  17975 
DH O = Cw exp  −10.90 − (23)
2 t
 T 
where T is in K and Cw is wt% of H2Ot. The H2O diffusivity in trachyte melt is slightly (by a
factor of about 3) higher than that in andesite melt with a similar SiO2 content. Combining this
observation with the small difference in the H2O diffusivity between metaluminous rhyolite and
peralkaline rhyolite, it appears that H2O diffusivity in silicate melts depends primarily on the
SiO2 content, and the effects of other components, such as alkalinity, are minor (though still
noticeable).
Self-diffusion of H2O. The above discussion is on the chemical diffusion or effective binary
diffusion of H2O, when there is a chemical concentration (or chemical potential) gradient in
H2O. Limited data on the self-diffusion of H2O, or exchange of 2H2O and 1H2O in a chemically
homogeneous system (i.e., without H2O chemical concentration gradients), are available (e.g.,
Nowak and Behrens 1997). The theoretical analysis by Zhang et al. (1991b) showed that the
self-diffusivity of H2O (D1H2O-2H2O) can be written as follows if H2Om is the diffusing species:
Xm Xm / X
D=
1
H 2 O- 2H 2 O
D=
H 2 Om DH2 Ot (24)
X dX m / dX
where X and Xm are mole fractions of H2Ot and H2Om on a single oxygen basis. The ratio (Xm/X)
can be calculated as follows (Eqn. 13 in Zhang 1999b):
Xm 8X
= (25)
X 8 X + K (1 − 2 X ) + K 2 (1 − 2 X )2 + 16 KX (1 − X )
where K is the equilibrium constant of Reaction (1). The calculation of dXm/dX can be found
from Equation (13). The values of (Xm/X)/(dXm/dX) are shown in Figure 3 for K = 0.5, and vary
from 0.5 at 0 wt% H2Ot to 0.67 at 7 wt% H2Ot. Hence, at low H2Ot, the D-H self-diffusivity
D1H2O-2H2O is about 0.5 times DH2Ot (Doremus 1969; Zhang et al. 1991b). At higher H2Ot, D1H2O-
2H O is closer to DH O (such as 0.7 times DH O ). The data of Nowak and Behrens (1997) are
2 2 t 2 t
consistent with the above results. This analysis shows that under some conditions, self-diffusivity
is smaller than chemical diffusivity, although some studies showed that self-diffusivity is often
greater than chemical (effective binary) diffusivity (Lesher 1990, 1994; Zhang 1993; Van der
Laan et al. 1994).
All natural silicate melts. For rhyolite to andesite melts in the calc-alkaline series, Behrens
et al. (2004) provided an expression of H2Ot diffusivity at 1 wt% H2Ot, Okumura and Nakashima
(2006) related the H2Ot diffusivity at 0.7 wt% H2Ot to NBO/T (ratio of non-bridging oxygen to
tetrahedrally coordinated cations) in the melt, and Ni et al. (2009a) presented a new diffusivity
expression specifically for intermediate temperatures (700-900 K). In all three studies, a single
parameter (silica content or NBO/T) is used to characterize the complicated melt composition
variation, which is clearly a simplification.
H2Ot diffusivities at 1 wt% H2Ot in various melts are summarized in Figure 11. Continuing
the effort of Behrens et al. (2004) and Ni et al. (2009a), we use H2Ot diffusivity data from 116
figure 11. Comparison of H2Ot

diffusivity at 1 wt% H2Ot in dif-
ferent melts at (a) intermediate
temperatures and ≤ 0.2 GPa;
and (b) high temperatures and
1 GPa. The trend from rhyolite
to dacite to andesite is the oppo-
site in the two cases. H2Ot dif-
fusivity is higher in peralkaline
melts than in metaluminous and
peraluminous melts. The lines
(of the same color in the on-
line version as the data points;
solid line for rhyolite melt; long
dashes for peralkaline rhyolite;
medium dashes for dacite; short
dashes for andesite; and dot line
for trachyte) are calculated for
the appropriate melt by Equa-
tion (26).
experiments by IRofprofiling
Fig. 11. Comparison method
H2Ot diffusivity except
at 1 wt% H2Ot infor basalt
different melts,
melts and arrivetemperatures
at (a) intermediate at the following
and ≤ 0.2 general
model for(b)Dhigh
GPa; and at 1 wt%and
H2Ottemperatures H21OGPa.
t: The trend from rhyolite to dacite to andesite is the opposite in the two
cases. H2Ot diffusivity is higher

1 wt%
ln Din = 12.06
peralkaline
H2 O t
+ 14.75AI
melts − 42.22Si
than in metaluminous and+peraluminous
2.716 P ×melts.
Si The lines (of (26)
42021
the same color as the data points) are calculated − 45407Si
for the appropriate + (7409Si
melt − 1593)P
by Eq. 26.
−
T
where D is in m2/s, T is in K, P is in GPa, Si is the cation mole fraction of Si, and AI = Na+K-Al
where Na, K and Al are cation mole fractions. One additional compositional parameter AI is
used in the above fitting compared to the models of Behrens et al. (2004) and Ni et al. (2009a).
The above expression reproduces all the literature data for rhyolite-peralkaline rhyolite-dacite-
andesite-haploandesite-trachyte composition within a factor of 2, covering 676-1629 K and
0-1.9 GPa (Fig. 11). Equation (26) can also be used to estimate the H2Ot diffusivity at <1
wt% H2Ot at intermediate temperatures and ≤ 3 wt% H2Ot at high temperatures with the aid
of proportionality assumption. The data on basalt melt (Zhang and Stolper 1991) cannot be
modeled well using a simple compositional dependence. The reason is not clear. The data of
Okumura and Nakashima (2004, 2006) are not included in the fitting because they are not based
on measured profiles.
190 Zhang & Ni
To develop a more general expression of H2O diffusivity in all natural silicate melts at
both high and low H2Ot for geological applications, more experimental data are necessary.
The critical data include diffusion in basalt melts at high H2Ot at both high and intermediate
temperatures, and in andesite melts at high H2Ot and intermediate temperatures. Furthermore,
it is necessary to resolve the dependence of DH2Ot on H2Ot content in high-temperature andesite
melts by careful experiments. More data on peralkaline melts from basanite to phonolite to
further confirm that the effect of peralkalinity is small would also greatly help. For specific
modeling involving a melt that is very different from studied melts (such as lunar basalt – Saal
et al. 2008; Zhang 2009), it is essential to obtain at least a few data points to verify that the
compositional difference does not lead to very different H2O diffusivities.
Some clarifying points. Below are some additional points of clarification:
(1) Although we often say that DH2Ot is proportional to H2Ot at low H2Ot (≤ a couple of
percent), meaning that DH2Ot approaches zero as H2Ot approaches zero, it is just an
approximation. In theory, DH2Ot approaches DOH as H2Ot approaches zero (d[H2Om]/
d[H2Ot] would be so small so that OH diffusion will dominate). That is, if species
equilibrium is reached, DH2Ot is linear with H2Ot with a very small intercept DOH, but
DOH is so small that it has not been resolved experimentally in natural melts discussed
above. However, in a soda lime silicate melt (74 mol% SiO2, 10 mol% CaO, and 16
mol% Na2O) and a float melt (72.5 mo% SiO2, 0.4 mol% Al2O3, 3.3 mol% MgO,
9.8 mol% CaO, and 13.7 mol% Na2O), DH2Ot is independent of H2Ot concentration
at 0.02 to 0.25 wt% H2Ot (Behrens 2006), possibly indicating that the DOH limit is
reached, as well as fairly large DOH values for these sodic melts.
(2) The proportionality relation at low H2Ot and Equation (13) holds only when there
is rough equilibrium between H2Om and OH for Reaction (1). Because the reaction
rate constant increases with increasing temperature and H2Ot, equilibrium is reached
at relatively high temperature but not low temperature. For a given temperature, the
proportionality applies at high H2Ot but not very low H2Ot; what can be considered
as very low H2Ot depends on the temperature. For example, Jambon et al. (1992)
showed that for rhyolite melt, at 0.114 wt% H2Ot and 783 K, Equation (13) and the
proportionality relation do not hold. Therefore, regarding point (1), in many cases the
proportionality relation may already be inapplicable before H2Ot becomes so low that
DOH becomes important.
(3) H2Om diffusivity is a modeling parameter that depends on the accuracy of the
equilibrium constant K, which in turn depends on the accuracy of the infrared
calibration to measure species concentrations. On the other hand, H2Ot diffusivity
is constrained by the length and shape of the measured H2Ot concentration profiles.
Hence, if K values in a melt are improved in the future, one cannot simply use the
new K values in the above equations in conjunction with DH2Om above to calculate
DH2Ot because the H2Om diffusivities themselves are based on the specific K values.
That is, internal consistency must be maintained for accurate calculation of DH2Ot.
The expressions of K values listed above should still be used to calculate DH2Ot unless
a new DH2Om expression is obtained by using the new K values.
(4) If different experimental methods are applied to characterize DH2Ot as a function of
H2Ot content (i.e., using concentration profiles), the diffusivity values at given T-P-
H2Ot and melt composition conditions should be independent of the method (such
as diffusion couple, dehydration, hydration, etc.) However, some methods can only
obtain average diffusivities (e.g., mass loss or gain methods), and different averages
may have different meanings, leading to different diffusivities. For example, under
the same conditions, the average DH2Ot value during dehydration experiments (during
which H2O diffuses out) and that during hydration experiments (during which H2O
diffuses in) are different (Moulson and Roberts 1961; Zhang et al. 1991a; Wang et al.
1996; Zhang 1999b; Figure 3-33 in Zhang 2008). If the surface H2Ot during hydration
experiments is the same as the initial H2Ot during dehydration experiments, and
the minimum H2Ot concentration is zero, and if DH2Ot is proportional to H2Ot, then
(Zhang et al. 1991a; Wang et al. 1996):
Din = 0.619DXs (27a)
Dout = 0.347DXi (27b)
Din = 1.78Dout (27c)
where DXi is DH2Ot at the initial H2Ot content of the sample during dehydration, and
DXs is DH2Ot at the surface H2Ot content of the sample during hydration. Ni and Zhang
(2008) showed how to derive the general relation between Dout and DXi.
There have also been many studies on obsidian hydration near room temperatures
(e.g., Friedman and Long, 1976; Cole an Chakraborty 2001; Riciputi et al. 2002;
Anovitz et al. 2004, 2006, 2008). These studies are not discussed here because (i)
they do not seem to fit neatly into the diffusion data trends by higher-temperature
data, probably due to effects of disequilibrium that cannot be treated simply, and (ii)
our focus here is on silicate melts at high temperatures to intermediate temperatures
(e.g., > 600 K), not surface processes at room temperatures.
MOleCUlAR H2 DIffUSION
In terrestrial magmas, hydrogen is typically oxidized and hence in the form of the hydrous
component. Under extremely reducing conditions, the hydrogen component may be present in
the elemental form as molecular H2. Diffusion of molecular H2 is different from that of H2O
and may play a role under very low fO2. In this section, diffusion of molecular H2 is reviewed
for completeness of coverage, and also for discussing the possible role of H2 diffusion relative
to H2O diffusion in natural silicate melts.
H2 has not been detected as a dissolved gas species in terrestrial silicate melt because it is
easily oxidized in natural melts when there is ferric iron present. To understand H2 diffusion,
it is necessary to avoid H2 oxidation so that pure molecular H2 (that is, the ratio of H2/H2O is
>> 1) diffusion can be characterized. Therefore, H2 diffusion experiments must be conducted
under very reducing conditions. Under such reducing conditions, iron in natural melts would be
significantly reduced to metallic Fe, which would complicate the experiments. Hence, extensive
and reliable H2 as well as D2 diffusion data are available only for pure silica glass. There are
more data on D2 diffusion (Lee et al. 1962; Lee 1963; Perkins and Begeal 1971; Shelby 1977)
than on H2 diffusion (Barrer 1941; Lee et al. 1962; Lee 1963; Shang et al. 2009).
Molecular H2 diffusion data on other melts and glasses are limited and their reliability is
uncertain. Chekhmir et al. (1985) tried to infer molecular H2 diffusivity in albite melt based on
reduction-induced color change (by reducing CO2 to graphite). Gaillard et al. (2002, 2003a,b)
investigated the kinetics of reduction of Fe3+ to Fe2+ by molecular H2 in rhyolite melts, and
tried to infer molecular H2 diffusivity (Gillard et al. 2003b). Because the mechanisms of these
complicated process (including solution, reactions and diffusion) are not well quantified (e.g.,
what the redox reaction kinetics are, whether the redox reaction plays a role in limiting the
entire process, how the process depends on H2 fugacity), it is not clear whether the inferred
molecular H2 diffusivity data are correct. The activation energy for “molecular H2 diffusion”
inferred by Chekhmir et al. (1985) is 113 kJ/mol (clearly too high, more likely reflect molecular
H2O diffusion), and by Gillard et al. (2003b) is about 40 kJ/mol (more reasonable). Even though
192 Zhang & Ni
it is not clear whether the estimated H2 diffusivities by Gaillard et al. (2003b) are correct, the
method used by Gaillard et al. (2003a,b) has great potential because they were able to measure
both molecular H2 and OH bands by infrared spectroscopy (the H2 peak at 415 mm−1 is weak in
IR spectra but detectable), and it may hence be possible to treat the problem quantitatively in a
similar fashion as in Zhang et al. (1991a) to understand both the redox reaction and diffusion.
In the discussion below, we focus on H2 and D2 diffusion in silica glass (in the glass science
literature, silica glass is sometimes referred to fused quartz, or even simply “quartz”), rather
than the unverified semi-quantitative data by Chekhmir et al. (1985) and Gillard et al. (2003b).
Most experimental data on H2 and D2 diffusion in silica glass are based on permeation
experiments, in which gas permeates from one side of a glass (in contact with hydrogen gas
of, e.g., 90 kPa) to the other side (a vacuum line connected to a mass spectrometer). It takes
some time for the gas to diffuse through the glass and reach a steady state. Hence, there is a pre-
steady-state stage, during which the gas flux builds up; there is also a post-steady-state stage,
during which the gas flux gradually decays to zero. Permeability (SI units: mol·m−1·s−1·Pa−1) is
directly obtained from the experimental flux and pressure gradient (pressure difference in two
chambers divided by the thickness of the glass wafer) at steady-state permeation. Diffusivity
may be calculated either from the measured permeability and solubility, or from the non-steady
state decay after the steady-state permeation terminates (or build-up at the beginning of the
permeation experiments). Because absolute concentrations are difficult to measure accurately,
the error in the solubility often dominates the error in the diffusivity if the solubility is needed
to estimate the diffusivity. In addition, pre-existing OH in the silica glass may contribute to the
permeation flux, and dissolved hydrogen molecules (including both H2 and D2) may react with
silica to form bound hydrogen such as OH or SiH (Shelby 1994). Hence, experimental data
must be carefully examined so that these effects are avoided. Shelby (1977) showed that once
these effects are accounted for, D2 (and H2) diffusivities can be obtained consistently.
Even though several groups investigated H2 permeation and diffusion in silica in 1910-
1940, as summarized in Barrer (1941), these earlier data were often an order of magnitude off
(Shang et al. 2009) and are not discussed here. Lee et al. (1962) measured the diffusivity of
H2 and D2 and found the data at T > 1073 K are scattered and do not follow the trend defined
by lower-temperature data. This was explained by a possible reaction of H2 or D2 with silica
network at high temperature. They also found that the diffusivity determined at the decay stage
of the permeation experiments is more reliable than that determined from the build-up stage,
likely because the sample is more fully annealed at the decay stage. Lee (1963) investigated the
effect of impurities in the silica glass and obtained new H2 and D2 diffusivity data at 600-1300
K. The effect of impurities including Al2O3 (ranging from 0.1 to 160 ppm) and OH (10 to 3000
ppm) in silica glass turned out to be small as long as care is taken to avoid possible reactions
and to use only the steady state and post-steady-state data. The data in Lee et al. (1962) are
not discussed further because the work was superseded by Lee (1963). Perkins and Begeal
(1971) reported D2 (and noble gas element) diffusivities in silica glass at 298-448 K. Shelby
(1977) determined D2 diffusivities in various silica glasses at 450-1000 K, and the differences
were found to be no more than 0.17 log units (0.38 ln units) (these are highly reproducible
experiments with internal precision better than 10% relative, or 0.04 log units). Shang et al.
(2009) developed a new technique and obtained H2 diffusivities at 296-523 K. All of these
studies were conducted at low pressures (often less than one atmosphere pressure).
Figure 12 compares all the H2 and D2 diffusion data in Lee (1963), Perkins and Begeal
(1971), Shelby (1977), and Shang et al. (2009). Data scatter is small; the maximum minus
minimum at a given temperature is about 0.4 in lnD. At high temperatures, H2 and D2 diffusion
was investigated by the same author (Lee 1963) on the same silica glass, and it was found
that H2 diffusivity is greater than D2 diffusivity, as expected, but by only about 20%, not the
theoretical 41% difference based on the square root of mass relation. The difference of 20% is
Diffusion of H, C, O Components
Youxue Zhang and Ni (Ch 5) Page 12 in Silicate Melts 193
figure 12. Diffusion data of H2 and D2 from literature (Lee 1963; Perkins and Begeal 1971; Shelby 1977;
Shang et al. 2009). Data of Lee (1963) and Shelby (1977) are read from figures in these papers. Nominal
errors Fig.
on the
12. data are smaller
Diffusion data of H2than theand
(in red) symbols. He and
D2 (in black) fromNe diffusion
literature (Leedata
1963;inPerkins
silicaand
glass (Swets
Begeal 1971;etShelby
al. 1961;
Frank et al. 1961; Perkins and Begeal 1971; Behrens 2010), and molecular H2Om diffusion in nominally
1977; Shang et al. 2009). Data of Lee (1963) and Shelby (1977) are read from figures in these papers. Nominal
dry rhyolite melt (Ni and Zhang 2008) are shown for comparison.
errors on the data are smaller than the symbols. He and Ne diffusion data in silica glass (Swets et al. 1961; Frank et
large enough to be distinguished

al. 1961; Perkins in the2010),
and Begeal 1971; Behrens sameandglass by the
molecular same
H 2O m author
diffusion whendry
in nominally self-consistency
rhyolite melt (Ni is
high, but roughly within error when
and Zhang 2008) are shown for comparison.
results from different silica glasses and different authors
are compared. At lower temperatures (<500 K), H2 diffusivity is apparently smaller than D2
diffusivity, and the difference can reach 20% relative (except for one point at room temperature
that deviates more). This difference is probably due to differences between various silica glasses
(such as different OH content), and accuracy of the solubility models used. Because there are
more data on D2 diffusion and they are all obtained by similar permeation methods with more
careful assessment of glass composition, the uncertainties on D2 diffusivities are smaller.
Fitting H2 and D2 diffusion data (Shang et al. 2009; Shelby 1977) by the Arrhenian relation
leads to:
DHsilica melt
=

exp  − (16.471 ± 0.070 ) –
( 5363 ± 33.4 )  (28)

2
 T 
DDsilica melt
=

exp  − (17.228 ± 0.042 ) –
( 4996 ± 21.8 )  (29)

2
 T 
where D is in m2/s, T is in K, and errors are given at the 2σ level. The activation energy for H2
or D2 diffusion is low, of the order 41-45 kJ/mol.
However, the relation between D2 diffusivity and temperature is slightly non-Arrhenian
(e.g., Shelby 1977), which can be captured well by the following expression:
 ( 4480.5 ± 16.1) 
DDsilica melt
=T ·exp 
 − ( 24.546 ± 0.031) –  (30)
2
 T 
194 Zhang & Ni
where D is in m2/s, and T is temperature in K. The above equation reproduces all experimental
D2 diffusion data in silica melt and glass within 24% (2σ error in lnD is 0.20, which is mainly
due to differences between different silica glasses). The implied Arrhenian activation energy
from the above equation depends somewhat on temperature as R(4480.5+T) based on the above
equation where R is the universal gas constant and T is temperature in K. That is, the activation
energy varies from 39.7 to 45.6 kJ/mol from 300 to 1000 K.
When compared with diffusivities of neutral noble gas molecules in silica melt (Swets al.
1961; Frank et al. 1961; Perkins and Begeal 1971), H2 and D2 diffusivities are between that of
He and Ne (Fig. 12). Furthermore, the activation energy for H2 and D2 diffusion is similar to that
for Ne diffusion and is greater than that for He diffusion.
Below, we evaluate the possible role of molecular H2 diffusion to the diffusion of the
hydrogen component (including H2Om and OH). In the presence of H2, one-dimensional
diffusion of total hydrogen component can be expressed as (following Zhang et al. 1991b):
∂Ctotal H ∂  2∂CH2 2∂CH2 Om 

=  DH2 + DH2 Om  (31)
∂t 
∂x  ∂x ∂x 

where C is concentration in mol/m3, Ctotal H is total H concentration (moles of H atoms per m3),
DH2 and DH2Om are molecular H2 and molecular H2O diffusion coefficients, and OH diffusion is
ignored as discussed earlier. The factor of 2 in the above equation is due to the fact that each H2
molecule or H2Om molecule contains two H atoms and the hydrogen atoms are used to represent
the hydrogen component. Comparing the above equation with
∂Ctotal H ∂  ∂C 
=  Dtotal H total H  (32)
∂t ∂x  ∂x 
where Dtotal H is the total H diffusivity, we obtain:
∂Ctotal H 2∂CH2 2∂CH2 Om
Dtotal=
H DH2 + DH2 Om (33)
∂x ∂x ∂x
Assuming local chemical equilibrium between various hydrogen species (including H2, H2Om,
and OH), then,
∂CH2 ∂CH2 Om
=
Dtotal H 2 DH2 + 2 DH2 Om (34)
∂Ctotal H ∂Ctotal H
Hence, the ratio of the molecular H2 diffusion contribution to the molecular H2Om contribution
is (again assuming local chemical equilibrium among species):
DH2 ∂CH2 DH2 CH2
=Ratio = (35)
DH2 Om ∂CH2 Om DH2 Om CH2 Om
The last equal sign in the above equation holds if H2 concentration is proportional to that of
H2Om (e.g., uniform oxygen fugacity). If the ratio equals 1, the two species contribute equally.
If the ratio is smaller than 1, molecular H2O diffusion contributes dominantly. If the ratio is
greater than 1, molecular H2 diffusion contributes dominantly. The contribution of H2 to total H
diffusion depends on the concentration ratio of H2/H2Om and the diffusivity ratio.
H2/H2Om in a given melt depends on the H2/H2O ratio in the gas or fluid phase (which
depends on oxygen fugacity fO2) and the solubility of H2 and H2O. First, we consider the
oxidation reaction in the gas phase:
H2O(g)  H2(g) + ½ O2(g) (36)
The equilibrium constant (KH2-H2O) can be written as:

fH2 fO1/2 2
K H2 -H2 O = (37)
fH 2 O
where f stands for fugacity. Using thermodynamic data in Robie and Hemingway (1995), the
equilibrium constant depends on temperature as follows:
 29435 
=
K H2 -H2 O exp  6.347 −  bar
1/ 2
(38)
 T 
where T is in K. At 298.15 K, KH2-H2O = 7.6×10−41 bar1/2; at 1800 K, KH2-H2O = 4.5×10−5 bar1/2.
Hence, at a given fO2, fH2/fH2O ratio increases with temperature. Oxygen fugacity in natural
magmas is often within 2 log units of Ni-NiO (NNO) Youxuebuffer (Carmichael
Zhang and Ni (Ch 5) and Ghiorso 1990).
Page 13
Figure 13a shows fH2/fH2O ratio as a function of temperature along the NNO buffer (plus or
minus 2 log units) using the thermodynamic data in Robie and Hemingway (1995) (the NIST-
JANAF Thermodaynamic Data Table does not contain data on NiO). It can be seen that fH2/fH2O
is often of the order 10−4 to 0.1 in the gas phase in equilibrium with natural melts. Note that the
fH2/fH2O ratio is different from the H2/H2Om concentration ratio in the melt, which also depends
on the solubility of H2 and H2O.
To relate the fugacity ratio to the molar concentration ratio of dissolved H2 to H2Om
concentrations in a melt, it is necessary to know the solubility of both H2 and H2Om (not H2Ot).
Note that resorting to solubility here does not necessarily mean the investigated melt is saturated
with respect to molecular or H2 or H2Om. No reliable data are available for both H2 and H2Om
solubility and diffusivity in any given melt. There are H2 solubility and diffusivity data on
figure 13. (a) Fugacity ratio of H2/H2O as

a function of temperature at oxygen buffers
of NNO, NNO-2 and NNO+2. (b) Dissolved
molecular H2 and H2Om concentration ratio
as a function of temperature at oxygen
buffers of NNO, NNO-2 and NNO+2.
(c) The ratio defined in Equation (35),
characterizing the relative contribution of
H2 diffusion relative and H2Om diffusion
to total hydrogen component diffusion as a
function of temperature at oxygen buffers
of NNO, NNO-2 and NNO+2.
Fig. 13. (a) Fugacity ratio of H2/H2O as a function of temperature at oxygen
(b) Dissolved molecular H2 and H2Om concentration ratio as a function of te

196 Zhang & Ni
silica melt and glass (Shackelford et al. 1972; Shelby 1977; Shang et al. 2009) but no H2Om
solubility and diffusivity data (though there are extensive H2Ot solubility and diffusivity data,
e.g., Moulson and Roberts 1961; Roberts and Roberts 1964; Doremus 1969, 1973; Tomozawa
1985). There are H2Om solubility and diffusivity data on rhyolite melt (Zhang et al. 1991a;
Zhang 1999b; Ni and Zhang 2008). For a rough estimation, we assume the H2 solubility and
diffusivity in silica melt are similar to those in rhyolite melt. The molecular solubilities of H2
(Shackelford et al. 1972) and H2Om (Zhang 1999b) are as follows:
 1068 
SHsilica melt
=exp  −13.662 + (39)
2
 T 
 3890 
S rhyolite melt
=exp  −14.193 + (40)
T 
H 2 Om

where the solubilities (SH2 and SH2Om) are in mol·m−3·Pa−1. Hence, at a given fO2 buffer, the molar
concentration ratio of [H2]/[H2Om] can be estimated from (fH2/fH2O)·(SH2/SH2Om), as shown in
Figure 13b (assuming H2 solubility in rhyolite is the same as that in silica melt). It can be seen
that under typical terrestrial conditions, [H2]/[H2Om] is low. Because the ratio of [H2]/[H2Om]
is low in terrestrial melts, the total hydrogen component is largely in the form of H2Om and
OH. It is hence appropriate to refer to the total hydrogen component as the hydrous component
(but molecular hydrogen can still play a role in the diffusion of the hydrous component, see
below). However, in lunar melts that are much more reducing, the [H2]/[H2Om] ratio may be
greater than 1, and the total hydrogen component is not simply the hydrous component.
Next, again assuming the H2 diffusivity in rhyolite melt is the same as that in silica melt,
using the molecular H2 diffusivity of Shang et al. (2009) (Eqn. 28) and the molecular H2Om
diffusivity of Ni and Zhang (2008) (Eqn. 14), the ratio (DH2[H2])/(DH2Om[H2Om]) as a function
of temperature and oxygen fugacity buffer is plotted in Figure 13c. It can be seen that when
oxygen fugacity is at NNO or higher, diffusion of the total hydrogen component is dominantly
due to H2Om diffusion except for temperatures below 600 K. Under more reducing conditions
(that can be encountered in terrestrial melts), at typical magmatic temperatures (≥ 1100 K),
H2Om diffusion still dominates the total hydrogen component flux, but molecular H2 diffusion
may dominate the diffusive flux of the total hydrogen component at temperatures below typical
magmatic temperatures. It is emphasized that this conclusion is tentative because of the many
assumptions involved. Previous experimental diffusion studies (e.g., Shaw 1974; Delaney and
Karsten 1981; Karsten et al. 1982; Zhang et al. 1991a; Zhang and Stolper 1991; Nowak and
Behrens 1997; Zhang and Behrens 2000; Behrens et al. 2004; Liu et al. 2004b; Okumura and
Nakashima 2004, 2006; Ni and Zhang 2008; Behrens and Zhang 2009; Ni et al. 2009a,b;
Wang et al. 2009) on the H2O component were conducted without detailed characterization
of the oxygen fugacity. Nonetheless, the consistency of these experimental data (e.g., Ni and
Zhang 2008) show that H2 diffusion probably did not play a major role in any of these studies.
Furthermore, simultaneous H2O and 18O diffusion studies in the same experiments (Behrens
et al. 2007) show that H2Om that carries O, not H2 that does not carry O, is the diffusing
species for the hydrous component. In the future, it will be important to investigate diffusion of
the total hydrous component under very reducing conditions with well-characterized oxygen
fugacity (e.g., oxygen fugacity that is uniform in the sample and independent of time).
In short, it is possible that molecular H2 diffusion could play a major role in transporting
the hydrous component under reducing conditions that may be encountered in terrestrial melts.
However, this inference is uncertain at present because of the many assumptions involved.
New experimental studies are necessary to resolve this issue.
DIffUSION Of THe CO2 COMpONeNT

Zhang et al. (2007) reviewed CO2 diffusion in silicate melts. Not much new work has
been published on natural melts since then. For completeness of this chapter and this review
volume, a brief summary is provided here. Dissolved CO2 is present in silicate melts in either
CO2 molecules or carbonate groups (Fine and Stolper 1985; Blank and Brooker 1994). Hence,
CO2 diffusion is also a multi-species diffusion. However, the presence of multiple species
seems to contrive to simplify the diffusion properties of CO2, contrary to the case of H2O
diffusion (Zhang et al. 2007).
Watson et al. (1982) were the first to investigate tracer diffusion of carbonate (using 14C
tracer diffusion) in silicate melts, and made the surprise discovery that CO2 diffusivity does
not depend on the anhydrous melt composition (from a haplobasalt to melt containing 30
wt% Na2O). More extensive 14C tracer diffusion studies by Watson (1991) confirmed this
conclusion but also showed that the diffusivity depends strongly on H2Ot. Watson et al. (1982)
and Watson (1991) determined concentration profiles using β-track maps made by exposing
nuclear emulsion plates to the quenched and sectioned diffusion capsules. As discussed in
Mungall (2002), the β-particle range is of the order of a hundred µm (International Commission
on Radiation Units and Measurements 1984), much more than initially thought. Hence, tracer
diffusion data using β-track maps may be compromised by this effect (those diffusivities
extracted from concentration profiles longer than 1 mm are less affected). Furthermore, Zhang
et al. (2007) showed that the data are inconsistent with effective binary diffusivities of CO2.
Hence, 14C tracer diffusion data of Watson et al. (1982) and Watson (1991) are not discussed
further in quantitative treatment.
Blank (1993), Sierralta et al. (2002) and Nowak et al. (2004) (as well as Fogel and
Rutherford 1990, Zhang and Stolper 1991, and Liu 2003) investigated effective binary diffusion
of CO2 at CO2 concentration levels of hundreds to thousands of ppm. For the purpose of
modeling natural magmatic processes (e.g., CO2 bubble growth in a basalt melt), the effective
binary diffusivities of total CO2 are the necessary diffusivities. These studies further confirmed
the rough independence of CO2 diffusivity on anhydrous melt composition. This independence
is even more surprising considering that CO2 is present as CO2 molecules in rhyolite melt
but as CO32− ion in basalt melt. Nowak et al. (2004) explained the dependence as follows:
Assume molecular CO2 is the diffusing species whereas CO32− is roughly immobile. From
rhyolite melt to basalt melts, molecular CO2 diffusivity increases (similar to Ar diffusivity),
which increases total CO2 diffusivity, but the fraction of dissolved CO2 present as molecular
CO2 decreases (the rest is CO32−), which decreases the total CO2 diffusivity. The two factors
roughly cancel each other, leading to a total CO2 diffusivity roughly independent of the
anhydrous melt composition. Baker et al. (2005) carried out three carbonate dissolution
experiments and measured total CO2 concentration profiles by difference-from-100% method,
and two diffusion couple experiments and measured CO2 concentration profiles by FTIR. The
compositions for which CO2 diffusion has been investigated are listed in Table 2.
Because of its rough independence of the anhydrous melt composition, CO2 is likely to
be the first component for which the diffusivity in various melts can be predicted based on
a limited number of studies, which can be compared to the large number of experimental
studies on H2O diffusion reviewed in the previous section. Existing experimental data on the
effective binary diffusion of CO2 cover 723-1623 K and ≤ 1 GPa. However, they are mostly on
dry melts, with only 4 data points on hydrous melts containing ≥ 1 wt% H2Ot, of which two
points with 5 wt% H2Ot by the reconnaissance experiments of Baker et al. (2005) apparently
showed that adding 5 wt% H2Ot does not affect CO2 diffusivity in a trachyte melt, contrary to
results by Sierralta et al. (2002) on albite melt and Watson (1991) on other melts. A possible
explanation of the discrepancy is that determination of CO2 concentration profiles by the
difference-from-100% method used by Baker et al. (2005) does not work well for hydrous
198 Zhang & Ni
Table 2. Chemical composition (wt% on dry basis) for CO2 diffusion studies.
ID Comp. SiO2 TiO2 Al2O3 feO MnO MgO CaO Na2O K2O p2O5 Ref
1 Rhyolite 76.45 0.08 12.56 1.02 .08 .06 .25 4.21 4.78 a
2 Rhyolite 77.5 0.07 13.0 0.56 0.04 0.05 0.52 4.10 4.18 b
3 Basalt 50.6 1.88 13.9 12.5 0.23 6.56 11.4 2.64 0.17 0.21 c
4 Ab 69.03 19.33 11.41 d
5 AbNa1 68.87 19.24 12.45 d
6 AbNa2 68.56 18.58 13.38 d
7 AbNa3 68.43 18.62 13.92 d
8 AbNa4 66.66 18.63 15.24 d
9 AbNa6 67.38 18.18 15.95 d
10 AbNa7 65.73 17.60 17.57 d
11 Rh (CO2) 75.09 0.22 13.83 1.47 1.40 3.86 4.10 e
12 Da (CO2) 71.78 0.40 15.31 2.26 3.09 4.52 2.97 e
13 DaAn (CO2) 66.13 0.68 16.18 4.59 4.79 4.26 1.90 e
14 An (CO2) 63.37 1.11 16.90 8.08 6.81 4.04 0.88 e
15 AnTh (CO2) 61.92 1.55 15.22 8.79 6.74 3.73 1.33 e
16 Th (CO2) 60.48 2.01 14.54 9.95 8.79 3.32 1.61 e
17 Ha (CO2) 53.55 3.74 16.93 11.32 9.19 4.71 1.44 e
18 Trachyte 59.9 0.39 18.0 0.89 0.12 3.86 2.92 4.05 8.35 0.21 f
The compositions are listed on the anhydrous basis.
References: a. Fogel and Rutherford (1990); b. Blank (1993); c. Zhang and Stolper (1991); d. Sierralta et al. (2002); e.
Nowak et al. (2004); f. Baker et al. (2005).
melts and hence CO2 diffusivities in hydrous trachyte melt are incorrect.
Because total CO2 diffusivity in all melts are similar to Ar diffusivity in silicic melts
(Zhang et al. 2007), and extensive Ar diffusion data in silicic melts are available, covering a
wide range of temperature, pressure and H2O contents (673-1773 K, ≤ 1.5 GPa, and ≤ 5 wt%
H2Ot), Zhang et al. (2007) used such data to derive the following equation, which is proposed
to be applicable for total effective binary diffusivity of CO2 in basalt to rhyolite melts:
all melts 17367 + 1944.8 P (855.2 + 271.2 P )
ln DCO =
−13.99 − + Cw (41)
2,total
T T
where D is in m2/s, T is in K, P is in GPa, and Cw is wt% of H2Ot.
Figure 14 compares experimental data and the above equation, especially the data by Baker
et al. (2005) because the comparison with Baker et al. (2005) was not made in Zhang et al.
(2007). It can be seen that the calculated line and the experimental data of the same color (the
online version) are in rough agreement, except for the two data points at 5 wt% H2Ot by Baker
et al. (2005) (purple short-dashed line and purple solid squares in Fig. 14; the color can be seen
in the online version). Excluding these two data points, the maximum uncertainty of the above
equation in predicting lnDCO2,total (Fogel and Rutherford 1990; Zhang and Stolper 1991; Blank
1993; Sierralta et al. 2002; Nowak et al. 2004; Baker et al. 2005) is 1.13. This 2σ uncertainty
is larger than typical experimental data uncertainty (with a 2σ of about 0.6 in lnD), because
DCO2,total depends weakly on melt compositions, which is ignored in the above treatment. New
experimental data on CO2 diffusion under conditions of high H2Ot content are necessary to
further constrain CO2 diffusivities.
figure 14. Comparison of experimental data of effective binary diffusivity of bulk CO2 with Equation (41)
under some conditions. The lines in matching color with data points (the colors can be seen in the online
version) are14.calculated
Fig. Comparison from Equation data
of experimental (41);ofsolid linebinary
effective corresponding
diffusivity ofto solid
bulk CO2circles;
with Eq.long dashes
41 under someto open
circles; medium dashes to squares with plus inside; short dashes (the uppermost line) to solid squares;
sparse conditions.
dot line toThesolid diamonds;
lines in matchingand dense
color dot points
with data line toareopen diamonds.
calculated For clarity,
from Eq.41. low-temperature
For clarity, low-temperature data
are not shown. Data sources: “0.1-0.2 wt% H2O; 0.05-0.25 GPa”: Blank (1993) and Fogel and Rutherford
data are not shown. Data sources: “0.1-0.2 wt% H2O; 0.05-0.25 GPa”: Blank (1993) and Fogel and Rutherford
(1990); “< 0.14 wt% H2O; 0.5 GPa”: Sierralta et al. (2002) and Nowak et al. (2004); “2 wt% H2O and 0.5
GPa”: Sierralta et al. (2002); “5 wt% H O; 1 GPa”: Baker et al. (2005); “≤ 0.2 wt% H O; 1 GPa”: Zhang
(1990); “<0.14 wt% H2O; 0.5 GPa”: Sierralta et al. (2002) and Nowak et al. (2004); “2 wt% H22O and 0.5 GPa”:
2
and Stolper (1991) and Baker et al. (2005); “1 wt% H2O; 1.2 GPa”: Baker et al. (2005).
Sierralta et al. (2002); “5 wt% H2O; 1 GPa”: Baker et al. (2005); “≤0.2 wt% H2O; 1 GPa”: Zhang and Stolper (1991)
Comparison
and Baker et al.of CO2“1 and
(2005); wt% HH 2O
2O; 1.2 diffusivities.
GPa”: Baker et al. Comparison
(2005). between DCO2,total, DH2Om, and
DH2Ot using data and equations summarized above reveal that DH2Om is greater than DCO2,total,
and DH2Ot is usually greater than DCO2,total (Fig. 15). This is not unexpected since the effective
radius of H2O molecules (1.37 Å, Shannon 1976; Zhang and Xu 1995) is smaller than the
effective radius of CO2 molecules (1.7 Å, Behrens and Zhang 2001; Zhang et al. 2007) that
dominates CO2 diffusion. However, because DCO2,total approaches a non-zero constant but DH2Ot
approaches essentially zero as H2Ot approaches zero, DCO2,total can become larger than DH2Ot at
magmatic temperatures at H2Ot < 0.2 wt% with this content depending temperature, pressure
and anhydrous melt composition. In natural silicate melts, H2Ot is usually greater than 0.2 wt%
and hence DH2Ot is greater than DCO2,total.
The diffusion of the two most important volatile components, H2O and CO2, shows
similarities and distinctions (Zhang et al. 2007). In both cases, speciation plays a critical
role, the neutral species (molecular H2O or CO2) is the dominant diffusing species, and the
diffusivity of the neutral species increases with H2Ot. However, there are also major differences.
Most importantly, H2O speciation results in DH2Ot that depends strongly on H2Ot (as well as
melt composition); while CO2 speciation results in CO2 diffusivity that is independent of CO2
concentration, and more surprisingly, even independent of the anhydrous melt composition,
which is a blessing for quantifying CO2 diffusion.
OxYgeN DIffUSION
Oxygen is the major constituent and a framework element in silicate melts. Therefore,
oxygen diffusivities are essential for characterizing reactions and transport in silicate melts.
200 Zhang & Ni
figure 15. Comparison of H2Om, H2Ot diffusivity and total CO2 diffusivity at 1400 K and 0.5 GPa. CO2,total
diffusivity is from Equation (41), H2OOmdiffusivity
Fig. 15. Comparison of H2Om, H
diffusivity in rhyolitic melt is from Equation (14), and H2Ot
and total CO2 diffusivity at 1400 K and 0.5 GPa. CO2,total
2 t
diffusivity in rhyolitic melt is calculated from Equations (13), (14), and (7a).
diffusivity is from Eq. 41, H2Om diffusivity in rhyolitic melt is from Eq. 14, and H2Ot diffusivity in rhyolitic melt is
calculated from Eqs. 13, 14, and 7a.

Because oxygen can be present in many different components and species, including molecular
H2O, O2, CO2, ionic CO32− and OH, as well as oxygen bonded to Si, Ti, Al, Fe, Mg, Ca, Na, K,
and P (often simplified as bridging oxygen, non-bridging oxygen and free O2−), oxygen diffusion
kinetics can be complicated but may also yield structural information for silicate melts. Hence,
oxygen diffusion has been investigated extensively in the geological, materials science, and glass
science literature. However, the various reports can be confusing. Some authors investigated
chemical (or effective binary) diffusion of various oxygen species and components, and others
investigated self-diffusion of oxygen, which is sometimes due to chemical diffusion of H2O. In
earlier studies, bulk exchange and analytical methods often using mass spectrometry are applied;
and in more recent studies, profiling methods typically using the ion microprobe are applied. As
a general rule, the studies employing profiling techniques provide more reliable data (e.g., Zhang
2008). In addition to studies by geologists, there is also a significant body of glass and materials
science literature on oxygen diffusion. Here, we focus on the geological literature.
Geochemists have investigated oxygen diffusion in melts with natural rock compositions as
well as mineral or mineral mixture compositions. Dacite, Di58An42, Na2Si4O9, jadeite and basalt
melts have been investigated more systematically. Limited data are available for diopside, CaO-
Al2O3-SiO2, nephelinite, andesite, and rhyolite melts. Because diffusion under wet conditions
and that under dry conditions have different mechanisms, they are discussed separately. 18O
self-diffusion and oxygen chemical diffusion are also discussed separately.
Self-diffusion of oxygen in silicate melts under dry conditions
There is a large literature on oxygen self-diffusion in dry silicate glasses by glass scientists,
which is not covered here. Muehlenbachs and Kushiro (1974) were the first in the geological
literature to explore self-diffusion of 18O (or 18O-16O exchange diffusion) in various melts at
high temperature and room pressure (about 0.1 MPa). Even though the diffusion data turned out
to have large errors (see discussion below), the method is introduced below because many other
authors used the same or a similar experimental technique. A melt blob suspended from a Pt
loop is first equilibrated with a gas (either CO2 or O2) of some normal δ18O in a one-atmosphere
furnace. This equilibrated melt is the initial state with known initial 18O/16O ratio. Then a gas
with a new and constant δ18O is continuously led into the furnace so that 18O and 16O in the gas
exchange with those in the melt blob through diffusion. That is, the experiments investigate
the interdiffusivities between 18O and 16O, which are also oxygen self-diffusivities. After some
duration, the sample is quenched and average δ18O in the whole melt blob is measured by mass
spectrometry. The result is compared with the analytical solution for diffusion in a sphere:
Mt 6 ∞ 1 n2 π2 Dt
1 − 2 ∑ 2 exp(−
= ) (42)
M∞ π n =1 n r2
where Mt and M∞ are the amount of 18O entering the sphere of radius r at time t and time ∞. The
specific equation is often written in the following form:
δ 18 Ot − δ 18 Ogas 6 ∞ 1  n2 π2 Dt 
= ∑
δ 18 O0 − δ 18 Ogas π2 n =1 n2
exp −
r2 
 (43)

where the subscripts 0 and t mean experimental time 0 (i.e., the initial state of equilibration)
and time t, and the subscript “gas” means the gas phase. The final δ18O in the melt is assumed
to be the same as that of the gas by treating gas as an infinite reservoir and ignoring isotopic
fractionation between the gas and the melt at high experimental temperatures. The above
equation is used to fit experimental data of δ18Ot versus t, from which the diffusivity is extracted.
Muehlenbachs and Kushiro (1974) reported diffusion data in basalt melt (Fig. 16) and
the mineral anorthite at room pressure and 1553-1803 K, as well as andesite, rhyolite, and
Di40An30Ab30 melts (and plagioclase, anorthite, diopside, enstatite, and forsterite minerals) at
1553 K. The diffusivity does not depend on whether the gas phase is O2 or CO2. For basalt melt,
the activation energy was found to be 377 kJ/mol, which is very high for a melt. Comparison
of different melts at 1553 K indicates that 18O diffusivity in basalt melt is highest, followed by
andesite melt and Di40An30Ab30 melt (about an order of magnitude lower than in basalt), then
rhyolite melt (about three orders of magnitude lower than in basalt) (Fig. 16). The mineral data
are not of concern here; they will be discussed in another chapter (Farver 2010, this volume).
-22
-24
lnD (D in m2/s)
-26
-28
-30 Basalt(M)
Basalt(C)
Basalt(L)
Andesite(M)
-32 Dacite(T)
Rhyolite(M) O self-diffusion
Calculated
-34
0.52 0.54 0.56 0.58 0.6 0.62 0.64
1000/T (T in K)
figure 16. Oxygen self-diffusivity in dry basalt, andesite, dacite and rhyolite melts. Data sources: Basalt(M),
Andesite(M) and Rhyolite(M): room pressure data by Muehlenbachs and Kushiro (1974); Basalt(C): room
pressure dataFig.
by16. Oxygen self diffusivity in dry basalt, andesite, dacite and rhyolite melts. Data sources: Basalt(M),
Canil and Muehlenbachs (1990); Basalt(L): 1 GPa data by Lesher et al. (1996); Dacite(T):
1 GPa data by Tinker and
Andesite(M) andRhyolite(M):
Lesher (2001). Calculated
room pressure line is forand
data by Muehlenbachs dacite at(1974);
Kushiro roomBasalt(C):
pressureroomusing Equation
pressure
(44) to compare with the data by Muehlenbachs and Kushiro (1974) and Canil and Muehlenbachs (1990).
data by Canil and Muehlenbachs (1990); Basalt(L): 1 GPa data by Lesher et al. (1996); Dacite(T): 1 GPa data by
Tinker and Lesher (2001). Calculated line is for dacite at room pressure using Eq. 44 to compare with the data by
Muehlenbachs and Kushiro (1974) and Canil and Muehlenbachs (1990).

202 Zhang & Ni
Dunn (1982) used the experimental and analytical approach of Muehlenbachs and Kushiro
(1974) and investigated 18O diffusion in Di58An42, Di40An60, and diopside melts at room pressure.
Canil and Muehlenbachs (1990) followed with an Fe-rich basalt melt. The activation energy for
18
O diffusion in dry basalt was estimated to be about 251 kJ/mol by Canil and Muehlenbachs
(1990), much lower than that of Muehlenbachs and Kushiro (1974), but the two data sets are
in good agreement (Fig. 16; compare Basalt(M) and Basalt(C)). This comparison shows that
activation energy based on limited data is not reliable especially when there is large data scatter.
Compared to more recent diffusion data based on the profiling method, these early data based
on bulk gain or loss methods show large scatter (2σ error is a factor of 3 in D, or about 1.1 in
lnD, Fig. 16) and often do not provide much quantitative constraint in our discussion below on
how oxygen diffusivities depend on various parameters.
Major advancements in 18O self-diffusion studies came with the development of the ion
microprobe for microscopic measurement of isotopic ratios, although with the requirements of
using materials highly enriched in 18O. Coles and Long (1974) and Hofmann et al. (1974) were
the first to apply the ion microprobe to study self-diffusion in minerals. Shimizu and Kushiro
(1984) were the first to determine 18O self-diffusivities in silicate melts by carrying out diffusion
couple experiments and by measuring 18O/16O ratios using an ion microprobe. They examined
oxygen self-diffusion (18O-16O exchange diffusion) in jadeite and diopside melts. Rubie et al.
(1993), Lesher et al. (1996), Liang et al. (1996), Poe et al. (1997), Reid et al. (2001), Tinker and
Lesher (2001), and Tinker et al. (2003) followed the approach of Shimizu and Kushiro (1984).
Rubie et al. (1993) and Poe et al. (1997) measured oxygen self-diffusivities in dry Na2Si4O9
melt at 1898-2800 K and 2.5-15 GPa, as well as in Na3AlSi7O17 and albite melts. Lesher et al.
(1996) determined oxygen (and silicon) self-diffusion in dry basalt melt at 1593-1873 K and
1-2 GPa. Liang et al. (1996) examined oxygen (as well as Ca, Al and Si) self-diffusivities in
various CaO-Al2O3-SiO2 melts at 1773 K and 1 GPa. Reid et al. (2001) studied oxygen (and
silicon) self-diffusion in dry diopside melt at 2073-2573 K and 3-15 GPa. Tinker and Lesher
(2001) investigated oxygen (and silicon) self-diffusion in dry dacite melt at 1628-1935 K and
1-5.7 GPa. Tinker et al. (2003) reported oxygen (and silicon) diffusivities in dry Di58An42 melt
at 1783-2037 K and 1-4 GPa. These new data generally have higher precision, with the possible
exception of Reid et al. (2001) (see below). Studies on wet melts are not summarized here and
will be discussed in a later section.
The melt that has been most systematically investigated for oxygen self-diffusion is dry
dacite even though the number of data points is not extensive (Tinker and Lesher 2001). Figure
17 shows all available experimental data. The data indicate that oxygen self-diffusivity exhibits
Arrhenian behavior within the T-P range investigated, and increases with increasing pressure
from 1 to 4 GPa and then decreases at pressures above 5 GPa. The decrease of the diffusivity
with pressure at ≥ 5 GPa may have structural implications, such as possible end of the formation
of highly coordinated Si or Al species (Tinker and Lesher 2001). For the purpose of predicting
oxygen self-diffusion in dacite melt, there are not enough data at ≥ 5 GPa to constrain the
relation. Hence, the data at 1628 to 1935 K and 1 to 4 GPa are fit to obtain the following
expression for the self-diffusivity of oxygen in dry dacite melt:
40830 − 5223P
− ( 6.57 + 2.148P ) −
dacite melt
ln D18 O = (44)
T
where T is in K and P is in GPa. The activation energy decreases with pressure as (339 − 43P)
kJ/mol. The activation volume is −RT∂(lnD)/∂P = −(43.4 − 0.018T)×10−6 m3/mol, a fairly large
negative activation volume, in contrast with the positive activation volumes for H2O and CO2
diffusion. The maximum error by Equation (44) to reproduce the experimental data at 1500-
1950 K and ≤ 4 GPa is 0.32 in terms of lnD. Equation (44) cannot be extrapolated at all to P >
4 GPa because the linearity between lnD and P does not hold at higher pressures.
Oxygen self-diffusion in dry Di58An42 melt has also been examined systematically (Dunn
1982; Tinker et al. 2003). Figure 18 shows all the available data. It can be seen that the data by
Dunn (1982) at room pressure are scattered and do not provide much constraint. The Di58An42
melt is often treated as a basalt-like melt or an FeO-free basalt. However, when oxygen self-
diffusivities in dry basalt (solid diamonds and a single plus in Fig. 18) are compared with
those in dry Di58An42, those in basalt are greater by about 1.2 lnD units. The diffusivity data on
Di58An42 melt by Tinker et al. (2003) at 1738 to 2037 K and 1 to 4 GPa can be expressed as:
-25
Oxygen self-diffusion
-26 in dacite melt
lnD (D in m2/s)
-27
-28
-29
1.0 GPa
2.0 GPa
4.0 GPa
-30 5.2 GPa
5.7 GPa
-31
0.5 0.52 0.54 0.56 0.58 0.6 0.62
1000/T (T in K)
figure
Youxue 17.andOxygen
Zhang Ni (Ch 5)self-diffusion data in 18
Page dacite melt (Tinker and Lesher 2001).
Fig. 17. Oxygen self diffusion data in dacite melt (Tinker and Lesher 2001).
-23
-24
-25
lnD (D in m2/s)
1.0 GPa
-26 2.0 GPa
3.5 GPa
4.0 GPa
-27 0.1 MPa
Basalt (1 GPa)
Basalt (2 GPa)
-28
-29
in Di58An42 melt
-30
0.5 0.55 0.6 0.65
1000/T (T in K)
figure 18. Oxygen self-diffusion data in Di58An42 melt compared with those in basalt. Data for Di58An42
melt are from Dunn (1982) (0.1 MPa; open diamonds) and Tinker et al. (2003) (1.0, 2.0, 3.5 and 4.0 GPa).
Fig. 18. Oxygen self diffusion data in Di58An42 melt compared with those in basalt. Data for Di58An42 melt are
Data for basalt are from Lesher et al. (1996).
from Dunn (1982) (0.1 MPa; open diamonds) and Tinker et al. (2003) (1.0, 2.0, 3.5 and 4.0 GPa). Data for basalt
are from Lesher et al. (1996) (solid diamonds and a single plus in green).
204 Zhang & Ni
Di An 42 melt 26331 − 240 P

ln D18 O58 =
−10.52 − (45)
T
where T is in K and P is in GPa (adding an additional P term such as that in Equation (44) does
not significantly improve the fit). The activation energy is (219 − 2.0P) kJ/mol. The activation
volume is −RT∂(lnD)/∂P = −2.0×10−6 m3/mol, an order of magnitude smaller than that for
oxygen diffusion in dacite melt. The maximum error by the above equation to reproduce the
data of Tinker et al. (2003) is 0.18 in terms of lnD. Extrapolation of the above equation to lower
pressures of 0-1 GPa is likely acceptable, but probably not to higher pressures of > 4 GPa.
Rubie et al. (1993) and Poe et al. (1997) studied oxygen self-diffusion in dry Na2Si4O9 (or
Na2O·4SiO2, NS4) melt. The data (Fig. 19) covering 1893-2800 K and 2.5-15 GPa can be fit by
12693 − 360 P
ln D18NaO2 Si4 O9 melt =
−17.19 − (46)
T
where D is in m2/s, T is in K. and P is in GPa. The activation energy is surprisingly low, only
about 100 kJ/mol (depending on pressure). The activation volume is negative and small, about
−3.0×10−6 m3/mol. The maximum error of the above equation in reproducing the experimental
data is 0.22 in terms of lnD.
Three papers explored oxygen self-diffusion in dry basalt melt (Muehlenbachs and
Kushiro 1974; Canil and Muehlenbachs 1990; Lesher et al. 1996). However, the earlier data
by Muehlenbachs and Kushiro (1974) and Canil and Muehlenbachs (1990) are scattered (Fig.
16). Lesher et al. (1996) reported three data points at 1 GPa and one datum at 2 GPa. Oxygen
self-diffusivities at 1593-1873 K and 1 GPa (solid squares in Fig. 16) can be represented by the
following equation (Lesher et al. 1996):
basalt melt 20447
ln D18 O =
−12.5 − (47)
T
where T is in K and D is in m2/s. The activation energy is 170 kJ/mol, smaller than those obtained
from other 18O diffusion data in basalt melt at room pressure (377 kJ/mol by Muehlenbachs
-20
-20.5 Oxygen self-diffusion

in Na 2Si 4O9 melt
-21
lnD (D in m 2/s)
-21.5
-22
2.5 GPa
4.0 GPa
-22.5 6.0 GPa
8.0 GPa
10 GPa
-23 12.5 GPa
15 GPa
-23.5
0.35 0.4 0.45 0.5 0.55
1000/T (T in K)
figure 19. Oxygen self-diffusivity in Na2Si4O9 melt with data from Rubie et al. (1993)
(with
Fig. 19. Oxygen self1000/T > in0.45)
diffusivity Na2Siand Poe et al. (1997) (with 1000/T < 0.45).
4O9 melt with data from Rubie et al. (1993) (with 1000/T > 0.45) and Poe et
al. (1997) (with 1000/T < 0.45).

and Kushiro 1974, and 251 kJ/mol by Canil and Muehlenbachs 1990). The large difference in
activation energy is likely due to data uncertainty rather than being real. Because the activation
volume is highly variable from dacite to Di58An42 melts (−11×10−6 to −2.0×10−6 m3/mol; see
also Tinker et al. 2003), it is not possible to use information from other melts to constrain the
pressure effect on oxygen diffusivity in basalt.
Three papers investigated oxygen self-diffusion in diopside melt (Dunn 1982; Shimizu and
Kushiro 1984; Reid et al. 2001). However, the temperature dependence was not constrained:
only at two pressures (0.1 MPa by Dunn 1982 and 3 GPa by Reid et al. 2001) were there
diffusion data at different temperatures. The data at 0.1 MPa are scattered and those at 3 GPa
are strange in that the diffusivity does not change much from 2073 to 2273 K, with an implied
activation energy of only 4 kJ/mol (Fig. 20). An error of a factor of 2.5 on individual D values is
needed to allow a more reasonable activation energy of about 250 kJ/mol. Hence, the precision
of the oxygen self-diffusion data by Reid et al. (2001) is not high. The reason is not clear.
Shimizu and Kushiro (1984) reported oxygen self-diffusion data in jadeite melt at 1673 to
1883 K at 1.5 GPa, and 0.5 to 2.0 GPa at 1673 K (Fig. 21). The diffusivity at 1.5 GPa can be
expressed as:
31815
ln D18jadeite melt
=
−10.84 − (48)
O
T
where D is in m2/s, and T is in K. The activation energy is 265 kJ/mol. At 1673 K, the activation
volume is −6.4×10−6 m3/mol. Assuming the activation volume is independent of temperature,
the P-T dependence of 18O diffusivity in jadeite melt may be written as:
32970 − 770 P
ln D18jadeite melt
=
−10.84 − (49)
O
T
where D is in m2/s, T is in K, and P is in GPa.
Liang et al. (1996) investigated the compositional effect of 18O self-diffusion in various
CaO-Al2O3-SiO
Youxue meltsandatNi 1773
2 Zhang (Ch 5) K and 1 GPa.
PageThey
20 found that 18O self-diffusivity increases
-20
-21
-22
lnD (D in m2/s)
-23
0.1MPa 9GPa
-24 1.0GPa 11GPa
1.7GPa 12GPa
3GPa 13GPa
-25 6GPa 14GPa
7GPa 15GPa
-26
O self-diffusion in dry diopside melt
-27
0.38 0.42 0.46 0.5 0.54 0.58
1000/T (T in K)
figure 20. Oxygen self-diffusivities in dry diopside melt. Data sources are: 0.1 MPa from Dunn (1982); 1.0
GPa and 1.7 GPa from Shimizu and Kushiro (1984); and the rest from Reid et al. (2001).
Fig. 20. Oxygen self diffusivities in dry diopside melt. Data sources are: 0.1 MPa from Dunn (1982); 1.0 GPa and
1.7 GPa from Shimizu and Kushiro (1984); and the rest from Reid et al. (2001).
206 Youxue Zhang and Ni (Ch 5)
Zhang & Ni
Page 21
-27.5
-28 in jadeite melt
lnD (D in m2/s) -28.5
-29
-29.5 0.5 GPa

1.0 GPa
-30 1.5 GPa
2.0 GPa
-30.5
0.53 0.54 0.55 0.56 0.57 0.58 0.59 0.6
1000/T (T in K)
figure 21. Oxygen self-diffusivities in jadeite melt (Shimizu and Kushiro 1984).
Fig. 21. Oxygen self diffusivities in jadeite melt (Shimizu and Kushiro 1984).
with decreasing silica and alumina content, as expected. As NBO/T increases from 0.3 to about
0.9 in CaO-Al2O3-SiO2 system, 18O self-diffusivity increases by about an order of magnitude.
Figure 22 compares oxygen self-diffusivity in all melts for which the temperature depen-
dence of diffusivity has been determined well at 1 GPa. Oxygen self-diffusivities increase with
decreasing SiO2 and Al2O3 contents, or from polymerized to depolymerized melts. From dacite
melt (NBO/T ≈ 0.1) to basalt melt (NBO/T ≈ 1), oxygen self-diffusivity increases by two orders
of magnitude. It may be inferred that oxygen self-diffusivities depend on oxygen speciation,
increasing from bridging oxygen (BO) to non-bridging oxygen (NBO) and then to free oxygen
(O2−). The diffusivity of each oxygen species may depend on the overall melt composition.
In terms of oxygen self-diffusion, basalt melt (NBO/T ≈ 1) is similar to diopside melt
(NBO/T = 2) and Na2Si4O9 melt (NS4 melt, NBO/T = 0.5), whereas dacite melt (NBO/T ≈ 0.1)
is similar to jadeite melt (NBO/T = 0) (Fig. 22). But the similarity in each group is not close
enough for diffusivities to be merged for a combined fitting. For different melts, there does not
seem to be a single compensation temperature where 18O diffusivities in all melts are the same.
There are not enough data yet to contemplate a general relation between oxygen self-
diffusivity and natural melt composition under dry conditions. To achieve such a goal, it is
necessary to investigate 18O diffusion in dry basalt, dry andesite, and dry rhyolite as a function
of temperature and pressure systematically (on par with the investigation of dacite melt), and to
examine the compositional dependence. In highly polymerized melts (rhyolite and pure silica),
it will be important to make the dry system very dry, e.g., less than 50 ppm H2Ot (depending
on temperature), so that the diffusive flux is due to true 18O self-diffusion, not H2O chemical
diffusion (see later sections). One may try to use self-diffusivities of other elements such as Si to
constrain those of oxygen. However, even though self-diffusivities of oxygen and silicon (both
are structural elements) in dry melts are often similar (e.g., Lesher et al. 1996; Poe et al. 1997;
Tinker et al. 2003), they may also be significantly different (e.g., Tinker and Lesher 2001; see
review by Lesher 2010 and Zhang et al. 2010).
It has been shown that oxygen self-diffusivity in dry melts is similar to the Eyring diffusivity
defined as D = kT/(λη) where λ is the jump distance and η is the melt viscosity (e.g., Shimizu
figure 22. Oxygen self-diffusion data in different melts at 1 GPa. Data sources are: Basalt: Lesher et al.
(1996); Diopside and Jadeite: Shimizu and Kushiro (1984); Di58An42: Tinker et al. (2003); and Dacite:
Fig. 22. Oxygen self diffusion data in different melts at 1 GPa. Data sources are: Basalt: Lesher et al. (1996);
Tinker and Lesher (2001). The long dashed line is for Na2Si4O9 extrapolated using Equation (46).
Diopside and Jadeite: Shimizu and Kushiro (1984); Di58An42: Tinker et al. (2003); and Dacite: Tinker and Lesher
and Kushiro (2001).

1984;TheTinker et al. 1994; Fig. 27a in this chapter). Therefore, even though 18O
long dashed line (in green) is for Na2Si4O9 extrapolated using Eq. 46.
self-diffusion in dry basalt, andesite and rhyolite melts has not been investigated extensively,
18
O self-diffusivity in these dry melts may be estimated using the Eyring equation with a jump
distance of 2.8×10−10 m as long as the melt viscosity can be estimated. In the last 10 years,
viscosity models for specific melts have been advanced to high precision (e.g., Zhang et al.
2003 and Hui et al. 2009 for rhyolite melt; Whittington et al. 2009 for dacite melt; Vetere et al.
2006 for andesite melt) and general viscosity models are also available (Hui and Zhang 2007;
Giordano et al. 2008) although with larger uncertainties (see Wang et al. 2009 for a comparison
of the models of Hui and Zhang 2007 and Giordano et al. 2008). Hence, 18O self-diffusivities in
dry melts can now be estimated with a precision similar to the precision of the viscosity models.
For hydrous melts, 18O “self” diffusivity is much greater than the Eyring diffusivity, and must
be understood and predicted in the context of H2O diffusion and the role of H2O in oxygen
diffusion (see discussion in a later section).
Chemical diffusion of oxygen under dry conditions
Chemical diffusion of oxygen under dry conditions is not well understood. Dunn (1983)
reported oxygen chemical diffusion data in nephelinite, alkali basalt and tholeiite melts at 1553
to 1723 K and 0.4 to 2 GPa. Oxidized spheres of glass with Fe3+/Fe2+ ratio of 3.7 to 5.8 were
packed with graphite powder in a graphite capsule in piston-cylinder experiments. After an
experiment, the spheres were retrieved and average FeO concentration in the bulk sample was
determined by wet chemical analyses (the Fe2O3 concentration is obtained by difference). Then
the diffusivity is estimated using an equation similar to Equation (42). The experimental data
of Dunn (1983) on basalt melt are shown in Figure 23. Dunn and Scarfe (1986) used the same
approach to obtain diffusivities in an andesite melt at 1623 K and 0.35 to 2 GPa. One possible
mechanism for the process in the two reports is as follows: Fe2O3 on the surface of the melt
sphere reacts with graphite as:
2Fe2O3(melt) + C(graphite)  4FeO(melt) + CO2(melt) (50)
Reduction of 1.1 wt% Fe2O3 to produce 1 wt% FeO is accompanied by the production of
208 Zhang & Ni
0.1 MPa (W)

-21 0.4 GPa (D)
0.6 GPa (D)
0.8 GPa (D)
1.2 GPa (D)
1.4 GPa (D)
-22 1.6 GPa (D)
2.0 GPa (D)
0.1 MPa (C)
lnD (D in m2/s)
1.0 GPa (L)

-23
-24
-25 Oxygen chemical

& self-diffusion
in basalt melt
-26
0.54 0.56 0.58 0.6 0.62 0.64 0.66 0.68 0.7
1000/T (T in K)
figure 23. Chemical diffusion and oxygen self-diffusion in basalt melt. Oxygen self-diffusion data (open
and solid circles) are: 0.1 MPa (C) from Canil and Muehlenbachs (1990); 1.0 GPa (L) from Lesher et al.
(1996).Fig.
The23.chemical diffusivities
Chemical diffusion are related
and oxygen to O2inmolecular
self diffusion basalt melt.diffusion ordiffusion
Oxygen self FeO effective
data (openbinary diffusion
and solid
(see text for discussion): 0.1 MPa (W) from Wendlandt (1991) based on weight gain (filled diamonds;
circles in red)
likely reflecting are: 0.1 MPa
molecular O2(C)diffusion
from Caniland
anddehydration),
Muehlenbachs (1990);
which1.0isGPa (L) a
about from Lesher
factor ofet30al. greater
(1996). The
than the
self-diffusivities; other data are from Dunn (1983), likely reflecting FeO diffusion.
chemical diffusivities are related to O molecular diffusion or FeO effective binary diffusion (see text for
2
discussion): 0.1 MPa (W) from Wendlandt (1991) based on weight gain (filled diamonds; likely reflecting molecular
0.153 wt% CO2. Then CO2 and FeO diffuse into the melt, and Fe2O3 in the interior of the melt
diffusesO2todiffusion
the surface of thewhich
and dehydration), melt. In this
is about scenario,
a factor the
of 30 greater thanchemical diffusivity
the self diffusivities; determined
other data are from by
Dunn (1983) and Dunn and Scarfe (1986) is related to interdiffusion between CO2+FeO and
Dunn (1983), likely reflecting FeO diffusion.
Fe2O3. Data scatter is considerable in these studies.
Wendlandt (1991) reported oxygen chemical diffusion data in basalt and andesite melts
at 1433 to 1633 K and 1 atmosphere with controlled oxygen fugacity (Fig. 23). Mass gain or
loss of the spherical sample at an imposed oxygen fugacity was monitored by a high precision
electrobalance. Diffusivities were obtained by fitting the variation of the total mass with time
using Equation (42). The reaction at the surface is likely 2FeO(melt) + (1/2)O2  Fe2O3(melt),
then FeO and Fe2O3 would interdiffuse in the melt. The diffusivity is hence oxygen chemical
diffusivity related to FeO and Fe2O3. The data precision is high because weight determination
is highly reproducible. However, the initial basalt sample contains 1.03 wt% H2Ot (the initial
andesite sample contains 0.44 wt% H2O). The presence of H2O in the initial sample may cause
two effects:
(1) Oxygen diffusivities may be enhanced due to H2O diffusion, which can carry oxygen
(see next section). Based on H2Ot diffusivity in basalt (Eqn. 22) and the effect of H2O
diffusion on oxygen diffusion, with 1.0 wt% H2Ot, the contribution of H2O diffusion
to oxygen self-diffusion would yield a lnDoxygen of −24.9 at 1633 K, and −25.6 at
1533 K, and −26.5 at 1433 K, which is not enough to account for the high oxygen
diffusivities by Wendlandt (1991) (Fig. 23).
(2) Because of the presence of H2O in the initial sample, there would be simultaneous
dehydration as O2 diffuses in or out of the sample. The effect of dehydration on the
weight data was not assessed by Wendlandt (1991). Hence, the accuracy of results in
this study is uncertain even though the precision is high.
In summary, the much higher apparent oxygen chemical diffusivity by Wendlandt (1991)
might be due to complications that are unaccounted for. Furthermore, even if only the data
by Dunn (1983) are considered, the scatter is still considerable, which makes it impossible to
infer how chemical oxygen diffusivity depends on pressure, or to infer the activation energy.
The large scatter in the data of Dunn (1983), Dunn and Scarfe (1986) and Wendlandt (1991)
may be partially related to the use of the bulk mass gain or mass loss method, which is prone
to complications. Furthermore, part of the difficulty is likely related to the different meanings
of oxygen chemical diffusion (such as molecular O2 diffusion, or FeO and Fe2O3 diffusion, or
CO2 or H2O diffusion, etc).
“Self” diffusion of oxygen in the presence of H2O
In experimental studies or natural systems, H2O is often present: 18O-enriched H2O may
enter the sample of interest from a fluid phase, or H2O may be initially in the sample. Because
H2O contains oxygen, the transport of H2O as molecular H2O (or as OH groups) would result
in an oxygen flux. Hence, oxygen diffusivity can be affected or controlled by H2O diffusion.
The relation between apparent 18O diffusivity in silicate melts and the presence of H2O
vapor (as well as H2O in the melt) has been explored. DeBerg and Lauder (1980) conducted
18
O-16O exchange experiments between melt spheres (composition: 62.2 wt% SiO2 and 37.2
wt% K2O prepared under a vacuum of 0.013 Pa) and an 18O-enriched O2 or H2O gas, and then
extracted 18O diffusivity. At 1175 K, 18O diffusivity does not depend on pressure of O2 from
7.5 to 56.3 kPa. On the other hand, the presence of 1.3 kPa of H2O vapor (in addition to 7.5
kPa of O2) yielded an 18O diffusivity about 2 times the dry 18O diffusivity at 1093 to 1143 K.
The solubility of H2O in this melt is not known. If we assume that the solubility in the potassic
silicate melt is roughly the same as that in rhyolite melt (Liu et al. 2005), the solubility at 1.3
kPa of H2O vapor pressure would be about 0.011 wt% (110 ppm). Hence, the observation by
DeBerg and Lauder (1980) would imply that when H2Ot concentration is of the order 100 ppm,
the contribution to 18O flux by H2O chemical diffusion is about the same as that of dry oxygen
diffusion for this potassic silicate melt at about 1100 K. As H2O partial pressure increases,
dissolved H2Ot in the melt increases, and 18O diffusivity increases. DeBerg and Lauder (1980)
concluded that the dissolution and diffusion of O2 in the melt do not play an important role (due
to the small solubility and small diffusivity of O2 molecules in the melt), but the dissolution and
diffusion of H2O in the melt play a critical role in 18O diffusion in addition to dry 18O diffusion.
Pfeffer and Ohring (1981) investigated 18O diffusion in silica in controlled water steam pressure
and reached a similar conclusion.
Zhang et al. (1991b) developed the general theory of diffusion of a multi-species
component, using the diffusion of H2O and 18O as specific examples. For 18O diffusion in the
presence of H2O in the sample (either high or low H2Ot), because both H2Om and OH carry
an oxygen atom, H2O diffusion also carries an 18O diffusion flux. Because it has been shown
from H2O diffusion studies that in silicate melts OH diffusion is negligible compared to H2Om
diffusion under almost all conditions, 18O diffusivity may be viewed as being enhanced by
H2Om diffusion alone. In such a case, the general equation for 18O diffusion can be written as
(Eqn. 6 in Behrens et al. 2007 plus the anhydrous diffusion term):
∂Ri ∂  ∂R  ∂  ∂( Ri ⋅ X m )  ∂  ∂X 
=  D 18O,anhydrous i  +  DH2 Om  − Ri  DH2 Om m  (51)
∂t ∂x  ∂x  ∂x  ∂x  ∂x  ∂x 
where Ri is the isotopic fraction of 18O, 18O/(16O+17O+18O), Xm is the mole fraction of H2Om
on a single oxygen basis, and D18O,anhydrous is 18O diffusivity of oxygen species not associated
with H in the presence or absence of H2O. The first term on the right hand side accounts for
18
O diffusive flux due to network oxygen diffusion (not associated with H); the second term
accounts for 18O diffusive flux due to H2Om diffusion; and the third term accounts for mass
210 Zhang & Ni
balance (so that at constant Ri there is no 18O flux, meaning that the sum of the second and
third terms is zero). D18O,anhydrous is expected to depend on H2Ot concentration because addition
of H2O would loosen the melt structure, which would likely cause a higher diffusivity of even
anhydrous oxygen species, such as those bonded to Si and Mg. Based on the above equation,
18
O diffusivity can often be estimated as (from Eqn. 15 in Zhang et al. 1991b by removing the
OH term):
D18 O ≈ D18 O,anhydrous + DH2 Om X m (52)
The DH2OmXm term in the above equation is the apparent 18O “self” diffusivity contributed by
H2O diffusion. Using the relation between DH2Om and DH2Ot (Eqn. 13), the above can be written
as:
dX
D18 O ≈ D18 O,anhydrous + DH2 Ot X m (53)
dX m
where X is the mole fraction of H2Ot on a single oxygen basis. The expression XmdX/dXm is
shown in Figure 3 for K = 0.5. At low H2Ot (e.g., when X ≤ 0.01), the above can be simplified
as (Eqn. 16 in Zhang et al. 1991b):
X
D18 O ≈ D18 O,anhydrous + DH O (54)
2 2 t
Zhang et al. (1991b) summarized literature data to evaluate the role of H2O diffusion in 18O
“self” diffusion. Behrens et al. (2007) experimentally investigated H218O sorption into a rhyolite
melt, from which both H2Ot and 18O diffusion profiles were measured. Their results show that
both H2Ot and 18O profiles indicate the same H2Om diffusivity (Fig. 24) under the assumption
that H2Om is the diffusing species and there is chemical and isotopic equilibrium. Thus, their
experimental data confirm that H2Om is the diffusing species in both H2Ot diffusion and 18O
“self” diffusion, as well as the quantitative theory of Zhang et al. (1991b) presented above.
In summary, H2O chemical diffusion (or effective binary diffusion) carries an 18O flux,
which contributes to an apparent 18O diffusivity. If the H2O content is uniform in the sample
and there is no chemical diffusion of H2O (e.g., diffusion couple with similar starting H2Ot
but different 18O/16O in the two halves, or sorption of 18O-enriched H2O into a sample already
containing the equilibrium concentration
of H2O), 18O-16O exchange can still be due to the
figure 24. Experimental data on H2Ot and Ri = 18O/(16O+17O+18O) profiles (data points) during hydration
using 18O-enriched H2O. The solid lines are fit by Equation (11) (for the H2Ot profile) and Equation (51)
Fig. 24. Experimental data on H Ot and Ri = 18O/(16O+17O+18O) profiles (data points) during hydration using 18O-
(for the Ri profile) assuming H2Om is the diffusing species and DH2Om = D0eaX2 with the same a and D0 values
for both profiles. From Behrens et al. (2007). enriched H2O. The solid lines are fit by Eq. 11 (for the H2Ot profile) and Eq. 51 (for the Ri profile) assuming H2Om
is the diffusing species and DH2Om = D0eaX with the same a and D0 values for both profiles. From Behrens et al.
(2007).
mobility (self-diffusion) of H2O, and the diffusion may be said to be self-diffusion. In both cases,
the diffusing species is H2Om. The resulting 18O diffusivities from H2O chemical diffusion and
from H2O self-diffusion are not much different. In the next section, we examine the conditions
for 18O diffusion in natural silicate melts to be true oxygen “self” diffusion or through H2O
diffusion, and quantify apparent 18O diffusivity in hydrous rhyolite and dacite melts.
“Self” diffusion of oxygen in natural silicate melts in natural environments
Using the theory presented above, if we know both the true 18O self-diffusivity under dry
conditions and the H2O diffusivity, we can compare D18O,anhydrous with DH2OmXm, or D18O,anhydrous
with DH2OtXmdX/dXm, to determine whether the true 18O self-diffusion or the apparent 18O flux
due to H2O diffusion dominates the 18O flux. Below, we employ data on H2O diffusion and dry
18
O self-diffusion to evaluate quantitatively the conditions when H2O diffusion dominates 18O
“self” diffusion.
(1) Dacite melt. This melt is considered first because extensive data are available. Figure
25 compares dry 18O self-diffusivity (Tinker and Lesher 2001) and apparent 18O diffusivity
contributed by H2O diffusion. First consider a numerical case at 1600 K and 1 GPa (which are
close to the covered experimental conditions of both 18O self-diffusion and H2O diffusion).
From Equation (44), dry 18O self-diffusivity is 3.5×10−14 m2/s. Based on Equations 18 and 53,
about 0.32 wt% H2Ot is required to contribute 3.5×10−14 m2/s to the oxygen diffusivity. Because
synthesized experimental dry melts typically contains only ≤ 0.1 wt% H2Ot, the experimental
dry 18O self-diffusivities at such high temperatures by Tinker and Lesher (2001) are true
D18O,anhydrous. Now consider 1200 K (a more reasonable magmatic temperature for natural
dacite) and 1 GPa. Data by Tinker and Lesher (2001) do not cover such a low temperature
(metastable melt). Dry 18O self-diffusivity extrapolated using Equation (44) is 2.1×10−17 m2/s.
Based on Equation (18),
Youxue Zhang and Niabout
(Ch 5) 0.05 wt% (500
Pageppm)
25 H2Ot is required to contribute 2.1×10−17
figure 25. Oxygen and H2Ot diffusion at 1 GPa in dacite melt. The open circles and the solid line are
H2Ot diffusivity at 0.95 to 1 GPa from Behrens et al. (2004) and Ni et al. (2009b). The two dashed lines
Fig. 25. Oxygen and H2Ot diffusion at 1 GPa in dacite melt. The open circles and the solid black line are H2Ot
are calculated oxygen diffusivities due to H2O diffusion based on the H2O diffusivity of Ni et al. (2009b).
The filled diamonds (in red
diffusivity at 0.95 in the
to 1 GPa fromonline
Behrens version)
et al. (2004)with
and Niaetbest-fit lineThe(indashed
al. (2009b). red black
in the online
lines version) are
are calculated
experimental 18O self-diffusion data at 1 GPa from Tinker and Lesher (2001). The dotted line is Eyring
diffusivity calculated usingdue
oxygen diffusivities thetoviscosity model
H2O diffusion based of HuiH2and
on the Zhangof(2007)
O diffusivity Ni et al.at 0.1 MPa
(2009b). to 1 diamonds
The filled GPa assuming
(in a
jumping distance of 2.8×10−10 m. 18
red) with a best-fit line (in red) are experimental O self diffusion data at 1 GPa from Tinker and Lesher (2001).
The dotted line (in blue) is Eyring diffusivity calculated using the viscosity model of Hui and Zhang (2007) at 0.1
MPa to 1 GPa assuming a jumping distance of 2.8x10-10 m.

212 Zhang & Ni
m2/s to the oxygen diffusivity. The difference at the two temperatures is due to lower activation
energy for H2O diffusion and higher activation energy for dry 18O diffusion. As the temperature
is lowered further, H2O diffusion would dominate the apparent 18O “self” diffusion (Fig. 25)
at even lower H2Ot. Because most natural dacite melts are expected to contain more than 0.2
wt% H2Ot, we conclude that apparent 18O diffusion in natural dacite melts is almost always
dominated by H2O diffusion.
(2) Basalt melt. Both 18O and H2O diffusion data are limited on basalt melt. Consider
1600 K and 1 GPa. From Equation (47), dry 18O self-diffusivity is about 1.1×10−11 m2/s. Based
on Equation (22), about 1 wt% H2Ot is necessary to contribute 1.1×10−11 m2/s to apparent
18
O diffusivity. That is, for basalt melt, due to high dry 18O self-diffusivity and magmatic
temperatures, about 1 wt% H2Ot is needed for H2O diffusion to dominate 18O “self” diffusion.
Therefore, in almost all mid-ocean ridge basalt melts (typically containing 0.2 to 0.7 wt% H2Ot,
but can be up to 1.2 wt%; Dixon et al. 1988; Michael 1988; Workman et al. 2006) and most ocean
island basalt melts (often more volatiles than MORB; Dixon et al. 1997; Hauri 2002; Workman
et al. 2006), 18O diffusivity is most likely dominated by true 18O self-diffusion. However, in pre-
eruptive island arc basalt (IAB) melts, and even in melt inclusions in mantle megacrysts likely
influenced by subduction, H2Ot concentration are 2-6 wt% (Stolper and Newman 1994; Wang
et al. 1999; Newman et al. 2000; Gurenko et al. 2005; Wallace 2005), meaning apparent 18O
self-diffusion in these melts is dominated by H2O diffusion.
(3) Rhyolite melt. There are only limited 18O diffusivity data in dry rhyolite melt at
1553 K and 0.1 MPa (Muehlenbachs and Kushiro 1974) and the quality of the data is not
high (see discussion in an earlier section). On the other hand, H2O diffusion in rhyolite melt
has been investigated extensively (e.g., comprehensive model by Ni and Zhang 2008). Figure
26 compares the contribution to 18O diffusion by the H2O flux with that by true network 18O
self-diffusion. Oxygen self-diffusivities by Muehlenbachs and Kushiro (1974) are higher by a
factor of about 24 than the Eyring diffusivity for dry melt calculated from the viscosity model
of Zhang et al. (2003). If the data by Muehlenbachs and Kushiro (1974) are accurate and the
melt was indeed dry (much less than 0.1 wt% H2Ot), oxygen flux due to chemical diffusion of
H2O when H2Ot content is about 0.1 wt% would be roughly the same as the oxygen flux due to
true oxygen self-diffusion. On one hand, oxygen self-diffusivities obtained by Muehlenbachs
and Kushiro (1974) are close to effective binary diffusivity of P, which seems to imply the
Eyring diffusivity limit. On the other hand, typical “dry” rhyolite glasses (either natural or
synthesized) contain ≥ 0.1 wt% H2Ot. If the rhyolite used by Muehlenbachs and Kushiro (1974)
also contained about 0.1 wt% H2Ot, their measurements would actually mean apparent 18O
“self” diffusivity due to H2O diffusion, implying true dry 18O self-diffusivities are yet to be
determined. Regardless how this issue is resolved, at typical rhyolite melt temperatures (such
as 1200 K), H2O diffusion would dominate oxygen transport at much lower H2Ot, such as 10 to
100 ppm level. That means, in natural rhyolite melt in which H2O content if often 4-6 wt% (e.g,
Wallace et al. 2003), diffusive transport of oxygen isotopes is through H2O diffusion.
There are no 18O diffusivity data in dry andesite melts, and hence similar quantitative
comparisons cannot be carried out. Nonetheless, it is expected that H2O diffusion plays a more
important role in transporting 18O flux as temperature is lowered and as the melt becomes
more silicic. Hence, the behavior of andesite melt is expected to be between that of basalt
melt and dacite melt. When H2Ot is high enough (e.g., ≥ 0.5 wt%), the apparent 18O diffusivity
in andesite melt is likely dominated by H2O diffusion. Because pre-eruptive natural andesite
melt often contains ≥ 0.5 wt% H2Ot, we expect the 18O flux in natural andesite melt to be often
dominated by H2O chemical diffusion. In summary, in nature, true 18O self-diffusion is the
diffusion mechanism only for relatively dry basalt melt (such as MORB and OIB).
The realization that diffusive transport of 18O in natural rhyolite and dacite melts is almost
always due to H2O diffusion means it is possible to predict 18O diffusive transport using Equation
figure 26. Oxygen and H2Ot diffusion at 0.1-100 MPa in rhyolite melt. The pressure effect in this small
pressure range is smaller than 0.2 lnD units. The solid line is the calculated H2Ot diffusivity. The two
dashed linesFig.(roughly
26. Oxygenparallel
and H2Otto the solid
diffusion line)MPa
at 0.1-100 areinoxygen diffusivities
rhyolite melt. The pressuredue toinHthis
effect
2O diffusion calculated
small pressure
at 0.1 GPa range
based on Ni and Zhang (2008). The Eyring diffusivity at 0-0.1 wt% H Ot is calculated
is smaller than 0.2 lnD units. The solid black line is the calculated H2Ot diffusivity. The2dashed black lines
from
viscosity data (solid circles; Neuville et al. 1993; Schulze et al. 1996) and the viscosity model (lines; Zhang
−10 18
et al. 2003)areassuming a jumping
oxygen diffusivities due todistance of 2.8×10
H2O diffusion calculated atm.
0.1 The two on
GPa based filled
Ni anddiamonds
Zhang (2008) arewith
experimental
a 2σ O
self-diffusion data at 0.1 MPa from Muehlenbachs and Kushiro (1974). The open squares and short dashed
line are effective binary
uncertainty of 0.49diffusion
in lnD. Thedata ofdiffusivity
Eyring P and fit at 0.8
at 0-0.1 wt%GPa
H2Otand 0.1 wt%
is calculated fromHviscosity
2Ot (Harrison and Watson
data (points;
1984), which are somewhat higher than the Eyring diffusivity.
Neuville et al. 1993; Schulze et al. 1996) and the viscosity model (lines; Zhang et al. 2003) assuming a jumping
distance of 2.8x10-10 m. The two filled diamonds (in red) are experimental 18O self diffusion data at 0.1 MPa from
(52) or (53) because H2O diffusion in these melts has already been investigated well (e.g.,
Muehlenbachs and Kushiro (1974). The open squares and short dashed line (both in red) are effective binary
Behrens et al. 2004; Liu et al. 2004b; Ni and Zhang 2008; Ni et al. 2009b). For the convenience
18 0.1 wt% H2Ot (Harrison and Watson, 1984), which is somewhat higher
of the readers, calculated
diffusion apparent
data of P and fit at 0.8 GPa andO diffusivities in hydrous rhyolite and dacite melts as
a function of T, P and H O
2 t
than the Eyring diffusivity. are listed in Tables 3 and 4. In the calculation, the diffusivity of
oxygen not associated with H is approximated by Eyring diffusivity. The calculated diffusivities
have a 2σ uncertainty of about 0.6 in lnD and may be applied in rough estimation of diffusion
rates. For calculation of the 18O diffusion profile, it is more accurate to use Equation (51).
Prediction of 18O diffusion in basalt and andesite melts require more experimental work in
terms of both H2O diffusion and dry 18O diffusion.
Contribution of CO2 diffusion to 18O transport in CO2-bearing melts
Because carbon species (CO2 molecule and CO32−) also carry oxygen, oxygen transfer
in natural systems may also be realized through the diffusion of these species, especially
molecular CO2. One-atmosphere experiments on 18O diffusion often use 18O-enriched CO2 gas
as the source for 18O (Muehlenbachs and Kushiro 1974; Canil and Muehlenbachs 1990). Not
withstanding the quality of such data, one may wonder whether the extracted diffusivity is
true 18O self-diffusivity, or just a reflection of CO2 diffusion carrying 18O into the sample. No
simultaneous investigation of 18O and CO2 diffusion has been carried out yet.
Although experimental data are lacking, the role of CO2 diffusion in transporting 18O
can be treated similarly as the role of H2O diffusion in transporting 18O. However, it is more
convenient to consider total CO2 diffusion rather than the diffusion of CO2 molecules because
DCO2,total is roughly independent of melt composition. The diffusion equation for the isotopic
fraction of 18O can be written as follows (comparing with Eqn. 51):
214 Zhang & Ni
Table 3. Calculated lnD18O (D in m2/s) in hydrous rhyolite melt.
H2Ot 0.1 gpa 0.5 gpa 1 gpa

(wt%)
1000 K 1300 K 1600 K 1000 K 1300 K 1600 K 1000 K 1300 K 1600 K
0.1 −37.36 −35.02 −33.43 −38.06 −35.38 −33.56 −38.93 −35.83 −33.72
0.3 −35.11 −32.78 −31.31 −35.80 −33.14 −31.46 −36.67 −33.59 −31.64
0.5 −34.02 −31.73 −30.28 −34.72 −32.08 −30.43 −35.59 −32.53 −30.61
1.0 −32.45 −30.24 −28.86 −33.15 −30.60 −29.01 −34.02 −31.05 −29.19
1.5 −31.43 −29.32 −28.02 −32.13 −29.68 −28.16 −33.00 −30.13 −28.34
2.0 −30.63 −28.63 −27.40 −31.33 −28.99 −27.55 −32.19 −29.44 −27.73
3.0 −29.34 −27.58 −26.51 −30.03 −27.94 −26.66 −30.90 −28.39 −26.84
4.0 −28.26 −26.76 −25.85 −28.95 −27.12 −25.99 −29.82 −27.56 −26.17
5.0 −27.99 −26.42 −25.46 −28.86 −26.86 −25.64
6.0 −27.1 −25.79 −25.00 −27.97 −26.24 −25.18
7.0 −26.26 −25.22 −24.59 −27.13 −25.67 −24.77
8.0 −25.47 −24.69 −24.22 −26.34 −25.14 −24.41
Each cell lists lnD18O values where D18O is calculated total 18O diffusivity in m2/s using Equation (52), with DH2Om from
Equation (14), K from Equation (7a), and D18O,anhydrous approximated by DEyring calculated using the viscosity model of
Zhang et al. (2003). Values of lnD at other temperatures can be obtained using the Arrhenius relation.
Table 4. Calculated lnD18O (D in m2/s) in hydrous dacite melt.

(wt%)
1000 K 1300 K 1600 K 1000 K 1300 K 1600 K 1000 K 1300 K 1600 K
0.1 −38.62 −34.60 −29.51 −39.21 −34.95 −29.53 −39.95 −35.33 −29.54
0.3 −36.33 −32.47 −28.54 −36.92 −32.86 −28.60 −37.65 −33.31 −28.65
0.5 −35.20 −31.42 −27.91 −35.79 −31.82 −27.99 −36.53 −32.28 −28.07
1.0 −33.54 −29.94 −26.93 −34.13 −30.34 −27.06 −34.87 −30.80 −27.18
1.5 −32.43 −29.02 −26.32 −33.02 −29.43 −26.47 −33.76 −29.90 −26.63
2.0 −31.54 −28.34 −25.87 −32.13 −28.75 −26.05 −32.87 −29.23 −26.22
3.0 −30.08 −27.30 −25.21 −30.67 −27.72 −25.42 −31.40 −28.22 −25.64
4.0 −28.83 −26.48 −24.72 −29.42 −26.91 −24.96 −30.16 −27.42 −25.21
5.0 −28.30 −26.21 −24.59 −29.04 −26.74 −24.86
6.0 −27.25 −25.58 −24.27 −27.99 −26.13 −24.57
7.0 −26.27 −25.01 −23.99 −27.00 −25.56 −24.31
8.0 −25.32 −24.47 −23.74 −26.06 −25.03 −24.07
Each cell lists lnD18O values where D18O is calculated total 18O diffusivity in m2/s using Equation (52), with DH2Om from
Equation (18), K from Equation (7b), and D18O,anhydrous approximated by DEyring calculated using the viscosity model of
Whittington et al. (2009).
∂R ∂2 R ∂ 2 ( R ⋅ X C0.5 Ototal ) ∂ 2 X C0.5 Ototal

=
D 18O,noncarbonated 2 + DCO2,total − DCO2,total R (55)
∂t ∂x ∂x 2 ∂x 2
where XC0.5Ototal is mole fraction of total C0.5Ototal (each C0.5Ototal contains a single oxygen and
two C0.5Ototal makes one CO2,total) on a single oxygen basis, and D18O,anhydrous is diffusivity of
18
O unassociated with carbon. XC0.5Ototal, is calculated as (Ccd/22.005)/[Ccd/22.005+(100-
Ccd)/W] where Ccd is wt% of CO2,total and W is the molar mass of the dry melt on a single
oxygen basis (Table 1). XCO2,total may be used instead of XC0.5Ototal, but a factor of 2 would have
to be incorporated because each CO2 molecule carries two oxygen atoms. DCO2,total does not
depend on CO2 concentration or anhydrous melt composition and hence can be taken out of
the differentials. Based on the above equation, the 18O diffusivity can often be estimated as
(comparing with Eqn. 52):
D18 O ≈ D18 O,noncarbonated + DCO2,total X C0.5 Ototal (56)
The DCO2,totalXC0.5Ototal term in the above equation is apparent 18O “self” diffusivity contributed by
CO2 diffusion. Because DCO2,total can be estimated from Equation (41), we examine quantitatively
possible contributions by CO2 to 18O diffusion below.
For rhyolite melt at 1553 K and 0.1 MPa (the condition at which 18O diffusivity was
determined in a CO2 gas by Muehlenbachs and Kushiro 1974), DCO2,total is 1.2×10−11 m2/s from
Equation (41). The measured 18O diffusivity is 4.5×10−15 m2/s. For CO2 diffusion to contribute
to such an 18O diffusivity, a CO2 concentration of 260 ppm is needed (equivalent to a CO2
partial pressure of 52 MPa). Typical natural rhyolite does not contain this much CO2. We hence
conclude that in this case, 18O diffusivity is not due to chemical CO2 diffusion. This is consistent
with the similarity of 18O diffusivities in CO2 gas and in O2 gas.
For the convenience of readers, contribution to apparent D18O by CO2 chemical diffusion
are calculated and listed in Table 5. Note that this table does not give the apparent D18O values
as in Tables 3 and 4 where the dry 18O self-diffusivity has been added or is assessed to be
negligible. Table 5 only lists the part of apparent D18O contributed by CO2 chemical diffusion
when the CO2 concentration is 100 ppm. The 2σ uncertainty in the calculated values is about 1.1
lnD units. For the full apparent 18O diffusivity, the anhydrous and non-carbonated true 18O self-
diffusivity and that contributed by CO2 diffusion must be added. Because DCO2,total (Equation 41)
does not depend on the anhydrous melt composition, Table 5 is roughly applicable to all silicate
melts. The contribution to the apparent 18O diffusivity is proportional to CO2,total concentration
(Eqn. 56), which can be used to estimate the contribution to D18O by CO2 chemical diffusion at
other CO2,total concentrations. Comparing Table 5 with Tables 3 and 4, the contribution by CO2
chemical diffusion to 18O diffusion at the 100 ppm CO2 level is not important to dacite melt or
rhyolite, with the possible exception of very dry rhyolite melt (≤ 0.2 wt% H2Ot) at very high
temperatures (≥ 1500 K).
Table 5. Calculated lnD18O (D in m2/s) contributed by CO2 diffusion at 100 ppm CO2,total.

(wt%) 1000 K 1300 K 1600 K 1000 K 1300 K 1600 K 1000 K 1300 K 1600 K
0 −39.27 −35.22 −32.69 −40.05 −35.82 −33.17 −41.02 −36.57 −33.78

0.1 −39.19 −35.15 −32.63 −39.95 −35.74 −33.11 −40.91 −36.48 −33.71
0.3 −39.01 −35.02 −32.52 −39.75 −35.59 −32.99 −40.69 −36.31 −33.57
0.5 −38.83 −34.88 −32.41 −39.56 −35.44 −32.86 −40.46 −36.13 −33.43
1.0 −38.39 −34.54 −32.14 −39.06 −35.06 −32.56 −39.9 −35.7 −33.08
1.5 −37.95 −34.2 −31.86 −38.57 −34.68 −32.25 −39.33 −35.27 −32.73
2.0 −37.51 −33.86 −31.59 −38.07 −34.3 −31.94 −38.77 −34.83 −32.37
3.0 −36.63 −33.18 −31.03 −37.08 −33.53 −31.32 −37.64 −33.97 −31.67
4.0 −35.74 −32.51 −30.48 −36.09 −32.77 −30.7 −36.52 −33.1 −30.97
5.0 −35.1 −32.01 −30.08 −35.39 −32.24 −30.26
Each cell lists lnD18O values contributed by CO2 chemical diffusion with D calculated as DCO2,totalXC0.5Ototal at 100 ppm
CO2,total and DCO2,total from Equation (41). To estimate D18O at other CO2 concentrations, multiply D18O (not lnD18O) by
the concentration ratio. For example, for 500 ppm CO2, multiply D18O values in this table by 5. Add D18O here to D18O in
Tables 3 or 4 (do not add lnD18O) would give the full D18O in hydrous and carbonated rhyolite and dacite melts.
216 Zhang & Ni
Oxygen diffusion and viscosity: applicability of the eyring equation

Oishi et al. (1975) showed that the 18O diffusivity in 4Na2O · 3CaO · 18SiO2,
2CaO · Al2O3 · 2SiO2 , and Na2O · 4SiO2 melts is inversely proportional to melt viscosity.
Shimizu and Kushiro (1984) showed that the Eyring equation relates 18O diffusivity well with
viscosity for dry diopside and jadeite melts (Fig. 27a). 18O diffusion data in other dry melts
(e.g., Liang et al. 1996; Tinker and Lesher 2001; Tinker et al. 2004) also roughly follow the
Eyring relation (Fig. 27a) even though in detail there are differences (e.g., Liang et al. 1996;
Tinker et al. 2004). Figure 25 shows that the Eyring relation roughly applies to dry dacite
melt at 1 GPa (rough agreement between solid diamonds and dotted line). However, Figure 26
indicates either limited experimental 18O diffusion data in dry rhyolite melt do not represent true
D18O,anhydrous, or the Eyring relation does not apply to dry rhyolite (comparing solid diamonds
and solid circles). Overall, available data largely indicate that the Eyring equation is applicable
to 18O diffusion in dry silicate melts.
For hydrous silicate melts, the available data indicate that the apparent 18O diffusivity is
often orders of magnitude greater than the Eyring diffusivity (or Stokes-Einstein diffusivity or
any of the other models of inverse proportionality between diffusivity and viscosity) (Behrens
figure 27. Comparison of the Ey-

ring diffusivity with 18O diffusiv-
ity in (a) dry melts, and (b) hydrous
rhyolite melts. Note that the scale is
base-10 logarithm, not base-e loga-
rithm. Data sources for (a): Shimu-
zu84 = Shimizu and Kushiro (1984);
Tinker04 = Tinker et al. (2004);
Liang96 = Liang et al. (1996). Data
sources for (b): Viscosity model used
for the calculation of rhyolite melt
viscosity is from Zhang et al. (2003).
H2Om, H2Ot and 18O diffusivities are
from Behrens et al. (2007) and Ni
and Zhang (2008).
Fig. 27. Comparison of the Eyring diffusivity and Einstein diffusivity with 18O diffusivity in (a) dry melts, and (b)
et al. 2007; Fig. 27b). Because natural silicate melts often contain a significant amount of H2O,
the results indicate that the application of the Eyring relation in geological melts is very limited.
18
O diffusion in terrestrial silicate melts in natural environments is often dominated by H2O flux
and cannot be predicted by the Eyring equation (but can be predicted from H2O diffusivities).
The inapplicability of the Eyring equation to 18O diffusion in hydrous silicate melts may
be explained as follows. In deriving the Eyring equation, one assumption is that the diffusion
and viscous flow are controlled by the same mechanism (motion). For the case of dry silicate
melts, viscous flow and diffusion of oxygen (the dominant ion in the melt) are both related to
structural mobility, and are hence controlled by the same mechanism. However, for the case
of wet silicate melts, viscous flow requires network oxygen mobility, but 18O diffusive flux is
carried by the mobility of neutral H2Om molecules, whose motion is not necessarily related to
network oxygen relaxation. In the context of this explanation, the Eyring equation cannot be
applied as a universal relation; it is necessary to know the diffusion mechanism and diffusing
species before assessing whether the equation can be applied. Because most natural melts are
hydrous, the applicability of the Eyring relation to natural melts is limited.
O2 DIffUSION IN pURe SIlICA MelT

Dissolved molecular O2 in silicate melts may also be referred to as physically dissolved
oxygen (meaning no reaction between O2 molecule and other components in the melt or glass
because essentially all melts are based on oxygen anions). In natural melts, the oxygen fugacity
is low. For example, oxygen fugacity at NNO buffer (about the average oxygen fugacity in
typical mantle derived silicate melts), fO2 is only about 3×10−7 Pa at 1200 K (calculation based
on thermodynamic data of Robie and Hemingway, 1995), lower than that in air by almost
11 orders of magnitude. Therefore, the dissolved O2 concentration in natural silicate melts
is expected to be extremely low. However, in industry and in high-temperature experiments
carried out in air or in pure oxygen gas, dissolved O2 concentration may be noticeable. Under
such conditions, pure molecular O2 diffusion may be studied, and molecular O2 diffusion may
contribute significantly to network oxygen diffusion.
We have shown that oxygen diffusion data in natural or nearly natural silicate melts are
complicated, but it is still possible to make sense out of most data in terms of contributions by
various oxygen species. Oxygen diffusion data in the glass and materials science literature are
even messier (e.g., Fig. 9 in the review by Lamkin et al. 1992), likely due to the contributions
by the various oxygen species discussed earlier (especially the hydrous component), plus the
additional complexity due to molecular O2. Because such data are on melts very different from
geological silicate melts, they are not the focus in this volume. Hence, we make no attempt
to quantitatively explain the complicated behavior of oxygen diffusivity in various silica and
silicate melts in the glass and materials science literature. Below, we briefly discuss molecular
O2 diffusion in silica melt for two purposes. First, such data allow us to roughly know the
molecular O2 diffusivity (for the completeness of this volume). Secondly, we may use the O2
diffusivity to assess the role of molecular O2 diffusion in oxygen diffusion in natural melts
(under oxygen fugacity of terrestrial igneous processes), which turns out to be unimportant
due to extremely low oxygen fugacity in natural silicate melts (see below). Molecular diffusion
of O2 in silica melt is best understood compared to other silicate melts due to the simple
composition of silica.
Even though there have been numerous studies on oxygen diffusion in silica melt/glass,
only a limited number of authors reported genuine molecular O2 diffusivities. Norton (1961)
carried out molecular O2 permeation experiments in silica glass and obtained molecular O2
diffusivities at 1223 and 1351 K. Hetherington and Jack (1964) made an order of magnitude
estimate of molecular O2 diffusivities in silica glass, and the values are consistent with the
218 Zhang & Ni
results of Norton (1961). Susa et al. (1990) reported diffusivities of molecular O2 in variously
prepared silica films at 1073-1273 K. Lamkin et al. (1992) reviewed oxygen diffusion data in
silica and silicate glasses. Kajihara et al. (2004, 2005) systematically investigated molecular O2
diffusion in silica glass with different OH contents. Kajihara et al. (2008) reviewed diffusion
and reaction of interstitial oxygen species in silica glass. Molecular O2 diffusion data in silica
glass are shown in Figure 28. Molecular O2 diffusivity in low-OH silica glass (Kajihara et al.
2004) can be expressed as follows:
  10727 ± 482  
D silica
O2
melt
=exp  − (19.22 ± 0.39 ) –   (57)
  T 
where D is in m2/s, T is in K and errors are given at the 2σ level. The activation energy for
molecular O2 diffusion is 89±4 kJ/mol, greater than that for Ne diffusion but slightly smaller
than that for H2Om diffusion. Network oxygen diffusivities (such as 18O diffusivities) in silica
melts are 5 to 9 orders of magnitude lower than molecular O2 diffusivity at 900-1700 K (Fig.
1A in Lamkin et al. 1992) and the activation energy is much higher.
To evaluate the role of molecular O2 in transporting oxygen in natural silicate melts,
we compare 2DO2[O2] and DO[O] where O2 means molecular oxygen and O means network
oxygen, and the factor 2 is because each molecular O2 carries two oxygen atoms. Because
molecular O2 solubility in silica melt is about 10−6 mol/m3 at 1 Pa of O2 pressure (Norton 1961),
the solubility at a typical oxygen fugacity of 10−7 Pa in natural silicate melt would be of the
order 10−13 mol/m3. The network oxygen concentration in silica melt is about 76560 mol/m3.
Hence the concentration ratio of 2[O2]/[O] is about 3x10−18. Because molecular O2 diffusivity is
no more than 9 orders of magnitude greater than network oxygen diffusivity, at typical oxygen
fugacities of natural silicate melts, the ratio of 2DO2[O2] to DO[O] is smaller than 3×10−9. Hence,
the contribution from molecular oxygen diffusion to oxygen transport is negligible in natural
silicate melts. For molecular O2 to play a significant role in transporting network oxygen, an
oxygen fugacity of about 100 Pa or greater is needed.
figure 28. Molecular O2 diffusivities in silica melts. Susa et al. (1990) and Kajihara et al. (2005) each
contain two data sets for slightly different silica melts; these are plotted separately.
Fig. 28. Molecular O2 diffusivities in silica melts. Susa et al. (1990) and Kajihara et al. (2005) each contain two
data sets for slightly different silica melts; these are plotted separately.
SUMMARY AND CONClUSIONS

Through decades of research, fairly extensive experimental data have been obtained on
H2O, CO2, and oxygen diffusion in silicate melts. The variation of diffusivities with temperature,
pressure, and melt composition (including water content) has led us to fundamental understanding
of diffusion mechanisms and melt structure. H2O diffusion and CO2 diffusion are both dominated
by the diffusion of the respective molecular species. The diffusivity of these two major volatile
components increases with temperature and water content, but decreases with pressure. Total
CO2 diffusivity increases exponentially with H2Ot, whereas H2Ot diffusivity first increases with
H2Ot proportionally, and then exponentially. While CO2 diffusivity does not depend much on
melt composition except for the H2Ot concentration, H2O diffusivity increases with decreasing
silica content at superliquidus temperatures, but the trend at intermediate temperatures is less
well defined. H2O diffusivity also increases slightly with increasing alkalinity. H2O speciation
causes proportionality between H2Ot diffusivity and H2Ot concentration at low H2Ot, whereas
CO2 speciation leads to independence of total CO2 diffusivity on CO2 concentration, and
also causes total CO2 diffusivity to be independent of the anhydrous melt composition. One
complexity that needs to be examined in detail in the future is the possible role of molecular H2
diffusion in transporting the hydrous component under reducing conditions.
Oxygen chemical diffusivity depends on the specific diffusion mechanisms and more
systematic studies are necessary to understand the various chemical oxygen diffusivities. In the
presence of H2O, oxygen “self” diffusion is often controlled by H2O diffusion. In typical natural
silicate melts, especially rhyolite, dacite and andesite melts, 18O diffusion is often carried by
H2O diffusion, and hence can be predicted from experimental H2O diffusivities that have been
well characterized in rhyolite and dacite melts. Predicted 18O diffusivities in hydrous rhyolite
and dacite melts are listed in tables and can be many orders of magnitude greater than the Eyring
diffusivity. Only in fairly dry basalt melts (such as MORB and OIB), would 18O diffusion occur
through true oxygen self-diffusion, in which case the 18O self-diffusivity is not much different
from the Eyring diffusivity. 18O flux carried by CO2 chemical diffusion is also quantitatively
evaluated, and the conclusion is that CO2 diffusion does not significantly enhance 18O diffusion
except under very special conditions. The contribution from molecular O2 diffusion is negligible
to network oxygen diffusion under oxygen fugacity encountered in natural silicate melts.
ACKNOwleDgMeNTS
This research is supported by NSF grants EAR-0711050 and EAR-0838127 and NASA
grant NNX10AH74G. Huaiwei Ni acknowledges financial support by the visitors program
of Bayerisches Geoinstitut. We thank Don Baker and an anonymous reviewer for careful and
constructive reviews, and Harald Behrens and Daniele Cherniak for comments.
RefeReNCeS
Acocella J, Tomozawa M, Watson EB (1984) The nature of dissolved water in sodium silicate glasses and its
effect on various properties. J Non-Cryst Solids 65:355-372
Anovitz LM, Elam JM, Riciputi LR, and Cole DR (2004) Isothermal time-series determination of the rate of
diffusion of water in Pachuca obsidian. Archaemetry 46:301-326
Anovitz LM, Ricuputi LR, Cole DR, Fayek M, Elam JM (2006) Obsidian hydration: a new paleothermometer.
Geology 34:517-520
Anovitz LM, Cole DR, Fayek M (2008) Mechanisms of rhyolitic glass hydration below the glass transition. Am
Mineral 93:1166-1178
Baker DR, Freda C, Brooker RA, Scarlato P (2005) Volatile diffusion in silicate melts and its effects on melt
inclusions. Ann Geophys 48:699-717
Barrer RM (1941) Diffusion in and through solids. Cambridge Univ Press Cambridge.
Behrens H (2006) Water diffusion in silicate glasses and melts. Adv Sci Tech 46:79-88
220 Zhang & Ni
Behrens H, Nowak M (2003) Quantification of H2O speciation in silicate glasses and melts by IR spectroscopy
– in situ versus quench techniques. Phase Transit 76:45-61
Behrens H, Yamashita S (2008) Water speciation in hydrous sodium tetrasilicate and hexasilicate melts:
constraints from high temperature NIR spectroscopy. Chem Geol 256:306-315
Pertrol 157:765-780
Behrens H, Zhang Y, Xu Z (2004) H2O diffusion in dacitic and andesitic melts. Geochim Cosmochim Acta
68:5139-5150
Blank JG (1993) An experimental investigation of the behavior of carbon dioxide in rhyolitic melt. PhD
Dissertation. California Institute of Technology, Pasadena
Blank JG, Brooker RA (1994) Experimental studies of carbon dioxide in silicate melts: solubility, speciation,
and stable carbon isotope behavior. Rev Mineral 30:157-186
Botcharnikov RE, Behrens H, Holtz F (2006) Solubility and speciation of C-O-H fluids in andesitic melt at
T=1100-1300°C and P=200 and 500 MPa. Chem Geol 229:125-143
Bottinga Y, Javoy M (1990) MORB degassing: Bubble growth and ascent. Chem Geol 81:255-270
Burn I, Roberts JP (1970) Influence of hydroxyl content on the diffusion of water in silica glass. Phys Chem
Glasses 11:106-114
Canil D, Muehlenbachs K (1990) Oxygen diffusion in an Fe-rich basalt melt. Geochim Cosmochim Acta
54:2947-2951
Carmichael ISE, Ghiorso MS (1990) The effect of oxygen fugacity on the redox state of natural liquids and their
crystallizing phases. Rev Mineral 24:191-212
Chekhmir AS, Persikov ES, Epel’baum MB, Bukhtiyarov PG (1985) Hydrogen transport through a model
magma. Geochem Int 22(8):61-65
Cherniak DJ, Hervig R, Koepke J, Zhang Y, Zhao D (2010) Analytical methods in diffusion studies. Rev
Cockram DR, Haider Z, Roberts GJ (1969) The diffusion of “water” in soda-lime glass within and near the
transformation range. Phys Chem Glasses 10:18-22
Cole DR, Chakraborty S (2001) Rates and mechanisms of isotopic exchange. Rev Mineral Geochem 43:83-223
Coles JN, Long JVP (1974) Ion-microprobe study of self-diffusion of Li+ in lithium-fluoride. Philos Mag
29:457-471
Crank J (1975) The Mathematics of Diffusion. 414 p. Clarendon Press Oxford
De Berg KC, Lauder I (1980) Oxygen tracer diffusion in a potassium silicate glass above the transformation
temperature. Phys Chem Glasses 21:106-109
Delaney JR, Karsten JL (1981) Ion microprobe studies of water in silicate melts: concentration-dependent water
diffusion in obsidian. Earth Planet Sci Lett 52:191-202
Dingwell DB, Webb SL (1990) Relaxation in silicate melts. Eur J Mineral 2:427-449
Dixon JE, Stolper EM, Delaney JR (1988) Infrared spectroscopic measurements of CO2 and H2O in Juan de
Fuca Ridge basaltic glasses. Earth Planet Sci Lett 90:87-104
Dixon JE, Clague DA, Wallace PJ, Poreda RJ (1997) Volatiles in alkalic basalts from the North Arch volcanic
field, Hawaii: extensive degassing of deep submarine-erupted alkalic series lavas. J Petrol 38:911-939
Doremus RH (1969) The diffusion of water in fused silica. In: Reactivity of Solids. Mitchell JW, Devries RC,
Roberts RW, Cannon P (eds) Wiley New York, p 667-673
Doremus RH (1973) Glass Science. Wiley New York
Doremus RH (1975) Interdiffusion of hydrogen and alkali ions in a glass surface. J Non-Cryst Solids 19:137-144
Drury T, Roberts JP (1963) Diffusion in silica glass following reaction with tritiated water vapor. Phys Chem
Glasses 4:79-90
Dunn T (1982) Oxygen diffusion in three silicate melts along the join diopside-anorthite. Geochim Cosmochim
Acta 46:2293-2299
Dunn T (1983) Oxygen chemical diffusion in three basaltic liquids at elevated temperatures and pressures.
Dunn T, Scarfe CM (1986) Variation of the chemical diffusivity of oxygen and viscosity of an andesite melt with
pressure at constant temperature. Chem Geol 54:203-215
Ernsberger FM (1980) The role of molecular water in the diffusive transport of protons in glasses. Phys Chem
Glasses 21:146-149
Farver JR (2010) Oxygen and hydrogen diffusion in minerals. Rev Mineral Geochem 72:447-507
Fine F, Stolper EM (1985) The speciation of carbon dioxide in sodium aluminosilicate glasses. Contrib Mineral
Petrol 91:105-121
Fogel RA, Rutherford MJ (1990) The solubility of carbon dioxide in rhyolitic melts: a quantitative FTIR study.
Am Mineral 75:1311-1326
Frank RC, Swets D.E, Lee RW (1961) Diffusion of Ne isotopes in fused quartz. J Chem Phys 35:1451-1459
Freda C, Baker DR, Romano C, Scarlato P (2003) Water diffusion in natural potassic melts. Geol Soc Spec Publ
213:53-62
Friedman I, Long W (1976) Hydration rate of obsidian. Science 191:347-352
Gaillard F, Scaillet B, Pichavant M (2002) Kinetics of iron oxidation-reduction in hydrous silicic melts. Am
Mineral 87:829-837
Gaillard F, Schmidt BC, Mackwell S, McCammon C (2003a) Rate of hydrogen-iron redox exchange in silicate
melts and glasses. Geochim Cosmochim Acta 67:2427-2441
Gaillard F, Pichavant M, Mackwell S, Champallier R, Scaillet B, McCammon CA (2003b) Chemical transfer
during redox exchanges between H2 and Fe-bearing silicate melts. Am Mineral 88:308-315
Gardner JE, Hilton M, Carroll MR (2000) Bubble growth in highly viscous silicate melts during continuous
decompression from high pressure. Geochim Cosmochim Acta 64:1473-1483
Giordano D, Russel JK, Dingwell D (2008) Viscosity of magmatic liquids: a model. Earth Planet Sci Lett
271:123-134
Gurenko AA, Belousov AB, Trumbull RB, Sobolev AV (2005) Explosive basaltic volcanism of the Chikurachki
Volcano (Kurile arc, Russia): insight on pre-eruptive magmatic conditions and volatile budget revealed
from phenocryst-hosted melt inclusions and groundmass glasses. J Volcanol Geotherm Res 147:203-232
Haller W (1963) Concentration-dependent diffusion coefficient of water in glass. Phys Chem Glasses 4:217-220
Hauri EH (2002) SIMS analysis of volatiles in silicate glasses, 2. isotopes and abundances in Hawaiian melt
inclusions. Chem Geol 183:115-141
Hetherington G, Jack KH (1964) The oxidation of vitreous silica. Phys Chem Glasses 5:147-149
Hofmann AW (1974) Strontium diffusion in a basalt melt and implications for Sr isotope geochemistry and
geochronology. Carnegie Inst Wash Yearbook 73:935-941
Houser CA, Herman JS, Tsong IST, White WB, Lanford WA (1980) Sodium-hydrogen interdiffusion in sodium
silicate glasses. J Non-Cryst Solids 41:89-98
Hui H, Zhang Y (2007) Toward a general viscosity equation for natural anhydrous and hydrous silicate melts.
Hui H, Zhang Y, Xu Z, Del Gaudio P, Behrens H (2009) Pressure dependnce of viscosity of rhyolitic melts.
Hui H, Zhang Y, Xu Z, Behrens H (2008) Pressure dependence of the speciation of dissolved water in rhyolitic
Ihinger PD, Zhang Y, Stolper EM (1999) The speciation of dissolved water in rhyolitic melt. Geochim
International Commission on Radiation Units and Measurements (1984) International Commission on
Radiation Units and Measurements (ICRU) Report 37: Stopping Powers for Electrons and Positrons.
ICRU, Bethesda, MD
Jambon A (1979) Diffusion of water in a granitic melt: an experimental study. Carnegie Inst Wash Yearbook
78:352-355
Jambon A, Zhang Y, Stolper EM (1992) Experimental dehydration of natural obsidian and estimation of DH2O at
low water contents. Geochim Cosmochim Acta 56:2931-2935
Kajihara K, Kamioka H, Hirano M, Miura T, Skuja L, Hosono H (2005) Interstitial oxygen molecules in
amorphous SiO2. III. measurements of dissolution kinetics, diffusion coefficient, and solubility by infrared
photoluminescence. J Appl Phys 98:013529
Kajihara K, Miura T, Kamioka H, Aiba A, Uramoto M, Morimoto Y, Hirano M, Skuja L, Hosono H (2008)
Diffusion and relations of interstitial oxygen species in amorphous SiO2: a review. J Non-Cryst Solids
354:224-232
Kajihara K, Miura T, Kamioka H, Hirano M, Skuja L, Hosono H (2004) Surface dissolution and diffusion of
oxygen molecules in SiO2 glass. J Ceram Soc Japan 112:559-562
Karsten JL, Holloway JR, Delaney JR (1982) Ion microprobe studies of water in silicate melts: temperature-
dependent water diffusion in obsidian. Earth Planet Sci Lett 59:420-428
Keppler H, Bagdassarov NS (1993) High-temperature FTIR spectra of H2O in rhyolite melt to 1300°C. Am
Mineral 78:1324-1327
Kohn SC, Dupree R, Smith ME (1989) A multinuclear magnetic resonance study of the structure of hydrous
albite glasses. Geochim Cosmochim Acta 53:2925-2935
222 Zhang & Ni
Lamkin MA, Riley FL, Fordham RJ (1992) Oxygen mobility in silicon dioxide and silicate glasses: a review. J
Eur Ceram Soc 10:347-367
Lanford WA, Davis K, Lamarche P, Laursen T, Groleau R, Doremus RH (1979) Hydration of soda-lime glass.
J Non-Cryst Solids 33:249-265
Lapham KE, Holloway JR, Delaney JR (1984) Diffusion of H2O and D2O in obsidian at elevated temperatures
and pressures. J Non-Cryst Solids 67:179-191
Lee RW, Frank RC, Swets DE (1962) Diffusion of hydrogen and deuterium in fused quartz. J Chem Phys
36:1062-1071
Lee RW (1963) Diffusion of hydrogen in natural and synthetic fused quartz. J Chem Phys 38:448-455
Lesher CE (1990) Decoupling of chemical and isotopic exchange during magma mixing. Nature 344:235-237
Liang Y, Richter FM, Davis AM, Watson EB (1996) Diffusion in silicate melts: I. Self diffusion in CaO-Al2O3-
SiO2 at 1500°C and 1 GPa. Geochim Cosmochim Acta 60:4353-4367
Liu Y (2003) Water in rhyolitic and dacitic melts. Ph.D. dissertation, University of Michigan
Liu Y, Behrens H, Zhang Y (2004a) The speciation of dissolved H2O in dacitic melt. Am Mineral 89:277-284
Liu Y, Zhang Y (2000) Bubble growth in rhyolitic melt. Earth Planet Sci Lett 181:251-264
Liu Y, Zhang Y, Behrens H (2004b) H2O diffusion in dacitic melt. Chem Geol 209:327-340
Liu Y, Zhang Y, Behrens H (2005) Solubility of H2O in rhyolitic melts at low pressures and a new empirical
model for mixed H2O-CO2 solubility in rhyolitic melts. J Volcanol Geotherm Res 143:219-235
Michael PJ (1988) The concentration, behavior and storage of H2O in the suboceanic upper mantle: implications
for mantle metasomatism. Geochim Cosmochim Acta 52:555-566
Moulson AJ, Roberts JP (1961) Water in silica glass. Trans Faraday Soc 57:1208-1216
Muehlenbachs K, Kushiro I (1974) Oxygen isotope exchange and equilibrium of silicates with CO2 or O2.
Carnegie Inst Wash Yearbook 73:232-236
Mungall JE (2002) Empirical models relating viscosity and tracer diffusion in magmatic silicate melts. Geochim
Navon O, Lyakhovsky V (1998) Vesiculation processes in silicate magmas. In: The Physics of Explosive
Volcanic Eruptions, Gilbert JS, Sparks RSJ (eds) Geological Society London, p 27-50
Neuville DR, Courtial P, Dingwell DB, Richet P (1993) Thermodynamic and rheological properties of rhyolite
and andesite melts. Contrib Mineral Petrol 113:572-581
Newman S, Stolper EM, Epstein S (1986) Measurement of water in rhyolitic glasses: calibration of an infrared
spectroscopic technique. Am Mineral 71:1527-1541
Newman S, Stolper EM, Stern R (2000) H2O and CO2 in magmas from the Mariana arc and back arc systems.
Geochem Geophys Geosys 1:1999GC000027
Ni H, Behrens H, Zhang Y (2009b) Water diffusion in dacitic melt. Geochim Cosmochim Acta 73:3642-3655
Ni H, Liu Y, Wang L, Zhang Y (2009a) Water speciation and diffusion in haploandesitic melts at 743-873 K and
100 MPa. Geochim Cosmochim Acta 73:3630-3641
Ni H, Zhang Y (2008) H2O diffusion models in rhyolitic melt with new high pressure data. Chem Geol 250:68-
78
Nogami M, Tomozawa M (1984a) Effect of stress on water diffusion in silica glass. J Am Ceram Soc 67:151-154
Nogami M, Tomozawa M (1984b) Diffusion of water in high silica glasses at low temperature. Phys Chem
Glasses 25:82-85
Norton FJ (1961) Permeation of gaseous oxygen through vitreous silica. Nature 191:701-701
Nowak M, Behrens H (1995) The speciation of water in haplogranitic glasses and melts determined by in situ
near infrared spectroscopy. Geochim Cosmochim Acta 59:3445-3450
Nowak M, Behrens H (1997) An experimental investigation on diffusion of water in haplogranitic melts. Contrib
Mineral Petrol 126:365-376
Nowak M, Behrens H (2001) Water in rhyolitic magmas: getting a grip on a slippery problem. Earth Planet Sci
Lett 184:515-522
Ohlhorst S, Behrens H, Holtz F, Schmidt BC (2000) Water speciation in aluminosilicate glasses and melts. In:
Applied Mineralogy in Research, Economy, Technology and Culture. Proc 6th Int Conf Appl Mineral.
Rammlmair D, Mederer J, Oberthur TH, Heimannund RB, Pentinghaus H (eds) Balkema, Rotterdam
1:193–196
Oishi Y, Terai R, Ueda H (1975) Oxygen diffusion in liquid silicates and relation to their viscosity. In: Mass
Transport Phenomena in Ceramics. Cooper A, Heuer A (eds) Plenum, p 297-310
Okumura S, Nakashima S (2004) Water diffusivity in rhyolitic glasses as determined by in situ IR spectroscopy.
Phys Chem Miner 31:183-189
Okumura S, Nakashima S (2006) Water diffusion in basaltic to dacitic glasses. Chem Geol 227:70-82
Persikov ES, Newman S, Bukhtiyarov PG, Nekrasov AN, Stolper EM (2010) Experimental study of water
diffusion in haplobasaltic and haploandesitic melts. Chem Geol 276:241-256
Pfeffer R, Ohring M (1981) Network oxygen exchange during water diffusion in SiO2. J Appl Phys 52:777-784
Poe BT, McMillan PF, Rubie DC, Chakraborty S, Yarger J, Diefenbacher J (1997) Silicon and oxygen self-
diffusivities in silicate liquids measured to 15 GPa and 2800 K. Science 276:1245-1248
Proussevitch AA, Sahagian DL (1996) Dynamics of coupled diffusive and decompressive bubble growth in
magmatic systems. J Geophys Res 101:17447-17455
Proussevitch AA, Sahagian DL (1998) Dynamics and energetics of bubble growth in magmas: analytical
formulation and numerical modeling. J Geophys Res 103:18223-18251
Reid JE, Poe BT, Rubie DC, Zotov N, Wiedenbeck M (2001) The self-diffusion of silicon and oxygen in diopside
(CaMgSi2O6) liquid up to 15 GPa. Chem Geol 174:77-86
Riciputi LR, Elam JM, Anovitz LM, Cole DR (2002) Obsidian diffusion dating by secondary ion mass
spectrometry: A test using results from Mound 65, Chalco, Mexico. J Arch Sci 29:1055-1075
Roberts GJ, Roberts JP (1964) Influence of thermal history on the solubility and diffusion of ‘water’ in silica
glass. Phys Chem Glasses 5:26-32
Roberts GJ, Roberts JP (1966) An oxygen tracer investigation of the diffusion of ‘water’ in silica glass. Phys
Chem Glasses 7:82-89
Rubie DC, Ross CR, Carroll MR, Elphick SC (1993) Oxygen self-diffuson in Na2Si2O9 liquid up to 10 GPa and
estimation of high-pressure melt viscosities. Am Mineral 78:574-582
Robie RA, Hemingway BS (1995) Thermodynamic properties of minerals and related substances at 298.15 K
and 1 bar (105 Pascals) pressure and at high temperatures. USGS Bulletin, Report 1452
Saal AE, Hauri EH, Cascio ML, Van Orman JA, Rutherford MJ, Cooper RF (2008) Volatile content of lunar
volcanic glasses and the presence of water in the Moon’s interior. Nature 454:192-196
Schulze F, Behrens H, Holtz F, Roux J, Johannes W (1996) The influence of H2O on the viscosity of a
haplogranitic melt. Am Mineral 81:1155-1165
chalcogenides. Acta Crystallogr A32:751-767
Shaw HR (1974) Diffusion of H2O in granitic liquids, I: experimental data; II: mass transfer in magma chambers.
In: Geochemical Transport and Kinetics. Hofmann AW, Giletti BJ, Yoder HS, Yund RA (eds) Carnegie Inst
Wash Publ. 634 Washington, p 139-170
Shackelford JF, Studt PL, Fulrath RM (1972) Solubility of gases in glass. II. He, Ne, and H2 in fused silica. J
Appl Phys 43:1619-1626
Shang LB, Chou IM, Lu WJ, Burruss RC, Zhang Y (2009) Determination of diffusion coefficients of hydrogen
in fused silica between 296 and 523 K by Raman spectroscopy and application of fused silica capilaries in
studying redox reactions. Geochim Cosmochim Acta 73:5435-5443
Shelby JE (1977) Molecular diffusion and solubility of hydrogen isotopes in vitreous silica. J Appl Phys
48:3387-3394
Shelby JE (1994) Protonic species in vitreous silica. J Non-Cryst Solids 179:138-147
Shelby JE (2008) A limited review of water diffusivity and solubility in glasses and melts. J Am Ceram Soc
91:703-708
Shen A, Keppler H (1995) Infrared spectroscopy of hydrous silicate melts to 1000 °C and 10 kbar: direct
observation of H2O speciation in a diamond-anvil cell. Am Mineral 80:1335-1338
Shimizu N, Kushiro I (1984) Diffusivity of oxygen in jadeite and diopside melts at high pressures. Geochim
Sierralta M, Nowak M, Keppler H (2002) The influence of bulk composition on the diffusivity of carbon dioxide
in Na aluminosilicate melts. Am Mineral 87:1710-1716
Silver L, Ihinger PD, Stolper EM (1990) The influence of bulk composition on the speciation of water in silicate
glasses. Contrib Mineral Petrol 104:142-162
Silver L, Stolper EM (1989) Water in albitic glasses. J Petrol 30:667-709
Smets J, Lommen TPA (1983) The role of molecular water in the leaching of glasses. Phys Chem Glasses
24:35-36
Sowerby JR, Keppler H (1999) Water speciation in rhyolitic melt determined by in-situ infrared spectroscopy.
Am Mineral 84:1843-1849
Sparks RSJ, Tait SR, Yanev Y (1999) Dense welding caused by volatile resorption. J Geol Soc 156:217-225
Stanton TR, Holloway JR, Hervig RL, Stolper EM (1985) Isotopic effect on water diffusivity in silicic melts.
Eos 66:1131
224 Zhang & Ni
Stolper EM (1982a) Water in silicate glasses: an infrared spectroscopic study. Contrib Mineral Petrol 81:1-17
Stolper EM (1982b) The speciation of water in silicate melts. Geochim Cosmochim Acta 46:2609-2620
Stolper EM (1989) Temperature dependence of the speciation of water in rhyolitic melts and glasses. Am
Mineral 74:1247-1257
Stolper EM, and Newman S (1994) The role of water in the petrogenesis of Mariana Trough magmas. Earth
Susa M, Shinohara H, Nagata K, Goto KS (1990) Diffusion of oxygen molecules in amorphous silica thin films
(in Japanese). J Metal Soc Japan 54:193-200
Swets DE, Lee RW, Frank RC (1961) Diffusion coefficients of helium in fused quartz. J Chem Phys 34:17-22
Tinker D, Lesher CE, Baxter GM, Uchida T, Wang Y (2004) High-pressure viscometry of polymerized silicate
melts and limitations of the Eyring equation. Am Mineral 89:1701-1708
Tinker D, Lesher CE, Hucheon ID (2003) Self-diffusion of Si and O in diopside-anorthite melt at high pressures.
Tomozawa M (1985) Concentration dependence of the diffusion coefficient of water in SiO2 glass. Am Ceram
Soc 68:C251-252
Tsong IST, Houser CA, Tong, SSC (1980) Depth profiles of interdiffusing species in hydrated glasses. Phys
Chem Glasses 21:197-198
Van der Laan SR, Zhang Y, Kennedy A, Wylie PJ (1994) Comparison of element and isotope diffusion of K and
Vetere F, Behrens H, Holtz F, Neuville DR (2006) Viscosity of andesitic melts - new experimental data and a
revised calculation model. Chem Geol 228:233-245
Wallace PJ (2005) Volatiles in subduction zone magmas: concentrations and fluxes based on melt inclusion and
volcanic gas data. J Volcanol Geotherm Res 140:217-240
Wallace PJ, Dufek J, Anderson AT, Zhang Y (2003) Cooling rates of Plinian-fall and pyroclastic-flow deposits in
the Bishop Tuff: inferences from water speciation in quartz-hosted glass inclusions. Bull Volcanol 65:105-
123
Wang L, Essene EJ, Zhang Y (1999) Mineral inclusions in pyrope crystals from Garnet Ridge, Arizona, USA:
implications for processes in the upper mantle. Contrib Mineral Petrol 135:164-178
Wang H, Xu Z, Behrens H, Zhang Y (2009) Water diffusion in Mount Changbai peralkaline rhyolitic melt.
Contrib Mineral Pertrol 158:471-484
Wang L, Zhang Y, Essene EJ (1996) Diffusion of the hydrous component in pyrope. Am Mineral 81:706-718
Wasserburg GJ (1988) Diffusion of water in silicate melts. J Geol 96:363-367
Watson EB (1991) Diffusion of dissolved CO2 and Cl in hydrous silicic to intermediate magmas. Geochim
Watson EB, Sneeringer MA, Ross A (1982) Diffusion of dissolved carbonate in magmas: experimental results
and applications. Earth Planet Sci Lett 61:346-358
Wendlandt RF (1991) Oxygen diffusion in basalt and andesite melts: experimental results and discussion of
chemical versus tracer diffusion. Contrib Mineral Pertrol 108:463-471
Whittington AG, Hellwig BM, Behrens H, Joachim B, Stechern A, Vetere F (2009) The viscosity of hydrous
dacitic liquids: implications for the rheology of evolving silicic magmas. Bull Volcanol 71:185-199
Withers AC, Behrens H (1999) Temperature-induced changes in the NIR spectra of hydrous albitic and rhyolitic
glasses between 300 and 100 K. Phys Chem Minerals 27:119-132
Withers AC, Zhang Y, Behrens H (1999) Reconciliation of experimental results on H2O speciation in rhyolitic
glass using in situ and quenching techniques. Earth Planet Sci Lett 173:343-349
Workman RK, Hauri EH, Hart SR, Wang J, Blusztajin J (2006) Volatile and trace elements in basaltic glass from
Samoa: implications for water distribution in the mantle. Earth Planet Sci Lett 241:932-951
Zhang Y (1994) Reaction kinetics, geospeedometry, and relaxation theory. Earth Planet Sci Lett 122:373-391
Zhang Y (1999a) A criterion for the fragmentation of bubbly magma based on brittle failure theory. Nature
402:648-650
Zhang Y (1999b) H2O in rhyolitic glasses and melts: measurement, speciation, solubility, and diffusion. Rev
Geophys 37:493-516
Zhang Y (2008) Geochemical Kinetics. Princeton University Press NJ
Zhang Y (2009) Degassing of lunar basalts. Geochim Cosmochim Acta 73:A1514 (abstr)
Zhang Y, Behrens H (1998) Resolving the controversy between quenched and in situ H2O speciation in silicate
melts and glasses. Eos Abstr 79:W123-W124
Zhang Y, Belcher R, Ihinger PD, Wang L, Xu Z, Newman S (1997a) New calibration of infrared measurement
of water in rhyolitic glasses. Geochim Cosmochim Acta 61:3089-3100
Zhang Y, Jenkins J, Xu Z (1997b) Kinetics of the reaction H2O + O = 2OH in rhyolitic glasses upon cooling:
geospeedometry and comparison with glass transition. Geochim Cosmochim Acta 61:2167-2173
Zhang Y, Stolper EM (1991) Water diffusion in basaltic melts. Nature 351:306-309
Zhang Y, Stolper EM, Ihinger PD (1995) Kinetics of reaction H2O + O = 2OH in rhyolitic glasses: preliminary
results. Am Mineral 80:593-612
Acta 55:441-456
Zhang Y, Xu Z (1995) Atomic radii of noble gas elements in condensed phases. Am Mineral 80:670-675
Zhang Y, Xu Z, Behrens H (2000) Hydrous species geospeedometer in rhyolite: improved calibration and
application. Geochim Cosmochim Acta 64:3347-3355
Zhang Y, Xu Z, Zhu M, Wang H (2007) Silicate melt properties and volcanic eruptions. Rev Geophys 45:RG4004,
doi:10.1029/2006RG000216
Vol. 72 pp. 311-408, 2010 8
Diffusion Data in Silicate Melts

Youxue Zhang
University of Michigan
Ann Arbor, Michigan 48109-1005, U.S.A.
youxue@umich.edu
Huaiwei Ni
Bayerisches Geoinstitut
University of Bayreuth
D-95440 Bayreuth, Germany
Yang Chen
Laboratoire Magmas et Volcans, OPGC
Universite Blaise Pascal, CNRS, IRD
5 Rue Kessler, 63038 Clermont-Ferrand Cedex, France
INtroDuCtIoN
Atomic scale diffusion in silicate melts plays a critical role in a variety of magmatic
processes, as long recognized by geologists (Bowen 1921). Geochemists have investigated
diffusion in silicate melts (including stable liquids and supercooled liquids and glasses) using a
variety of methods as micro-analytical tools becoming available (e.g., Medford 1973; Jambon
and Carron 1976; Magaritz and Hofmann 1978a,b; Hofmann 1980; Shimizu and Kushiro 1984;
Zhang et al. 1991a; Koepke and Behrens 2001; Saal et al. 2008). The investigations were aimed
at understanding the mechanisms of diffusion, the effect of temperature, pressure, H2O content,
fO2, and more generally melt composition on the diffusivities, and to measure the diffusivity
values for geological applications. A large database is now available. This chapter reviews
diffusion data relevant to magmatic systems. The large body of literature on diffusion in simple-
system melts, of interests to glass and ceramic scientists, is not included in this review, nor are
the calculated diffusivities from ab initio and molecular dynamics methods.
This review covers literature diffusion data (except multicomponent diffusion matrix data,
and those of H, C, O and noble gases) in multicomponent silicate melts (the term melts as used
in this chapter also includes glasses), including melts with mineral compositions and including
glasses. This chapter complements the other chapters on diffusion in silicate melts, which cover
H, C and O diffusion (Zhang and Ni 2010, this volume), noble gas diffusion (Behrens 2010,
this volume), theories and models of self and tracer diffusion (Lesher 2010, this volume), and
multicomponent diffusion (Liang 2010, this volume).
We have compiled experimental data on self, tracer and effective binary diffusion in Online
Supplementary Table 1 (found at http://www.minsocam.org/MSA/RIM), with about 3600 entries
of individual diffusivity values, together with experimental conditions and melt compositions.
Furthermore, the parameters in the Arrhenius equations (lnD = lnD0 – E/(RT), where D is
diffusivity, D0 is the preexponential factor, E is the activation energy, R is the gas constant, and
T is the absolute temperature; see Zhang 2010, this volume) in the literature on silicate melts
1529-6466/10/0072-0008$10.00 DOI: 10.2138/rmg.2010.72.8

312 Zhang, Ni, Chen
for geologic applications to predict such diffusivities are listed in Online Supplementary Table
2 (found at http://www.minsocam.org/MSA/RIM) in the form of lnD0 (where D0 is in m2/s) and
E/R (with unit of K). In the compiling process and in preparing this review, we also compare,
assess and model diffusivity data to obtain new predictive equations. Some experimental data
that are deemed erroneous or inaccurate (see discussion below) have been removed from the
main part of the database and placed at the end of Online Supplementary Table 1. Although we
made efforts, there may be omissions and inaccuracies in the database.
Although literature Arrhenius relations are listed in Online Supplementary Table 2, in this
review, when possible, we make new efforts to model the available diffusion data (often through
combining data from different works) as a function of temperature, pressure, H2O (total H2O)
content, as well as other compositional parameters. Furthermore, to help readers to choose
the “best” equations, we also recommend some equations based on our evaluations. These
equations are presented in the text of this chapter. In these equations, T is temperature in K,
P is pressure in GPa, D is diffusivity in m2/s, w is wt% of H2o (w = 1 means 1 wt%), and
errors are given at 2s level. As will be seen (and evaluated) later, a general empirical model to
predict cation diffusivities in silicate melts is available (Mungall 2002), and often it can predict
diffusivities to within 1 logD unit (or 2.3 lnD units), but sometimes the error is larger. Hence,
we will only present newly developed relations when the relation can predict diffusivities with
a precision better than 1 logD unit.
terminology
For different kinds of diffusion, we follow the definitions in Zhang (2008, 2010). Briefly,
self diffusion (SD) means the diffusion of an isotope in an iso-chemical system in which there
are no elemental concentration gradients and the only difference is in isotope fractions. Tracer
diffusion (TD) refers to diffusion of a tracer (either radioactive tracer or a trace element; here
“trace” means no more than 1 wt%) in a multicomponent system with the same base composi-
tion; that is, the concentration variation of all other components is due to the dilution effect of the
tracer. (Note that there are other definitions of tracer diffusivities, Lasaga 1998; Lesher 2010.)
Natural silicate melts are multicomponent systems (consisting of three or more components).
Complete quantitative treatment of multicomponent diffusion requires the use of a diffusivity
matrix (Zhang 2010, this volume) and is reviewed by Liang (2010) in this volume. However,
because such diffusivity matrices are very difficult to obtain, effective binary diffusion (Cooper
1968) is often adopted as an approximate treatment. In this treatment, the component of interest
is considered as one component, and all the other components combined are considered as the
other component. In other words, the off-diagonal cross diffusivities in a diffusion matrix are
not considered. The diffusivity is referred to as the effective binary diffusivity. Effective binary
diffusion approach works well in some cases (e.g., when the concentration gradient is largely
in the component of interest) but does not work well in other cases (e.g., when there is uphill
diffusion). At least three different cases of effective binary diffusion may be distinguished:
(1) Binary interdiffusion. When the concentration gradients are essentially between two
components in a multicomponent system (e.g., K-Na interdiffusion in alkali feldspar
melts), the diffusion is referred to as binary interdiffusion (ID). In such a case, the
diffusion of these two components in a multicomponent diffusion is similar to simple
binary diffusion and can be treated well using the effective binary approach. The other
components often show uphill diffusion and cannot be treated well by the effective
binary approach.
(2) The first type of effective binary diffusion. When the compositional variation is mostly
in one component of interest and the concentrations of other components vary only
due to the dilution effect of adding this component, the diffusion of this component
can be treated well by the effective binary approach and is referred to as the first type
Diffusion Data in Siliate Melts 313
of effective binary diffusion, with the acronym FEBD (the same acronym may also
mean the first type of effective binary diffusivity). In this definition, FEBD becomes
tracer diffusion when the concentration of the component of interest is below 1 wt%.
In other words, FEBD includes tracer diffusion. In a sense, tracer diffusion may also
be referred to as effective binary diffusion because whenever there is chemical activity
in one component, there would be chemical activity gradients in other components
(Gibbs-Duhem equation).
(3) The second type of effective binary diffusion. When there are concentration gradients
in several major components (such as a basalt-dacite diffusion couple), the diffusion
of a component is referred to as the second type of effective binary diffusion, or SEBD
(meaning either second type of effective binary diffusion or diffusivity). It is expected
that SEBD depends on the concentration gradients of major components (Cooper
1968) and hence cannot be directly compared with the other types of diffusivities.
Different cases of SEBD may be distinguished. When the concentration gradient
of the component is the greatest in the system, or only slightly smaller than the
greatest, then SEBD of the component is not dominated by cross diffusivities and is
hence more meaningful. When, on the other hand, the concentration gradient of the
component is small compared to other components, the cross diffusion effect might
dominate the diffusion of the component (even if it does not show uphill diffusion).
Consequently, the SEBD value does not indicate the intrinsic mobility and has only
limited applicability. In our discussion below, when possible, we will make efforts to
differentiate cases whether or not the SEBD is dominated by cross diffusion.
Regarding the different types of effective binary diffusion data, FEBD (including tracer
diffusivity) data are numerous, self-consistent, and widely applicable. Interdiffusivity data are
also expected to be self-consistent at the same bulk melt composition although less data are
available. SEBD data are less consistent and it is necessary to identify the specific conditions
(such as the sense of concentration gradients qualified as, e.g., olivine dissolution in basalt melts,
in addition to temperature, pressure and bulk composition) for SEBD values to be applicable.
General comments about experimental methods to extract diffusivities
Early experimental studies of diffusion (especially in the 1970’s and 1980’s) mostly
used radioactive tracers reflecting the technological availability at the time. The experimental
method is referred to as the thin-source (or thin-film) technique (Watson and Dohmen 2010,
this volume). A glass cylinder is first prepared and a base surface is polished. Then on the
polished base, a very thin layer of solution, which contains the radioactive tracer of interest, is
deposited. After drying, the sample is heated up for some duration so that the radioactive tracer
diffuses into the sample. The heating duration must be short enough so that diffusion does not
affect the other end. The sample is then quenched. The diffusion profile can be measured in two
ways. One is to cut a section of the sample and measure the concentration (often it is relative
concentration) profile of the radioactive tracer using β-track mapping (e.g., Watson et al. 1982).
The equation to describe the concentration variation as a function of distance is the solution to
the diffusion problem of an instantaneous plane source in half space (Crank 1975):
2
C = C0 e - x /(4 Dt )
(1)
where x is distance measured from the surface, t is time, D is the diffusion coefficient, and
C0 is the concentration at the surface (x = 0), which can be expressed as M/(πDt)1/2 with M
being the total deposited mass of the radioactive tracer per unit area. Because radioactivity (A)
rather than concentration (C) is often measured and A is proportional to C, Equation (1) can
be written as:
2
A = A0e - x /(4 Dt )
(2)
314 Zhang, Ni, Chen
The fitting procedure can be either a nonlinear fit of A versus x, or a linearized fit of lnA versus
x2. The nonlinear fit is preferable if available. The linearized fit is more visual but requires
careful handling of measurement errors (Zhang 2008, 2010). In the early years, it was thought
that β-particle range is on the order of several micrometers, and the β-track mapping method
was thought to be accurate. However, it was found later that the β-particle range is actually two
orders of magnitude greater (International Commission on Radiation Units and Measurements
1984; Mungall 2002), leading to a significant convolution effect of a few hundred micrometers.
Hence, diffusivities based on β-track mapping method were often overestimated unless the
profiles are at least a few mm long.
In the second method, the total residual radioactivity rather than activity at a single
location is measured by successively removing thin layers of the sample starting from the
original deposition surface (x = 0) (Jambon and Carron 1976). Hence, each measurement is the
integration of the activity from x to infinity. Denoting the integrated activity as B, then
∞ ∞ x
∫=
Cdx′ ∫
2
=B C0e - x ' /(4=
Dt )
dx′ B0erfc (3)
x x
4 Dt
The diffusivity can be obtained through either a nonlinear fit of B versus x, or a linearized inverse
error function fit of erfc−1(B/B0) versus x. Again, the linearized fit requires careful consideration
of measurement errors (even more so than in the case of fitting Eqn. 2).
The most common experimental method to obtain all kinds of diffusivities is the diffusion
couple method (e.g., Medford 1973), in which two cylinders of roughly equal diameter and
height are juxtaposed. Each portion of the diffusion couple is initially uniform in composition,
and the interface between them is kept horizontal to minimize convection. At high temperatures,
the two halves are melted and diffusion generates a concentration profile. The initial two halves
may be either two glass cylinders or two layers of rock powders. Using glass cylinders would
produce more reliable data because (i) the powders can shift and penetrate each other during
compression, (ii) air in pore space of powders must be dissolved during an experiment (otherwise
air bubbles would interfere with diffusion), and (iii) upon melting at high temperature, crystals
in rock powders need some time to be diffusively homogenized with the rest of the melt. The
experimental heating duration must be long enough to produce a measurably long profile and
short enough so that diffusion does not reach either end of the diffusion couple. After quenching
to glass, concentration profiles are measured by microbeam analyses (e.g., electron microprobe,
ion microprobe, or infrared spectroscopy; Cherniak et al. 2010, this volume). Assuming D does
not vary across a profile, the resulting profile is fit by an error function (Crank 1975)
CL + CU CU - CL x
=C + erf (4)
2 2 4 Dt
where x is distance measured from the Matano interface (the position of the Matano interface
is defined by mass balance so that the diffusant loss from one side is equal to the diffusant gain
on the other side), t is time, C is concentration, CL and CU are the concentrations in the initial
lower half and upper half. Again, the nonlinear fitting directly using Equation (4) is preferable
to linearized fitting.
Tracer diffusivity and FEBD of a volatile component can also be obtained by sorption or
desorption methods. A glass cylinder with uniform initial composition is heated up to high
temperature in a gas medium so that the surface concentration of the volatile component of
interest is constant. In sorption, the volatile component diffuses from the gas medium into the
melt. In desorption, the volatile diffuses from the melt into the gas medium. If D is constant
across the profile and the experimental duration is not too long, the concentration profile is as
follows:
x
C =Ci + (Cs - Ci )erfc (5)
4 Dt
where x is distance measured from the surface, Ci is the initial concentration in the melt (glass)
cylinder, and Cs is the surface concentration.
Another method to obtain SEBD is through diffusive crystal dissolution experiments.
In such experiments, a single crystal wafer and a glass cylinder are juxtaposed similar to a
diffusion couple (but this is not a diffusion couple per se because there is negligible diffusion
in the crystal wafer). To avoid convection in the melt, the mineral-melt interface must be
horizontal and the interface melt must be gravitationally stable (Zhang et al. 1989; Shaw
2000). At high temperatures when the crystal becomes undersaturated in the melt, the crystal
dissolves, and the extra mass at the crystal-melt interface diffuses into the melt. For major
equilibrium-determining components (Zhang et al. 1989; Chen and Zhang 2008, 2009), the
resulting diffusion profile developed in the melt layer can be treated as an effective binary
process and fit by:
 x-L 
(Cs - Ci )erfc  
C= Ci +  4 Dt  (6)
 -L 
erfc  
 4 Dt 
where x is distance measured from the crystal-melt interface, Ci is the initial concentration
in the melt (glass), Cs is the concentration in the melt at the crystal-melt interface, and L
is the melt growth distance (crystal dissolution distance multiplied by the ratio of crystal
density to melt density). The early experiments (1980’s) often packed a few crystal spheres
inside glass powder. This experimental design leads to the sinking or rising of the crystal
grain in the melt, inducing convection. Even in the absence of crystal motion, non-horizontal
melt-crystal interfaces can result in convective removal of the interface melt. In such cases,
extracted diffusivities are often incorrect (see discussion in Zhang et al. 1989). These data are
not included in review below.
Grouping of the elements
Our review of self, tracer and effective binary diffusivities of all elements (except for H,
C, O and noble gases which are discussed by Zhang and Ni 2010; Behrens 2010, this volume)
in geologically relevant silicate melts are arranged in 15 groups according to their chemical
properties and positions in the Periodic Table. It is noteworthy that no data is currently available
for a number of elements (indicated by italics below).
(1) the alkalis (Li, Na, K, Rb, Cs, Fr) (9) Ti, Zr, Hf
(2) the alkali earth (Be, Mg, Ca, Sr, Ba, Ra) (10) V, Nb, Ta
(3) B, Al, Ga, In, Tl (11) Cr, Mo, W
(4) Si, Ge, Sn, Pb (12) Mn, Fe, Co, Ni, Cu, Zn
(5) N, P, As, Sb, Bi (13) Tc, Ru, Rh, Pd, Ag, Cd
(6) S, Se, Te, Po (14) Re, Os, Ir, Pt, Au, Hg
(7) F, Cl, Br, I, At (15) Ac, Th, Pa, U
(8) Sc, Y, REE
Data compilation
Diffusion data in geologically relevant silicate melts are compiled in Online Supplementary
Table 1, including experimental temperature, pressure, melt compositions and other conditions.
If the melt is said to be of a mineral composition (such as jadeite melt) but no composition is
316 Zhang, Ni, Chen
reported in the paper, then the ideal mineral composition is assumed for the melt in Online
Supplementary Table 1. Comments (in the “comments” column) are added for data of suspect
quality. If the data are known to be incorrect or have large uncertainties, the data are removed to
the end of the Table, and are excluded in quantitative modeling. The excluded data are mostly
one of the following three categories:
(1) In some mineral dissolution experiments, convection is known or likely to be
present, but the effective binary diffusivity values were extracted without correctly
considering the convection effect. These data include Brearley and Scarfe (1986),
Finnila et al. (1994), Wolf and London (1994), Linnen et al. (1995, 1996), and Baker
and Rutherford (1996). Other diffusion data, including Harrison and Watson (1984)
and Baker et al. (2002), might also have suffered from this complication.
(2) In some early radioactive tracer diffusion studies, the concentration profiles were
determined using β-track mapping (e.g., 45Ca diffusion data of Hofmann and
Magaritz 1977; diffusion data in Watson 1979a,b, 1981, 1991a; Watson et al. 1982).
As discussed earlier, these data are compromised by the long β-particle range.
Hence, concentration profiles based on β-track mapping are placed in the end of the
Table as inaccurate data. Radioactive tracer diffusivities using the residual activity
measurements with serial sectioning are not affected.
(3) If the error of a diffusivity value is large (about an order of magnitude), or if the
diffusivity value is an order of magnitude estimate, the data are not included.
Furthermore, sometimes a whole data set seems to have large errors. For example,
Baker (1989) determined SEBD of various elements in dry rhyolite7-dacite1 diffusion
couples at 1573 K and 1 GPa by the Boltzmann-Matano method. Si effective binary
diffusivities under the same conditions can vary by an order of magnitude, indicating
experimental or analytical difficulties (such as convection in the experiments). Baker
(1990; see Baker 2010 correction) reported more SEBD in dry rhyolite7-dacite1
diffusion couples at 1473-1873 K and 0.1 MPa and 1 GPa, including those in Baker
(1989), with similarly large scatter. van der Laan and Wyllie (1993) roughly estimated
SEBD of Na, K, Mg, Al and Si in a diffusion couple of dry rhyolite and wet basalt
melts at 1193 K and 1 GPa, which are not meant to be accurate. Data in these papers
are excluded from quantitative treatment because of the large scatter.
In the text and figures below, melt compositions will be referred to using simple terms
such as basalt, andesite, dacite, and rhyolite. To simplify the terms, granite melt, rhyolite
melt and obsidian melt as used in the literature will all be referred to as rhyolite melt. To be
more specific, compositionally different rhyolites will be distinguished as rhyolite1, rhyolite2,
HR1 (haplorhyolite1), etc. Compositions of some melts, for which the diffusion of multiple
elements has been investigated, are listed in Table 1 for easy reference. A melt often contains
some H2O, which may have a large effect on the diffusivity. Hence, effort is made in this
review to specify H2O content, but this is not straightforward. In the literature, sometimes H2O
content is not measured but simply indicated as dry. These melts are also indicated as dry in
this review, which often means H2O is ≤ 0.2 wt% (but could be higher). H2O concentrations, if
measured, can be found in Online Supplementary Tables 1 and 2. However, H2O concentrations
reported in the literature in the early years (prior to 1982 in all laboratories, and after 1982 in
some laboratories) are often inaccurate because they are based on weight loss or how much
water is added to the system without direct measurements. These problems are worse at low
H2O concentrations. For example, one sample (rhyolite15 in Table 1) was initially reported to
contain 0.38 wt% H2O (Magaritz and Hofmann 1978a,b; Jambon 1982), but it was found later
to contain only 0.11 wt% H2O (Jambon et al. 1992).
Quantification of D as a function of T, H2o, P, fo2 and melt composition

Diffusivities in silicate melts depend on temperature, H2O content, pressure, oxygen
fugacity (for elements with variable valences), and melt composition (H2O is part of the
melt composition but is often treated separately due to its large effect on diffusivity). The
dependence on temperature is the Arrhenius relation (lnD is linearly dependent on 1/T) even
when the glass transition is crossed.
The dependence on H2O is more complicated: for neutral molecular species, lnD increases
linearly with H2O (Behrens and Zhang 2001; Zhang and Behrens 2000). For most elements
reviewed in this chapter, the relation between lnD and H2O will be examined when possible.
Sometimes, the relation is linear, especially for rapidly diffusing species. As it will be seen,
sometimes, it appears that lnD depends linearly on the square root of H2O, especially for
slowly diffusing species.
The pressure effect is examined based on the data. Often, lnD depends linearly on P in
a small pressure range (a couple of GPa). The effect of oxygen fugacity for elements with
variable valences has been noted but not quantified.
The diffusivity of a given element often depends strongly on the bulk melt composition
from basalt to rhyolite. This dependence is the least understood and often it is difficult to
quantify.
In this work, we make an effort to evaluate the various dependences of diffusivities and to
fit diffusion data by empirical equations. In the fitting, clear outliers are excluded. The equations
to describe the compositional effects are tentative but are a step towards understanding how the
melt composition affects the diffusivity.
DIffuSIoN of INDIvIDual eleMeNtS

Diffusion of major elements versus minor and trace elements
During diffusion between two different melts, there are concentration gradients in major
elements (or major oxides). For example, between dacite and rhyolite melts (e.g., Baker 1991),
SiO2 concentration may vary from 64 to 76 wt%, TiO2 from 0.6 to 0.1 wt%, Al2O3 concentration
from 18 to 13 wt%, total FeO from 4.9 to 0.8 wt%, MgO from 1.7 to 0.1 wt%, CaO from 5.2
to 0.5 wt%, Na2O from 4.7 to 4.3 wt%, and K2O from 1.8 to 4.7 wt%. The concentration
gradient of one component would affect the diffusive flux (or diffusion profile) of the other
components due to cross terms in the diffusion matrix (DeGroot and Mazur 1962; Lasaga
1998; Zhang 2008). The component with the largest concentration gradient (which is SiO2 in
this example, and also in many cases in natural melt systems) would affect other components
the most. The component with the smallest concentration difference (which is Na2O in this
example) would be affected the most by other components. Hence, accurate description of
multicomponent diffusion requires accounting for the cross terms using the diffusion matrix
approach (Liang 2010, this volume). Nonetheless, it is expected that the component with the
largest concentration gradient is least affected by the cross terms, and the effective binary
approach would work roughly. As the concentration gradient of a component decreases, its
diffusion is increasingly affected by the other components, meaning that the extracted effective
binary diffusivity is less and less meaningful for applications to diffusion of the component in
a melt in general. When the cross terms dominate, producing uphill diffusion, effective binary
diffusivities cannot be extracted at all, though other approximate treatments might work (such
as the modified effective binary approach by Zhang 1993).
When there are strong major oxide concentration gradients in a diffusion couple, minor
and trace elements with intrinsic mobility (or self diffusivities) much higher (e.g., 10 or more
table 1. Composition (wt%) of some melts in diffusion studies on anhydrous basis.
318
ID Composition Sio2 tio2 al2o3 feot Mno Mgo Cao Na2o K 2o P2o5 ref
1 LP1 36.76 14 7.51 23 0.1 10.2 8.57 0.59 1
2 nephelinite 39.19 3.6 18.01 4.18 6.65 20.21 4.09 1.17 1.05 2
3 basanite 42.86 2.98 11.6 11.82 0.2 14.24 10.07 3.72 1.51 0.98 3
4 LP2 44.26 3.6 9.08 21.7 0.6 11 8.76 0.37 1
5 basalt1 45.9 3.72 16.55 11.76 0.18 6.3 9.62 3.72 1.66 0.52 4-9
6 LB1 47.1 0.88 12.6 20.2 0.25 6.31 12.8 0.2 0.02 10
7 basalt2 47.14 1.65 14.68 9.7 0.16 10.49 10.71 2.81 1.19 0.5 11
8 basalt3 47.48 3.3 14.07 12.93 7.2 10.2 2.12 0.37 12
9 HB1 47.7 16 16.4 19.2 13,14
10 basalt4 48 1.7 16.8 10.4 0.2 6.23 10.8 3.95 1.79 0.54 15
11 basalt5 48.1 1.1 15 12.2 0.2 8.1 12.2 2.3 0.1 0.1 16
12 basalt6 48.5 2.7 13.8 12.65 7.55 10.9 2.5 0.41 17
13 CMAS1 48.8 15.6 10.7 24.6 18
14 basalt7 48.88 0.92 17.5 7.77 0.16 9.6 12.22 2.36 0 0.06 3
15 basalt8 49.31 1.1 16.85 8.33 8.82 10.98 2.77 0.25 2
16 basalt9 49.53 1.14 16.97 12.33 0.21 5.14 10.22 2.39 0.42 0.16 11
17 basalt10 49.68 2.59 12.7 11.05 0.17 9.46 10.79 2.31 0.48 6-9
18 basalt11 50.23 1.94 13.61 12.06 0.22 7.05 10.75 2.73 0.17 0.1 19,20
19 limburgite 50.48 4.55 14.96 1.52 10.09 13.39 2.19 1.56 0.73 2
20 mugearite1 51.52 1.95 17.73 7.73 0.19 3.39 10.04 4.48 2.12 0.58 21
21 basalt12 51.53 2.04 15.17 9.85 0.22 5.66 10.89 3.07 0.46 22
22 NCMAS1 52.79 24.43 12.88 9.9 23
23 mugearite2 52.87 2 18.19 7.93 0.2 3.48 7.43 4.6 2.17 0.6 21
24 NMAS1 54.5 15.8 12.1 0.5 17.6 18
25 NCMAS2 54.97 31.09 9.22 4.73 23
Zhang, Ni, Chen
26 K-phonolite 55.8 0.2 22.4 2.26 0.1 0.21 2.67 6.46 9.89 24
27 leuconorite 55.99 0.52 22.96 2.81 1.42 9.26 4.89 0.97 2
28 andesite1 56.5 1.24 18 6.71 0.13 3.96 7.73 3.75 1.7 0.38 25
29 Na-phonolite 57.5 0.2 22.9 1.84 0.5 0.07 0.45 11.6 4.97 24
30 andesite2 57.76 0.85 19.14 6.28 0.11 3.45 6.27 4.31 1.79 0.2 4-9
31 andesite3 59 0 17.3 7.2 0 3.17 6.74 4.12 1.58 26
32 phonolite1 59.8 0.78 19.52 2.95 0.66 3.86 5.5 7.31 27
33 NCMAS3 60.76 23.79 6.27 4.36 4.82 23
34 andesite4 61.42 0.8 17.27 5.26 0.1 3.23 6.05 4.2 1.18 22
35 andesite5 61.47 0.53 17.02 4.64 0 1.64 5.05 7.31 1.71 28
36 andesite6 61.69 1.37 18.48 4.53 1.5 4.45 2.64 4.11 0.59 2
37 HA1 62.1 19.47 2.5 11 4.25 1 29
38 trachyte1 62.37 0.48 18.61 3.68 0.24 1.73 5.43 7.25 27
39 dacite1 63.51 0.56 17.61 4.91 0 1.68 5.24 4.67 1.77 30-32
40 trachyte2 64.03 0.71 15.92 6.53 0.29 0.45 1.36 5.94 4.5 0.13 33,34
41 dacite2 64.7 0.5 16.6 4.7 0.1 2.6 5 3.5 1.7 0.2 15
42 dacite3 64.92 0.46 17.54 3.92 0.09 2.14 5.19 3.01 2.66 0.07 11
43 nordmarkite 65.57 0.97 16.99 2.88 0.22 0.82 0.84 6.53 5.96 2
44 HR6+Na 66.15 9.62 20.6 3.64 35
45 NCMAS4 67.63 16.4 6.48 4.51 4.98 23
46 HD2 68.28 15.57 0.64 7.94 4.87 33,34
47 trachyte3 68.48 0.69 15.72 2.05 0.73 1.44 4.58 5.67 2
48 pantellerite0 68.73 0.53 8.63 9.52 0.31 0.11 0.43 6.19 4.34 0 33,34
49 dacite4 69 1.02 13.11 4.92 0.13 1 2.7 4.21 2.72 0.24 6-9
50 pantellerite1 69.2 0.44 10.75 7.8 0.3 0.11 0.56 6.32 4.14 0.02 6-9
51 HD1 69.38 0.54 16.87 3.36 3.04 3.91 3.92 36
52 pantellerite2 69.95 0.3 8.11 7.14 0.23 0.32 7.16 4.35 0.54 2
53 trondhjemite 70.94 0.23 14.99 1.35 0.72 3.46 5.96 1.44 37
54 HR1 71.55 19.1 0.63 8.75 38
55 rhyolite1 71.56 0.26 7.69 8.06 0.32 0.01 0.27 7.11 4.25 0 33,34
56 rhyolite2 72.1 0.2 14.86 2.2 0.47 2.03 4.13 3.2 12
57 rhyolite3 72.8 0 14.5 1.65 0 0.34 1.44 3.68 5.07 26
58 FaLcQ 72.8 8.37 10.75 0.07 7.14 22
59 rhyolite4 72.81 0.05 16.63 0.12 0.06 0.32 4.45 3.94 0.41 2
60 rhyolite5 73.01 0.2 12.57 2.58 0.05 0.54 0.41 5.05 4.6 39,44
61 rhyolite6 73.29 0.17 14.31 1.03 0.391 0.91 4.65 1.85 37
62 rhyolite7 73.61 0.11 12.65 0.67 0.08 0.06 0.52 6.87 4.54 0 28,33,34
63 rhyolite8 73.68 0.3 13.64 1.82 0.1 0.22 0.99 4.57 4.09 40
64 NCMAS5 74.02 8.37 3.31 9.21 5.09 23
65 HR2 74.85 15.8 5.33 4.03 41
66 HR3 75.07 15.56 5.27 4.1 41
67 HR4 75.2 15.36 5.38 4.07 41
68 rhyolite9 75.45 0.12 15.85 0.55 0.16 0.48 2.84 1.8 37
69 rhyolite10 75.59 0.05 12.27 0.85 0.08 0.02 0.29 4.43 4.52 2
70 rhyolite11 75.64 0.1 13.62 0.7 0.07 0.35 3.34 2.26 37
71 HR5 75.81 14.6 5.45 4.14 41
72 rhyolite12 76.01 0.11 13 0.7 0.09 0.52 3.65 4.81 42,43
73 rhyolite13 76.12 0.07 13.31 0.91 0.06 0.21 1.14 4.62 2.78 0.04 11
74 rhyolite14 76.37 0.11 13.35 0.79 0.06 0.07 0.51 4.31 4.67 30-32
75 rhyolite15 76.87 0.07 12.2 0.94 0.05 0.05 0.3 4.38 4.4 6-9,44
76 rhyolite16 77 0.22 12.5 3.18 0.06 0.07 1.66 4.13 2.75 26
77 rhyolite17 77.9 0.1 12.6 0.8 0.06 0.62 3.88 4.03 45
78 HR6 78.92 9.5 6.05 5.53 46
79 HR7(HPG8) 79.43 11.64 4.77 4.16 35
80 HR8 79.7 11.55 3.92 4.15 45
Diffusion Data in Siliate Melts
81 HR9 80.55 11.05 4.38 4.02 46

82 NCMAS6 80.68 14.12 5.2 23
The compositions are listed on the anhydrous basis and arranged by SiO2 wt%. CMAS: CaO-MgO-Al2O3-SiO2 system; NMAS: Na2O-MgO-Al2O3-SiO2 system; NCMAS: Na2O-CaO-MgO-
Al2O3-SiO2 system; FaLcQ: fayalite-leucite-quartz system (Lesher and Walker 1986); HA: haploandesite; HB: haplobasalt; HD: haplodacite; HR: haplorhyolite (HR7 is also named HPG8, a
composition in Na2O-CaO-Al2O3-SiO2 system); LB: lunar basalt; LP1 or LP2: lunar picrite. All compositional acronyms are followed by a numeral to distinguish from other acronyms. For
example, even though there is only one LB composition, it is named LB1.
References: 1. Morgan et al. 2006; 2. Lesher and Walker 1991; 3. Lundstrom 2003; 4. Lowry et al. 1981; 5. Lowry et al. 1982; 6. Hofmann and Magaritz 1977; 7. Magaritz and Hofmann
1978a; 8. Cunningham et al. 1983; 9. Henderson et al. 1985; 10. Van Orman and Grove 2000; 11. Koyaguchi 1989; 12. Lesher 1994; 13. LaTourrette et al., 1996; 14. LaTourrette and
Wasserburg 1997; 15. Alletti et al. 2007; 16. MacKenzie and Canil 2006; 17. Lesher et al. 1996; 18. MacKenzie et al. 2008; 19. Chen and Zhang 2008; 20. Chen and Zhang 2009; 21. Medford
1973; 22. Lesher and Walker 1986; 23. Ruessel and Wiedenroth 2004; 24. Balcone-Boissard et al. 2009; 25. Zhang et al. 1989; 26. van der Laan et al. 1994; 27. Behrens and Hahn 2009; 28.
Baker and Bossanyi 1994; 29. Koepke and Behrens 2001; 30. Baker 1989; 31. Baker 1990; 32. Baker 1992; 33. Baker and Watson 1988; 34. Baker 1993; 35. Mungall et al. 1999; 36. Tinker
and Lesher 2001; 37. Hayden and Watson 2007; 38. Dunn and Ratliffe 1990; 39. Jambon and Semet 1978; 40. Perez and Dunn 1996; 41. Wolf and London 1994; 42. Watson 1981; 43. Rapp
and Watson 1986; 44. Jambon 1982; 45. Bai and Koster van Groos 1994; 46. Gabitov et al. 2005.
319
320 Zhang, Ni, Chen
times) than the SiO2 diffusivity (or Eyring diffusivity) often display concentration profiles with
lengths similar to SiO2 concentration profiles, which has been explained as quasi-equilibrium
“partitioning” between the two end-member liquids due to rapid diffusion (e.g., Watson 1982;
Zhang 1993). Minor and trace elements with intrinsic mobilities comparable to SiO2 diffusivity
often show uphill diffusion though there may also be “normal” monotonic diffusion profiles
depending on the initial concentration distribution of the element (Zhang 1993).
H diffusion
H diffusion is reviewed by Zhang and Ni (2010, this volume).
the alkalis (li, Na, K, rb, Cs, fr)
Li diffusion. Five papers reported Li diffusion data in silicate melts (46 points). Jambon
and Semet (1978) explored Li tracer diffusivities in albite melt, Ab39Or61 (though it was called
orthoclase) melt, and rhyolite5 (Table 1) melt at 570-1183 K and in air, with concentrations
measured by ion microprobe. Even though Jambon and Semet (1978) reported 0.52 wt%
H2O (including H2O molecules and hydroxyl groups) in rhyolite5 based on loss-on-ignition,
the H2O content is treated as unknown and probably much lower than 0.52 wt% because
(i) determination of H2O content using loss-on-ignition is not accurate, (ii) another rhyolite
(rhyolite15) that was determined to contain 0.38 wt% H2O using the same method (Magaritz
and Hofmann 1978a,b; Jambon 1982) turned out to contain only 0.11 wt% H2O (Jambon et al.
1992), and (iii) some H2O would be lost during the experiment (heating at one atmosphere).
Lowry et al. (1981) examined Li tracer diffusion in nominally dry basalt1 melt at 1569-1673
K and 0.1 MPa (in air). Cunningham et al. (1983) investigated Li tracer diffusion in andesite2
and dacite4 melts at 1571-1673 K and 0.1 MPa (in air). Roselieb et al. (1998) determined Li
tracer diffusion in jadeite melt at 758-1173 K and 0.1 MPa (in air). Lundstrom (2003) reported
an SEBD value of Li in a diffusion couple made of basalt7 and basanite at 1723 K and 0.9 GPa.
Li diffusivity values are summarized in Figure 1.
Tracer diffusivity of Li does not depend much on melt composition; it increases slightly
from basalt to andesite to dacite to rhyolite and then to Ab39Or61 to jadeite and albite melts
(some of the trends can only be seen when Figure 1 is enlarged). That is, as melt viscosity
increases, Li diffusivity also increases, opposite to the trends predicted by the various
diffusivity-viscosity relations such as the Einstein equation (Einstein 1905) or the Eyring
equation (Glasstone et al. 1941). In dry rhyolite melt, Li tracer diffusivities at 570-1182 K and
0.1 MPa can be fit as (Fig. 1):
rhyolite5  (10966 ± 855) 

DLi TD = exp ( -13.04 ± 1.06) -  (7)
 T 
The maximum error of Equation (7) in fitting the data is 0.39 in lnD. The activation energy for
Li diffusion is low, about 91.2±7.1 kJ/mol, similar to H2O diffusion (e.g., Karsten et al. 1982;
Zhang et al. 1991a). Tracer diffusivities of Li in basalt, andesite, dacite, jadeite and albite
melts can be estimated by multiplying D from Equation (7) by 0.56, 0.68, 0.84, 3.2 and 4.0,
respectively.
After some effort, we fit the compositional dependence of Li tracer diffusivities in dry
albite, Ab39Or61, rhyolite, dacite, andesite, and basalt melts at 570-1673 K and 0.1 MPa as
follows:
 (11918 ± 536) - (7684 ± 1361) X ) 
DLiall dry
TD
melts
=exp  -13.36 -  (8)
 T 
where X = mole fraction of NaAlO2 + KAlO2. The maximum error of Equation (8) in
reproducing the diffusivity data is 0.76 in lnD (or 0.33 in logD) except for an outlier point
Zhang et al. (Ch 8) Diffusion data in silicate melts
figure 1. Fig.
Li 1.
tracer diffusivities
Li tracer diffusivities inin
drydry silicate
silicate melts atmelts atexcept
0.1 MPa, 0.1 MPa, except
for the b-b for the b-b
(basalt7-basanite) (basalt7-basanite)
couple, for which
couple, for which the diffusivity is SEBD and the pressure is at 0.9 GPa. Source of data: albite, Ab39Or61
the diffusivity is SEBD and the pressure is at 0.9 GPa. Source of data: albite, Ab39Or61 and rhyolite5 (Jambon
and rhyolite5 (Jambon and Semet 1978); jadeite (Roselieb et al. 1998); dacite4 and andesite2 (Cunningham
et al. 1983);
andbasalt1 (Lowry
Semet 1978); et(Roselieb
jadeite al. 1981);
et al.b-b (Lundstrom
1998); 2003). (Cunningham et al. 1983); basalt1 (Lowry et
dacite4 and andesite2
al. 1981); b-b (Lundstrom 2003).
for albite melt, which is off by 1.64 lnD units. One single datum of Li SEBD from a basalt7-
basanite diffusion couple (Fig. 1) at high pressure (0.9 GPa) by Lundstrom (2003) is lower
than the predicted Li tracer diffusivity at 0.1 MPa using Equation (8) by a factor of 3.2 (or 1.2
in lnD). Li (also Na) diffusivities are orders of magnitude greater than diffusivities of other
ionic species.
Na diffusion. Eight papers reported Na diffusion data in silicate melts (87 points),
including tracer diffusivities and effective binary diffusivities. (Na-K interdiffusivities are
discussed separately and not included here because they may also be counted as K diffusion
data). The radioactive tracer diffusion data include both 22Na and 24Na tracers. Because the
difference in their diffusivities is expected to be small (no more than 5% relative based on the
square root of mass difference), the precision of diffusion data in melts is not high enough to
resolve the difference. Hence, they are treated together. Jambon and Carron (1976) explored
Na diffusion in essentially pure orthoclase and albite
1 melts (including glasses) at 623-1068 K
and 0.1 MPa. Magaritz and Hofmann (1978a) determined Na tracer diffusivities (2 points) in
rhyolite15 melt at 970-1076 K and 0.1 MPa. Sample rhyolite15 was reported to contain 0.38
wt% H2O by weighing before and after heating overnight to 1573 K, but later Jambon et al.
(1992) found that the H2O content is actually 0.11 wt%. Watson (1981) characterized the effect
of H2O on Na diffusion in rhyolite12 melt at 973-1173 K, 0.21 GPa, and 0 and 3.5 wt% H2O.
Jambon (1982) investigated tracer diffusion of Na in dry rhyolite5 melt at 411-775 K and 0.1
MPa. Lowry et al. (1982) examined Na radioactive tracer diffusion in basalt1 and andesite2
melts at 1561-1675 K and 0.1 MPa. Henderson et al. (1985) determined Na diffusivities in
dacite4 and pantellerite1 melts at 1470-1675 K and 0.1 MPa. Behrens (1992) studied Na tracer
diffusion in plagioclase melts at 673-1273 K and 0.1 MPa. Chen and Zhang (2009) reported
Na SEBD data in basalt11 melt during clinopyroxene dissolution.
Figure 2 shows Na tracer diffusivities in dry melts at ≤ 0.21 GPa. There is good consistency
between the data. For example, Na diffusivities in plagioclase melt decrease systematically as
An# increases from 0 to 90%, and the activation energy increases with An#. Tracer diffusivity
322 Zhang, Ni, Chen
figure 2.Fig.
Na2. tracer
Na tracerdiffusivity
diffusivity inin silicate
silicate melts melts at 0.0001-0.21
at 0.0001-0.21 GPa.areThe
GPa. The melts meltsdryare
nominally nominally
except for the wetdry except
for the wet rhyolite12 that contains 3.5 wt% H2O. Li tracer diffusivity in rhyolite5 (dashed line) is shown
rhyolite12 that contains 3.5 wt% H2O. Li tracer diffusivity in rhyolite5 (dashed line) is shown for comparison.
for comparison. Source of data: albite and orthoclase (Jambon and Carron 1976); An50Ab50-An90Ab10
(Behrens Source
1992); basalt1
of data: albite and andesite2
and orthoclase (Lowry
(Jambon et al.1976);
and Carron 1982); dacite4 and pantellerite1
An50Ab50-An90Ab10 (Behrens 1992);(Henderson
basalt1 et al.
1985); rhyolite5 (Jambon 1982; Watson 1981); rhyolite15 (Magaritz and Hofmann 1978a); wet rhyolite12
and andesite2 (Lowry et al. 1982); dacite4 and pantellerite1 (Henderson et al. 1985); rhyolite5 (Jambon 1982;
with 3.5 wt% H2O (Watson 1981).
Watson 1981); rhyolite15 (Magaritz and Hofmann 1978a); wet rhyolite12 with 3.5 wt% H2O (Watson 1981).
data on dry rhyolite15 melt at 0.1 MPa (Magaritz and Hofmann 1978a), those on dry rhyolite5
melt at 0.1 MPa (Jambon 1982), and a single diffusivity datum in dry rhyolite12 melt at 0.21
GPa by Watson (1981) lie in the same trend. Na tracer diffusivity in “dry” rhyolite5 and
rhyolite15 melts at 411-1173 and ≤ 0.2 GPa can be described by the following equation (Fig. 2):
 (10210 ± 243) 
rhyolite5 & rhyolite15
DNa TD = exp ( -13.52 ± 0.39) -  (9)
 T 
The maximum error of Equation (9) in reproducing the experimental data is 0.20 in lnD. The
activation energy for Na tracer diffusion is very low, about 84.9±2.0 kJ/mol, comparable with
(even slightly lower than) that for Li diffusion. Na tracer diffusivity in dry rhyolite melts (solid
line in Fig. 2) is similar to Li tracer diffusivity in dry rhyolite5 melt (dashed line in Fig. 2).
Adding 3.5 wt% H2O to rhyolite melt (crosses2 in Fig. 2) roughly doubles the Na diffusivity,
a relatively small effect (compared to the effect of H2O on Cs diffusion, for example). At
high temperatures (1470-1675 K), Na diffusivity increases from basalt to dacite to andesite to
rhyolite melt, similar to the behavior of Li diffusion, again opposite to the famous diffusivity-
viscosity relations. Na diffusivity in basalt1, andesite2, dacite4, and pantellerite1 melts can be
roughly estimated by multiplying D values from Equation (9) by 0.23, 0.27, 0.41, and 0.65,
respectively.
The extensive Na tracer diffusion data in dry plagioclase melts (Jambon and Carron 1976;
Behrens 1992) at 623-1273 K and 0.1 MPa show that Na diffusivity is a curved function of
the fraction of anorthite (Behrens 1992). All the data in plagioclase melts can be fit as follows:
plag melt  (6158 + 5769 X An + 12480 X An

2
)
DNa TD =exp  -16.87 + 5.318 X An -  (10)
 T 
where XAn is the mole fraction of anorthite component in plagioclase melts. The fits are shown
in Figure 3a and the 2s error in predicting lnD values by Equation (10) is 0.65. Equation (10) is
expected to be applicable to Na tracer diffusivity in all plagioclase melts. The implied activation
energy is 51 kJ/mol for albite melt to 203 kJ/mol in anorthite melt. The huge variation in the
activation energy of Na diffusion from albite melt to anorthite melt is surprising.
The only SEBD data for Na are by Chen and Zhang (2009) for a MORB melt (basalt11,
Table 1) during
Zhang clinopyroxene dissolution
et al. (Ch 8) Diffusion in themelts
data in silicate melt at 1544-1790 K and 0.47-1.9 GPa.
The data are shown in Figure 3b. Because the concentration gradient of Na in melts during
figure 3. Na diffusivities in various melts. The lines in (a) are calculated from Equation (10). The lines in
Fig.linear
(b) are simple 3. Nafits.
diffusivities in various
Data sources: tracer melts. The in
diffusivities lines in (a) aremelts
plagioclase calculated from
(Jambon andEq. 3a. The
Carron 1976;lines in (b) are simple
Behrens 1992); SEBD in basalt11 at 0.47-1.9 GPa (Chen and Zhang 2009); Na-K interdiffusivities at XAb
= 0.5 and 1linear fits. Data
GPa (Freda sources:
and Baker tracerFor
1998). diffusivities
the other in plagioclase
data, the sourcemelts
can (Jambon
be foundand Carron
in the 1976;
caption of Behrens 1992); SEB
Figure 2.
in basalt11 at 0.47-1.9 GPa (Chen and Zhang 2009); Na-K interdiffusivities at XAb = 0.5 and 1 GPa (Freda and
Baker 1998). For the other data, the source can be found in the caption of Fig. 2.
324 Zhang, Ni, Chen
clinopyroxene dissolution is small, Na SEBD values are expected to depend on the various
major element concentration gradients (i.e., multicomponent diffusion effects by other major
concentration gradients such as Ca and Mg may cause large variation in Na SEBD). To the first
order, data from 0.47 to 1.9 GPa roughly follow a single trend, indicating negligible pressure
dependence. However, the SEBD values are clearly lower than tracer diffusivities in basalt
melts, by a factor of about 16. They are also much lower than Na-K interdiffusivities in albite-
orthoclase melts to be discussed below.
K diffusion. Six papers reported K diffusion data in silicate melts (40 points). Jambon
and Carron (1976) and Jambon (1982) explored K tracer diffusion in albite, orthoclase and
rhyolite5 melts at 610-1262 K and 0.1 MPa. Note that tracer diffusion of K in orthoclase is
also self or isotopic diffusion because other K isotopes are present (Zhang 2010). Watson and
Jurewicz (1984) investigated SEBD of K during granite dissolution and melting into a basalt
melt (close to basalt7 in Table 1) at 1548 K and 1 GPa. van der Laan et al. (1994) carried
out diffusion experiments on rhyolite3-rhyolite16 and andesite3-rhyolite16 couples at 1523 K
and 1.0 GPa, from which K isotopic effective binary diffusivities and SEBD were extracted.
Roselieb and Jambon (1997) reported tracer diffusion data of K in jadeite melts at 0.1 MPa
(with two data points for albite melt) by loading KCl solution onto the polished surface of a
glass cylinder. Hence the diffusion may be Na-K interdiffusion (because there is Na in jadeite
and albite). Lundstrom (2003) reported basalt7-basanite diffusion couple experiments at 1723
K and 0.9 GPa, from which K SEBD values were extracted. The diffusion data are summarized
in Figure 4.
At 1200 K, diffusivities of K in albite, orthoclase, jadeite and rhyolite5 melts are about the
same. K tracer diffusivity in “dry” rhyolite5 melt at 645-1118 K and 0.1 MPa can be expressed
as (Jambon 1982):
 (12775 ± 927) 
DKrhyolite5
TD = exp ( -14.87 ± 1.11) -  (11)
 T 
activation energy is 106±8 kJ/mol.
SEBD values of K are not consistent with K tracer diffusivities. For example, in rhyolite3-
rhyolite16 diffusion couple, K SEBD values are lower than tracer diffusivities by a factor of
6. The difference may be explained as follows. Because K is a relatively minor component in
natural melts, even in the rhyolite3-rhyolite16 diffusion couple (both halves are rhyolite but
with SiO2 content of 72.8 wt% and 77.0 wt%), K diffusion may still be significantly affected
by cross diffusivity terms so that the SEBD is not close to tracer diffusivities. There are not
enough K SEBD data to quantify the temperature or compositional dependence. Nonetheless,
the limited data indicate that SEBD of K decreases from the basalt7-basanite couple to the
rhyolite3-rhyolite16 couple, to basalt during granite dissolution (similar to andesite-basalt
couple), and to the andesite3-rhyolite16 couple. That is, the SEBD of K decreases as the
mean SiO2 concentration increases (meaning viscosity increases), but also as the difference
in SiO2 between the two ends increases. The latter is a clear indication of cross effects of
multicomponent diffusion (contribution to K diffusion flux due to SiO2 concentration gradient).
Na-K interdiffusion, and inconsistency between tracer diffusion and interdiffusion data.
Freda and Baker (1998) systematically investigated Na-K interdiffusion in dry albite-orthoclase
melts at 1073-1740 K, 1 and 2 GPa, and at 0 and 5.5 wt% H2O. For true interdiffusion, there
is only one diffusivity for both Na and K. However, Freda and Baker (1998) reported both
Na and K diffusivities at XAb = 0.25, 0.50 and 0.75. The differences between the Na and K
diffusivities are relatively small and may be viewed as the error in diffusivity determination.
Hence, we average their reported Na and K diffusivities as Na-K interdiffusivities. The Na-K
figure 4. Fig.
K diffusivities
4. K diffusivitiesinin various melts.Data
various melts. Data sources
sources and conditions:
and conditions: Jadeite
Jadeite (Roselieb (Roselieb
and Jambon 1997);and
albiteJambon
1997); albite (Jambon and Carron 1976; Roselieb and Jambon 1997); orthoclase (Jambon and Carron
(Jambon and Carron 1976; Roselieb and Jambon 1997); orthoclase (Jambon and Carron 1976); rhyolite5 (Jambon
1976); rhyolite5 (Jambon 1982); rhy3-rhy16 (rhyolite3-rhyolite16) and and3-rhy16 (andesite3-rhyolite16)
(van der Laan
1982);et al. 1994,
rhy3-rhy16 1 GPa); b-b (basalt7-basanite)
(rhyolite3-rhyolite16) (Lundstrom 2003,
and and3-rhy16 (andesite3-rhyolite16) (Van Der0.9Laan
GPa); granite-in-basalt
et al. 1994, 1 GPa);
(Watson 1982, 1 GPa); Na-K interdiffusivities in dry albite-orthoclase melt at Ab50-Or50 (Freda and Baker
1998, 1-2 b-b
GPa).(basalt7-basanite) (Lundstrom 2003, 0.9 GPa); granite-in-basalt (Watson 1982, 1 GPa); Na-K interdiffusivities
in dry albite-orthoclase melt at Ab50-Or50 (Freda and Baker 1998, 1-2 GPa).
interdiffusivities in dry and wet Ab-Or systems are shown in Figure 5a. A pressure increase
from 1 to 2 GPa does not have a significant effect on the interdiffusivities. As Or component
increases, Na-K interdiffusivities increase slightly (by a factor of 1.6 from Ab75Or25 to Ab25Or75).
The diffusion data covers only two H2O contents: 0 and 5.5 wt%. The addition of 5.5 wt% H2O
increases Na-K interdiffusivities only slightly, by a factor of about 3. Assuming lnD is linear to
H2O content, the data at 1073-1773 K, 1-2 GPa and 0-5.5 wt% H2O can be fit as follows:
 (16976 - 1189w) 
ID = exp ( -11.51 - 0.6067w + 0.95 X Or ) -
Ab-Or
DNa-K  (12)
 T 
where XOr is mole fraction of orthoclase in Ab-Or feldspar. The fits are shown in Figure 5a.
activation energy in dry feldspar melt does not change
4 much with alkali feldspar composition
and is 141±8 kJ/mol. The activation energy in feldspar melt containing 5.5 wt% H2O is about
87 kJ/mol.
Because Na trace element (or tracer) diffusivity in orthoclase melt (Jambon and Carron
1976) is the Na-K interdiffusivity at Or100, and K trace element (or tracer) diffusivity in albite
melt (Jambon and Carron 1976; Roselieb and Jambon 1997) is the Na-K interdiffusivity at
Ab100, they can be plotted together with the interdiffusivity data of Freda and Baker (1998).
The data are compared in Figure 5b, with the tracer diffusivities (i.e., interdiffusivities at
Ab100 and Or100) as triangles (in blue in the online version). The data appear to be inconsistent
(e.g., the interdiffusivities at intermediate compositions are not between the interdiffusivities
at Ab100 and Or100) and hence no general equation was found to describe the compositional
dependence. The reason for such inconsistency is not clear, but might be due to (i) pressure
effect (1-2 GPa for Na-K interdiffusivities Na and K by Freda and Baker 1998, versus 0.0001
GPa for tracer diffusivities by Jambon and Carron 1976, and Roselieb and Jambon 1997), (ii)
326 Zhang, Ni, Chen
figure 5. Na-K interdiffusivities

Fig. 5. in albite-orthoclase
Na-K interdiffusivities binary. Data
in albite-orthoclase source: Data
binary. dry and wet Ab
source: 75Or
dry 25,wet
and Ab50Ab75Or25,
Or50 and Ab50Or50 and
Ab25Or75 (Freda and Baker 1998); Ab100 and Or100 (Jambon and Carron 1976; Roselieb and Jambon 1997).
Ab25Or75 (Freda and Baker 1998); Ab100 and Or100 (Jambon and Carron 1976; Roselieb and Jambon 1997).
complexity in the relation between interdiffusivity, tracer diffusivity and composition (e.g.,
highly non-ideal mixing of the binary melts), (iii) the compositions are not perfect Ab-Or
binaries, (iv) inter-laboratory inconsistency, and (v) tracer diffusivity is different from the trace
element diffusivity.
Rb diffusion. Seven papers reported Rb diffusion data in silicate melts (49 points). Jambon
and Carron (1976) and Jambon (1982) explored Rb tracer diffusion in albite, orthoclase and
rhyolite5 melts at 693-1268 K and 0.1 MPa. Roselieb and Jambon (1997) reported tracer
diffusivities of Rb in jadeite melts at 1073-1293 K and 0.1 MPa,
5 plus two data points in albite
melt. Nakamura and Kushiro (1998) characterized Rb tracer diffusion in jadeite melt at 1673 K
and 1.25-2.0 GPa and in diopside melt at 1863 K and 1-1.25 GPa. Koepke and Behrens (2001)
determined Rb tracer diffusivities in a wet haploandesite (HA1 in Table 1) melt containing 4.5-
5.2 wt% H2O at 1373-1673 K and 0.5 GPa, and one single datum in dry haploandesite at 1673
K and 0.5 GPa. Lundstrom (2003) reported basalt7-basanite diffusion couple experiments at
1723 K and 0.9 GPa, from which Rb SEBD values were extracted. Behrens and Hahn (2009,
and correction) investigated Rb tracer diffusion in dry and wet (≤ 1.7 wt% H2O) trachyte1 melt
at 1323-1527 K and 0.5 GPa and in dry phonolite1 melt at 1423-1523 K and 0.5 GPa. Figure
6 shows some diffusion data.
Rb tracer diffusivity values in jadeite melt at 1673 K and 1.25-2.0 GPa by Nakamura
and Kushiro (1998) indicate negligible pressure dependence, and the data are also roughly
consistent with the extrapolation of the low-temperature data at 0.1 MPa by Roselieb and
Jambon (1997). On the other hand, Rb TD values using radioactive tracer 86Rb in albite melt
obtained by Jambon and Carron (1976) are different from Rb tracer diffusivities using elemental
Rb tracer by Roselieb and Jambon (1997) (compare albite1 and albite2 in Fig. 6): data obtained
by Roselieb and Jambon (1997) are smaller than those of Jambon and Carron (1976) by a
factor of 2 to 4 (0.7 to 1.4 lnD units). The difference may be attributed either to experimental
uncertainty or to difference between radioactive 86Rb tracer and elemental Rb tracer.
Rb diffusivities in melts of dry orthoclase, wet HA1 (4.5-5.2 wt% H2O) and wet trachyte1
(1.1-1.7 wt% H2O) are similar and high, followed by dry phonolite1 melt, then dry rhyolite5,
albite, jadeite1, jadeite2, and dry trachyte1 and HA1. Rb tracer diffusivity in dry rhyolite5 melt
at 693-1218 K and 0.1 MPa can be expressed as follows:
 (15299 ± 249) 
rhyolite5
DRb TD = exp ( -15.42 ± 0.27) -  (13)
 T 
activation energy is 127±2 kJ/mol.
figure 6. Fig.
Rb 6.diffusivities
Rb diffusivitiesinin various melts.Data
various melts. Data sources
and conditions: dry jadeite1 dry jadeite1
and albite2 at 0.1and
MPa:albite2
Roseliebat 0.1
MPa: Roselieb and Jambon (1997); dry jadiete2 at 1.25-2.0 GPa: Nakamura and Kushiro (1998); dry
albite1 andand
orthoclase at 0.1
Jambon (1997); dryMPa: Jambon
jadiete2 and
at 1.25-2.0 Carron
GPa: (1976);
Nakamura rhyolite5:
and Kushiro (1998);Jambon
dry albite1(1982); dry and
and orthoclase wet (4.5-
at 0.1
5.2 wt% HMPa:
2O) HA1 at 0.5 GPa: Koepke and Behrens (2001); dry and wet (1.1-1.7 wt%
Jambon and Carron (1976); rhyolite5: Jambon (1982); dry and wet (4.5-5.2 wt% H2O) HA1 at 0.5 GPa:H 2 O) trachyte (tra1)
at 0.5 GPa: Behrens and Hahn (2009); dry phonolite (pho1) at 0.5 GPa: Behrens and Hahn (2009); SEBD
in basalt1-basanite
Koepke and(b-b)
Behrenscouple
(2001);at dry0.9
andGPa: Lundstrom
wet (1.1-1.7 (2003).
wt% H2O) trachyte (tra1) at 0.5 GPa: Behrens and Hahn (2009);
dry phonolite (pho1) at 0.5 GPa: Behrens and Hahn (2009); SEBD in basalt1-basanite (b-b) couple at 0.9 GPa:
Lundstrom (2003).
328 Zhang, Ni, Chen
For dry and wet trachyte1 melt, the limited data are not enough to define a clear relation
between lnD and H2O. Assuming that lnD increases linearly with H2O, the following expression
may be obtained for trachyte melt at 1323-1528 K and 0-1.7 wt% H2O,
 (19137 ± 8358) - (3458 ± 602)w 
trachyte1; 0.5 GPa
DRb TD = exp ( -13.82 ± 5.78) -  (14)
 T 
The maximum error to reproduce the experimental data is 0.71 lnD units.
Assuming similar relation for dry and wet HA1 melt containing up to 5.2 wt% H2O, Rb
tracer diffusivities at 1373-1673 K and 0.5 GPa can be expressed as follows (Koepke and
Behrens 2001):
 (22875 ± 6672) - (1155 ± 254)w 
dry&wet HA1
DRb TD ≈ exp ( -11.77 ± 3.92) -  (15)
 T 
The maximum error of Equation (15) in reproducing the experimental data is 0.24 in lnD.
Comparing Equations (14) and (15), Rb diffusivity depends on H2O much more strongly in
trachyte melt than in haploandesite melt.
Cs diffusion. Six papers reported Cs diffusion data in silicate melts (83 points) (excluding
data based on β-track mapping). Jambon and Carron (1976) explored Cs tracer diffusion in
albite (978-1313 K) and orthoclase (855-1313 K) melts at 0.1 MPa. Jambon (1982) investigated
Cs tracer diffusion in rhyolite5 (875-1188 K and 0.0001-0.3 GPa) and rhyolite15 (1063-1573
K and 0.1 MPa) melts. He showed that a pressure increase from 0.1 MPa to 0.3 GPa does
not change Cs tracer diffusivity noticeably. Lowry et al. (1982) examined Cs radioactive
tracer diffusion in basalt1 and andesite2 melts at 1569-1675 K and 0.1 MPa. Henderson et al.
(1985) studied Cs tracer diffusivities in dacite4 (1573-1672 K) and pantellerite1 (1473-1575
K) melts at 0.1 MPa. Roselieb and Jambon (1997) characterized Cs tracer diffusion in jadeite
melts at 1073-1293 K and 0.1 MPa, plus two data points in albite melt. Mungall et al. (1999)
determined Cs tracer diffusivities in dry HR7 (1410-1873 K, 0.1 MPa), wet HR7 (3.6 wt%
H2O, 1573 and 1873 K, 1 GPa), and dry HR7 + Na (1083-1473 K, 0.1 MPa) melts. The data on
synthetic simple melts are summarized in Figure 7a and the data on natural silicate melts and
their analogs are summarized in Figure 7b.
Cs tracer diffusivities in albite melt obtained by Jambon and Carron (1976) using a 134Cs
radioactive tracer are four times those of Roselieb and Jambon (1997) using a Cs elemental
tracer (compare albite1 and albite2 in Fig. 7a). Cs tracer diffusivity in rhyolite15 is lower than
that in rhyolite5, due to lower SiO2 and/or higher H2O in rhyolite5.
Cs diffusivity increases from rhyolite to dacite to andesite to basalt melt (Fig. 7b),
consistent with the direction of viscosity decrease but opposite to the trends for Li and Na
diffusivity. Cs tracer diffusivities in dry rhyolite15 melt and in dry HR7 melt are the lowest
among all the melts investigated and can be expressed as follows
 (26406 ± 1329) 
dry rhyolite15 & HR7
DCs TD = exp ( -11.98 ± 0.99) -  (16)
 T 
Equation (16) is applicable to dry rhyolite15 and HR7 melts at 1063-1873 K and ≤ 0.3 GPa,
with maximum error of 0.46 in lnD in reproducing the experimental data. The activation energy
is 220±11 kJ/mol. Cs tracer diffusivities in rhyolite5 melt are about 2.5 times those in Equation
(16) (Fig. 7b).
Because extensive data are available, we made an attempt to fit the temperature, pressure
and compositional dependence of Cs tracer diffusivities in dry melts, obtaining the following:
 (51068 - 27975 X1 - 68760 X 2 + 1649P ) 

DCs TD = exp 16.53 - 32.43 X1 - 37.3 X 2 -
all dry melts
 (17)
 T 
where X1 is the sum of cation mole fractions of Si + Ti + Al + P, and X2 = max(Na + K-Al,0) (i.e.,
X2 = Na + K-Al
ZhangifetNa
al.+(Ch
K-Al8) > 0, and Xdata
Diffusion 2 = 0inifsilicate
Na + K-Al
melts< 0). Equation (17) fit experimental
data to within 0.72 lnD units including the following dry melts: albite and jadeite of Roselieb
and Jambon (1997), basalt1 and andesite2 of Lowry et al. (1982), dacite4 and two points of
figure 7. Cs diffusivities in (a) synthetic silicate melts (a) and (b) natural silicate melts or analogs. The
Fig.Equation
lines are from 7. Cs(17)
diffusivities in diffusion
(fit to all dry (a) synthetic
data) silicate melts
except for (a) and
orthoclase and(b)wet
natural
melts. silicate melts or analogs. The
Data sources:
albite1 (Jambon and Carron 1976); albite2 and jadeite (Roselieb and Jambon 1997); HR7+Na (Mungall
et al. 1999); orthoclase
from Eq. 7b(Jambon
(fit to allanddryCarron 1976);
diffusion HR7
data) (a haplorhyolite
except melt)and
for orthoclase andwet
wet melts.
HR7 (with 3.6sources: albite1 (Ja
Data
wt% H2O) (Mungall et al. 1999); rhyolite5 (rhy5) and rhyolite15 (rhy15) (Jambon 1982); dacite4 and
pantellerite1 (Henderson
Carron 1976);et albite2
al. 1985);andandesite2
jadeite and basalt1 and
(Roselieb (Lowry et al. 1982).
Jambon 1997); HR7+Na (Mungall et al. 1999); orthoclas
and Carron 1976); HR7 (a haplorhyolite melt) and wet HR7 (with 3.6 wt% H2O) (Mungall et al. 1999); r
330 Zhang, Ni, Chen
three on pantellerite1 (Henderson et al. 1985), dry HR7 and HR7 + Na (Mungall et al. 1999),
and rhyolite15 at 0.1 GPa (Jambon 1982). One D value in pantellerite1 at 1575 K is higher
than that predicted by Equation (17) by 1.0 lnD unit. However, diffusion data for albite by
Jambon and Carron (1976) are off by 0.1 to 1.6 lnD units, and those for orthoclase by Jambon
and Carron (1976) are off by 4.5 to 7.8 lnD units. Cs diffusivities in rhyolite5 melt at 0.1 GPa
(Jambon 1982) are higher than those predicted by Equation (17) by 0.6 to 1.9 lnD units, likely
due to higher H2O in the melt.
Fr diffusion. No Fr diffusion data in silicate melts are known.
Some comments on alkali diffusivities in silicate melts. Diffusivities of alkali elements
in rhyolite5 melt at 0.0001-0.1 GPa are summarized in Figure 8. The small pressure difference
is not expected to cause noticeable variation in the diffusivities. In the alkali series, smaller
cations usually diffuse more rapidly (Li ≈ Na > K > Rb > Cs) in silicate melts. This is similar
to noble gas diffusion: the diffusivity of noble gases increases as the size decreases, from
Xe to Kr to Ar to Ne to He (e.g., Perkins and Begeal 1971; Roselieb et al. 1995; Behrens
2010, this volume). This trend may sound intuitive, but cannot be explained by the Stokes-
Einstein equation (see Zhang 2010, this volume) because the diffusivity variation is orders of
magnitude, much larger than the relative change in the ionic radius. Furthermore, this trend
cannot be generalized to other elemental series: e.g., in alkali earth elements, REE, and most
other series, the trend can be opposite.
the alkali earths (Be, Mg, Ca, Sr, Ba, ra)
Be diffusion. Only one study (Mungall et al. 1999) reported Be tracer diffusion data on
three melt compositions: HR7, wet HR7 with 3.6 wt% H2O, and HR7 + Na (Table 1). The data
are shown in Figure 9, together with the line for Cs diffusivities in dry HR7 for comparison.
Even though Be is a small cation, Be diffusivities are very low, much lower than the lowest
diffusivities of alkali elements (Cs). As will be seen later, Be diffusivities are also the smallest
among the alkali earth elements. Be tracer diffusivity in dry HR7 at 1410-1873 K and 0.1 MPa
can be expressed as follows:
 (38690 ± 4236) 
TD = exp ( -8.38 ± 2.62) -
HR7
DBe  (18)
 T 
The maximum error of Equation (18) in reproducing the three experimental diffusivities is
0.21 in lnD. The activation energy is high, 322±35 kJ/mol.
Mg diffusion. Mg is a major element in most natural silicate melts. Extensive Mg
diffusion data in silicate melts are available, with 13 papers and 166 points (of which 15 are
for hydrous melts). Zhang et al. (1989) determined Mg SEBD in dry andesite1 melt during
crystal dissolution experiments. Kubicki et al. (1990) studied SEBD of Mg in dry diopside-
anorthite melts. Sheng et al. (1992) explored Mg self diffusion in dry CMAS melts at 1534-
1826 K and in air. Baker and Bossanyi (1994) examined the effect of H2O and F on Mg SEBD
in rhyolite7-dacite1 diffusion couple at 1373-1673 K and 1 GPa. van der Laan et al. (1994)
obtained Mg SEBD in dry rhyolite3-rhyolite16 and andesite3-rhyolite16 diffusion couples at
1523 K and 1 GPa. LaTourrette et al. (1996) and LaTourrette and Wasserburg (1997) studied
Mg self diffusion in dry HB1 melt at 1623-1773 K and in air. Mungall et al. (1999) examined
Mg tracer diffusion in dry HR7 and dry HR7 + Na at 0.1 MPa and wet HR7 (3.6 wt% H2O)
melts at 1 GPa. Van Orman and Grove (2000) reported a datum of Mg SEBD in lunar basalt
(LB1 in Table 1) during clinopyroxene dissolution. Roselieb and Jambon (2002) investigated
Mg tracer diffusion in dry albite and jadeite melts at 1073-1293 K and 0.1 MPa. Lundstrom
(2003) acquired SEBD of Mg in dry basalt7-basanite couple at 1723 K and 0.9 GPa. Chen
and Zhang (2008, 2009) determined MgO SEBD in basalt11 (MORB) melt during olivine and
clinopyroxene dissolution. Mg diffusion data are shown in Figure 10.
figure 8. Li, Na, K, Rb and Cs tracer diffusivities in rhyolite and albite melts. See Figures 1-7 for data
sources. In Fig. 8. and
(b), Rb Li, Cs
Na,diffusion
K, Rb andin Cs tracer
albite diffusivities
melts in rhyolite
were studied and albite
in two papers melts.andSee
(Jambon Figs.1976;
Carron 1-7 for data sources. In
Roselieb and Jambon 1997), with small differences.
(b), Rb and Cs diffusion in albite melts were studied in two papers (Jambon and Carron 1976; Roselieb and Jambon
1997), with small differences.

Examination of the data shows the following. First, the pressure effect at ≤ 2 GPa is
negligible (e.g., open circles, triangles, squares and diamonds in Fig. 10a). Second, differences
between self-diffusivities and SEBD are less than 1 lnD unit (e.g., self diffusivities in HB1
and SEBD in basalt11 in Fig. 10a) when SiO2 concentration difference across the profile is ≤ 6
wt%, although there are no data at exactly the same bulk composition for a direct comparison.
(LB1, even though also a basalt, is compositionally very different from terrestrial basalts.)
Thirdly, Figure 10b shows that Mg diffusivities vary significantly
8 with melt composition (and
temperature), especially from andesite1 to HR7.
Mg SEBD in dry basalt11 (MORB) melt (containing 0.2 to 0.4 wt% H2O) during olivine
332 Zhang, Ni, Chen
figure 9. Be
Fig.tracer diffusivities
9. Be tracer diffusivitiesinindry HR7at at
dry HR7 0.10.1 MPa,
MPa, wet(containing
wet HR7 HR7 (containing 3.6
3.6 wt% H2O) at 1wt%
GPa, H 2O)
and dryat 1 GPa, and
dry HR7 + Na at 0.1 MPa. All data are from Mungall et al. (1999). The short dashed (and purple) line with
HR7+Na at 0.1 MPa. All data are from Mungall et al. (1999). The short dashed (and purple) line with no
no corresponding points are for Cs diffusion in dry HR7, shown for comparison.
corresponding points are for Cs diffusion in dry HR7, shown for comparison.
dissolution at 1543-1753 K and 0.5-1.4 GPa can be expressed as (Chen and Zhang 2008):
 (26222 ± 2470) 
basalt11 (oliv diss)
DMg SEBD = exp ( -7.92 ± 1.50) -  (19)
 T 
The maximum error of Equation (19) in reproducing the diffusion data is 0.27 in lnD.
Mg SEBD values in the same basalt11 melt during clinopyroxene dissolution at 1544-
1790 K and 0.5-1.9 GPa are lower than that during olivine dissolution by about 0.37 in lnD
and can be expressed as follows (Chen and Zhang 2009)
 (28922 ± 2420) 
basalt11 (cpx diss)
DMg SEBD = exp ( -6.65 ± 1.46) -  (20)
 T 
Although the same initial melt composition (basalt11)
9 is used in the olivine and clinopyroxene
dissolution studies by Chen and Zhang (2008, 2009), the interface melt during clinopyroxene
dissolution is more SiO2-rich than that during olivine dissolution. Therefore, it is expected
that Mg SEBD values during clinopyroxene dissolution are lower than those during olivine
dissolution.
In andesite1 melt, Mg SEBD during olivine dissolution at 1488-1673 K and 0.5-1.5 GPa
can be expressed as:
 (35049 ± 6283) 
andesite1 (oliv diss)
DMg SEBD = exp ( -3.53 ± 3.96) -  (21)
 T 
The maximum error of Equation (21) in reproducing the experimental diffusivities is 0.34 in
lnD. Mg SEBD values in the same andesite1 melt during clinopyroxene dissolution are not
significantly different from those during olivine dissolution.
In highly silicic melt HR7, Mg tracer diffusivities can be expressed as follows:
figure 10. Mg diffusivities in silicate melts. basalt11(ol) means D in basalt11 melt during olivine
dissolution;Fig. 10. Mg diffusivities
LB1(cpx) means D ininlunar
silicate melts.
basalt basalt11(ol)
during means dissolution;
clinopyroxene D in basalt11b-b
meltmeans
duringbasalt7-
olivine dissolution;
basanite couple. Data sources can be found in the text and Table 1. Pressure is not shown in (b) because (a)
LB1(cpx) means D in lunar basalt during clinopyroxene dissolution; b-b means basalt7-basanite couple. Data
shows that the pressure effect is negligible.
sources can be found in the text and Table 1. Pressure is not shown in (b) because (a) shows that the pressure
effect is negligible.  (29341 ± 7457) 

TD = exp ( -11.77 ± 4.51) -
HR7
DMg  (22)
 T 
The maximum error of Equation (22) in reproducing the experimental diffusivities at 1410-
1873 K and 0.1 MPa is 0.49 in lnD.
10
We made an effort to model the compositional effect on Mg diffusivities (including self
and tracer diffusivities, and SEBD) in natural and nearly natural dry basalt to rhyolite melts
(rather than melts with mineral compositions), including SEBD in andesite1 melt (Zhang et
334 Zhang, Ni, Chen
al. 1989), self diffusivities in HB1 melt (LaTourrette et al. 1996; LaTourrette and Wasserburg
1997), tracer diffusivities in HR7 (Mungall et al. 1999), SEBD in lunar basalt (Van Orman and
Grove 2000), SEBD in basalt7-basanite melts (Lundstrom 2003), and SEBD in basalt11 (Chen
and Zhang 2008, 2009). We also included the SEBD data by Kubicki et al. (1990) on Di-An
melts and self diffusion data by Sheng et al. (1992) on “basic” melts because the compositions
of these melts are not far from the HB1 melt of LaTourrette et al. (1996) and LaTourrette
and Wasserburg (1997). Data by van der Laan et al. (1994) are not included because the H2O
content is fairly high (0.3-1.2 wt%). For SEBD values, there is a range of compositions along
a profile. For the modeling purpose, the melt composition is taking to be the average of the
two diffusion ends if a diffusion couple profile is fit by assuming a constant diffusivity, or the
average of the far-field and interface melts if the experiment method is crystal dissolution. The
diffusion data can be well fit by the following:
10993 + 17839 XSA
SD, TD & SEBD = -5.17 - 11.37 XSi - 2.16 X FM -
dry basalt to rhyolite
ln DMg (23)
T
where X is cation mole fraction, XFM = Fe + Mn + Mg, and XSA = Si + Al. Fits of the data
by Equation (23) are shown in Figure 11, with a reproducibility within 0.76 lnD units (or
0.33 logD units) except for the four outlier points (one point on basalt7-basanite couple by
Lundstrom 2003 is off by 0.9 lnD units, one point for lunar basalt melt by Van Orman and
Grove 2000 is off by 1.9 lnD units, one single point for andesite melt during quartz dissolution
by Zhang et al. 1989 is off by 1.2 lnD units, and one point in Kubicki et al. 1990 is off by 1.2
lnD units). The maximum error in using Equation (23) to predict Mg tracer diffusivities is 0.7
lnD units for HR7 + Na melt, 1.1 lnD units for jadeite melt, 1.6 lnD units for albite melt.
Mg diffusion data in hydrous melts are limited, with only 15 points. Adding 1 wt% H2O
increases the diffusivity by about 2 lnD units (van der Laan et al. 1994); adding 3.6 wt%
H2O increases the diffusivity by about 4 lnD units (Mungall et al. 1999); and adding 3.2
figure 11.Fig.
Fitting MgMg
11. Fitting diffusion
diffusion data
data inin natural
natural and nearly
and nearly natural natural basalt melts
basalt to rhyolite to rhyolite melts
by Eq. 11. by Equation (23).
basalt11(ol)
basalt11(ol) means D in basalt11 melt during olivine dissolution. Data sources: LB1 (LaTourrette et al.
means D in basalt11 melt during olivine dissolution. Data sources: LB1 (LaTourrette et al. 1996; LaTourrette and
1996; LaTourrette and Wasserburg 1997); HR7 (Mungall et al. 1999); basalt11(ol) (Chen and Zhang 2008);
basalt11(cpx) (Chen1997);
Wasserburg and HR7
Zhang 2009);
(Mungall et al.andesite1(ol) (Zhang
1999); basalt11(ol) et al.
(Chen and 1989);
Zhang Dibasalt11(cpx)
2008); 50An50 (Kubicki
(Chen andetZhang
al. 1990).
(2009); andesite1(ol) (Zhang et al. 1989); Di50An50 (Kubicki et al. 1990).
wt% and 6.0 wt% H2O increases the diffusivity by 3.1 and 3.5 lnD units, respectively (Baker
and Bossanyi 1994). The limited data indicate that adding H2O increases MgO diffusivity
significantly, but the effect cannot be quantified yet.
Ca diffusion. Sixteen papers reported Ca diffusion data in silicate melts (129 points, of
which 20 are for hydrous silicate melts). Medford (1973) explored Ca SEBD in mugearite1
and mugearite2 melts (Table 1) using the diffusion couple method at 1503-1696 K and 0.1
MPa. Hofmann and Magaritz (1977) studied Ca tracer diffusion in dry basalt3 melt at 1523-
1723 K and 0.1 MPa. Jambon (1982) acquired Ca tracer diffusivities in rhyolite5 melt at
905-1204 K and 0.1 GPa. Harrison and Watson (1984) reported Ca SEBD in rhyolite12 melt
during apatite dissolution at 1473-1673 K, 0.8 GPa, and 0.1 and 1.0 wt% H2O. Zhang et
al. (1989) extracted Ca SEBD in andesite1 melt during diffusive clinopyroxene dissolution
experiments. Kubicki et al. (1990) studied SEBD of Ca in binary systems of dry diopside-
anorthite melts at 1633-1963 K and 0.0001 to 2 GPa, with most data at 0.2 GPa. Behrens
(1992) reported Ca tracer diffusivities in dry Ab40An60 melt at 993-1123 K and 0.1 MPa. Baker
and Bossanyi (1994) examined the effect of H2O and F on Ca SEBD in the same diffusion
couple at 1373-1673 K and 1 GPa. van der Laan et al. (1994) obtained Ca isotopic effective
binary diffusivities in rhyolite3-rhyolite16 and andesite3-rhyolite16 diffusion couples at
1523 K and 1 GPa. LaTourrette et al. (1996) characterized Ca self diffusion in dry HB1 melt
(Table 1) at 1623-1773 K and in air. Liang et al. (1996a) explored Ca self diffusion in various
dry CaO-Al2O3-SiO2 melts at 1773 K and 1 GPa. Mungall et al. (1999) examined Ca tracer
diffusion in dry HR7 and dry HR7 + Na at 0.1 MPa and wet HR7 (3.6 wt% H2O) melts at 1
GPa. Roselieb and Jambon (2002) studied tracer diffusion of Ca in dry jadeite and albite melts
at 923-1293 K and 0.1 MPa. Lundstrom (2003) obtained SEBD of Ca in dry basalt7-basanite
couple at 1723 K and 0.9 GPa. Gabitov et al. (2005) examined Ca SEBD in two haplorhyolite
(HR6 and HR9 in Table 1) melts during fluorite dissolution at 1173-1273 K, 0.075-0.1 GPa,
and 1.2-4.8 wt% H2O. Morgan et al. (2006) measured Ca SEBD in low-Ti and high-Ti lunar
picrite melts (labeled LP1 and LP2 in Table 1) during anorthite dissolution at 1673 K and 0.6
GPa. Chen and Zhang (2008, 2009) determined Ca SEBD in basalt11 melts during olivine and
clinopyroxene dissolution. During olivine dissolution, the Ca concentration gradient is small,
and Ca diffusion is dominated by cross-diffusivities, and sometimes even show uphill diffusion
(e.g., Zhang et al. 1989). Hence, Ca effective binary diffusivities extracted by Chen and Zhang
(2008) may be useful in some aspects but are not comparable with Ca diffusion controlled
by its own concentration gradient. Ca SEBD values during olivine dissolution are hence not
modeled quantitatively below; only Ca SEBD values during clinopyroxene dissolution are
considered. Melt compositions are listed in Table 1 except for diopside-anorthite and CaO-
Al2O3-SiO2 melts.
Ca diffusion data in dry natural and nearly natural melts are shown in Figure 12. SEBD
data of Ca (close to FEBD) in mugearite melt at 1503-1696 K and 0.1 MPa by Medford (1973)
are scattered and do not define a good Arrhenius trend (Fig. 12a), reflecting the experimental
and analytical difficulties in the early years of diffusion studies.
Ca self diffusivities in HB1 melts at 1623-1773 K and 0.1 MPa (LaTourrette et al. 1996),
SEBD in lunar picrite melts during anorthite dissolution at 1673 K and 0.6 GPa (Morgan et
al. 2006), and SEBD in basalt11 during clinopyroxene dissolution at 1509-1790 K and 0.5-1.9
GPa (Chen and Zhang 2009) are similar and independent of pressure, and can be fit by the
following equation:
(19138 ± 2349)
SD & SEBD =
-(11.79 ± 1.43) -
basalt
ln DCa (24)
T
The maximum error of Equation (24) in reproducing diffusivities in the three experimental
studies (LaTourrette et al. 1996; Morgan et al. 2006; Chen and Zhang 2009) is 0.30 in lnD.
336 Zhang, Ni, Chen
figure 12. Ca diffusivity in dry (a) mafic melts (SiO2 ≤ 60 wt%) and (b) silicic melts (SiO2 > 60 wt%).
The legend Fig. 12. Ca
contains diffusivity
information in dry
about (a) mafic
the type melts (SiO
of diffusion, 2 ≤ 60
the melt wt%) and(Mu
composition (b) means
silicic mugearite;
melts (SiO2 > 60 wt%). The legend
b-b means basalt7-basanite couple; and1=andesite1; rhy5=rhyolite5; rhy12(ap) means D in rhyolite 12
during apatite dissolution;
contains r-r means
information rhyolite3-rhyolite16
about couple;the
the type of diffusion, a-r melt
meanscomposition
andesite3-rhyolite16
(Mu meanscouple),
mugearite; b-b means basalt7-
pressure, and authors (M73: Medford 1973; L96: LaTourrette et al. 1996; L03: Lundstrom 2003; M06:
Morgan et al. 2006; C09:
basanite Chenand1=andesite1;
couple; and Zhang 2009; rhy5=rhyolite5;
Z89: Zhang et al. 1989; J82: Jambon
rhy12(ap) means 1982; H84: Harrison
D in rhyolite 12 during apatite dissolution; r-r
and Watson 1984; V94: van der Laan et al. 1994; M99: Mungall et al. 1999).
means rhyolite3-rhyolite16 couple; a-r means andesite3-rhyolite16 couple), pressure, and authors (M73: Medford
However, two CaL96:
1973; SEBD values in
LaTourrette basalt7-basanite
et al. diffusion
1996; L03: Lundstrom couple
2003; M06:(Lundstrom 2003)
Morgan et al. 2006;are
C09: Chen and Zhang 2009
greater than those predicted by Equation (24) by 1.5 to 2.0 lnD units (Fig. 12a), and those in
mugearite melt
Z89:(Medford 1973)
Zhang et al. 1989;are lower
J82: by up
Jambon to 1.2
1982; lnD
H84: units. and Watson 1984; V94: Van Der Laan et al. 1994; M99
Harrison
Ca diffusivities in rhyolite melts are more variable than in basalt melts. In dry HR7 melt
Mungall et al. 1999).
(among the most silicic rhyolite), Ca tracer diffusivity at 1410-1873 K and 0.1 MPa (Mungall
et al. 1999) can be expressed as:
12
(29830 ± 4356)
TD = -(10.58 ± 2.63) -
dry HR7
ln DCa (25)
T
The maximum error Equation (25) in reproducing the four experimental diffusivities is 0.28
in lnD.
Adding H2O enhances Ca diffusivity significantly.
Sr diffusion. Ten papers reported Sr diffusion data in silicate melts (110 points, all except
for one are tracer or self diffusivities). Hofmann and Magaritz (1977) studied Sr tracer diffusion
in dry basalt10 melt at 1532-1719 K and 0.1 MPa. Magaritz and Hofmann (1978a) determined
Sr tracer diffusivities in dry rhyolite15 melt at 948-1226 K and 0.1 MPa. Lowry et al. (1982)
explored tracer diffusion of 85Sr in dry basalt1 and andesite2 melts at 1566-1676 K and 0.1
MPa. Lesher (1994) investigated Sr self diffusion in dry rhyolite2 melt at 1528-1738 K and 1
GPa, and in dry basalt3 melt at 1528 K and 1 GPa. Perez and Dunn (1996) examined Sr tracer
diffusion in rhyolite8 melt (containing < 0.8 wt% H2O, but H2O was not reported for every
sample) at 1273-1723 K and 1 GPa. Nakamura and Kushiro (1998) reported tracer diffusivities
of Sr in jadeite and diopside melts at a single temperature (1673 K for jadeite and 1863 K for
diopside) and some pressures. Mungall et al. (1999) investigated Sr tracer diffusion in dry
HR7 at 1410-1873 K and 0.1 MPa, wet HR7 (3.6 wt% H2O) at 1 GPa and two temperatures
(1573 and 1873 K), and dry HR7 + Na at 1083-1773 K and 0.1 MPa. Roselieb and Jambon
(2002) examined Sr tracer diffusivities in albite and jadeite melts at 918-1293 K and 0.1 MPa.
Lundstrom (2003) obtained a single SEBD of Sr between basalt7-basanite melt at 1723 K and
0.9 GPa. Behrens and Hahn (2009) characterized Sr tracer diffusion in dry and wet trachyte
and phonolite melts at 1323-1573 K and 0.5 GPa. Sr diffusivity data are shown in Figure 13.
Sr diffusivities are the lowest in dry HR7 melt among the melts investigated. Sr self
diffusivities in dry rhyolite2 are similar to Sr tracer diffusivities in rhyolite8 containing < 0.8
wt% H2O and in dry rhyolite15. Self and tracer diffusivities of Sr increase from dry HR7, to
dry natural metaluminous rhyolite and dry trachyte, then dry andesite, then dry basalt and dry
phonolite, then wet trachyte, and then wet phonolite. Some of the equations are given below.
Self and tracer diffusivity of Sr in rhyolite2 at 1 GPa (Lesher 1994), rhyolite8 at 1 GPa
(Perez and Dunn 1996) and rhyolite15 at 0.1 MPa (Magaritz and Hofmann 1978a) can be fit
as follows:
≤1 GPa  (18745 ± 824) 

DSrdrySDrhyolites,
& TD = exp ( -14.80 ± 0.65) -  (26)
 T 
The maximum error of Equation (26) in reproducing the diffusion data is 0.89 in lnD (0.39 in
logD). The activation energy is 156±7 kJ/mol.
Sr tracer diffusivities in dry basalt1 and basalt10 melts at 1567-1675 K and 0.1 MPa
(Hofmann and Magaritz 1977; Lowry et al. 1982) can be fit as follows:
 (20361 ± 3881) 
DSrdryTDbasalt = exp ( -11.40 ± 2.40) -  (27)
 T 
In dry and wet trachyte melts, Sr tracer diffusivities at 1323-1527 K, 0.5 GPa and ≤ 1.7
wt% H2O (Behrens and Hahn 2009) are roughly linear to H2O content, and can be fit as follows:
 (22435 ± 7571) - (3430 ± 545)w 
DSrdryTDtrachyte1 = exp ( -12.71 ± 5.23) -  (28)
 T 
338 Zhang, Ni, Chen
Fig. 13.figure 13. Srin diffusivities

Sr diffusivities silicate melts. in
Thesilicate melts.isThe
default pressure default
0.1 MPa. pressure
Data is be
sources can 0.1found
MPa.in the text.
Data sources can be found in the text.
There are not enough data to resolve how Sr tracer diffusivities in phonolite melts depend
on H2O. Assuming the dependence is similar to trachyte melts, then Sr tracer diffusivities in
dry and wet phonolite at 1373-1528 K, 0.5 GPa and ≤ 1.9 wt% H2O (Behrens and Hahn 2009)
can be fit as follows:
 (24708 ± 6624) - (1644 ± 299)w 
DSrphonolite1
TD = exp ( -8.98 ± 4.56) -  (29)
 T 
Ba diffusion. Twelve papers reported Ba diffusion data in silicate melts (117 points),
most of which are tracer or self diffusivities. Hofmann and Magaritz (1977) explored Ba tracer
diffusion in dry basalt10 melt at 1523-1723 K and 0.1 MPa. Magaritz and Hofmann (1978a)
determined Ba tracer diffusivities in dry rhyolite15 melt at 973-1226 K and 0.1 MPa. Jambon
(1982) reported Ba tracer diffusivities in rhyolite5 melt at 905 and 1083 K and 0.1 GPa. Lowry
et al. (1982) characterized Ba tracer diffusion in basalt1 and andesite2 melts at 1572-1673 K
13
and 0.1 MPa. Henderson et al. (1985) studied Ba tracer diffusion in dacite2 (1573-1672 K)
and pantellerite1 (1470-1575 K) melts at 0.1 MPa. LaTourrette et al. (1996) examined Ba self
diffusion in dry HB1 melt at 1623-1773 K and in air. Nakamura and Kushiro (1998) obtained
Ba tracer diffusivities in dry jadeite and diopside melts at 1573-1723 K and 0.75-2.0 GPa.
Mungall et al. (1999) characterized Ba tracer diffusion in dry HR7 and dry HR7 + Na at 0.1
MPa and wet HR7 (3.6 wt% H2O) melts at 1 GPa. Koepke and Behrens (2001) measured Ba
tracer diffusivities in wet HA1 melt (4.5-5.2 wt% H2O) and one datum for dry HA1. Roselieb
and Jambon (2002) investigated Ba tracer diffusion in dry albite and jadeite melts at 1073-
1293 K and 0.1 MPa. Lundstrom (2003) acquired SEBD of Ba in dry basalt7-basanite couple
at 1723 K and 0.9 GPa. Behrens and Hahn (2009) examined Ba tracer diffusion in dry and wet
trachyte and phonolite melts. Ba diffusion data are shown in Figure 14.
Ba diffusivities increase with decreasing viscosity (from polymerized silicic melt to
depolymerized melt, and from dry to wet melt). Ba tracer and self diffusivities in dry basalt1
(Lowry et al. 1982), basalt10 (Hofmann and Magaritz 1977) and HB1 (LaTourrette et al. 1996)
melts are consistent and can be fit by the following equation:

 (15691 ± 3800) 
dry basalt; 0.1 MPa
DBa SD and TD = exp ( -14.68 ± 2.31) -  (30)
 T 
Equation (30) can reproduce the data to within 0.56 lnD units.
Experimental data in trachyte and phonolite melts are not enough to resolve how D
depends on H2O. We assume that lnD increases linearly with H2O to fit the data. For dry and
figure 14. Ba diffusivities in selected melts. Melt compositions can be found in Table 1. (a) Dry melts. (b)
ComparisonFig.of14.
dryBa diffusivities
and in selected
wet melts. The melts. in Melt
H2O contents compositions
wet melts can bewet
are as follows: found
HA1:in4.5-5.2
Table 1.
wt%;(a)wet
Dry melts. (b)
tra (trachyte1): 1.1-1.7 wt%; wet pho (phonolite1): 1.6-1.9 wt%; wet HR7: 3.6 wt%.
Comparison of dry and wet melts. The H2O contents in wet melts are as follows: wet HA1: 4.5-5.2 wt%; wet tra
(trachyte1): 1.1-1.7 wt%; wet pho (phonolite1): 1.6-1.9 wt%; wet HR7: 3.6 wt%.
340 Zhang, Ni, Chen
hydrous trachyte1 and phonolite1 melts, Ba tracer diffusivities at 1323-1528 K, 0.5 GPa, 0-1.9
wt% H2O can be fit as:
 (20885 ± 6797) - (3065 ± 489)w 
trachyte1; 0.5GPa
DBa TD = exp ( -14.34 ± 4.67) -  (31)
 T 
 (25996 ± 3412) - (1805 ± 190)w 
phonolite1; 0.5GPa
DBa TD = exp ( -8.63 ± 3.35) -  (32)
 T 
Equations (31) and (32) can reproduce the data to within 0.60 and 0.23 lnD units, respectively.
Ra diffusion. No Ra diffusion data in silicate melts are known.
Summary of alkali earth diffusion. For the diffusion of alkali earth elements in dry
HR7 + Na melt (Mungall et al. 1999), the diffusivity decreases from Ba to Sr, to Ca, to Mg,
and to Be. That is, the diffusivity sequence for the alkali earth elements is: Ba > Sr > Ca > Mg
> Be. In dry HR7 melt (Mungall et al. 1999), the diffusivity sequence is: Ba ≈ Sr > Ca > Mg >
Be (Fig. 15a). In wet HR7 melt containing 3.6 wt% H2O (Mungall et al. 1999), the diffusivity
sequence is: Sr > Ba ≈ Ca > Mg > Be (Fig. 15b). In basalt10 melt (Hofmann and Magaritz
1977), the diffusivity sequence is: Ca ≈ Co > Sr > Ba, opposite to the trend in HR7 + Na. In
HB1 melt (LaTourrette and Wasserburg 1996), the diffusivity sequence is: Mg > Ca > Ba,
similar to the trend in basalt10. In albite and jadeite melts (Roselieb and Jambon 1995), the
diffusivity sequence is: Sr ≈ Ca > Ba > Mg.
Hence, for the alkali earth elements, there is no simple relation between diffusivity and
cation size: the diffusivity may increase or decrease with increasing size, depending on the
melt composition and other factors. That larger cations with the same valence may diffuse
more rapidly in some melts may be counterintuitive to some. Nonetheless, this trend will be
encountered in diffusion of other isovalent series (REE, B, Al, Ga, Si-Ge, etc.). This and other
trends will be discussed in a later summary section.
B, al, Ga, In, and tl
B diffusion. Three papers reported B diffusion data in silicate melts (70 points). Baker
(1992a) investigated B tracer diffusion in dacite1 and rhyolite14 melts at 1573-1873 K and 1
GPa. Chakraborty et al. (1993) studied FEBD of B in diffusion couples with one half made
of HR7 and the other half made of HR7 plus 5 wt% or 10 wt% B2O3 at 1473-1873 K and 0.1
MPa. Mungall et al. (1999) examined B tracer diffusion in dry HR7 melt at 0.1 MPa and 1410
and 1673 K, dry HR7 + Na melt at 1083-1473 K and 0.1 MPa, and wet HR7 (3.6 wt% H2O) at
1573-1873 K and 1 GPa. The data are shown in Figure 16.
As shown by Chakraborty et al. (1993), FEBD values of B decrease as SiO2 concentration
increases along a diffusion couple. When FEBD values of B at the HR7 end with 75-79 wt%
SiO2 (Chakraborty et al. 1993) are compared with tracer diffusivities of B in HR7 (Mungall
et al. 1999), they are consistent (Fig. 16). B tracer diffusivities in rhyolite14 (76 wt% SiO2)
are similar to those in HR7 + B melts containing 70-75 wt% SiO2. B diffusivities (FEBD by
Chakraborty et al. 1993 and tracer diffusivities by Mungall et al. 1999) in dry HR7 melt at
1410-1873 K, 0.1 MPa, and B2O3 concentration ≤ 10 wt% can be fit as:
 (39664 - 1162C ) 
DBdryTD&FEBD
HR7, 0.1 MPa
= exp  -8.56 - 0.486C -  (33)
 T 
where C is wt% of B2O3. All B diffusion data in Chakraborty et al. (1993) and Mungall et al.
(1999) can be reproduced by Equation (33) to within 0.62 lnD units. B diffusivity in rhyolite14
melt at 1573-1873 K and 1 GPa (Baker 1992) is about 6 times the diffusivity calculated using
Equation (33).
figure 15. Comparison

Fig. 15. Comparison of Be,
of Be, Mg, Ca, Mg,BaCa,
Sr and Sr and Baindiffusivities
diffusivities dry and wetinHR7 melt (Mungall et al. 1999).
dry and wet HR7 melt (Mungall et al. 1999).
When compared with other cations, B diffusivities are similar to diffusivities of trivalent
cations Lu and Ga in HR7 melts.
Al diffusion. Twelve papers reported Al diffusion data in silicate melts (127 points). Cooper
and Kingery (1964) explored Al diffusion (SEBD) in CaO-Al2O3-SiO2 melt during sapphire
dissolution at 1618-1823 K and 0.1 MPa. Cooper and coworkers also investigated diffusion
in other synthetic systems of interests to ceramic and glass scientists, such as K2O-SrO-SiO2,
etc. These works are not covered here because we focus on geologically relevant silicate
15
melts. Baker and Watson (1988) investigated Al diffusion (SEBD) in rhyolite1-rhyolite8 and
HD2-rhyolite8 couples (Table 1) at 1171-1273 K and 0.01 GPa and 1373-1473 K and 0.2-1
GPa. Zhang et al. (1989) determined Al SEBD in andesite1 melt during dissolution of olivine,
clinopyroxene, spinel and quartz at 1488-1673 K and 0.55-2.15 GPa. Because the interface
342 Zhang, Ni, Chen
figure 16. B diffusivities

Fig. in silicate
16. B diffusivities in silicatemelts.
melts. Data sources:
Data sources: FEBD
FEBD at 71-79
at 71-79 wt%
wt% SiO SiO
2 at 0.1 2 at(Chakraborty
MPa 0.1 MPa (Chakraborty
et al.
et al. 1993); tracer diffusivities in dry HR7 at 0.1 MPa, in HR7 + Na at 0.1 MPa, and in wet HR7 (3.6 wt%
1993); tracer diffusivities in dry HR7 at 0.1 MPa, in HR7+Na at 0.1 MPa, and in wet HR7 (3.6 wt% H2O) at 1 GPa
H2O) at 1 GPa (Mungall et al. 1999); and in rhyolite14 and dacite1 at 1 GPa (Baker 1992a).
(Mungall et al. 1999); and in rhyolite14 and dacite1 at 1 GPa (Baker 1992).
melt composition changes from basalt (during olivine dissolution) to rhyolite (during quartz
dissolution), it is important to include the interface melt composition variation to understand
the data. Kubicki et al. (1990) obtained Al SEBD in diopside-anorthite (Di-An) melts at 1633-
1923 K and 0.1-2 GPa. Most of their data are at 0.2 GPa, and compositional change from
Di100-Di80An20 couple to Di60An40-Di40An60 couple does not affect Al diffusivities in a major
way. Baker and Bossanyi (1994) examined the effect of H2O and F on Al SEBD in rhyolite8-
dacite1 diffusion couples at 1373-1673 K and 1 GPa. van der Laan et al. (1994) reported Al
SEBD in rhyolite3-rhyolite16 and andesite3-rhyolite16 diffusion couples. Liang et al. (1996a)
investigated Al self diffusion in CaO-Al2O3-SiO2 systems at 1773 K and 1 GPa. Van Orman
and Grove (2000) obtained a single datum for Al SEBD in a lunar basalt melt (LB1 in Table
1) during clinopyroxene dissolution at 1623 K and 1.3 GPa. Lundstrom (2003) obtained two
data points for Al SEBD in basalt7-basanite couple. Morgan et al. (2006) determined Al SEBD
in LP1 and LP2 melts during anorthite dissolution at 1673 K and 0.6 GPa. Chen and Zhang
(2008, 2009) obtained Al SEBD data in basalt11 melt (with 0.3-0.4 wt% H2O) during olivine
16
and diopside dissolution at 1543-1790 K and 0.5-1.9 GPa.
Despite the numerous papers on Al diffusivities, most Al diffusivity data were obtained as
side-products. Except for Liang et al. (1996a) who investigated Al self diffusion in a ternary
system at a single temperature (1773 K), other studies are all on SEBD of Al in various melts.
No tracer diffusivities or FEBD on Al are available. Furthermore, Al diffusion data in some
papers are scattered. Hence, Al diffusion in natural silicate melts is not very well constrained.
Experimental Al SEBD data in dry melts are shown in Figure 17. Al diffusivities decrease
from lunar picrites and basalt-basanite couple, to basalt11, and to andesite1. For basalt to
andesite melts, the data are less scattered and the trends with SiO2 content is consistent. Al
SEBD data for Di-An melts (ranging from Di100-Di80An20 couple to Di60An40-Di40An60 couple)
vary more widely, but are roughly the same as those in basalt11 melt.
Al diffusivities (SEBD) in basalt11 melt (containing about 0.3-0.4 wt% H2O) during both
olivine and clinopyroxene dissolution at 1509-1790 K and 0.5-1.9 GPa (Chen and Zhang 2008,
2009) are similar and roughly independent of pressure, and can be expressed as:
figure 17.Fig.
SEBD of Al
17. SEBD in in
of Al drydrymelts.
melts. Indicated
Indicated inin
thethe legends
legends are theare
meltthe melt composition
composition listed in Tablelisted
1 (b-b in Table 1 (b-b
means
means basalt7-basanite couple with 46 wt% average SiO2), pressure, and references (C08: Chen and Zhang
basalt7-basanite couple with 46 wt% average SiO2), pressure, and references (C08: Chen and Zhang 2008; C09:
2008; C09: Chen and Zhang 2009; L03: Lundstrom 2003; M06: Morgan et al. 2006; V00: Van Orman and
Grove 2000;
ChenK90: Kubicki
and Zhang 2009; et al.Lundstrom
L03: 1990; Z89:2003;Zhang et al.et1989).
M06: Morgan al. 2006; V00: Van Orman and Grove 2000; K90:
Kubicki et al. 1990; Z89: Zhang et al. 1989).
 (31293 ± 2599) 
SEBD = exp ( -5.75 ± 1.71) -
basalt11
DAl  (34)
 T 
Equation (34) reproduces the experimental data to within 0.44 lnD units. Furthermore,
Equation (34) can also roughly predict Al SEBD in Di-An melts (Fig. 17) with a maximum
error of 1.1 lnD units.
Al diffusivities (SEBD) in andesite1 melt (containing about 0.04 wt% H2O) during
olivine, clinopyroxene and spinel dissolution at 1488-1673 K and 0.5-2.1 GPa (Zhang et al.
1989) are similar and roughly independent of pressure, and can be expressed as:
 (37649 ± 5848) 
SEBD = exp ( -2.52 ± 4.95) -
andesite1
DAl  (35)
 T 
Equation (35) reproduces the experimental data17to within 0.79 lnD units. Al diffusivity during
quartz dissolution (Zhang et al. 1989) is lower than that predicted by Equation (35) by 2.05
lnD units, attributed to a large difference in the interface melt composition during quartz
dissolution compared to dissolution of the other minerals.
Data in Figure 17 show systematic dependence of Al SEBD on SiO2, although the data are
scattered. We fit all SEBD data on natural silicate melts as follows:
 (23111 + 5918 XSi ) 
basalt to andesite
DAl SEBD = exp ( -0.88 - 18.02 XSi ) -  (36)
 T 
Equation (36) reproduces the experimental data of Zhang et al. (1989), Van Orman and Grove
(2000), Lundstrom (2003), Morgan et al. (2006), and Chen and Zhang (2008, 2009) to within
1 lnD unit (or 0.43 logD units).
Al is a network-forming element. When compared to the diffusion of the most common
major network-forming element Si, Al diffusivity is often slightly higher than Si diffusivity,
344 Zhang, Ni, Chen
about 1.5 to 2.5 times the Si diffusivity (Zhang et al. 1989; Liang et al. 1996a; Chen and Zhang
2008, 2009). Hence, when Al diffusivity in a melt is not known but Si diffusivity in the melt is
known, Al diffusivity can be roughly estimated to be two times the Si diffusivity.
Ga diffusion. Two papers reported Ga tracer diffusion data in silicate melts (18 points).
Baker (1992a) investigated Ga diffusion in dacite1 and rhyolite14 melts (Table 1) at 1573-1873
K and 1 GPa. Baker (1995) examined Ga diffusion in albite melt at 1427-1775 K and 0.1 MPa.
The data are
Zhangsummarized in Figure
et al. (Ch 8) Diffusion 18.melts
data in silicate Mungall (2002) dismissed Ga diffusion data by Baker
(1992a).
figure
Fig. 18. Ga tracer 18. Ga
diffusivities tracermelts.
in silicate diffusivities in silicate
Data sources: melts.
albite (Baker Data
1995); sources:
rhyolite14 and dacite1 (Baker
albite (Baker 1995); rhyolite14 and dacite1 (Baker 1992a).
1992).
In diffusion. No In diffusion data in silicate melts are known.

Tl diffusion. Only one paper reported Tl diffusion data in silicate melts. MacKenzie and
Canil (2008) studied Tl tracer diffusion in dry CMAS1 and NMAS1 (Table 1) melts using
devolatization (desorption) experiments at 1473-1623 K and 0.1 MPa. The diffusion data are
summarized in Figure 19. There is considerable scatter in the data.
Comparison of diffusivities of B, Al, Ga, In, and Tl. There are not enough data to compare
diffusivities of B, Al, Ga, In and Tl because (i) there are no data on In diffusion; (ii) the only Tl
diffusion data are for CMAS1 and NMAS1 systems, different from melt compositions studied
for the other elements; (iii) the only common melts that have been investigated for B and Ga
diffusion are dacite1 and rhyolite14 melts; and (iv) Al diffusion data in dacite and rhyolite melts
are SEBD data and are highly scattered. Hence, only B and Ga diffusivities are compared in
Figure 20. In dacite1 melt, diffusivities of the two elements are similar. However, in rhyolite15
melt, B diffusivity is lower than Ga diffusivity by a factor of 2 to 4. The latter is consistent with
the trend shown by the alkali earth elements with 18 smaller cations having smaller diffusivity.
Nonetheless, the difference in diffusivity is small. It is likely that Al diffusivities lie between
those of B and Ga, meaning that B, Al and Ga all have similar diffusivities. Because Tl is
volatile, its diffusivities are expected to be higher than those of B, Al and Ga.
figure 19. Tl diffusivi-

ties in silicate melts. Data
source: MacKenzie and
Canil (2008).
Fig. 19. Tl diffusivities in silicate melts. Data source: MacKenzie and Canil (2008).

of B and Ga diffusivities
in dacite1 and rhyolite14
melts.
19
Fig. 20. Comparison of B and Ga diffusivities in dacite1 and rhyolite14 melts.
C, Si, Ge, Sn and Pb

C diffusion. C diffusion in silicate melts in the form of dissolved molecular CO2 and
carbonate ion CO32− is reviewed in another chapter (Zhang and Ni 2010). Data for carbon
diffusion in other forms (such as dissolved molecular CO) are not available.
Si diffusion. Si is the second most major element in silicate melts (next to oxygen), and it
controls the melt structure. Hence, its diffusion has been investigated extensively, with about 23
papers and more than 262 diffusivity values (some of the diffusivities are reported as smooth
trends rather than data points). Lesher and Walker (1986, 1991) obtained Si SEBD from Soret
diffusion experiments on nephelinite, basalt8, limburgite, basalt11, leuconorite, nordmarkite,
trachyte, pantellerite2, FaLcQ (fayalite-leucite-quartz), rhyolite4, and rhyolite10 melts (Table
1). Baker and Watson (1988) investigated Si SEBD in rhyolite1-rhyolite8 and HD2-rhyolite8
couples at 1171-1273 K and 0.01 GPa and 1273-1473 K and 0.2-1 GPa. Koyaguchi (1989)
reported Si SEBD in basalt-dacite and basalt-rhyolite couples at 1473-1773 K, 1 GPa, and 0.39-
346 Zhang, Ni, Chen
13.77 wt% H2O. The compositions used by Koyaguchi (1989) include basalt2, basalt9, dacite3,
and rhyolite13. Zhang et al. (1989) obtained Si SEBD in andesite1 melt during dissolution of
olivine, spinel, rutile and quartz at 1488-1673 K and 0.55-1.5 GPa. Because the interface melt
composition changes from basalt (during olivine dissolution) to rhyolite (during quartz dissolu-
tion), it is important to include the interface melt composition variation to understand the data.
Baker (1991, 1993) and Baker and Bossanyi (1994) examined the effect of H2O, F and Cl on
Si SEBD in rhyolite8-dacite3 diffusion couples at 1373-1673 K and 1 GPa, and found that the
effect of F and Cl is negligible (D increases by less than a factor of 2 when F and Cl concentra-
tions are 1 wt% or less), and that of H2O is significant. Kubicki et al. (1990) reported Si SEBD
in diopside-anorthite melts at 1873 K and 0.2 GPa. Baker (1992a) studied Si tracer diffusion
in rhyolite8 and dacite1 melts at 1573-1773 K and 1 GPa. In these experiments, a layer of
sodium silicate glass powder (with 73 mol% SiO2 and 27 mol% Na2O, or close to Na6Si8O19)
with thickness < 0.1 mm enriched in 30Si (a stable silicon isotope, meaning that the experiments
are more like self diffusion experiments) is loaded on a dacite or rhyolite glass cylinder. The
sample is loaded into a piston-cylinder assemblage and heated up. Because the typical Si dif-
fusion length in the experiments of Baker (1992a) is about 0.05 mm (Fig. 1 in Baker 1992a),
which might be thinner than the thickness of the loaded film (< 0.1 mm), the experiments may
not be true tracer diffusion experiments, but are close to diffusion couple experiments between
a synthetic sodium silicate on one half and either dacite or rhyolite on the other half, with an Si
isotopic gradient between the two halves. The extracted diffusivities are likely isotopic effec-
tive binary diffusivities (IEBD), although such diffusivities based on isotopic fraction profiles
are often not too far off self diffusivities (Zhang 1993; Lesher 1994; van der Laan et al. 1994).
van der Laan et al. (1994) reported Si SEBD in rhyolite3-rhyolite16 and rhyolite16-
dacite3 diffusion couples. Baker (1995) investigated Si SD in albite melt at 1438-1831 K and
0.1 MPa. Lesher et al. (1996) investigated Si self diffusion in basalt6 melt (Table 1) at 1593-
1873 K and 1 GPa, and 1673 K and 2 GPa. Liang et al. (1996a) investigated Si self diffusion
in the CaO-Al2O3-SiO2 system. Liang et al. (1996b) extracted multicomponent diffusivity
matrices in the CaO-Al2O3-SiO2 system (see review in Liang 2010, this volume), and also
SEBD of Si, but the SEBD data were only shown in a figure without the melt composition. Poe
et al. (1997) measured Si self diffusivity in NS4 (Na2Si4O9) melt at 2100 to 2800 K and 10-15
GPa. Van Orman and Grove (2000) estimated Si SEBD in a lunar basalt melt (LB1 in Table
1) during clinopyroxene dissolution at 1623 K and 1.3 GPa. Reid et al. (2001) studied Si self
diffusion in diopside melt at 2073-2573 K and 3-15 GPa. Tinker and Lesher (2001) examined
Si self diffusion in dry HD2 melt at 1628-1935 K and 1.0-5.7 GPa. Lundstrom (2003) obtained
one datum of Si SEBD in basalt7-basanite couple. Tinker et al. (2003) investigated Si SD
in diopside-anorthite (Di52An48) melt at 1783-2037 K and 1-4 GPa. Morgan et al. (2006)
determined Si SEBD in two lunar picrite melts (LP1 and LP2 in Table 1) during anorthite
dissolution at 1673 K and 0.6 GPa. Chen and Zhang (2008, 2009) obtained Si SEBD in
nominally dry basalt11 melt (about 0.3-0.4 wt% H2O) during olivine and clinopyroxene
dissolution at 1543-1790 K and 0.5-1.9 GPa.
Si self diffusivities. In dry silicate melts, silicon self diffusivity is similar to and often
slightly lower than oxygen self diffusivity (Fig. 21). The maximum difference is about 1.5
lnD units and this difference occurs in silicic melts with 69 wt% SiO2. The limited data
apparently indicate that when diffusivities are low (<10−12 m2/s), silicon self diffusivities can
be significantly lower than oxygen self diffusivities. Discussion is abundant in the literature
about the similarity of oxygen self diffusivity with the Eyring diffusivity, at least in dry melts
(e.g., Shimizu and Kushiro 1984; Tinker et al. 2004) but clearly not in wet melts (Behrens et
al. 2007; Zhang and Ni 2010, this volume). It is possible that silicon self diffusivity matches
better with the Eyring diffusivity in all melts, whereas oxygen self diffusivity is similar to the
Eyring diffusivity only in dry melts containing < 65 wt% SiO2. Further experimental study is
necessary to verify this hypothesis.
figure 21.Fig.
Comparison
21. Comparisonof of
silicon
silicon self diffusivity
self diffusivity and and oxygen
oxygen self diffusivity
self diffusivity in dry melts.in Data
dry sources
melts.and
Data sources and
conditions: basalt6 melt (Lesher 1996; 1593-1873 K and 1-2 GPa); CAS (Liang et al. 1996a; CaO-Al2O3-
SiO2 meltsconditions:
at 1773 basalt6
K andmelt 1 (Lesher
GPa); 1996;
NS41593-1873 1-2 GPa); CAS (Liang et al. 1996a; CaO-Al2O3-SiO2 melts
(Poe et Kal.and1997; Na2Si4O9 melt, 2500-2800 K and 10-15 GPa);
diopside melt (Reid
at 1773 K andet1 al. 2001;
GPa); 2073-2573
NS4 (Poe et al. 1997;KNaand
Si O
2 4 9
3-15
melt,GPa); HD2
2500-2800 (Tinker
K and and Lesher
10-15 GPa); 2001;
diopside melt haplodacite
(Reid et
melt; 1628-1935 K and 1-5.7 GPa); Di58An42 melt (Tinker et al. 2003; 1783-2037 K and 1-4 GPa).
al. 2001; 2073-2573 K and 3-15 GPa); HD2 (Tinker and Lesher 2001; haplodacite melt; 1628-1935 K and 1-5.7
GPa); Di58An42 melt (Tinker et al. 2003; 1783-2037 K and 1-4 GPa).
Si self diffusion in HD2 melt has been investigated systematically (Tinker and Lesher
2001). The dependence with pressure is not monotonic: as the pressure increases from 1 to 4
GPa, the diffusivity increases; but as the pressure increases further, the diffusivity decreases.
There are not enough data at P > 4 GPa to constrain how Si self diffusivity varies with pressure.
Hence, restricting our attention to 1-4 GPa, Si self diffusivity in dry HD2 melt at 1628-1935
K can be fit as follows:
 (54245 ± 690) - (8523 ± 2984)P 
DSidrySDHD2 = exp ( -3.913 ± 1.603)P -  (37)
 T 
Equation (37) can reproduce the experimental data at 1-4 GPa to within 0.66 lnD units. It can
be extrapolated to 0 GPa, but cannot be extrapolated to ≥4 GPa.
Effective binary diffusivities of Si. Most Si21diffusivities are SEBD values from diffusion
couples made of rhyolite-basalt, rhyolite-dacite, or basalt7-basanite, or mineral dissolution
experiments in basalt and andesite melts. In these experiments, as well as in nature, the SiO2
gradient is often the largest gradient, or one of the largest. Hence, even though it is expected
that the gradients of other concentration gradients would affect SiO2 diffusion due to cross
diffusion effects, such effects are not expected to completely overwhelm SiO2 diffusion.
Figure 22a compares different kinds of Si diffusivities and the maximum difference is 1.6
lnD units (0.7 logD units). Therefore, SiO2 SEBD in most geological cases is expected to be
roughly the same as the self diffusivity. However, for synthetic melts (such as the CaO-Al2O3-
SiO2 system, Liang et al. 1996a,b) where SiO2 gradient may be small compared to gradients
of other components, the difference between self diffusivities and SEBD can be very large.
Experimental SEBD values of Si in diffusion couple or Soret diffusion experiments usually
decrease from basalt to rhyolite, and lnD versus SiO2 for typical natural melts is roughly linear
(Fig. 22b) (Lesher and Walker 1986, 1991; Koyaguchi 1989). Lesher and Walker (1986, 1991)
showed based on Soret diffusion experiments that lnDSi decreases linearly with increasing
348 Zhang, Ni, Chen
figure 22. (a) Comparison between SD (self diffusivities) and SEBD of Si. Data sources: SD in basalt6
Fig. 22. (a) Comparison between SD (self diffusivities) and SEBD of Si. Data sources: SD in basalt6 (Lesher
(Lesher et al. 1996); SEBD in basalt12 (Lesher and Walker 1986); SEBD in b-d (basalt2-dacite3; Koyaguchi
1989); SEBD in b-b (basalt7-basanite; Lundstrom 2003); SEBD in basalt11 (Chen and Zhang 2008, 2009).
(b) SEBD 1996); SEBD
of Si versus in2 basalt12
SiO (Lesher
content. Data and LW86
sources: Walker(Soret
1986); SEBDresults
diffusion in b-dof(basalt2-dacite3; Koyaguchi 1989); SEBD in
Lesher and Walker
1986); K89 (diffusion couple results of Koyaguchi 1989).
(basalt7-basanite; Lundstrom 2003); SEBD in basalt11 (Chen and Zhang 2008, 2009). (b) SEBD of Si versus SiO
SiO2: in dry melts atData

content. 1748sources:
K, lnDSiLW86 m2/s) diffusion
(D in(Soret decreasesresults
from of
−24.2 in aand
Lesher basalt with1986);
Walker 51 wt%K89 (diffusion couple resul
SiO2 to −26.8 in a rhyolite with 75 wt% SiO2 (a factor of 13, or 1.1 orders of magnitude).
KoyaguchiKoyaguchi 1989). from diffusion couple experiments that lnDSi decreases linearly
(1989) showed
with increasing SiO2: in dry melts at 1773 K, lnDSi (D in m2/s) decreases from −24.2 in a
basalt with 50 wt% SiO2 to −27.6 in a rhyolite with 74 wt% SiO2 (a factor of 30, or 1.5 orders
of magnitude).
22
Baker (1993) and Baker and Bossanyi (1994) examined the effect of H2O, F and Cl on Si
SEBD. Adding 3 wt% H2O increases Si SEBD by a factor of about 10. In dry and wet melts,
adding 1 wt% of F or Cl increases Si SEBD by less than a factor of 2, within experimental
error of the diffusion data by Baker and coworkers. Even though the data seem to indicate a
significant effect on the activation energy of Si SEBD by F and Cl, the activation energy based
on a small temperature range and scattered data is not very reliable (e.g., see discussion in
Zhang and Ni 2010). Overall, the effect of F and Cl can be ignored unless Si diffusion data
accuracy can be significantly improved or F and Cl concentrations are much higher than 1 wt%.
Ge diffusion. Only one paper (Mungall et al. 1999) reported Ge tracer diffusion data
in silicate melts (9 points), in dry HR7 and HR7 + Na, and wet HR7 (Table 1). The data are
summarized in Figure 23. Ge diffusivity in HR7 can be expressed as:
 (37414 ± 8558) 
TD = exp ( -11.27 ± 5.16) -
dry HR7
DGe  (38)
 T 
In dry HR7Zhang
melt,
et al.Ge diffusivities
(Ch 8) Diffusion data in are similar
silicate melts to those of Hf, and they are similar to Eyring
diffusivities (Fig. 23).
figure 23.
Fig. Ge diffusion
23. Ge dataininsilicate
diffusion data silicate
melts.melts.
Data Data
source:source: Mungall
Mungall et al. (1999).et Eyring
al. (1999). Eyring
diffusivity line isdiffusivity
line is calculated from the viscosity data of Hess et al. (1995) on dry HR7 melt.
calculated from the viscosity data of Hess et al. (1995) on dry HR7 melt.
Sn diffusion. Only one paper reported good-quality Sn tracer diffusion data in silicate
melts (Behrens and Hahn 2009). Two other papers reported highly scattered SEBD of Sn
(Linnen et al. 1995, 1996) obtained from cassiterite dissolution in haplorhyolite melts, but it
was likely that the data were compromised by convection. Sn in silicate melts may be in the
form of Sn2 + or Sn4 + . Hence, Sn diffusivity may depend on oxygen fugacity. Because the data
of Linnen et al. (1995, 1996) are the only data examining the effect of fO2, they are shown in
Figure 24a. Their data indicate a decrease of Sn diffusivity with increasing fO2, meaning that
Sn2 + diffuses more rapidly than Sn4 + , as expected.
Behrens and Hahn (2009) characterized Sn tracer diffusivities in dry and wet trachyte1
and phonolite1 melts (Fig. 24b). The oxygen fugacity in these experiments was near the MnO-
Mn3O4 buffer but somewhat variable. Assuming DSn increases linearly with H2O, Sn tracer
350 Zhang, Ni, Chen
diffusivities in dry and wet trachyte1 and phonolite 1 melts at 1323-1528 K, 0.5 GPa and ≤ 1.9
wt% H2O (Behrens and Hahn 2009) can be fit as follows:
 (27290 ± 10275) - (2994 ± 740)w 
trachyte1
DSn TD = exp ( -11.08 ± 7.10) -  (39)
 T 
 (30799 ± 8555) - (1492 ± 478)w 
phonolite1
DSn TD = exp ( -6.34 ± 5.98) -  (40)
 T 
Equations (39)
Zhangand
et al.(40) reproduce
(Ch 8) experimental
Diffusion data data to within 0.88 and 0.57 lnD units,
in silicate melts
respectively.
figure 24. Sn diffusivities in silicate melts. Data in (a) are from Linnen et al. (1995, 1996) and might be incorrect
Fig. 24. Sn diffusivities in silicate melts. Data in (a) are from Linnen et al. (1995, 1996) and might be incorrect due
due to convection in the mineral dissolution experiments. Data in (b) are from Behrens and Hahn (2009).
to convection in the mineral dissolution experiments. Data in (b) are from Behrens and Hahn (2009).
Pb diffusion. Four papers reported Pb self and tracer diffusion data in silicate melts (40
points). Jambon (1982) obtained one datum of Pb tracer diffusivity in dry rhyolite15 melt
(Table 1). LaTourrette et al. (1996) reported one datum of Pb self diffusion in HB1 melt. Perez
and Dunn (1996) investigated Pb tracer diffusivities in rhyolite8 melt (containing < 0.8 wt%
H2O, but H2O is not always measured) at 1273-1723 K and 1 GPa. MacKenzie and Canil
(2008) characterized tracer diffusion of Pb on dry CMAS1 and NMAS1 at 1473-1623 K and
0.1 MPa. The diffusion data are shown in Figure 25.
For dry melts, the diffusivity increases from rhyolite to basalt and CMAS1 to NMAS1
melts. Pb tracer diffusivities in dry and wet rhyolite8 melts at 1273-1723 K, 1 GPa, and ≤ 2.6
wt% H2O (Perez and Dunn 1996) can be expressed as:
 28512 - 7900 w 
rhyolite8, 1 GPa
DPb TD = exp  -9.08 - 4.32 w -  (41)
 T 
The fits by Equation (41) to Pb diffusivities in dry and wet rhyolite are shown in Figure 25.
Except for two outliers, Equation (41) reproduces experimental data to within 0.72 lnD units.
Pb diffusion data of Perez and Dunn (1996) on rhyolite8 at 1 GPa and those of Jambon (1982)
on rhyolite5 at 0.1 MPa lie roughly on the same trend.
Comparison of diffusivities of C, Si, Ge, Sn and Pb. Diffusion data of C, Si, Ge, Sn,
and Pb are not directly comparable. C in silicate melt is tetravalent and diffuses as a linear
molecule (CO2) in most situations even when there is significant CO32− (Nowak et al. 2004;
Zhang et al. 2007), whereas Si and Ge are tetravalent and often in tetrahedral sites, but Sn
and Pb are often divalent. Hence, one may compare Si and Ge diffusivities as a group, and
Sn and Pb diffusivities as another group. For Si and Ge, the only Ge diffusion data are tracer
diffusivities in dry HR7, wet HR7, and dry HR7 + Na melts, for which Si diffusion data are
not available. Hence, no direct comparison can be made. However, because both Ge and Si
diffusivities areetsimilar
Zhang to the Eyring
al. (Ch 8) Diffusion diffusivity,
data in silicate melts Si and Ge diffusivities are similar. For Sn and
Pb diffusion, no direct comparison can be made.
figure 25.Fig.
Pb25.diffusion
Pb diffusiondata
datainin silicate melts.If If
silicate melts. notnot indicated,
indicated, the meltthe melt
is dry, the is dry, the
pressure is 0.1pressure is 0.1 MPa and
MPa and the
the diffusivities
diffusivities are tracer diffusivities. Data sources: dry and wet rhyolite8 at 1 GPa (Perez and Dunn 1996); Dunn 1996);
are tracer diffusivities. Data sources: dry and wet rhyolite8 at 1 GPa (Perez and
rhyolite15 (Jambon 1982); self diffusivity in HB1 melt (LaTourrette et al. 1996); CMAS1 and NMAS1
(MacKenzie and Canil
rhyolite15 (Jambon2008).
1982);The lines areincalculated
self diffusivity from Equation
HB1 melt (LaTourrette (41)
et al. 1996); for dry
CMAS1 and and wet(MacKenzie
NMAS1 rhyolite8.
and Canil 2008). The lines are calculated from Eq. 25 for dry and wet rhyolite8.
352 Zhang, Ni, Chen
N, P, as, Sb, Bi
N diffusion. No N diffusion data in natural silicate melts are known.
P diffusion. Four papers reported P diffusion data (SEBD) in silicate melts (35points).
Harrison and Watson (1984) investigated SEBD of P in dry and wet (up to 10 wt% H2O)
rhyolite12 melt (Table 1) during apatite dissolution at 1373-1773 K and 0.8 GPa. Rapp and
Watson (1986) examined SEBD of P in wet rhyolite12 melt (containing 1-6 wt% H2O) during
monazite dissolution at 1273-1673 K and 0.8 GPa. Pichavant et al. (1992) obtained two SEBD
values of P in wet synthetic haplorhyolite melt during apatite dissolution at 1173 K and 0.2
GPa; the data are likely compromised by convection. Lundstrom (2003) reported two SEBD
values in basalt7-basanite diffusion couple at 0.9 GPa. The data are summarized in Figure 26.
Interestingly, as noted by Rapp and Watson (1986), P diffusivity during apatite dissolution
in rhyolite melt is significantly faster than that during monazite dissolution in the same melt
(comparing solid and open symbols in Figure 26; e.g., at 1273 K, 0.8 GPa, and 6 wt% H2O,
P diffusivity in rhyolite melt is ~4.3×10−14 m2/s during apatite dissolution, but only 1.5×10−15
m2/s during monazite dissolution, with a difference of a factor of 28), even though one might
expect that they would be similar. The data are from the same laboratory, and are hence free of
inter-laboratory inconsistencies. One possibility is that the difference in the diffusing species
of P during monazite versus apatite dissolution may lead to the difference in the diffusivity.
The dependence of P diffusivities on H2O content is shown in Figure 27a. It is not clear
whether the relation between lnDP and H2O is linear or curved; some of the scatter is likely due
to uncertainty in determining H2O content in the early years.
Because P is a network former and diffuses slowly, one might expect that DP is the same
as DEyring calculated from viscosity η. Diffusivity of P is indeed among the lowest of all
figureFig.26.
26. P diffusivity
P diffusivity in silicate
in silicate melts. melts. Thepoints
The various various points
at 1000/T at are
= 0.978 1000/T = 0.978
for hydrous are
rhyolite for(Wolf
melts hydrous rhyolite
melts and
(Wolf and London 1994). Open symbols represent data during apatite dissolution;
London 1994). Open symbols represent data during apatite dissolution; solid symbols represent data during
solid symbols
represent data during monazite dissolution. Data sources: “rhy12(Ap)”: Apatite dissolution in rhyolite 12
by Harrison
monazite and WatsonData
dissolution. (1984);
sources:“rhy12(Mon)”: Monazite
"rhy12(Ap)": Apatite dissolutiondissolution inHarrison
in rhyolite 12 by rhyolite12 by Rapp and Watson
and Watson
(1986); b-b couple: basalt7-basanite couple by Lundstrom (2003); “Haplorhyolite(Ap)”:
(1984); "rhy12(Mon)": Monazite dissolution in rhyolite12 by Rapp and Watson (1986); b-b couple: basalt7-basanite
apatite dissolution
in haplorhyolites by Pichavant et al. (1992).
couple by Lundstrom (2003); "Haplorhyolite(Ap)": apatite dissolution in haplorhyolites by Pichavant et al. (1992).
elements, but is different from the Eyring diffusivity. For example, Figure 27b compares lnDP
versus lnDEyring, showing that there is no unique correlation between the two: DP is equal to or
greater than DEyring during apatite dissolution, but is often smaller than DEyring during monazite
dissolution. The difference between DP and DEyring can be almost two orders of magnitude.
Hence, it is impossible to use viscosity or Eyring diffusivity to predict P diffusivity.
Although DP ≠ DEyring, it is possible to relate P diffusivity to viscosity or Eyring diffusivity
under specific
Zhangconditions.
et al. (Ch 8) For example,
Diffusion data infor P diffusivity
silicate melts in wet rhyolite containing 6 wt%
dry rhyolite12
H2O during monazite dissolution, ln DP SEBD; apatite diss = (20.64±2.26) + (1.509±0.064)lnDEyring
figure 27. P diffusivities in rhyolite12 melts. In the legend of (b), A means apatite and M means monazite;
Fig. 27. P diffusivities in rhyolite12 melts. In the legend of (b), A means apatite and M means monazite; solid
solid symbols represent data during apatite dissolution; open symbols represent data during monazite
dissolution. DEyring is calculated using the viscosity model of Zhang et al. (2003) and a jump distance of
symbols represent data during apatite dissolution; open symbols represent data during monazite dissolution.
2.8×10−10 m.
DEyring is calculated using the viscosity model of Zhang et al. (2003) and a jump distance of 2.8x10-10 m.
354 Zhang, Ni, Chen
where viscosity needed to estimate DEyring is from Zhang et al. (2003). However, no universal
relation exists between P diffusivity and Eyring diffusivity. Hence, such relations would have
to be established at each specific H2O contents, meaning that there is not much advantage for
doing so.
As diffusion. No As diffusion data in silicate melts are known.
Sb diffusion. Two papers reported Sb tracer diffusion data in silicate melts (18 points).
Koepke and Behrens (2001) investigated Sb diffusion in HA1 (Table 1) with one datum in dry
HA1 and 4 points in wet HA1 (4.5-5.2 wt% H2O). MacKenzie and Canil (2008) studied Sb dif-
fusion in dry CMAS1 and NMAS1 melts. The data are summarized in Figure 28. Sb diffusivity
in wet HA1 melt at 1373-1673 K, 0.5 GPa, and 4.5-5.2 wt% H2O can be described by:
 (16423 ± 1934) 
wet HA1
DSb TD = exp ( -12.56 ± 1.28) -  (42)
 T 
Sb diffusivity in andesite melt is slightly greater than (about 1.5 to 2 times) Ba diffusivity in
the same melt.
figure
Fig. 28. Sb tracer diffusivities 28. Sbmelts.
in silicate tracer diffusivities in silicate melts.
Bi diffusion. No Bi diffusion data in silicate melts are known.

o, S, Se, te, Po
O diffusion. Oxygen diffusion in silicate melts is reviewed in another chapter (Zhang and
Ni 2010).
S diffusion. Sulfur in silicate melts can be present in various species controlled mainly by
oxygen fugacity (the valence of sulfur can be −2, + 4, and + 6, with corresponding species of
S2−, SO2 and SO42 + ). Sulfur diffusivity is expected to depend on the sulfur species and hence
on oxygen fugacity, which is a main complication. Three papers reported S diffusion data in
silicate melts (41 points). In a review paper, Watson (1994) previewed sulfur diffusivity in a
variety of melts, including rhyolite, dacite, haplodacite, haploandesite, and a lunar ultramafic
melt using (i) the devolatization technique (obtaining tracer diffusivity or FEBD), (ii) the
diffusion couple technique (obtaining tracer diffusivity or FEBD), and (iii) the thin source
technique (obtaining tracer diffusivity). Increasing oxygen fugacity leads to decreasing sulfur
SEBD. Winther et al. (1998) characterized S tracer diffusion in dry albite melt at 1573-1773 K,
1 GPa, andZhang
oxidized conditions. Freda et al. (2005) investigated S tracer diffusion in Etna and
et al. (Ch 8) Diffusion data in silicate melts
Stromboli basalt melts. Some diffusion data are summarized in Figure 29.
figure 29. Fig.

S tracer diffusion
29. S tracer ininsilicate
diffusion melts.Data
silicate melts. Data sources
and conditions: dry
dry and wet andmelts
basalt wet(from
basalt
Etnamelts
and (from
Etna and Stromboli) at reduced conditions of QFM-3 (Freda et al. 2005); dry albite under “oxidized”
Stromboli) at reduced conditions of QFM-3 (Freda et al. 2005); dry albite under “oxidized” condition (Winther et al.
condition (Winther et al. 1998).
1998).
Due to the complexity of S speciation as well as the limited number of studies, sulfur
diffusion in silicate melts is not well understood. Furthermore, errors for S diffusion data
are larger than those of other elements, likely also due to the various sulfur species. Zhang
et al. (2007) obtained the following equation to describe sulfur tracer diffusivity in Etna and
Stromboli basalt melts (ignoring the small compositional difference between the two) at 1498-
1723 K, 0.5-1 GPa (with negligible pressure effect in this range), oxygen fugacity of 3 log
units below QFM (hence sulfur is in the form of S2−), and 0-4 wt% H2O:
 27692 - 651.6 w 
TD = exp  -8.21 -
DSbasalts  (43)
 T 
The maximum error of Equation (43) in reproducing the experimental diffusivities is 0.77 in
lnD. The activation energy is about 230 kJ/mol in dry basalt melt, decreasing to 209 kJ/mol in
wet basalt melt containing 4 wt% H2O.
29
Se diffusion. No Se diffusion data in silicate melts are known.
Te diffusion. Only one experimental study reported Te diffusion data (SEBD). Te can
exist in different valences, similar to S. Hence, its diffusivity may depend on the oxygen
fugacity. MacKenzie and Canil (2008) carried out devolatization (desorption) experiments of
CMAS1 melt (Table 1) at 1523-1623 K and in air, and extracted Te diffusivities. The data are
shown in Figure 30. As can be seen, there is considerable scatter in the data (about a factor
of 5 difference in diffusivity values at 1573 K), similar to the large scatter in S diffusivity.
Because CMAS1 composition is close to a haplobasalt, the Te diffusion data in CMAS1 melt
356 Zhang, Ni, Chen
figure 30.Fig.
Te30.
tracer diffusivities
Te tracer inCMAS1
diffusivities in CMAS1 melt melt
(Table (Table 1) under
1) under oxidized oxidized
conditions conditions
(MacKenzie (MacKenzie
and Canil 2008) and
Canil 2008) compared with S tracer diffusivities in dry Etna and Stromboli basalt melts under reduced
compared with S tracer diffusivities in dry Etna and Stromboli basalt melts under reduced conditions (Freda et al.
conditions (Freda et al. 2005).
2005).
are compared in Figure 30 with S diffusion data in basalt melt, though the condition was
oxidized for Te diffusion and reduced for S diffusion. The two sets of data are comparable.
Po diffusion. No Po diffusion data in silicate melts are known.
Summary of O, S, Se and Te diffusivities. The diffusivities of O, S, Se and Te are not
readily comparable because of limited and scattered data on S and Te and no data on Se, as
well as different speciation and oxidation states of O, S and Te.
f, Cl, Br, I, at
F diffusion. Five papers reported F diffusion data in silicate melts (84 points). Dingwell
and Scarfe (1984, 1985) explored FEBD of F in jadeite and albite melts (containing up to
6.3 wt% F) using the diffusion couple method at 1473-1673 K and 1.0-1.5 GPa by packing
F-bearing and F-free jadeite powders into a Pt capsule. Dingwell and Scarfe (1985) reported
more FEBD values of F in jadeite, albite and a peraluminous silicate melt (45.5 wt% O; 5.4
wt% F; 31.58 wt% Si; 10.87 wt% Al; and 6.14 30 wt% Na) using devolatization experiments
at 1473-1673 K and 0.1 MPa of pure oxygen gas. Gabitov et al. (2005) studied SEBD of F
during fluorite dissolution into a haplorhyolite (similar to HR7 in Table 1) melt at 1173-1273
K, 0.1 GPa, and 1.2 to 4.8 wt% H2O, and showed that dissolved H2O enhances F diffusivity
significantly. Alletti et al. (2007) characterized F tracer diffusion in dry and wet basalt4 melts at
1523-1723 K and 0.5-1.0 GPa. Balcone-Boissard et al. (2009) investigated F tracer diffusion in
Na-phonolite and K-phonolite melts at 1473-1723 K and 0.5-1 GPa and examined the effect of
Na/K ratio and H2O content. In addition, Baker and Balcone-Boissard (2009) reviewed halogen
diffusion in silicate melts (also Baker et al. 2005). Figure 31 shows F diffusion data in dry
jadeite, albite and basalt melts.
F diffusivities in jadeite melt at 0.1 MPa are greater than those in albite melt by 3.4 lnD units.
In the pressure range of 1.0 to 1.5 GPa, F diffusivities in jadeite melt are almost independent of
pressure within 0.27 lnD units, but are significantly (1.8 lnD units) higher than those at 0.1 MPa.
The independence of F diffusivity on pressure at high pressure but the dependence on pressure
from 0.1 MPa to 1 GPa may imply either a structural change in jadeite melt with pressure, or a
figure 31.Fig.F 31.

diffusivities
F diffusivities in dryalbite,
in dry albite, jadeite,
jadeite, andmelts.
and basalt4 basalt4
Data melts. Data 1.0-1.5
sources: jadeite sources: jadeite 1.0-1.5
GPa (Dingwell and GPa
(Dingwell Scarfe
and Scarfe 1984); jadeite and albite at 0.1 MPa (Dingwell and Scarfe 1985); basalt4
1984); jadeite and albite at 0.1 MPa (Dingwell and Scarfe 1985); basalt4 at 1 GPa (Alletti et al. 2007).
at 1 GPa
(Alletti et al. 2007).
change in the diffusion species from the devolatization method in 0.1 MPa pure O2 gas to the
diffusion couple method in Pt capsule (Baker and Balcone-Boissard 2009), or experimental
error. Another surprise is that F diffusivities in polymerized jadeite melt are similar to those
in less polymerized basalt melt (for comparison, the O self diffusivity in dry basalt melt is
greater than that in jadeite melt by about 2 orders of magnitude, Shimizu and Kushiro 1984;
Lesher et al. 1996). Taking together, it is possible that F diffusivities in jadeite melt at 1.0-1.5
GPa (Dingwell and Scarfe 1984) are compromised somehow, e.g., because powders (rather
than glass cylinders) were packed to make the diffusion couples or because of the presence of
unknown amounts of H2O in high-pressure melts (Baker and Balcone-Boissard 2009).
F diffusivities in dry basalt4 and jadeite melts can be expressed as (Alletti et al. 2007):
 (22967 ± 5384) 
DFdryTDbasalt4; 1 GPa = exp ( -9.46 ± 3.32) -  (44)
 T 
31
Equation (44) can reproduce the experimental data to within 0.5 lnD units. The activation
energy is 191±45 kJ/mol. A single datum at 0.5 GPa shows that the pressure effect within this
range is negligible. Adding 3 wt% H2O in basalt melt increases lnD by 0.54 units.
Figure 32 summarizes F diffusion data in natural silicate melts. Experimental F diffusion
data at 0.5 and 1 GPa (Balcone-Boissard et al. 2009) show that the pressure effect is negligible.
Diffusion data in dry and wet melts suggest that lnD is linear to H2O. The diffusion data in
dry and wet Na-phonolite and K-phonolite melts at 1473-1723 K, 1 GPa (likely applicable to
0.5-1.5 GPa), and 0-5 wt% H2O can be fit as follows:
 (17624 ± 2347) - (312 ± 68)w 
DFNa-phonolite;
SEBD
1 GPa
= exp ( -13.04 ± 1.47) -  (45)
 T 
 (21110 ± 4350) - (481 ± 123)w 
DFK-phonolite;
SEBD
1 GPa
= exp ( -11.43 ± 2.71) -  (46)
 T 
358 Zhang, Ni, Chen
figure 32. Fig.

F effective binary
32. F effective diffusivities
binary diffusivities inin basalt
basalt and phonolite
and phonolite melts.
melts. Data Datadrysources:
sources: dry and
and wet basalt4 wet
at 0.5 to 1basalt4
at 0.5 to 1 GPa: Alletti et al. (2007); dry and wet Na and K phonolites at 0.5 to 1 GPa: Balcone-Boissard et
GPa: Alletti et al. (2007); dry and wet Na and K phonolites at 0.5 to 1 GPa: Balcone-Boissard et al. (2009); and wet
al. (2009); and wet HR7 at 0.1 GPa (Gabitov et al. 2005).
HR7 at 0.1 GPa (Gabitov et al. 2005).
Equations (45) and (46) can reproduce the experimental data to within 0.34 and 0.47 lnD units,
respectively.
Cl diffusion. Four papers reported Cl diffusion data in silicate melts (98 points). In
addition, there was an early abstract by Watson and Bender (1980) reporting some results,
which are not included in this review because no details are available. Bai and Koster van Groos
(1994) investigated SEBD (close to tracer diffusion or FEBD) of Cl in rhyolite17 and HR8
melts (Table 1) in contact with NaCl melt or NaCl solution at 0.0001 to 0.46 GPa. Lundstrom
(2003) reported a SEBD datum of Cl from basalt7-basanite diffusion couple experiment at
1723 K and 0.9 GPa. Alletti et al. (2007) examined Cl tracer diffusion in dry and wet basalt4
melts at 1523-1723 K and 0.5-1.0 GPa. Balcone-Boissard et al. (2009) determined Cl tracer
diffusivities in dry and wet Na-phonolite and K-phonolite melts at 1473-1723 K and 0.5-1 GPa.
Furthermore, Baker and Balcone-Boissard (2009) reviewed halogen diffusion in silicate melts
(also Baker et al. 2005).
32
Bai and Koster van Groos (1994) inferred that lnDCl in rhyolite17 increases rapidly with
H2O from 0 to 2 wt%, and then does not increase much as H2O increases from 2 to 6 wt%
(Fig. 33a). However, Balcone-Boissard et al. (2009) deduced that lnDCl in Na- and K-phonolite
melts is almost linear to H2O from 0 to 5 wt% H2O (Fig. 33a). Both studies used the method
of difference-from-100% in electron microprobe analyses to estimate H2O. Hence, there is
some uncertainty in H2O contents in both studies. On the other hand, in some experiments
by Bai and Koster van Groos (1994), the pressure is too low to keep H2O content in the
rhyolite17 melt (e.g., 6.9 wt% H2O at a pressure of 0.1 GPa), indicating possible problems in
the experiments. Tentatively, we suggest that DCl depends linearly on H2O based on Balcone-
Boissard et al. (2009) and that the inference by Bai and Koster van Groos (1994) is incorrect.
For dry and wet Na-phonolite and K-phonolite melts at 1523-1723 K, 1 GPa and 0-5 wt%
H2O, Cl diffusivity can be expressed as follows:
 (17963 ± 5144) - (605 ± 155)w 
DClNa-phonolite;
TD
1 GPa
= exp ( -14.57 ± 3.22) -  (47)
 T 
figure 33. Cl diffusivities in various melts as a function of H2O and temperature. Data sources: dry and wet
Fig. 33. melt:
rhy17 (rhyolite17) Cl diffusivities in various
Bai and Koster melts (1994);
van Groos as a function
Na andof K
H2pho
O and temperature.
(phonolites): Data sources: dry and wet rhy17
Balcone-Boissard
et al. (2009); dry basalt4: Alletti et al. (2007).
(rhyolite17) melt: Bai and Koster Van Groos (1994); Na and K pho (phonolites): Balcone-Boissard et al. (2009); dr
basalt4: Alletti et al. (2007)

 (15429 ± 5319) - (547 ± 168)w 
DClK-phonolite;
TD
1 GPa
= exp (-15.88 ± 3.32) -  (48)
 T 
Equations (47) and (48) can reproduce the experimental data to within 0.68 and 0.62 lnD units,
respectively. When the data on K-phonolite were fit, three outlier points (one is a factor of 10
off, see Fig. 33b, likely due to a typographical error) were excluded. The D values at 0.5 GPa
can also be described well by Equations (47) and (48) (for 133 GPa).
When different melts are compared, Cl diffusivities in dry basalt4 melt are higher than
those in phonolite and rhyolite17 melts by a factor of 2 to 10 at temperature ≥ 1500 K. Those
360 Zhang, Ni, Chen
in dry K-phonolite melt are higher than those in dry rhyolite17 and dry Na-phonolite melts by
0.4 lnD units. Cl diffusivities in dry rhyolite17 melt are similar to those in dry Na-phonolite
melt. The similarity in Cl diffusivities between dry Na-phonolite and dry rhyolite melts is
unexpected considering (i) the large compositional difference between Na-phonolite and
rhyolite17 and (ii) the significant difference in DCl between compositionally similar Na-
phonolite and K-phonolite (similar composition).
Br diffusion. Only one paper (Alletti et al. 2007) reported Br tracer diffusion data (10
points) in basalt4 melt (Table 1) at 1523-1723 K and 1.0 GPa. Figure 34 shows the data, and
considerable scatter can be seen. Linearity between lnD and 1/T is not perfect.
I and At diffusion. No iodine and astatine diffusion data in silicate melts are known.
Comparison of F, Cl, and Br diffusivities. Figure 35 compares diffusivities of F, Cl and
Br. The diffusivity increases as the size of halogen ions decreases, similar to the univalent
cations. The total increase from Br to F diffusivity is a factor of about 2 (with much scatter) in
basalt. In dry Na-phonolite, the increase from Cl to F diffusivity is a factor of about 6. In dry
K-phonolite and in wet Na-phonolite, the increase is smaller. The bottom line is that F, Cl and
Br diffusivities vary with anion size, but the difference is not large, at least at the temperatures
investigated.
He, Ne, ar, Kr, Xe, rn
Diffusion of noble gas elements in silicate melts is reviewed by another chapter in this
volume (Behrens 2010).
Sc, Y, ree
Sc diffusion. Only one paper reported Sc diffusion data in silicate melts (13 points). Lowry
et al. (1982) investigated Sc tracer diffusion in basalt1 and andesite2 melts at 1570-1676 K
and 0.1 MPa. The data are summarized in Figure 36. When compared to REE diffusivities, Sc
tracer diffusivity in basalt1 melt is identical to Yb self diffusivity in HB1 melt; and Sc tracer
diffusivity in andesite2 melt is slightly higher than La, Nd, Y, Er and Gd diffusivities in a
haploandesite (HA1) melt (Fig. 36). It appears that Sc diffusivity is similar to Yb diffusivity.
figure
Fig. 34. Br 34.
tracerBr tracer diffusivities
diffusivities in basalt melt. inData
basalt melt.
are from Data
Alletti are(2007).
et al. from Alletti et al. (2007).
figure
Fig.35.
35.Comparison
ComparisonofofF,F,ClCland
andBrBrdiffusivities
diffusivitiesininbasalt
basaltand
andNaNaand KK
and phonolite (pho).
phonolite (pho). Data sources: basalt4
Data sources: basalt4 (Alletti et al. 2007); phonolites (Balcone-Boissard et al. 2009).
(Alletti et al., 2007); phonolites (Balcone-Boissard et al., 2009)
Y diffusion. Five papers reported Y diffusion data in silicate melts (43 points). Baker and
Watson (1988) investigated SEBD of Y (close to FEBD or tracer diffusion) in dry rhyolite1-
rhyolite8 and HD1-rhyolite8 diffusion couples (Table 1) at 1171-1673 K and 0.01-1 GPa.
Nakamura and Kushiro (1998) obtained two tracer diffusivity values of Y in jadeite melt at
1673 K and 1.25 and 1.5 GPa (the two values differ by two orders of magnitude and hence
cannot both be correct), as well as two values in diopside melt at 1863 K and 1 and 1.25
GPa. Mungall et al. (1999) studied Y tracer diffusion in dry HR7 and HR7 + Na and wet HR7
containing 3.6 wt% H2O, with D values in wet HR7 (3.6 wt% H2O) and HR7 + Na (20% Na2O)
roughly on the same trend. Koepke and Behrens (2001) reported35 Y tracer diffusivities in wet
HA1 melt (4.5-5.2 wt% H2O) and one datum in dry HA1 melt. Behrens and Hahn (2009)
362 Zhang, Ni, Chen
figure 36. Sc tracer dif-

fusivities in dry basalt1 and
andesite2 melts (Lowry et al.
1982) compared with REE
diffusivities in dry HB1 (La-
Tourrette et al. 1996) and dry
HA1 (Koepke and Behrens
2001).
Fig. 36. Sc tracer diffusivities in dry basalt1 and andesite2 melts (Lowry et al. 1982) compared with REE
characterized
diffusivities Y traceret al.
in dry HB1 (LaTourrette diffusion inHA1
1996) and dry dry(Koepke
and wet trachyte1
and Behrens 2001). and phonolite1 melts. The diffusion
data are summarized in Figure 37.
SEBD data are limited and not enough to model how they vary with temperature, melt
composition and concentration gradients. H2O has a large effect on Y diffusivity. For example,
for tracer diffusion in trachyte1, adding 1.7 wt% H2O increases DY by a factor of 28. Assuming
that lnD increases linearly with the square root of H2O for each type of melt, Y tracer diffusivities
in dry and wet trachyte1 and phonolite1 melts at 1323-1528 K, 0.5 GPa, and ≤ 1.9 wt% H2O
can be fit as follows:
 (23941 ± 5364) - (4717 ± 568) w 
DYtrachyte1
TD = exp ( -15.22 ± 3.68) -  (49)
 T 
 (31693 ± 5159) - (2132 ± 275) w 

DYphonolite1
TD = exp ( -7.58 ± 3.51) -  (50)
 T 
Equations (49) and (50) can reproduce experimental data to within 0.43 and 0.36 lnD units,
respectively. 36
La diffusion. Five papers reported La diffusion data in silicate melts (50 points). Rapp
and Watson (1986) examined SEBD of LREE (La, Ce, Nd, and Sm, assumed to have the
same diffusivity) during monazite dissolution in wet rhyolite12 melts (containing 1-6 wt%
H2O) (Table 1) at 1273-1673 K and 0.8 GPa. Nakamura and Kushiro (1998) studied La tracer
diffusion in jadeite and diopside melts at 1673-1863 K and 0.75-2.0 GPa. Koepke and Behrens
(2001) investigated La tracer diffusion in a wet haploandesite melt (4.5-5.2 wt% H2O) and
reported one datum in dry haploandesite melt at 0.5 GPa. Lundstrom (2003) obtained a SEBD
value for La between a basalt7-basanite diffusion couple at 1723 K and 0.9 GPa. Behrens and
Hahn (2009) characterized La tracer diffusion in dry and wet trachyte and phonolite melts at
0.5 GPa. The diffusion data are summarized in Figure 38.
For SEBD in rhyolite12, from 1 to 4 wt% H2O, lnDLa increases significantly, but from 4
to 6 wt% H2O, there is almost no increase (Rapp and Watson 1986). The data do not conform
to a linear relation between lnD and H2O or linear relation between lnD and the square root of
363
figure 37.Fig.
Y diffusivities in silicate
38. La diffusivities in silicatemelts. Except for
melts; percentage dry
in the b-b couple
legends forofwhich
means wt% the sources:
H2O. Data data arejadeite
SEBD,and other
data are tracer diffusivities. Data sources: jadeite and diopside melts (Nakamura and Kushiro 1998); dry
HR7, wet diopside
HR7 with(Nakamura and Kushiro 1998); SEBD in rhy12 (rhyolite12) during monazite dissolution (1, 2, 4, and 6
3.6 wt% H2O, and HR7 + Na (Mungall et al. 1999); b-b (basalt7-basanite) couple
(Lundstrom 2003);
wt% dry and
H2O) (Rapp and wet
Watson(4.5-5.2 wt%
1986); dry andH 2O)
wet HA1
HA1 (Koepke
(4.5-5.2 wt% H2and Behrens
O) (Koepke and2001);
Behrens dry and
2001); drywet
and trachyte1
wet
and phonolite1 (Behrens and Hahn 2009).
trachyte1 (1.1-1.7 wt% H2O) (Behrens and Hahn 2009); dry and wet phonolite1 (1.8 wt% H2O) (Behrens and Hahn
2009); SEBD in b-b (basalt7-basanite) couple (Lundstrom 2003).
38
figure 38.Fig.
La39.diffusivities ininsilicate
Ce diffusivities melts;
silicate melts, percentage
including FEBD inin
drythe legends
jadeite means
and diopside wt%
melts of H2O.
(Nakamura and Data sources:
Kushiro
jadeite and1998),
diopside (Nakamura and Kushiro 1998); SEBD in rhy12 (rhyolite12) during
TD in dry rhy15 (rhyolite15) melt (Jambon 1982), and SEBD in wet rhy12 (rhyolite12) melts with 1, 2, 4,
monazite
dissolution (1, 2, 4, and 6 wt% H2O) (Rapp and Watson 1986); dry and wet HA1 (4.5-5.2 wt% H2O)
(Koepke andand 6Behrens
wt% H2O 2001);
(Rapp anddry and1986).
Watson wet trachyte1 (1.1-1.7 wt% H2O) (Behrens and Hahn 2009); dry
and wet phonolite1 (1.8 wt% H2O) (Behrens and Hahn 2009); SEBD in b-b (basalt7-basanite) couple
(Lundstrom 2003).
364 Zhang, Ni, Chen
H2O content. Trying to relate D to viscosity does not lead to a simple relation either. Although
the observed lnD versus H2O trends could be real, it is also possible (even likely) that the H2O
contents were not known accurately in these early experimental studies, complicating efforts
to quantify how diffusivity depends on H2O. The same comment applies to Ce, Nd, Sm and P
diffusion in rhyolite12 investigated by Rapp and Watson (1986).
Adding H2O increases La tracer diffusivity significantly. For example, for tracer diffusion
in phonolite1, adding 1.7 wt% H2O increases DLa by 2.4 lnD units (1.06 logD units). Based
on diffusion data in dry and wet trachyte1 melt, it appears that lnD increases linearly with the
square root of H2O content. Using such a relation, tracer diffusivities in dry and wet trachyte1
and phonolite1 melts with H2O ≤ 1.9 wt% at 1323-1528 K and 0.5 GPa can be fit as:
 (25128 ± 4896) - (4810 ± 519) w 
trachyte1
DLa TD = exp ( -14.13 ± 3.36) -  (51)
 T 
 (33135 ± 4870) - (2941 ± 412) w 

phonolite1
DLa TD = exp ( -6.44 ± 3.34) -  (52)
 T 
Equations (51) and (52) can reproduce experimental data to within 0.38 and 0.40 lnD units,
respectively.
Ce diffusion. Three papers reported Ce diffusion data in silicate melts (30 points). Jambon
(1982) investigated Ce tracer diffusion in dry rhyolite15 (Table 1) at 1148-1573 K and 0.1
MPa. Rapp and Watson (1986) obtained SEBD of Ce in wet rhyolite12 melts during monazite
dissolution (assumed to be exactly the same as La diffusivities). Nakamura and Kushiro (1998)
studied Ce tracer diffusion in jadeite and diopside melts at 1673-1863 K and 0.75-2.0 GPa.
Ce in silicate melts can be present as either trivalent or tetravalent cations, but in all of these
studies Ce is likely mainly trivalent, and its diffusivities are similar to La diffusivities. The data
are shown in Figure 39.
A single Ce tracer diffusivity in dry rhyolite15 at 1573 K does not lie in the same trend as
data at 1148-1373 K (red filled circles and solid line in Fig. 39). On the other hand, Ce tracer
diffusivities in dry rhyolite15 at 1148-1373 K and 0.1 MPa are coincidentally similar to SEBD
of Ce during monazite dissolution in rhyolite12 melt containing 1 wt% H2O at 1273-1673 K
and 0.8 GPa; both data sets can be fit by a single equation:
 (58413 ± 2536) 
dry rhyolite15
DCe TD ≈ DCe SEBD = exp (5.78 ± 1.87) -
wet rhyolite12 (1 wt% H 2 O)
 (53)
 T 
The maximum error in reproducing the experimental data is 0.55 in terms of lnD. The Ce
tracer diffusivity in dry rhyolite15 at 1573 K by Jambon (1982) is a clear outlier, off the trend
by 5.9 lnD units (2.6 logD units).
Pr diffusion. Only one paper reported Pr diffusion data in silicate melts (8 points).
Nakamura and Kushiro (1998) investigated Pr tracer diffusion in jadeite and diopside melts.
The data are shown in Figure 40. Pr tracer diffusivities in jadeite melt increases with increasing
pressure. At 1573-1723 K and 0.75-2 GPa, the diffusion data can be fit as:
 (40807 ± 9501) - (1697 ± 704)P 
DPrjadeite
TD
melt
= exp ( -4.57 ± 5.65) -  (54)
 T 
Equation (54) reproduces experimental data to ≤ 0.20 lnD units. The activation energy at 1.5
GPa is 318±71 kJ/mol and the activation volume at 1673 K is −(14±6)×10−6 m3/mol.
figure 39. Fig.

Ce 39.
diffusivities ininsilicate
Ce diffusivities melts,
silicate melts, including
including FEBDFEBD in dry
in dry jadeite andjadeite
diopside and
meltsdiopside
(Nakamura melts (Nakamura
and Kushiro
and Kushiro 1998), TD in dry rhy15 (rhyolite15) melt (Jambon 1982), and SEBD in wet rhy12 (rhyolite12)
1998), TD in dry rhy15 (rhyolite15) melt (Jambon 1982), and SEBD in wet rhy12 (rhyolite12) melts with 1, 2, 4,
melts with 1, 2, 4, and 6 wt% H2O (Rapp and Watson 1986).
and 6 wt% H2O (Rapp and Watson 1986).
39
figure
Fig. 40. Prtracer
40. Pr tracer diffusivities
diffusivities inand
in jadeite jadeite and
diopside diopside
melts melts
(Nakamura (Nakamura
and Kushiro 1998) and Kushiro 1998).
Nd diffusion. Nine papers reported Nd diffusion data in silicate melts (90 points). Rapp
and Watson (1986) examined SEBD of Nd during monazite dissolution in wet rhyolite12
melts (treated as the same as La diffusivities). Lesher (1994) obtained Nd self diffusivities
in rhyolite2 and basalt3 melts at 1528-1738 K and 1 GPa. LaTourrette et al. (1996) and
LaTourrette and Wasserburg (1997) determined Nd self diffusivities in HB1 melt at 1623-
1773 K and 0.1 MPa (in air and in oxygen fugacity corresponding to the Fe-FeO buffer). Perez
and Dunn (1996) measured Nd tracer diffusivities in rhyolite8 melt at 1448-1673 K, 1 GPa
and ≤0.8 wt% H2O (but H2O is not always measured). Nakamura and Kushiro (1998) studied
366 Zhang, Ni, Chen
Nd tracer diffusion in jadeite and diopside melts at 1673-1863 K and 0.75-2.0 GPa. Mungall
et al. (1999) investigated Nd tracer diffusion in dry HR7 and HR7 + Na melts at 0.1 MPa,
and a wet HR7 melt with 3.6 wt% H2O at 1 GPa. Koepke and Behrens (2001) reported Nd
tracer diffusivities in wet HA1 melt (4.5-5.2 wt% H2O) and one datum in dry HA1 melt at 0.5
GPa. Behrens and Hahn (2009) characterized Nd tracer diffusion in dry and wet trachyte1 and
phonolite1 melts at 0.5 GPa. Melt compositions can be found in Table 1 except for jadeite and
diopside. Zhang
Figure 41 shows some diffusion data.
et al. (Ch 8) Diffusion data in silicate melts
figure 41.Fig.Nd
41. diffusivities
Nd diffusivities in some
in some melts.
melts. In legend,
In the the legend,
"HB1-r"“HB1-r”
means HB1means HB1 under
under reducing reducing
conditions; "rhy" conditions;
“rhy” means rhyolite; “pho” means phonolite; “tra” means trachyte. Melt compositions are listed in Table
means rhyolite; "pho" means phonolite; "tra" means trachyte. Melt compositions are listed in Table 1. Data
1. Data sources: SD in HB1 and HB1-r (LaTourrette et al. 1996; LaTourrette and Wasserburg 1997); SD
in basalt3sources:
and rhy2
SD in(Lesher 1994);
HB1 and HB1-r TD in HA1
(LaTourrette et al. (Koepke and Behrens
1996; LaTourrette 2001);
and Wasserburg TDSD
1997); in inrhy8 (Perez
basalt3 and Dunn
and rhy2
1996); TD in HR7 (Mungall et al. 1999); TD in dry and wet pho1 and tra1 (Behrens and Hahn 2009); TD
in jadeite(Lesher 1994); TD in HA1 (Koepke and Behrens 2001); TD in rhy8 (Perez and Dunn 1996); TD in HR7 (Mungall et
(Nakamura and Kushiro 1998).
al. 1999); TD in dry and wet pho1 and tra1 (Behrens and Hahn 2009); TD in jadeite (Nakamura and Kushiro 1998).
Nd self diffusivities do not depend on oxygen fugacity from air to Fe-FeO buffer
(LaTourrette and Wasserburg 1997), which is expected because the valence of Nd is not
expected to change. In jadeite melt, Nd tracer diffusivities depend on pressure according to
Nakamura and Kushiro (1998). However, Nd self diffusivities in HB1 melt with no Fe, Ti and
Na (Table 1) at 0.1 MPa (LaTourrette and Wasserburg 1997) and a single Nd self diffusivity
in basalt3 with significant Fe, Ti and Na at 1 GPa (Lesher 1994) are similar, either suggesting
negligible pressure dependence and good choice of HB1 melt composition by LaTourrette and
Wasserburg (1997), or due to coincidence. Nd self diffusivities in these two melts at 1528-
1773 K and 0.001-1 GPa can be expressed as (Fig. 41):
 (24001 ± 2803) 
dry basalt
DNd SD = exp ( -10.11 ± 1.66) -  (55)
 41 T 
Equation (55) reproduces all experimental data to within 0.3 lnD units.
In the highly silicic melt HR7 at 1410-1873 K and 0.1 MPa, Nd tracer diffusivities
are orders of magnitude lower than self diffusivities in basalt melt, and can be described as
(Mungall et al. 1999):
 (35847 ± 1488) 
TD = exp ( -9.90 ± 0.92) -
dry HR7
DNd  (56)
 T 
Equation (56) reproduces three experimental data to within 0.08 lnD units.
In wet haploandesite melt containing 4.5-5.2 wt% H2O, Nd tracer diffusivities at 1373-
1673 K and 0.5 GPa can be expressed as (Koepke and Behrens 2001):
 (20280 ± 220) 
TD = exp ( -11.44 ± 0.15) -
wet HA1
DNd  (57)
 T 
Equation (57) reproduces four experimental data to within 0.01 lnD units.
In dry and wet trachyte and phonolite melts with ≤1.9 wt% H2O at 1323-1528 K and 0.5
GPa (Behrens and Hahn 2009), the tracer diffusion data can be fit as follows:
 (24381 ± 5125) - (4615 ± 543) w 
TD = exp ( -14.57 ± 3.52) -
trachyte1
DNd  (58)
 T 
 (31697 ± 3401) - (2937 ± 288) w 

FEBD = exp ( -7.45 ± 2.33) -
phonolite1
DNd  (59)
 T 
Equations (58) and (59) reproduce experimental data to within 0.36 and 0.23 lnD units,
respectively.
Pm diffusion. No Pm diffusion data in silicate melts are known.
Sm diffusion. Four papers reported Sm diffusion data in silicate melts (49 points).
Rapp and Watson (1986) studied SEBD of Sm in wet rhyolite12 (treated as the same as La
diffusivities). Nakamura and Kushiro (1998) investigated Sm tracer diffusion in jadeite melt
at 1573-1723 K and 0.75-2.0 GPa and in diopside melt at 1863 K and 1.0-1.25 GPa. Koepke
and Behrens (2001) reported Sm tracer diffusivities in a wet haploandesite (Table 1) melt with
4.5-5.2 wt% H2O at 1373-1673 K an 0.5 GPa, and one datum in dry haploandesite melt at
1673 K and 0.5 GPa. Behrens and Hahn (2009) examined Sm tracer diffusion in dry and wet
trachyte and phonolite melts (Table 1) at 0.5 GPa. The diffusion data are shown in Figure 42.
Sm diffusivity increases from polymerized melt to depolymerized melt, and increases with
increasing H2O.
Tracer diffusivities in dry and wet trachyte1 melt with H2O ≤ 1.7 wt% can be fit as:
 (24485 ± 4149) - (4790 ± 439) w 
TD = exp ( -14.70 ± 2.83) -
trachyte1
DSm  (60)
 T 
Equation (60) can reproduce experimental data to within 0.31 lnD units.
Using the same relation to fit tracer diffusion data in dry and wet phonolite1 melt
containing ≤ 1.9 wt% H2O, we obtain:
 (33207 ± 4055) - (3149 ± 343) w 
phonolite1
DSm TD = exp ( -6.62 ± 2.78) -  (61)
 T 
Equation (61) can reproduce experimental data to within 0.32 lnD units.
Eu diffusion. Eight papers reported Eu diffusion data in silicate melts (76 points). Because
Eu in natural melts can be present as Eu2 + and/or Eu3 + , and because Eu2 + and Eu3 + have
different diffusivities, it is important to control the oxygen fugacity in Eu diffusion studies.
368 Zhang, Ni, Chen
figure 42.Fig.
Sm 42.tracer diffusivities
Sm tracer insilicate
diffusivities in silicate melts.
melts. DataData sources:
sources: dryHA1
dry and wet andwith
wet4.5-5.2
HA1wt%with 4.5-5.2
H 2O wt% H2O
at 0.5 GPa
at 0.5 GPa (Koepke and Behrens 2001); dry and wet trachyte1 (tra1) and phonolite1 (pho1) (Behrens and
(Koepke and Behrens 2001); dry and wet trachyte1 (tra1) and phonolite1 (pho1) (Behrens and Hahn 2009); and
Hahn 2009); and jadeite and diopside melts (Nakamura and Kushiro 1998).
jadeite and diopside melts (Nakamura and Kushiro 1998).
Magaritz and Hofmann (1978b) explored Eu tracer diffusion in basalt10 and rhyolite12 melts
(Table 1) in air. They found that the data (7 points) below the liquidus behaved irregularly,
possibly due to convection induced by newly formed crystals. These data are excluded from
the compilation and are not used in the discussion here. Jambon (1982) investigated Eu tracer
diffusion in dry rhyolite5 melt at 973-1573 K and 0.0001-0.4 GPa. Henderson et al. (1985)
studied Eu tracer diffusivities in basalt1, andesite2, dacite2 and pantellerite1 melts at 1475 to
1673 K and in air. LaTourrette and Wasserburg (1997) examined the effect of oxygen fugacity
on Eu self diffusivity in dry HB1 melt by conducting experiments in air and in Fe-FeO buffer
at 1673-1773 K and 0.1 MPa. Nakamura and Kushiro (1998) investigated Eu tracer diffusion
in jadeite melt at 1573-1723 K and 0.75-2.0 GPa and in diopside melt at 1863 K and 1.0-1.25
GPa with uncontrolled oxygen fugacity although attempt was made to change the oxygen
fugacity. Koepke and Behrens (2001) reported Eu tracer diffusivities in wet (4.5-5.2 wt% H2O)
and dry HA1 melt at 1373-1673 K, 0.5 GPa, and logfO2 of about NNO + 3. Behrens and Hahn
(2009) investigated Eu tracer diffusion in dry and wet trachyte and phonolite melts at 0.5 GPa
42
in which the oxygen fugacity was varied.
Figure 43 shows Eu diffusion data. The individual Arrhenius relations can be found in
Online Supplementary Table 2. LaTourrette and Wasserburg (1997) determined Eu diffusivities
in air (where Eu is expected to be mostly Eu3 + ) and in the oxygen fugacity of the Fe-FeO buffer
(where Eu is expected to be mostly Eu2 + ). They found that Eu diffusivities at the reduced con-
dition of the Fe-FeO buffer are higher by 42% than those at oxidized conditions in air (the dif-
ference in lnD is 0.35), a relatively small difference (compare “HB1” and “HB1-r” in Fig. 43).
Based on multi-species diffusion treatment (Zhang et al. 1991a,b), assuming that the
fraction of Eu2 + does not depend on Eu concentration, DEu can be expressed as follows:
=DEu X Eu2+ DEu2+ + X Eu3 + DEu3 + (62)
where DEu2 + and DEu3 + are the diffusivities of Eu2 + and Eu3 + , XEu2 + = Eu2 + /(Eu2 + + Eu3 + ), and
XEu3 + = Eu3 + /(Eu2 + + Eu3 + ). Hence DEu lies between DEu3 + and DEu2 + . DEu3 + is expected to lie
figure 43. Eu diffusivities in various melts. (a) All are in air except for HB1-r (reduced at Fe-FeO buffer).
Fig. 43. Eu diffusivities in various melts. (a) All are in air except for HB1-r (reduced at Fe-FeO buffer). Dat
Data sources: pantellerite1 (pant1), dacite4, andesite2, and basalt1: Henderson et al. (1985); basalt10:
Magaritz and Hofmann (1978b); HB1 in air and in reduced conditions: LaTourrette and Wasserburg (1997).
sources: pantellerite1 (pant1), dacite4, andesite2, and basalt1: Henderson et al. (1985); basalt10: Magaritz and
(b) Diffusivities at high pressures. Data sources: rhy15 (rhyolite15): Jambon (1982); dry and wet HA1:
Koepke and Behrens (2001); dry and wet trachyte1 (tra1) and phonolite1 (pho1): Behrens and Hahn (2009).
Hofmann (1978); HB1 in air and in reduced conditions: LaTourrette and Wasserburg (1997). (b) Diffusivities
high pressures. Data sources: rhy15 (rhyolite15): Jambon (1982); dry and wet HA1: Koepke and Behrens (200
between DSm and DGd, which are similar (Fig. 44a). If the difference between DEu3 + and DEu2 + is
indeed so small
dryasand
shown by LaTourrette
wet trachyte1 (tra1) andand Wasserburg
phonolite1 (pho1):(1997),
Behrensthen DEu should
and Hahn (2009).not be much
greater than DSm and DGd. However, Figure 44a shows that DEu may be greater than DSm and DGd
by almost an order of magnitude. Hence, the conclusion that Eu2 + diffusivity is only slightly
greater than Eu3 + diffusivity by LaTourrette and Wasserburg (1997) is not general.
Behrens and Hahn (2009) also varied oxygen fugacity in determining the tracer diffusivities
of Eu in dry and wet trachyte and phonolite melts. Their data did not allow them to infer tracer
diffusivities of both Eu2 + and Eu3 + . Rather, they assumed DEu3 + is the average of DGd and DSm
of Sr2 + and Eu2 + are 0.118 and
(this should work well), and DEu2 + is the same as DSr (ionic radii 43
0.117 nm in octahedral sites), and estimated XEu2 + in each experiment. Figure 44b compares Eu
370 Zhang, Ni, Chen
figure 44. Comparison of Eu diffusivities with those of (a) Sm and Gd, and (b) Sr and Gd in some melts.
Fig. 44. Comparison of Eu diffusivities with those of (a) Sm and Gd, and (b) Sr and Gd in some melts. Eu
Eu diffusivities are higher by variable amount than both Sm and Gd diffusivities, but are between Sr and Gd
diffusivities. Data sources: wet HA1 melt at 0.5 GPa with 4.5-5.2 wt% H2O (Koepke and Behrens 2001);
jadeite melt atdiffusivities are higher by variable amount than both Sm and Gd diffusivities, but are between Sr and Gd
1.5 GPa (Nakamura and Kushiro 1998); dry trachyte1 (tra1) and phonolite1 (pho1) at 0.5
GPa (Behrens and Hahn 2009).
diffusivities. Data sources: wet HA1 melt at 0.5 GPa with 4.5-5.2 wt% H2O (Koepke and Behrens 2001); jadeite
melt at
diffusivities with 1.5 GPa
those (Nakamura
of Gd and Sr. and
Eu Kushiro 1998);indry
diffusivities trachyte1
these cases(tra1) andwithin
do lie phonolite1 (pho1) of
the range at 0.5 GPa (Behrens and
those of Gd andHahnSr,2009).
consistent with the assumption of Behrens and Hahn (2009). Nonetheless,
it is necessary to directly verify that DEu2 + is similar to DSr. To estimate Eu diffusivity in natural
melts, it is critical to evaluate the Eu2 + /Eu ratio because Eu2 + diffusivity can be an order of
magnitude higher than that of Eu3 + (Fig. 44).
Gd diffusion. Four papers reported Gd diffusion data in silicate melts (40 points).
Magaritz and Hofmann (1978b) investigated Gd (and Eu) tracer diffusion in basalt10 melt
in air. They concluded that tracer diffusivities of Gd and those of Eu are identical. The data
(7 points) below the liquidus of basalt10 are somewhat problematic
44 and are excluded from
the compilation as well as the discussion. Nakamura and Kushiro (1998) investigated Gd
tracer diffusion in jadeite melt at 1573-1723 K and 0.75-2.0 GPa and in diopside melt at 1863
K and 1.0-1.25 GPa. Koepke and Behrens (2001) reported Gd tracer diffusivities in a wet
haploandesite (Table 1) melt with 4.5-5.2 wt% H2O at 1373-1673 K and 0.5 GPa, and one
datum in dry haploandesite melt at 1673 K and 0.5 GPa. Behrens and Hahn (2009) investigated
Gd tracer diffusion in dry and wet trachyte and phonolite melts (Table 1) at 0.5 GPa. The
diffusion data
Zhangare
et al. shown in Figure
(Ch 8) Diffusion 45. melts
data in silicate
figure 45.Fig.
Gd45.tracer diffusivities
Gd tracer insilicate
diffusivities in silicate melts.
sources: jadeite
jadeite and and
diopside diopside
melts melts
(Nakamura (Nakamura and
and Kushiro
Kushiro 1998), dry and wet HA1 with 4.5-5.2 wt% H2O at 0.5 GPa (Koepke and Behrens 2001); dry and
1998), dry and wet HA1 with 4.5-5.2 wt% H2O at 0.5 GPa (Koepke and Behrens 2001); dry and wet trachyte1 (tra1)
wet trachyte1 (tra1) and phonolite1 (pho1) (Behrens and Hahn 2009); dry basalt10 at 0.1 MPa (Magaritz
and Hofmann 1978b).(pho1) (Behrens and Hahn 2009); dry basalt10 at 0.1 MPa (Magaritz and Hofmann 1978).
and phonolite1
Gd diffusivity increases from polymerized melt to depolymerized melt. Adding H2O

increases Gd diffusivity significantly. Gd tracer diffusivities in wet haploandesite melt
containing 4.5-5.2 wt% H2O at 1373-1673 K and 0.5 GPa can be expressed as:
 (19867 ± 1520) 
FEBD = exp ( -11.99 ± 1.00) -
wet HA
DGd  (63)
 T 
Assuming lnD increases linearly with H2O, Gd tracer diffusivities in dry and wet trachyte1
and phonolite1 melts (≤1.9 wt% H2O) at 1323-1527 K and 0.5 GPa can be fit as:
 (26366 ± 3986) - (5021 ± 453) w 
trachyte1
DGd TD = exp ( -13.60 ± 2.73) -  (64)
 T 
 (33656
45 ± 5536) - (3294 ± 468) w 
phonolite1
DGd TD = exp  ( -6.55 ± 3.79) -  (65)
 T 
Equations (64) and (65) reproduce diffusion data to within 0.31 and 0.51 lnD units, respectively.
Tb diffusion. Only one paper reported Tb diffusion data in silicate melts (8 points).
Mungall et al. (1999) investigated Tb tracer diffusion in HR7, HR7 + Na, and wet HR7 melts
(Table 1). The data are summarized in Figure 46. Tb diffusivities in dry HR7 melt at 1410-
372 Zhang, Ni, Chen
Fig. 46. Tbfigure 46.inTb

diffusivities drydiffusivities in dry
HR7, dry HR7+Na, and HR7, dry
wet HR7 (3.6HR7
wt% +
H2Na, and(Mungall
O) melts wet HR7et al. 1999).
(3.6 wt% H2O) melts (Mungall et al. 1999).
1873 K and 0.1 MPa can be expressed as:

 (38248 ± 1948) 
TD = exp ( -8.95 ± 1.20) -
dry HR7
DTb  (66)
 T 
Equation (66) reproduces experimental data to within 0.10 lnD units.
Dy diffusion. Only one paper reported Dy diffusion data in silicate melts (8 points).
Nakamura and Kushiro (1998) investigated Dy tracer diffusion in jadeite melt at 1573-1723
K and 0.75-2.0 GPa and in diopside melt at 1863 K and 1.0-1.25 GPa. The data are shown in
Figure 47. Assuming constant activation volume, Dy diffusivities in dry jadeite melt at 1573-
1723 K and 0.75-2.0 GPa can be expressed as:
 46240 - 2040 P 
dry jadeite
DDy TD ≈ exp  -1.93 -  (67)
 T 
Equation (67) reproduces experimental data to within
46 0.21 lnD units.
Ho diffusion. No Ho diffusion data in silicate melts are known.
Er diffusion. Two papers reported Er diffusion data in silicate melts (13 points). Nakamura
and Kushiro (1998) investigated Er tracer diffusion in jadeite melt at 1573-1723 K and 0.75-
2.0 GPa and in diopside melt at 1863 K and 1.0-1.25 GPa. Koepke and Behrens (2001) studied
Er tracer diffusion in dry and wet HA1 melts (Table 1). The data are shown in Figure 48.
Assuming constant activation volume, Er diffusivities in dry jadeite melt at 1573-1723 K and
0.75-2.0 GPa can be expressed as:
 47186 - 1953P 
DErdryTDjadeite ≈ exp  -1.30 -  (68)
 T 
For wet haploandesite containing about 4.5-5.2 wt% H2O at 1373-1673 K and 0.5 GPa, Er
diffusivities can be expressed as (Koepke and Behrens 2001):
 (17895 ± 2876) 
DErwetTDHA1 (5 wt%) = exp ( -13.34 ± 1.90) -  (69)
 T 
Tm diffusion. No Tm diffusion data in silicate melts are known.
Yb diffusion. Four
Zhang et al. (Ch papersdata
8) Diffusion reported Yb diffusion data (38 points). LaTourrette et al. (1996)
in silicate melts
studied Yb self diffusion in dry HB1 melt (Table 1) at 1623-1773 K and in air. Nakamura and
figureFig.
47.47.Dy
Zhang al.tracer
etDy tracer diffusivities
(Ch 8)diffusivities
Diffusion inin
in dry
data dry and
jadeite
silicate jadeite andmelts
diopside
melts diopside melts
(Nakamura and (Nakamura
Kushiro 1998). and Kushiro 1998).
47
figure
Fig. 48.48. Er diffusivities
Er diffusivities in dry
in dry jadeite andjadeite
diopsideand
meltsdiopside
(Nakamuramelts (Nakamura
and Kushiro and
1998), and dryKushiro 1998),
and wet HA1melts
and dry and wet HA1melts (Koepke and Behrens 2001).
(Koepke and Behrens 2001).
374 Zhang, Ni, Chen
Kushiro (1998) examined Yb tracer diffusion in jadeite melt at 1573-1723 K and 0.75-2.0 GPa
and in diopside melt at 1863 K and 1.0-1.25 GPa. Koepke and Behrens (2001) studied Yb
tracer diffusion in a wet HA1 melt (4.5-5.2 wt% H2O) at 0.5 GPa. Behrens and Hahn (2009)
investigated Yb tracer diffusion in dry and wet trachyte1 and phonolite1 melts at 0.5 GPa. The
diffusion data are shown in Figure 49.
Yb diffusivity increases from polymerized melt to depolymerized melt. Adding H2O
increases Yb diffusivity significantly. Yb tracer diffusivities in wet haploandesite melt
containing 4.5-5.2 wt% H2O at 1373-1673 K and 0.5 GPa can be expressed as:
 (20310 ± 10329) 
TD = exp ( -12.34 ± 6.06) -
wet HA1
DYb  (70)
 T 
In dry and wet trachyte and phonolite melts with ≤1.9 wt% H2O at 1323-1528 K and 0.5
GPa (Behrens and Hahn 2009), the tracer diffusion data can be fit as follows:
 (25867 ± 4109) - (4892 ± 467) w 
TD = exp ( -14.19 ± 2.81) -
trachyte1
DYb  (71)
 T 
 (33718 ± 5399) - (3570 ± 457) w 

FEBD = exp ( -6.96 ± 3.70) -
phonolite1
DYb  (72)
 T 
Equations (71) and (72) reproduce experimental data to within 0.31 and 0.40 lnD units, respec-
tively.
Lu diffusion. Two papers reported Lu diffusion data in silicate melts (15 points).
Nakamura and Kushiro (1998) characterized Lu tracer diffusion in jadeite melt at 1573-1723
K and 0.75-2.0 GPa and in diopside melt at 1863 K and 1.0-1.25 GPa. Mungall et al. (1999)
figure 49.
Fig.Yb49.diffusivities
Yb diffusivities in silicate
in silicate melts.
sources: self diffusivities
self diffusivities in atdry
in dry HB1 melt HB1(LaTourrette
0.1 MPa melt at 0.1 MPa
(LaTourrette et al. 1996); tracer diffusivities in jadeite and diopside melts (Nakamura and Kushiro 1998),
et al. 1996); tracer diffusivities in jadeite and diopside melts (Nakamura and Kushiro 1998), wet HA1 melt with 4.5-
wet HA1 melt with 4.5-5.2 wt% H2O at 0.5 GPa (Koepke and Behrens 2001); dry and wet trachyte1 (tra1)
and phonolite1
5.2 wt%(pho1) (Behrens
H2O at 0.5 GPa (Koepke andandHahn 2009).
Behrens 2001); dry and wet trachyte1 (tra1) and phonolite1 (pho1) (Behrens
and Hahn 2009).

investigated Lu tracer diffusion in dry HR7, dry HR7 + Na, and wet HR7 melts (Table 1). The
data are shown in Figure 50. Lu diffusivities in HR7 (3 points) are low and can be expressed as:
 (39949 ± 5206) 
TD = exp ( -8.07 ± 3.22) -
dry HR7
DLu  (73)
 T 
Equation (73) reproduces experimental data to within 0.26 lnD units. Adding 3.6 wt% H2O or
20% wt% Na2O increases DLu by roughly 3 orders of magnitude, depending on temperature.
Summary of Y and REE diffusion. There are a total of 423 diffusivity values determined
for Y and REE. Y and REE diffusion data in selected melts are compared in Figure 51. Because
the ionic radius of Y3 + (0.0900 nm) is similar to that of Ho3 + (0.0901 nm), and there are no
Ho diffusivity data, in the plots, Y is placed in the position of Ho. Under reduced conditions,
Eu diffusivity is often significantly higher than the rest of the REE, attributable to the presence
of Eu2 + . Other REEs mostly form a smooth trend. Data that are slightly off the trend are
attributed to experimental errors. Based on the data in various melts (basalt to rhyolite, and dry
to hydrous melts), there is a slight decrease of diffusivity from La to Lu, about 0.5 lnD units.
That is, diffusivities of the REE decrease as the size decreases, contrary to the alkali cations.
Overall, the rare earth elements as a group behave well in terms of diffusion properties. With
the smooth behavior, even though there are no Ho and Tm diffusion data, their diffusivities
can be readily predicted from the other REE diffusion data. Eu3 + tracer diffusivities can also
be well predicted. On the other hand, predicting Eu diffusivity at a given temperature, pressure
and oxygen fugacity requires knowing the Eu3 + /Eu2 + ratio and Eu2 + diffusivity.
ti, Zr, Hf
Ti diffusion. Nine papers reported Ti diffusion data in silicate melts (59 points). Most
of the data were SEBD obtained as a byproduct of crystal dissolution experiments. Zhang et
al. (1989) reported a single Ti diffusivity (SEBD) in dry andesite melt (Table 1) during rutile
dissolution at 1648 K and 1.5 GPa. LaTourrette et al. (1996) investigated Ti self diffusion in
dry HB1 melt (Table 1) at 1623-1773 K and in air. Mungall et al. (1999) studied Ti tracer
diffusion in dry HR7, wet HR7 (3.6 wt% H2O), and dry HR7 + Na melt (Table 1) at 1083-1873
figure 50.
Fig. 50. Lu Lu diffusion
diffusion datamelts.
data in silicate in silicate melts.
Data for jadeiteData for jadeite
and diopside and
melts are diopside
from Nakamuramelts are
and Kushiro
from Nakamura and Kushiro (1998), and the rest are from Mungall et al. (1999).
(1998), and the rest are from Mungall et al. (1999).
376
figure 51. Diffusivities of

rare earth elements in various
silicate melts. Data sources
are: jadeite (Jd) and diopside
(Di) melts: Nakamura and
Kushiro (1998); rhyolite12
(rhy12): Rapp and Watson
(1986); dry and wet HR7:
Mungall et al. (1999); dry
and wet haploandesite (HA1):
Zhang, Ni, Chen
Koepke and Behrens (2001);

dry and wet phonolite1 (pho):
Behrens and Hahn (2009).
Fig. 51. Diffusivities of rare earth elements in various silicate melts. Data sources are: jadeite (Jd) and diopside
(Di) melts: Nakamura and Kushiro (1998); rhyolite12 (rhy12): Rapp and Watson (1986); dry and wet HR7: Mungall
et al. (1999); dry and wet haploandesite (HA1): Koepke and Behrens (2001); dry and wet phonolite1 (pho): Behrens
K, 0.0001 and 1 GPa. Van Orman and Grove (2000) extracted a Ti diffusivity value (SEBD)
in lunar basalt melt (LB1 in Table 1) during clinopyroxene dissolution at 1623 K and 1.3
GPa. Lundstrom (2003) obtained two SEBD values of Ti from dry basalt7-basanite diffusion
couple experiment at 1723 K and 0.9 GPa. Morgan et al. (2006) acquired SEBD values of Ti
in two lunar picrite melts (LP1 and LP2 in Table 1) during anorthite dissolution at 1673 K and
0.6 GPa. Hayden and Watson (2007) extracted Ti diffusivities (SEBD) in hydrous rhyolite11,
trondhjemite, and rhyolite6 (an intermediate mixture between rhyolite11 and trondhjemite)
melts during rutile dissolution, with H2O content by difference-from-100% method. Chen and
Zhang (2008, 2009) determined Ti diffusivities (SEBD) in basalt11 melt during olivine and
clinopyroxene dissolution.
As discussed in Zhang et al. (1989) and Zhang (2008, 2010), SEBD values in melts
from crystal dissolution experiments are more self-consistent (meaning cross-terms in
multicomponent diffusion play a smaller role) if the component is the major component in the
crystal that determines the crystal-melt equilibrium. Hence, it is expected that SEBD values
of Ti during rutile dissolution (Zhang et al. 1989; Hayden and Watson 2007) are more reliable
and consistent than other experimental data. On the other hand, Ti SEBD values in melts
obtained from olivine and clinopyroxene dissolution (Van Orman and Grove 2000; Morgan
et al. 2006; Chen and Zhang 2008, 2009) are affected or even dominated by cross diffusion
terms (sometimes even showing uphill diffusion as during olivine dissolution in andesite melt;
Zhang et al. 1989) and are hence less consistent from one condition to another. Nonetheless,
as data are compared in Figure 52, the SEBD data in basalt melts are roughly consistent with
self diffusivities in HB1 melt. From HB1 to mid-ocean ridge basalt to lunar picrites to basalt7-
basanite couple during dissolution of various minerals, Ti self diffusivities and SEBD vary by
no more than a factor of 5. The scatter in Ti SEBD in hydrous rhyolite to trondhjemite with
unspecified H2etOal.between
Zhang 5 todata
(Ch 8) Diffusion 12inwt%
silicateis also a factor of 5 (Hayden and Watson 2007). It is not
melts
figure 52.Fig.Ti52.diffusivities
Ti diffusivitiesin dryHB1
in dry HB1meltmelt
at 0.1at
MPa0.1(HB,
MPa (HB, LaTourrette
LaTourrette et al. 1996), dryet al. 1996),
basalt11 dry olivine
melt during basalt11 melt
during olivine dissolution at 0.5-1.4 GPa (Chen and Zhang 2008), dry basalt11 melt during clinopyroxene
dissolution at 0.5-1.9
dissolution GPaGPa
at 0.5-1.4 (MORB2, Chen2008),
(Chen and Zhang and dry Zhang 2009),
basalt11 dry clinopyroxene
melt during LB1 melt dissolution
during clinopyroxene
at 0.5-1.9 and
anorthite dissolution
GPa (MORB2, Chen and Zhang 2009), dry LB1 melt during clinopyroxene and anorthite dissolution (Van Orman couple
(Van Orman and Grove 2000; Morgan et al. 2006), dry b-b (basalt7-basanite)
(Lundstrom 2003), dry andesite1 melt during rutile dissolution (Zhang et al. 1989), hydrous rhyolite to
trondhjemite melts2000;
and Grove withMorgan
5-12 etwt% H2Odry
al. 2006), during rutile dissolution
b-b (basalt7-basanite) couple (rhy-tron, Hayden
(Lundstrom 2003), and Watson
dry andesite1 melt 2007), and
dry HR7, during
wet HR7 rutile (3.6 wt%(Zhang
dissolution H2O)etand dry HR7+Na
al. 1989), melts
hydrous rhyolite (Mungall et
to trondhjemite al.with
melts 1999).
5-12 wt% H2O during
rutile dissolution (rhy-tron, Hayden and Watson 2007), and dry HR7, wet HR7 (3.6 wt% H2O) and dry HR7+Na
melts (Mungall et al. 1999).

378 Zhang, Ni, Chen
clear whether this is due to insensitivity of Ti diffusivity with respect to H2O at such high H2O,
or due to uncertainties in the H2O content. Ti diffusivities in some melts are given below:
 (24600 ± 3800) 
DTidrySDHB1; 0.1 MPa = exp ( -10.08 ± 2.23) -  (74)
 T 
 (39776 ± 4960) 
DTidryTDHR7 = exp ( -9.16 ± 2.99) -  (75)
 T 
Ti SEBD values in basalt11 melt during clinopyroxene dissolution is about 1.6 times, and
those during olivine dissolution is about 4 times the self diffusivities in dry HB1 (Eq. 52a).
Ti SEBD in dry andesite1 melt during rutile dissolution at 1.5 GPa is 0.6 times the diffusivity
from Equation (74). Adding 3.6 wt% H2O or 20 wt% Na2O to HR7 melt increases Ti diffusivity
by roughly the same amount. Ti diffusivity increases from HR7 (a haplorhyolite) melt to basalt
melt by about 4 orders of magnitude. The effect of 5-12 wt% H2O on Ti duffusivity in silicic
melts is greater than that in mafic melts.
Zr diffusion. Nine papers reported Zr diffusion data in silicate melts (100 points).
Harrison and Watson (1983) extracted SEBD (close to FEBD and tracer diffusivities) of Zr in
rhyolite12 melt (Table 1) during zircon dissolution at 1473-1673 K, 0.8 GPa and 0.1-6.3 wt%
H2O. Baker and Watson (1988) determined SEBD (close to FEBD) of Zr using the rhyolite1-
rhyolite7 diffusion couple at 1171-1673 K and 0.01-1 GPa. Baker et al. (2002) studied SEBD
(close to FEBD) of Zr in rhyolite14 melt and two other rhyolite melts with slightly different
concentrations of FeO (difference of 1.2 wt%), F (difference of 1.2 wt%) and Cl (difference
of 0.35 wt%) during zircon dissolution at 1323-1673 K, 1 GPa, and ≤ 5.1 wt% H2O. No
difference was found in Zr diffusivity among the three rhyolite melts used by Baker et al.
(2002). LaTourrette et al. (1996) investigated Zr self diffusion in dry HB1 melt at 1623-1773
K and in air. Nakamura and Kushiro (1998) characterized Zr tracer diffusion in jadeite melt at
1673 K and 1.25-2 GPa. Mungall et al. (1999) examined Zr tracer diffusion in dry HR7, wet
(3.6 wt% H2O) HR7, and dry HR7 + Na melt (Table 1) at 1083-1873 K, 0.0001 and 1 GPa.
Koepke and Behrens (2001) measured Zr tracer diffusivities in wet HA1 melt with about 4.5-
5.2 wt% H2O at 1373-1673 K and 1 GPa and one datum in dry HA1 melt. Lundstrom (2003)
reported a single SEBD value of Zr from a dry basalt7-basanite diffusion couple experiment at
1723 K and 0.9 GPa. Behrens and Hahn (2009) investigated Zr tracer diffusion in dry and wet
trachyte1 and phonolite1 melts at 1323-1528 K and 1 GPa.
Zr diffusion in rhyolite melts has been investigated extensively (Harrison and Watson
1983; Baker and Watson 1988; Baker et al. 2002) and hence inter-laboratory comparisons
can be made. The data for rhyolite melts are shown in Figure 53. Some data are difficult
to reconcile. For example, the huge pressure effect from 0.01 to 1 GPa in the Zr diffusion
data of Baker and Watson (1988) (solid squares and open triangles) are difficult to explain.
In dry rhyolite melts of similar compositions, Zr diffusivities in samples containing < 0.7
wt% halogens (Baker and Watson 1988) are about 2 orders of magnitude higher than those
obtained by Harrison and Watson (1983). Baker and Watson (1988) attributed the difference
to the effect of halogens, but a later study (Baker et al. 2002) indicated that adding even 1.2
wt% F has only a minor effect on Zr diffusivities. Assuming that the more recent paper by the
same lead author is more likely correct, it appears that the data in Baker and Watson (1988)
are erroneous, and the effect of F and Cl at the level of ≤ 1 wt% is minor on Zr diffusion. For
zircon dissolution in wet rhyolite melts (solid triangles and circles with center dot), there is no
consistency either: Zr diffusivities at 6.0 wt% H2O (Harrison and Watson 1983) are smaller
than those at 4.2-4.8 wt% H2O (Baker et al. 2002). Two possible explanations can be advanced
although it is difficult to judge which data are more reliable. First, there might be convection in
the zircon dissolution experiments (Harrison and Watson 1983; Baker et al. 2002), which could
have compromised the diffusion data (see Zhang et al. 1989), although the solubility of zircon
Zhang et al. (Ch 8) Diffusion data in silicate melts 379
figure 53.
Fig. Zr diffusivities
53. Zr diffusivities inin rhyolite
rhyolite (rhy) (rhy) and
and HPG HPGPercentage
melts. melts. inPercentage in thewt%
the legend indicates legend indicates
of H2O. Data wt% of
H2O. Data sources: HW83 (Harrison and Watson 1983); BW88 (Baker and Watson 1988); B02 (Baker et
al. 2002);sources:
M99 HW83 (Harrison and Watson 1983); BW88 (Baker and Watson 1988); B02 (Baker et al. 2002); M99
(Mungall et al. 1999).
(Mungall et al. 1999).
is small and hence less likely to cause convection. Secondly, in earlier studies (Harrison and
Watson 1983), uncertainties in H2O concentration were large. Due to these inconsistencies, at
the moment, we cannot confidently assess Zr diffusivities in rhyolite melts even with so many
data. More experimental data minimizing convection and with accurate determination of H2O
contents are necessary.
Tracer diffusivities of Zr in dry and wet (3.6 wt% H2O) haplorhyolite (HR7) melt
(Mungall et al. 1999) are also shown in Figure 53 for comparison with natural rhyolite melts.
Zr diffusivities in dry HR7 are lower than those in dry natural rhyolite melts. Zr diffusivities in
wet HR7 melt containing 3.6 wt% H2O seem to be more consistent with the data of Baker et
al. (2002) than with the data of Harrison and Watson (1983), considering that diffusivities are
expected to increase with H2O contents.
Zr self diffusivities in dry HB1 melt, tracer diffusivities in dry and wet HA1, phonolite1
and trachyte1 melts are shown in Figure 54. Based on Figures 53 and 54 and the original
papers, Zr diffusivities in some systems are as53
follows:
 (42366 ± 6397) 
DZrdryTDHR7 = exp ( -7.55 ± 3.87) -  (76)
 T 
 (25865 ± 6299) 
DZrdrySDHB1 = exp ( -9.82 ± 3.70) -  (77)
 T 
 (22736 ± 547) 
DZrwetTDHA1; 5% = exp ( -11.18 ± 0.36) -  (78)
 T 
where % in the superscript indicates wt% of H2O.
Hf diffusion. Three papers reported Hf diffusion data in silicate melts (31 points), all on
tracer diffusion. Mungall et al. (1999) investigated Hf diffusion in dry HR7, wet HR7 (3.6 wt%
380 Zhang, Ni, Chen
figure 54.Fig.Zr54.self andand

Zr self tracer
tracer diffusivities. Data
diffusivities. Data sources:
sources: HB1 (LaTourrette
HB1 (LaTourrette et al. 1996);et al.(Koepke
HA1 1996);and
HA1 (Koepke and
Behrens
Behrens 2001); tra1 (trachyte1) and pho1 (phonolite1) (Behrens and Hahn 2009); jadeite and diopside
2001); tra1 (trachyte1) and pho1 (phonolite1) (Behrens and Hahn 2009); jadeite and diopside melts (Nakamura and
melts (Nakamura and Kushiro 1998); dry and wet HR7 and HR7 + Na (Mungall et al. 1999).
Kushiro 1998); dry and wet HR7 and HR7+Na (Mungall et al. 1999).
H2O) and dry HR7 + Na melts (Table 1). Koepke and Behrens (2001) examined Hf diffusion in
a wet haploandesite melt (4.5-5.2 wt% H2O). Behrens and Hahn (2009) studied Hf in dry and
wet trachyte1 and phonolite1 melts. The diffusion data are shown in Figure 55. For dry melts,
Hf diffusivities increase from HR7 to trachyte and phonolite to HR7 + Na. Adding 3.6 wt%
H2O to HR7 has roughly the same effect as adding 20 wt% Na2O to HR7. Adding ~1.8 wt%
H2O to phonolite melt increases diffusivity by a factor of about 50. Hf diffusivities in some
systems are as follows:
 (42614 ± 13496) 
TD = exp ( -7.90 ± 8.14) -
dry HR7
DHf  (79)
 T 
 (20586 ± 3891) 
wet HA1; 5%
DHf TD = exp ( -12.55 ± 2.57) -  (80)
 T 
where % in the superscript indicates wt% of H254
O.
Figure 56 compares Ti, Zr, and Hf diffusion data in four melts. In three of the four melts
(dry HB1, dry and wet HR7 melts), Ti, Zr and Hf diffusivities (self and tracer diffusivities)
are similar (often within a factor of 3). However, in wet rhyolite melts, Ti and Zr diffusivities
(SEBD) differ by more than two orders of magnitude (Hayden and Watson 2007). We
tentatively attribute the difference to either the unknown difference in H2O content, or the
more variable nature of SEBD values, or some experimental difficulties such as convection in
the experiments. Hence, it is tentatively concluded that Ti, Zr and Hf diffusivities are similar
to each other in silicate melts.
v, Nb, ta
V diffusion. No V diffusion data in silicate melts are known.
Nb diffusion. Six papers reported Nb diffusion data in silicate melts (42 points). Baker and
Watson (1988) explored SEBD (close to FEBD) of Nb using diffusion couple of dry rhyolite1
381
figure 55.
Fig.Hf tracer
55. Hf tracer diffusivities inHR7,
diffusivities in dry drywetHR7,
HR7, wet
HR7, wet
dry HR7+Na, dryHA1,
HR7dry+and
Na,wetwet
tra1 HA1, dryandand
(trachyte1) dry wet tra1
(trachyte1) and dry and wet pho1 (phonolite1) melts with wt% of H2O indicated for wet melts. Data
and wet pho1 (phonolite1) melts with wt% of H2O indicated for wet melts. Data sources: M99 (Mungall et al.
sources: M99 (Mungall et al. 1999); KB01 (Koepke and Behrens 2001); BH09 (Behrens and Hahn 2009).
1999); KB01 (Koepke and Behrens 2001); BH09 (Behrens and Hahn 2009).
55

Fig. 56. ComparisonofofTi, Zrand
Ti, Zr andHfHf diffusivities
diffusivities inHB1
in (i) dry (i) dry HB1
melt at melt(ii)atdry
0.1 MPa, 0.1HR7
MPa,
melt (ii)
at 0.1dry HR7
MPa, wet melt at
0.1 MPa, wet HR7 melt with 3.6 wt% H2O at 1 GPa, and (iv) wet rhy-tron (rhyolite-trondhjemite) melt with
HR7 melt with 3.6 wt% H2O at 1 GPa, and (iv) wet rhy-tron (rhyolite-trondhjemite) melt with 5-12 wt% H2O at 1
5-12 wt% H2O at 1 GPa and wet rhyolite melt with 6 wt% H2O at 0.8 GPa.
GPa and wet rhyolite melt with 6 wt% H2O at 0.8 GPa.
and rhyolite7 (Table 1) melts at 1171-1673 K and 0.01-1 GPa. Nakamura and Kushiro (1998)
examined Nb tracer diffusion in jadeite melt at 1673 K and 1.25-2.0 GPa. Mungall et al.
(1999) studied Nb tracer diffusion in dry HR7, wet HR7 (3.6 wt% H2O), and dry HR7 + Na
melts. Koepke and Behrens (2001) reported Nb tracer diffusivities in a wet haploandesite melt
at 1373-1673 K and 0.5 GPa, plus one datum for dry haploandesite melt. Lundstrom (2003)
obtained a single datum of SEBD of Nb between basalt7-basanite melts. Behrens and Hahn
382 Zhang, Ni, Chen
(2009) characterized Nb diffusion in dry and wet trachyte1 and phonolite1 melts at 1323-1528
K, 0.5 GPa, and ≤ 1.9 wt% H2O. Figure 57 shows Nb diffusion data in various melts.
Nb diffusivities increase from dry HR7 melt at 0.1 MPa, to dry rhyolite, trachyte and
phonolite melts, to wet HA1 melt. The effect of adding 3.6 wt% H2O to HR7 is about the same
as adding 20 wt% Na2O in terms increasing Nb diffusivity. All tracer diffusion data of Nb in
dry silicate melts can be fit as follows:
 (71534 XSA + 22351X NK - 28958 - 290 P ) 
all dry melts
DNb TD = exp ( -10.92 -  (81)
 T 
where XSA and XNK are the cation mole fractions of Si + Al and Na + K, respectively. Equation
(81) reproduces all data in dry melts (≤0.1 wt% H2O) to within 0.74 lnD units except for one
clear outlier (trachyte
Zhang et al. (Ch 8)melt).
DiffusionEquation
data in silicate(81)
melts cannot reproduce SEBD data of Nb.
figure 57.Fig.Nb
57.diffusivities
Nb diffusivities in dryr-rr-r
in dry (rhyolite1-rhyolite7)
(rhyolite1-rhyolite7) couple
couple at 0.01 and 1at 0.01
GPa andand1Watson
(Baker GPa (Baker
1988), dryand
HR7Watson
1988), dry HR7 melt at 0.1 MPa, wet HR7 with 3.6 wt% H2O at 1 GPa, dry HR7 + Na at 0.1 MPa (Mungall
melt at 0.1 MPa, wet HR7 with 3.6 wt% H2O at 1 GPa, dry HR7+Na at 0.1 MPa (Mungall et al. 1999), dry jadeite at
et al. 1999), dry jadeite at 1.25-2.0 GPa (Nakamura and Kushiro 1998), dry and wet (4.5-5.2 wt% H2O)
HA1 at 0.5 GPa GPa
1.25-2.0 (HA, Koepke
(Nakamura andand Behrens
Kushiro 2001),
1998), dry and wetdry b-b wt%
(4.5-5.2 (basalt7-basanite) couple
H2O) HA1 at 0.5 GPa (Lundstrom
(HA, Koepke and 2003),
and dry and wet (1.1-1.7 wt% H2O) tra1 (trachyte1) and dry and wet (1.7-1.9 wt% H2O) pho (phonolite1)
melts at 0.5 GPa2001),
Behrens (Behrens
dry b-band Hahn 2009).
(basalt7-basanite) couple (Lundstrom 2003), and dry and wet (1.1-1.7 wt% H2O) tra1
(trachyte1) and dry and wet (1.7-1.9 wt% H2O) pho (phonolite1) melts at 0.5 GPa (Behrens and Hahn 2009).
Ta diffusion. Only one paper (Mungall et al. 1999) reported Ta tracer diffusion data in
silicate melts (7 points). Figure 58 shows the data. Ta diffusivities in HR7 melt (Table 1) can
be fit as (Mungall et al. 1999):
 (33792 ± 4683) 
TD = exp ( -12.41 ± 2.84) -
dry HR7
DTa  (82)
 T 
Equation (82) reproduces experimental data to within 0.24 lnD units. Adding 3.6 wt% H2O
or 20 wt% Na2O increases DTa by roughly the same factor (3 to 5 orders of magnitude,
depending on temperature). The limited data indicate that Ta diffusivities are within a factor
2 of Nb diffusivities but can be either larger or smaller than Nb diffusivities (probably within
experimental uncertainties).
57
figure
Fig. 58. Ta 58. Ta diffusivities
diffusivities in dry
in dry HPG melt HPG
at 0.1 MPa,melt at 0.1 at
dry HR7+Na MPa, dryand
0.1 MPa, HR7wet+HR7
Na (3.6
at 0.1
wt%MPa,
H2O) at 1
and wet HR7 (3.6 wt% H2O) at 1 GPa. Data from Mungall et al. (1999).
GPa. Data from Mungall et al. (1999).
Cr, Mo, W
Cr diffusion. Three papers reported Cr diffusion data in silicate melts (18 points). Zhang
et al. (1989) reported a single Cr SEBD value in andesite1 melt during olivine dissolution at
1563 K and 0.55 GPa. Koepke and Behrens (2001) reported Cr tracer diffusivities in wet HA1
containing 4.5-5.2 wt% H2O at 1373-1673 K and 0.5 GPa, and one single datum at 1673 K
and 0.5 GPa for dry HA1. Behrens and Hahn (2009) investigated Cr tracer diffusion in dry and
wet trachyte1 and phonolite1 melts at 1373-1528 K, 0.5 GPa and ≤ 1.9 wt% H2O. The data are
shown in Figure 59.
The single Cr SEBD datum in dry andesite1 melt and the single Cr tracer diffusivity in
dry HA1 melt are roughly consistent and lie roughly in the same trend as Cr tracer diffusivities
in dry phonolite1 melt. Cr tracer diffusivities in dry trachyte melt are lower than those in dry
phonolite melt by a factor of 5. Cr tracer diffusivities in wet HA1 melt containing 4.5-5.2
wt% H2O and those in wet phonolite1 melt containing 1.7 wt% H2O are similar. Cr tracer
diffusivities in wet HA1 melt containing 4.5-5.2 wt% H2O at 0.5 GPa can be expressed as:
58
 (27244 ± 6343) 
DCrwetTDHA1; 4.5-5.2% = exp ( -7.00 ± 4.19) -  (83)
 T 
where % in the superscript indicates wt% of H2O.
Mo diffusion. No Mo diffusion data in silicate melts are known.
W diffusion. One paper (Mungall et al. 1999) reported W tracer diffusion data in silicate
melts (6 points), on dry HR7 and HR7 + Na melts. The data are shown in Figure 60. The three
points on dry HR7 do not line up well in an Arrhenius plot. Diffusivities in HR7 + Na are 3 to
4 orders of magnitude higher. W diffusivities in dry HR7 melts are similar to Nd diffusivities
(Fig. 60). However, in HR7 + Na melt, this similarity does not hold any more.
Mn, fe, Co, Ni, Cu, Zn
Mn diffusion. Three papers reported Mn diffusion data in silicate melts (28 points). Lowry
et al. (1982) investigated Mn tracer diffusivities in dry basalt1 and andesite2 melts at 1567-
384 Zhang, Ni, Chen
figure 59.Fig.
Cr59.
tracer
Zhang et al.
Cr diffusivities
(Ch
tracer 8) Diffusionin
diffusivities dry
indata
dry and
inand wetwet
silicate HA1
melts melts
melts
HA1 (Koepke(Koepke and
and Behrens Behrens
2001), 2001),melt
dry andesite1 dry(Zhang
andesite1
et melt
(Zhang etal.
al.1989),
1989), and dry and wet trachyte1 and phonolite1 melts
and dry and wet trachyte1 and phonolite1 melts (Behrens and Hahn 2009).
(Behrens and Hahn 2009).
59
figure
Fig. 60. W tracer 60. Wintracer
diffusivities diffusivities
dry HR7 and HR7+Na in dry HR7
melts. and HR7
Nd diffusivities are+shown
Na melts.
for comparison.
Nd diffusivities are shown for comparison.
1676 K and 0.1 MPa. Henderson et al. (1985) studied Mn tracer diffusivities in dry dacite4
and pantellerite1 melts at 1473-1673 K and 0.1 MPa. Baker and Watson (1988) investigated
Mn SEBD (close to FEBD) in dry rhyolite1-rhyolite7 melts at 1218-1672 K and 0.01 and 1.0
GPa. Figure 61 displays all the data. Except for one outlier point (Fig. 61), all TD and SEBD
data of Mn at ≤ 0.02 GPa can be fit by:
all dry melts; ≤ 0.02 GPa  23993 

DMn = exp 1.15 - 15.44 X1 - 39.92 X 2 - (84)
T 
TD & SEBD

figure 61.Fig.
Mn 61.diffusivities
Mn diffusivities(tracer diffusivities
(tracer diffusivities and and SEBD)
SEBD) in dryand
in dry basalt1 basalt1 and
andesite2 andesite2
melts (Lowry etmelts (Lowry
al. 1982), dry et al.
1982), dry dacite4 and pantellerite1 melts (Henderson et al. 1985), and dry r-r (rhyolite1-rhyolite7) couple
(Baker anddacite4
Watson and pantellerite1 melts (Henderson et al. 1985), and dry r-r (rhyolite1-rhyolite7) couple (Baker and Watson
1988).
1988).
where X1 is the sum of cation mole fractions of Si + Ti + Al + P, and X2 = max(Na + K-Al,0) (i.e.,
X2 = Na + K-Al if Na + K-Al > 0, and X1 = 0 if Na + K-Al < 0, this is because peralkalinity
seems to affect diffusivity more than peraluminity). The available data cannot resolve the
compositional dependence of the activation energy. The maximum error in reproducing the
experimental data by Equation (84) is 0.37 lnD units (Fig. 61). The Mn SEBD data in rhyolite
at 1 GPa (Baker and Watson 1988) deviate from Equation (84) (e.g., the data would indicate a
negative activation energy), probably due to experimental or analytical problems.
Mn and Co diffusivities are similar in all the melts that have been investigated (basalt,
andesite, dacite and pantellerite1).
Fe diffusion. Eleven papers reported Fe diffusion data in silicate melts (165 points). In
silicate melts, Fe can be present as ferrous (Fe2 + ) or ferric (Fe3 + ). Diffusivities of the two
different states are likely significantly different. Hence, it is important to characterize Fe
oxidation state before the experiment (e.g., whether the two halves of a diffusion couple have
the same ferric/ferrous ratio), control oxygen fugacity
61 during the experiments, and/or measure
the oxidation state of Fe after the experiment. Lowry et al. (1982) examined Fe radioactive
tracer diffusion in basalt1 and andesite2 melts under oxidized condition at 1570-1675 K
and in air. Henderson et al. (1985) studied Fe diffusivities in dacite4 and pantellerite1 melts
under oxidized condition at 1475-1672 K and in air. Baker and Watson (1988) investigated Fe
SEBD in rhyolite1-rhyolite7 and HD2-rhyolite7 couples (Table 1) at 1171-1273 K and 0.01
GPa and 1373-1473 K and 1 GPa (fO2 not characterized). The pressure effect is insignificant
compared to data scatter. Dunn and Ratliffe (1990) investigated tracer diffusion of ferrous
iron in a peraluminous sodium aluminosilicate (HR1 in Table 1) at 1516-1716 K and 0.1
MPa using a CO-CO2 gas mixture to control oxygen fugacity to be near the FMQ buffer, and
at 1523-1723 K and 0.6-2 GPa in graphite capsules. Watson (1991b) examined Fe transport
in slightly melted dunite. Because the extracted diffusivity is not for pure liquid, the data
will not be used further due to limitations of the scope of this chapter. Baker and Bossanyi
(1994) examined the effect of H2O and fluorine on Fe SEBD in dacite1-rhyolite7 diffusion
couples at 1373-1673 K and 1 GPa. van der Laan et al. (1994) reported a single Fe diffusivity
386 Zhang, Ni, Chen
(SEBD) in a rhyolite3-rhyolite16 couple at 1523 K and 1 GPa. Koepke and Behrens (2001)
obtained Fe tracer diffusivities in hydrous HA1 melt at 1373-1583 K and 0.5 GPa with oxygen
fugacity roughly at NNO + 3. Lundstrom (2003) obtained two Fe SEBD values in basalt7-
basanite couple. Ruessel and Wiedenroth (2004) characterized the compositional effect on Fe
tracer diffusivities in Na2O-CaO-MgO-Al2O3-SiO2 melts (Table 1) in air and 1300-1873 K, but
diffusivity values were only reported at 1573 K. Morgan et al. (2006) determined Fe SEBD in
lunar picrites (LP1 and LP2) during anorthite dissolution at 1673 K and 0.6 GPa.
Due to the presence of both Fe2 + and Fe3 + in typical geological conditions, with Fe3 +
likely having lower diffusivities (due to higher bond strength than Fe2 + ), Fe diffusivity is
expected to lie between that of Fe2 + and Fe3 + . If oxygen fugacity is uniform in the melt,
equilibrium between Fe2 + and Fe3 + would lead to uniform ferric to ferrous ratio. Following
Zhang et al. (1991b), Fe diffusivities can be expressed as follows:
=DFe X Fe2+ DFe2+ + X Fe3 + DFe3 + (85)
where XFe2 + = Fe2 + /(Fe2 + + Fe3 + ). Based on valence and size consideration, it can be argued
that Fe2 + diffusivities lie between that of Mn2 + and Co2 + (ionic radii of Mn2 + , Fe2 + and Co2 +
in octahedral sites at high spin are 0.083, 0.078 and 0.0745 nm, Shannon 1976). If Fe3 + is
in the tetrahedral site, Fe3 + diffusivity is likely similar to that of Ga3 + (ionic radii of Fe3 +
and Ga3 + in tetrahedral sites are 0.049 and 0.047 nm). If Fe3 + is in the octahedral sites, Fe3 +
diffusivity is likely similar to that of Cr3 + , or V3 + , or Ga3 + (ionic radii of Fe3 + , Cr3 + , V3 + , and
Ga3 + in octahedral sites are 0.0645, 0.0615, 0.062, and 0.64 nm).
In experiments and nature, Mn and Co are divalent, and Ga and Cr are trivalent. Hence for
Mn, Co, Cr and Ga, we do not need to worry about the oxidation state. Furthermore, Mn and
Co diffusivities are similar (Fig. 62). Hence, Fe diffusivities in melts are expected to lie between
those of Mn and Ga, or of Co and Ga; Figure 62 is consistent with this expectation. Furthermore,
Figure 62 shows that Fe2 + diffusivity is about 6 times Fe3 + diffusivity in dacite melts.
Oxygen fugacity was not always controlled in experimental studies. For example, no care
was taken to control oxygen fugacity in the experiments of Baker and Watson (1988), Baker
and Bossanyi (1994) and van der Laan et al. (1994) because the main goal of their studies was
not on Fe diffusion. Some Fe diffusion data are shown in Figure 63.
Lowry et al. (1982) and Henderson et al. (1985) conducted the tracer diffusion
experiments in air. There might still be some uncertainty about the oxidation state of the
diffusion experiments because the initial glass was not equilibrated in air before diffusion
experiments. Nonetheless, Fe diffusivities are significantly lower (by a factor of about 3)
than both Mn and Co diffusivities, and only slightly higher than Ga diffusivities (Fig. 62),
consistent with the expectation that Fe is mostly trivalent in these experiments. Fe diffusivities
(TD) in metaluminous basalt1-andesite2-dacite4 melts by Lowry et al. (1982) and Henderson
et al. (1985) in air can be expressed as:
 29879 
basalt-andesite-dacite; air
DFe = exp 5.30 - 17.64 X1 - (86)
T 
TD

where X1 is the sum of cation mole fractions of Si + Ti + Al + P. The available data cannot resolve
the compositional dependence of the activation energy (about 248 kJ/mol). The maximum
error in reproducing the experimental data by Equation (86) is 0.42 lnD units (Fig. 63).
Dunn and Ratliffe (1990) studied Fe2 + diffusion in HR1 melt (a haplorhyolite melt) by
using reducing conditions at 0.0001 and 0.6-2 GPa so that Fe2 + is the dominant iron species.
They found that the pressure effect is insignificant from 0.6 to 2 GPa (diffusivity variation
is within 0.3 lnD units). However, Fe tracer diffusivity values at 0.6-2 GPa using graphite
capsules is 5 times those at 0.1 MPa at FMQ buffer. Dunn and Ratliffe (1990) attributed the
figure 62.
Zhang
Fig. et
62.al.Tracer
Tracer (Ch 8)diffusivities
Diffusion data
diffusivities ofofMnin silicate
Mn and
and CoComelts
inin dacite4
dacite4 melt melt (Henderson
(Henderson et those
et al. 1985), al. 1985),
of Ga inthose
dacite1ofmelt
Ga in dacite1
melt (Baker(Baker
1992a) compared with those of Fe in dacite4 melt in air (Henderson
1992) compared with those of Fe in dacite4 melt in air (Henderson et al. 1985). et al. 1985).
62
figure 63.
Fig.Some experimental
63. Some dataonon
experimental data Fe diffusion
Fe diffusion in silicate
in silicate melts.
melts. Data Data
for dry forand
basalt1 drydrybasalt1 and
andesite2 dryareandesite2
melts
melts are from Lowry et al. (1982); dry dacite4 and pantellerite1 melts are from Henderson et al. (1985);
from Lowry et al. (1982); dry dacite4 and pantellerite1 melts are from Henderson et al. (1985); dry HR1 at 0.1 MPa
dry HR1 at 0.1 MPa and 0.6-2 GPa are from Dunn and Ratliffe (1990); and wet HA1 are from Koepke and
Behrens and
(2001).
0.6-2 GPa are from Dunn and Ratliffe (1990); and wet HA1 are from Koepke and Behrens (2001).
difference to a sudden jump from 0.0001 to 0.6 GPa and then no additional pressure effect
above 0.6 GPa, a somewhat ad hoc explanation. More likely, the sudden jump is due to
experimental difficulties.
The oxygen fugacity in the experiments of Koepke and Behrens (2001) using hydrous
haploandesite (4.9-5.2 wt% H2O) was estimated to be about NNO + 3, at which the ferric to
ferrous ratio is roughly 1. Hence, these data are roughly the average Fe2 + and Fe3 + diffusivities
388 Zhang, Ni, Chen
and can be expressed as:

 (24421 ± 10053) 
wet HA1; 0.5 MPa; NNO+3; 4.9-5.2%
DFe TD = exp ( -7.83 ± 6.84) -  (87)
 T 
where % in the superscript indicates wt% of H2O. The maximum error in reproducing the three
experimental data by Equation (87) is 0.28 lnD units.
Ruessel and Wiedenroth (2004) produced extensive Fe tracer diffusion data in Na2O-
CaO-MgO-Al2O3-SiO2 melts at 1300-1873 K. The experiments were implicitly conducted in
air, though it was not clearly specified. Hence, the diffusivities are close to Fe3 + diffusivities.
Ruessel and Wiedenroth (2004) reported activation energy and pre-exponential factors for
melts with various compositions, but only reported diffusivity values at 1573 K. The activation
energy varies from 209 to 284 kJ/mol as the melt composition varies. Due to errors in extracted
activation energies and pre-exponential factors, it is easier to treat original diffusion data than
to model how activation energies and pre-exponential factors depend on melt composition.
The available data of Fe diffusion, though numerous, are not systematic enough to un-
derstand both Fe2 + and Fe3 + diffusion. In order to better understand Fe diffusion, it is neces-
sary to conduct Fe diffusion studies under a range of oxygen fugacities so that diffusivities
of both Fe2 + and Fe3 + and their dependence on temperature, pressure and composition can be
extracted. Then Fe diffusivity at any given fO2 (or given ferric/ferrous ratio) can be inferred
using Equation (85).
Co diffusion. Three papers reported Co diffusion data in silicate melts (29 points).
Hofmann and Magaritz (1977) studied Co tracer diffusion in dry basalt10 melt (Table 1) at
1553-1713 K and 0.1 MPa. Lowry et al. (1982) investigated Co tracer diffusion in dry basalt1
and andesite2 melts at 1567-1672 K and 0.1 MPa. Henderson et al. (1985) studied Co tracer
diffusion in dry dacite4 and pantellerite1 melts at 1473-1672 K and 0.1 MPa. No EBD data of
Co are available.
Co tracer diffusion data in two different dry basalts (basalt1 by Lowry et al. 1981 and
basalt10 by Hofmann and Magaritz 1977) are in good agreement (Fig. 64) and can be fit as
follows:
 (20211 ± 3706) 
dry basalts; 0.1 MPa
DCo TD =exp  -(11.05 ± 2.29) -  (88)
 T 
The maximum error in reproducing the experimental data by Equation (88) is 0.16 lnD units.
The activation energy is 168±31 kJ/mol.
Co tracer diffusivity increases from pantellerite1 to dacite4 to andesite2 to basalt1 melts;
the increase from pantellerite to dacite is a little strange because increasing alkalinity is usually
thought to cause an increase in diffusivity. All diffusion data are shown in Figure 64 and can be
roughly fit by the following equation:
 (4352 + 25116 X1 ) 
dry melts; 0.1 MPa
DCo TD =exp  -11.02 - 42.07 X 2 -  (89)
 T 
where X1 is the sum of cation mole fractions of Si + Ti + Al + P, and X2 = max(Na + K-Al,0)
(i.e., X2 = Na + K-Al if Na + K-Al>0, and X1 = 0 if Na + K-Al <0, this is because peralkalinity
seems to affect diffusivity more than peraluminity). The maximum error in reproducing the
experimental data by Equation (89) is 0.48 lnD units (Fig. 64). Mn and Co tracer diffusivities
are similar.
Ni diffusion. Two papers reported Ni diffusion data in silicate melts (18 points). Koepke
and Behrens (2001) investigated Ni tracer diffusion in a wet haploandesite containing 4.5-5.2
figure
Fig. 64. 64.diffusivities
Co tracer Co tracerin diffusivities
dry melts at 0.1 in
MPadry melts
(Lowry at 1982;
et al. 0.1 MPa (Lowry
Henderson et al. et al. 1982;
1982), and fit of all the
Henderson et al. 1982), and fit of all the data by Equation (89).
data by Eq. 64b.
wt% H2O at 1373-1673 K and 0.5 GPa. Behrens and Hahn (2009) studied Ni tracer diffusion
in dry and wet trachyte and phonolite melts at 1323-1528 K, 0.5 GPa and ≤ 1.9 wt% H2O.
The data are shown in Figure 65. Ni diffusivities in wet trachyte (1.1 wt% H2O) and wet
phonolite (1.7-1.9 wt% H2O) are greater than in the respective dry melts by a factor of about
20, indicating strong dependence on H2O.
Cu diffusion. No Cu diffusion data in silicate melts are known.
Zn diffusion. Three papers reported Zn diffusion data in silicate melts (25 points). Baker
and Watson (1988) investigated Zn effective binary diffusion (SEBD) in rhyolite1-rhyolite7
and HD2-rhyolite7 diffusion couples (Table 1) at 1171-1673 K and 0.01-1 GPa. Koepke and
Behrens (2001) reported Zn tracer diffusivities in dry and wet HA1 melt at 1373-1673 K, 0.5
GPa, 0.06 and ~5 wt% H2O. Behrens and Hahn (2009) examined Zn tracer diffusion in dry
and wet trachyte1 and phonolite1 melts at 1323-1528 K, 0.5 GPa, and 0-1.9 wt% H2O. The
data are summarized in Figure 66. Some of the data (such as diffusivity in dry trachyte and dry
phonolite) apparently have large errors.
64
tc, ru, rh, Pd, ag, Cd
No diffusion data on Tc, Ru, Rh, Pd, and Ag in silicate melts are known. Cd diffusion has
been investigated only in one paper (MacKenzie and Canil 2008), reporting 8 tracer diffusivity
values of Cd in a dry CMAS1 melt (Table 1) at 1523-1623 K and 0.1 MPa. The data are shown
in Figure 67.
re, os, Ir, Pt, au, Hg
In this group of elements, no diffusion data on Os, Ir, Pt, Au, and Hg in silicate melts are
known, and only Re diffusion data are available. Two papers reported Re diffusion data in
silicate melts (42 points). MacKenzie and Danil (2006, 2008) investigated Re tracer diffusion
in a number of dry melts through devolatization (desorption) experiments. Re diffusivities
appear to be similar to those of Mn.
The concentration profiles in some of the Re devolatization/diffusion experiments show
390 Zhang, Ni, Chen
figure 65.
Fig.Ni
65.tracer
Ni(Ch diffusivities
tracer diffusivities inin
wetwet haploandesite
haploandesite (HA) melt (Koepke and Behrens 2001), dry and
Zhang et al. 8) Diffusion data in silicate melts (HA) melt (Koepke and Behrens 2001), dry and wet trachyte
wet trachyte (tra) and phonolite (pho) melts (Behrens and Hahn 2009), all at 0.5 GPa.
(tra) and phonolite (pho) melts (Behrens and Hahn 2009), all at 0.5 GPa.
65
figure 66.Fig.
Zn66.diffusivities in silicate
Zn diffusivities in silicatemelts.
melts. The default
The default conditions
conditions are
are dry dryand
melts melts and 0.5
0.5 GPa. DataGPa. Data sources:
sources:
SEBD in r-r (rhyolite1-rhyolite7) couple (Baker and Watson 1988); tracer diffusivities in dry and wet HA1
SEBD in r-r (rhyolite1-rhyolite7) couple (Baker and Watson 1988); tracer diffusivities in dry and wet HA1 (Koepke
(Koepke and Behrens 2001); dry and wet trachyte1 and phonolite1 (Behrens and Hahn 2009).
and Behrens 2001); dry and wet trachyte1 and phonolite1 (Behrens and Hahn 2009).
peculiar features where the whole profile seems to move into the melt with an increasingly
thick surface layer void of Re. Currently there is no satisfactory explanation for such a behavior.
Because the behavior is not well understood, it is not clear whether the extracted diffusivities
from the profiles are reliable. Furthermore, there is much data scatter. New experimental
studies are necessary to understand Re diffusion.
Fig.figure 67. diffusivities

67. Cd tracer Cd tracerindiffusivities in CMAS1
CMAS1 melt (MacKenzie andmelt
Canil(MacKenzie
2008). and Canil 2008).
ac, th, Pa, u

Ac diffusion. No Ac diffusion data in silicate melts are known.
Th diffusion. Four papers reported Th diffusion data in silicate melts (24 points).
LaTourrette et al. (1996) and LaTourrette and Wasserburg (1997) investigated Th tracer
diffusion in HB1 melt (Table 1) at 1673-1773 K, 0.1 MPa, and two fO2 conditions (air and Fe-
FeO buffer). Mungall and Dingwell (1997) carried out an extensive study of Th tracer diffusion
in HR7 melts: dry melt at 1473-1873 K and in air, as well as 1673 K and high pressures of
1-3.5 GPa, and wet melt containing 1.7 to 5 wt% H2O at 1273-1673 K and 1 GPa. Nakamura
and Kushiro (1998) examined Th tracer diffusion in jadeite melt at 1673 K and 1.25-2 GPa.
The diffusion data are summarized in Figure 68. Varying oxygen fugacity from air to Fe-FeO
buffer does not affect Th diffusivities (within 5% in D or 0.05 lnD units) (LaTourrette et al.
1996; LaTourrette and Wasserburg 1997).
Th tracer diffusivities in air and in Fe-FeO buffer are the same and can be described by
(LaTourrette and Wasserburg 1997):
67
 (26554 ± 4037) 
TD = exp ( -9.30 ± 2.37) -
HB1
DTh  (90)
 T 
The lnDTh values in dry and wet HR7 are not linear to H2O content (Mungall and Dingwell
1997), but roughly linear to square root of H2O content. Hence, we fit Th tracer diffusion data
in dry and wet HR7 melt at 1273 to 1873 K, ≤ 3.5 GPa, and ≤ 5 wt% H2O by Mungall and
Dingwell (1997) as follows:
 44682 - 1370 P - 7281 w 
TD = exp  -7.02 - 1.303 w -
HR7
DTh  (91)
 T 
392 Zhang, Ni, Chen
Fig. 68.figure 68. Thin diffusivities

Th diffusivities jadeite, dry HR7,inwet
jadeite, dry5 HR7,
HR7 (with wt% H2wet HR7
O), and HB1(with
melts. 5 Data
wt%sources
H2O),can be
and HB1 melts. Data sources can be found in text.
found in text.
Self and tracer diffusivity values of Th are similar to those of U in both HB1 and jadeite
melts (within 0.25 lnD units). In dry and wet HR7 melts, the difference between Th and U
diffusivities are within 0.56 lnD units except for an outlier with difference of 1.3 lnD units.
Pa diffusion. No Pa diffusion data in silicate melts are known.
U diffusion. Four papers reported U diffusion data in silicate melts (25 points). U may
be present as U4 + and/or U6 + in silicate melts. Hence, the diffusivity is expected to depend on
oxygen fugacity. LaTourrette et al. (1996) and LaTourrette and Wasserburg (1997) investigated
U self diffusion in HB1 melt (Table 1) at 1623-1773 K, 0.1 MPa, and two fO2 conditions
(air and Fe-FeO buffer). They found that U self diffusivities under reduced condition of the
Fe-FeO buffer is about 20% greater than those in air (oxidizing condition), resolvable only
with the best data. Mungall and Dingwell (1997) carried out an extensive study of U tracer
diffusion in HR7 melt (Table 1): dry melt at 1473-1873 K and in air, as well as 1673 K and
high pressures of 1-3.5 GPa, and wet melt containing 1.7 to 5 wt% H2O at 1273-1673 K and 1
GPa. Nakamura and Kushiro (1998) examined U tracer diffusion in jadeite melt at 1673 K and
1.25-2 GPa. The data are summarized in Figure6869.
The lnDU values in HR7 are not linear to H2O content (Mungall and Dingwell 1997), but
roughly linear to square root of H2O content. Hence, U tracer diffusivities in dry and wet HR7
melt at 1273 to 1873 K, ≤ 3.5 GPa, and ≤ 5 wt% H2O can be expressed as follows:
 44729 - 1093P - 8944 w 
FEBD = exp  -6.37 - 2.647 w -
DUHPG8  (92)
 T 
Th and U diffusivities are among the lowest, similar to Zr, Hf and Si, and lower than
diffusivities of most other elements (e.g., LaTourrette and Wasserburg showed that Th and U
diffusivities are lower than Si and O diffusivities by about 1.2 lnD units).
figure 69.Fig.
U 69.
diffusivities ininjadeite,
U diffusivities HB1,
jadeite, HB1, dry dry andHR7
and wet wetmelts.
HR7Datamelts. Data
sources: sources:
jadeite jadeite
(Nakamura (Nakamura and
and Kushiro
Kushiro 1998); dry and wet HR7 (Mungall et al. 1999); HB1 (LaTourrette et al. 1996; LaTourrette and
1998); dry and wet HR7 (Mungall et al. 1999); HB1 (LaTourrette et al. 1996; LaTourrette and Wasserburg 1997).
Wasserburg 1997).
DISCuSSIoN
the empirical model by Mungall (2002)
Mungall (2002) reviewed earlier models for predicting diffusivities (e.g., Baker 1992b)
and presented an empirical model to predict cation diffusivity in silicate melts. He divided
cations into three groups: (i) alkalis, (ii) intermediate field strength elements (IFSE), defined
as those with ionic strength z2/r between 1 and 10 where z is the valence of the cation and r
is the radius of the cation in Å; and (iii) high field strength elements (HFSE), defined as those
with z2/r > 10. For the alkalis, Mungall (2002) fit the diffusivity empirically as a function of
ionic radius, and two compositional parameters (in addition to the temperature dependence):
42.6(ri - 1.03)2 + 6.63Y1 - [1239Y2 + 27424(ri - 1.03)2 + 6975]
lnDi =
-16.16 + (93)
T
where ri is in Å, Y1 = M/O = [FeO + MnO +69MgO + CaO + 2(Na2O + K2O + H2O)]/
[2(SiO2 + TiO2) + 3Al2O3 + FeO + MnO + MgO + CaO + Na2O + K2O + H2O + 5P2O5],
and Y2 = Al/(Na + K + H), where Al, Na, K, and H are cation mole fractions, and SiO2, TiO2,
Al2O3, FeO, MnO, MgO, CaO, Na2O, K2O, H2O and P2O5 are oxide mole fractions. Note that
Equation (93) is modified from that in Mungall (2002) in two aspects (i) Mungall (2002) used
cm2/s as the diffusivity unit, but we use m2/s; and (ii) Mungall used the base-10 logarithm
and we use natural logarithm for consistency with other parts of this chapter. The use of Al/
(Na + K + H) as a parameter means that the melt must contain significant Na, K and H.
For IFSE, the diffusivity is empirically related to viscosity η (in Pa·s), temperature, zi2/r,
and Y1(M/O) by Mungall (2002):
 zi 2 
 - 12.42  ( 3730Y1 - 2102 )
 Di η  0.564 zi 2  r 
ln  =-37.8 + + (94)
 T  r T
394 Zhang, Ni, Chen
For HFSE, the diffusivity is empirically related to viscosity, temperature and M/O by
Mungall (2002):
 Dη
ln  i  =
-29.2 - 6.61Y1 (95)
 T 
That is, all HFSE have the same diffusivities according to this model.
The model of Mungall (2002) has been tested by Behrens and Hahn (2009) using directly
measured diffusivity and viscosity values, and the model-predicted diffusivities are 0 to 1.3
logD (0 to 3 lnD) units greater than the experimental values. To further test the model of
Mungall et al. (2002), we use our compiled self and tracer diffusivities and FEBD values to
compare with calculated diffusivities using the model of Mungall (2002), focusing on silicic
melts. The viscosity is calculated from the general model of Hui and Zhang (2007), which
has been shown to work well for silicic melts (Wang et al. 2009). We avoid the multi-valent
elements so that the field strength can be calculated accurately. To be consistent with Mungall
(2002), we used the same radii sent to us by J. Mungall (personal communication), which are
crystal radii of Shannon 1976 with coordination number of 6 and high spin, except for the
following: coordination number is 8 for Sr2 + , Ba2 + , Nd3 + , and Tb3 + , 4 for B3 + , Si4 + , and Ga3 + ,
and 3 for Be2 + ; and low spin for Mn2 + and Co2 + .
Figure 70 compares calculated diffusivities with experimental diffusivity values for various
silicic melts. It can be seen that the scatter is quite large, as much as 2.8 logD units (6.4 lnD
units). The following diffusivity data are off by more than 1.3 log units (a factor of 20): Sr and
Ba tracer diffusion data in rhyolite15 melt by Magaritz and Hofmann (1978) and Ce diffusion
data in rhyolite15 by Jambon (1982) (partially this is because we adopted the corrected H2O
concentration for this rhyolite melt from 0.38 wt% to 0.11 wt% by Jambon et al. 1992); B
diffusion data in HR7 melt by Chakraborty et al. (1993), and U, Th and Be diffusion data in
HR7 melt by Mungall et al. (1997, 1999). For viscosity of rhyolie15 and HR7, we used the
specific viscosity model for rhyolites (Zhang et al. 2003; Hui et al. 2009) and confirmed that
the large difference between calculated and experimental diffusivities are not due to the choice

figure 70. Comparison of calculated diffusivity based on the model of
Fig. 70. Mungall
Comparison (2002) and
of calculated experimental
diffusivity based on thediffusivity (horizontal
model of Mungall (2002) andaxis).
experimental diffusivity
(horizontal axis).
of the viscosity model of Hui and Zhang (2007). Although some of these discrepancies can
be explained away (see Mungall 2002) and some are because Mungall (2002) overlooked a
corrected H2O concentration (more than 10 years after the original publications), we concur
with Behrens and Hahn (2009) that the empirical model of Mungall (2002) can be used as
an order of magnitude estimate for tracer diffusivities, but not accurate enough for practical
applications.
effect of ionic size on diffusivities of isovalent ions
For alkali elements or univalent cations (Fig. 8), the diffusivities increase from Cs to Rb
to K and to Na in rhyolite and albite melts. That is, the diffusivity increases as the ionic size
decreases. But the trend is not perfect: from Na to Li, the diffusivity may increase or decrease
or stay the same. For alkali earth elements or divalent cations (Fig. 15), the diffusivities increase
from Be to Mg to Ca and to Sr in HR7 and HR7 + Na melts. That is, the diffusivity decreases
as the ionic size decreases in these melts. Again, the trend is not perfect: from Sr to Ba, the
diffusivities may increase or decrease or stay the same.
For trivalent REE species, the diffusivities increase smoothly from La to Lu in melts
(Fig. 51). The smoothness makes the REE data excellent references when comparing with
diffusivities of other elements. In Figure 71, we examine whether diffusivities continue to
decrease to smaller trivalent cations, including B3 + (the smallest trivalent cation) and Cr3 + .
Figure 71 shows that the trend does not continue: at some ionic radius smaller than that of Lu3 + ,
trivalent cation diffusivities do not decrease any more with decreasing ionic radius.
In Figure 72, we compare tetravalent cation diffusivities to examine the size effect for this
group in HR7 melt. The ionic radii of Si4 + , Ge4 + , and Ti4 + in tetrahedral sites increase from
0.026, to 0.039 to 0.042 nm (Shannon 1976). Those of Ti4 + , Hf4 + , Zr4 + , U4 + , and Th4 + in octa-
hedral sites increase from 0.0605, 0.071, 0.072, 0.089, and 0.094 nm (Shannon 1976). Because
Si diffusivity is not available in HR7 melt, the Eyring diffusivity is shown in Figure 72 as the
proxy for Si diffusivity. The differences among the diffusivities of these tetravalent cations are
small, but there is minor increase from diffusivity of Si (Eyring) to Ge≈Th to Hf (except for an
outlier with low diffusivity) to Ti≈Zr≈U.
Based on the experimental data discussed above, for alkali elements and halogen elements
(Fig. 35), the diffusivities of each isovalent series decrease as the size of the ion increases,
and the activation energy increases as the size increases. In a similar fashion, diffusivities of
noble gas elements and other neutral molecules also decrease as the size of the neutral atom or
molecule increases (Zhang and Xu 1995; Behrens 2010, this volume). On the other hand, for
divalent, trivalent and tetravalent cations, the trend is mostly opposite: diffusivities of isovalent
ions increase as the size of the ion increases. The observations might be puzzling or even
counter-intuitive to some and appear to be unexplained previously. We attribute these opposite
trends to the interplay of bond strength and ionic size as follows.
Diffusion requires a species to detach from the original site (breaking the bonds between
the species and surrounding particles), and then move (or jump) from the old site to a new site,
meaning to squeeze through an orifice formed by neighboring ions to arrive at a new site. That
is, detachment and going through some orifice are two necessary and sequential steps. Hence,
the slowest step determines the overall diffusion rate. The slowest step depends on both the
size and ionic valences as follows:
(1) If the particle is bonded weakly to other particles, e.g., for neutral atoms and molecules
(0-valence “ions”), or for univalent cations and anions, the first step (detachment) is
easy and the second step (going through an orifice) is rate limiting. Hence, a smaller
particle diffuses more rapidly because it is easier to get through an orifice.
(2) If the particle is bonded strongly to other particles, meaning higher valence ions
396 Zhang, Ni, Chen
figure 71. Trivalent cation diffusivities. Data sources: HR7 (Mungall et al. 1999); HA1 (Koepke and
Fig. 71. Trivalent cation diffusivities. Data sources: HR7 (Mungall et al. 1999); HA1 (Koepke and Behrens 2001).
Behrens 2001). The Eyring diffusivity is calculated from viscosity data of Hess et al. (1995) on HR7 melt.
The Eyring diffusivity is calculated from viscosity data of Hess et al. (1995) on HR7 melt.
(divalent and higher valence), the first step (detaching from other particles) is rate
determining. Hence, ions with weaker bonds, meaning low charge to radius ratio, i.e.,
larger ions in an isovalent series, detach more easily and diffuse more rapidly.
The above explanation can be pushed further to explain why Li and Na diffusivities are
similar rather than following the trend of increasing diffusivity with decreasing size from Cs
to Rb to K to Na. Li + is small so that the bond strength is fairly high, meaning that from Na +
to Li + , even though the size becomes smaller leading to high diffusivity, the bond strength
71 similar diffusivity between Li +
increases leading to lower diffusivity, with the net effect being
+
and Na . Similarly, the fact that Ba diffusivity sometimes does not continue the trend of Be
to Mg to Ca to Sr for which the diffusivity increases as the bond strength decreases (or size
figure
Fig.72.
72. Tetravalent cation
Tetravalent cation diffusivities.
diffusivities. Diffusivity
Diffusivity data
data are from are from
Mungall Mungall
et al. (1999). TheetEyring
al. (1999). The
diffusivity is
Eyring diffusivity is calculated from actual viscosity data of Hess et al. (1995) on HR7 melt.
calculated from actual viscosity data of Hess et al. (1995) on HR7 melt.
increases) can also be explained: Ba is so large that the bond strength is weak so that the size
effect is also significant, leading to Ba diffusivity similar to or even smaller than Sr diffusivity.
Exactly when the bond strength effect or the size effect would be more important is
expected to depend on the melt composition and structure. In more densely packed melts
(mafic or depolymerized melts), there is less free space (low ionic porosity) for particles to
move around. Hence, the size effect is expected to be more important than the bond strength
effect, so that larger ions would diffuse more slowly. In less densely packed melts (felsic or
polymerized melts), more free space is available so that the bond strength effect is expected to
be more important, meaning that smaller ions with greater bond strength tend to diffuse more
slowly. Hence, the exact ionic size for the trend of diffusivity versus ionic radius to reverse
in an isovalent series depends on the melt composition, as shown below in the discussion of
diffusivity sequence.
Another complexity is that the diffusivities of different species depend on temperature,
and above a certain temperature (the compensation temperature), the trend would be reversed.
Hence, the above explanation may be better applied 72 to the activation energy, which is expected
to increase with the size of the diffusing species for neutral and univalent ions, but to decrease
with size (or increase with bond strength) for ions with valence of two or more. Data in Figure
8 support this hypothesis, but the activation energy does not vary much from Mg to Ca to Sr
in Figure 15. It is possible that greater temperature range is necessary to resolve the difference
in the activation energy.
Dependence of diffusivities on melt composition
For noble gas and alkali elements, when the particle size is small, such as He (Jambon
and Shelby 1980), Ne (Frank et al. 1961; Perkins and Beagal 1971; Matsuda et al. 1989),
Li (Fig. 1), and Na (Fig. 3), the diffusivity (tracer diffusivities and FEBD) increases from
depolymerized melts (which are densely packed) to polymerized melts (which are less densely
packed), meaning available free space is more important for the diffusion of these elements.
The compositional dependence of He, Ne, Li and Na diffusivities means that they increase as
melt viscosity increases, opposite to the famous diffusivity-viscosity relations that diffusivity
is inversely related to viscosity (e.g., the Einstein relation and Eyring relation). As the particle
398 Zhang, Ni, Chen
size increases, tracer diffusivity values of K and Rb (Figs. 4 and 6) do not vary strongly
with melt composition, and Ar tracer diffusivities (Zhang et al. 2007) increase slightly from
polymerized to depolymerized melts. For other elements, especially the divalent and higher
valence ions, the diffusivity increases from polymerized to depolymerized melts, suggesting
that flexibility (or rigidity) of the structure plays a greater role in controlling diffusion of these
elements. The dependence on the melt composition and structure can again be explained by
the two necessary steps for diffusion (detachment and squeezing through an orifice).
For small and weakly bonded species (such as He, Ne, Li and Na), detaching from the
original site is easy. As long as the structure provides large holes (i.e., high ionic porosity),
these species can move around easily. However, when the structure is dense (i.e., low ionic
porosity), then these species cannot move so rapidly. Hence, for these species, the diffusivity
increases as the ionic porosity increases. For silicate melts, ionic porosity increases from
depolymerized (also low-viscosity) melts to polymerized (also high-viscosity) melts, leading
to the observed positive dependence of diffusivity of these species on melt viscosity.
On the other hand, for large and weakly bonded species (such as Ar, Kr, Xe, and Cs), the
size of the orifices in all melts, even for the melt with high ionic porosity, are not large enough.
Hence, the species must push particles away to expand an orifice significantly so that it can
move through the orifice. If the melt structure is more rigid (polymerized melt), it is more
difficult to push aside particles, meaning diffusion rate would decrease from depolymerized
melt to polymerized melt.
For higher valence ions, the bonds between the diffusing species and other particles are
strong. It is more difficult to break the bonds if the melt structure is more rigid (polymerized
melt). Hence, diffusivity also decreases from depolymerized melt to polymerized melt.
Experimental data show that adding water always increases the diffusivity of an element
without exception. The magnitude of increase is greater for elements with low diffusivities.
Dissolved H2O in a melt results in at least two effects to increase diffusivity: one is to increase
the ionic porosity so that the melt is less densely packed (so that diffusivity would increase
even for small noble gases and alkalis), and the other is to depolymerize the melt so that
the viscosity is decreased and the structure becomes less rigid. Because both effects would
increase the diffusivity, adding water always increases the diffusivity.
Diffusivity sequence in various melts
For a given melt, we rank the elements by their diffusivity and define the ranking as a
diffusivity sequence (by analogy to the incompatible element sequence in presenting data
in “spidergrams”). Such a sequence is useful in rough estimation of diffusion behavior in
a silicate melt. The concept of diffusivity sequence in a given melt is best applied to self
diffusivity, tracer diffusivity and FEBD because they are more closely related to the intrinsic
mobility, whereas binary interdiffusivity and SEBD depend on the concentration gradient of
other components (e.g., binary interdiffusivity depends on which component is the exchanging
component). The sequence may not be universal because diffusing species may change as melt
composition or oxygen fugacity or other conditions change. Furthermore, the sequence is for
diffusion below the compensation temperature. If there were a single perfect compensation
temperature, the diffusivity sequence would become just the opposite above the compensation
temperature. Because experimental data often show that there is no perfect compensation
temperature, at high temperatures such as above 1700 K, the diffusivity sequence could be
irregular. The following sequences do not include multi-valent elements such as Fe and Eu
because their diffusivities depend on the oxygen fugacity, which would change their position
in the diffusivity sequence.
Diffusivity sequence in rhyolite and related melts. Self and tracer diffusivities in dry
rhyolite melt and HR7 (haplorhyolite) melt are shown in Figure 73.
figure 73. Diffusivities in dry HR7 and rhyolite melts. Data for HR7 melts are from Mungall and Dingwell
(1997) andFig.
Mungall et al. (1999). in
73. Diffusivities Data
dryforHR7
rhyolite
and melts are melts.
rhyolite from Jambon
Data and
for Semet (1978),
HR7 melts areJambon (1982), and Dingwell (1997
from Mungall
Harrison and Watson (1983, 1984), Blank (1993), Lesher et al. (1996), and Bai and Koster van Groos (1994).
and Mungall et al. (1999). Data for rhyolite melts are from Jambon and Semet (1978), Jambon (1982), Harriso
In dryand
rhyolite5
Watson and albite
(1983, melts,
1984), Blankdiffusivities ofetalkali
(1993), Lesher elements
al. (1996), canandbeKoster
and Bai ranked asGroos
Van follows
(1994).
(Fig. 8):
Cs < Rb < K < Na ≈ Li
In dry HR7 melt (a haplorhyolite melt) (Mungall et al. 1999), the diffusivities of alkali earth
elements can be ranked as follows (Fig. 15a):
73
Be < Mg < Ca < Sr ≈ Ba
400 Zhang, Ni, Chen
Synthesizing the data in Mungall et al. (1999) and Figure 73 on HR7 and rhyolite5 melts as
well as other information, we obtain (Fig. 73):
P ≈ Si ≈ Th ≈ Ge ≈ Ti ≈ Hf ≈ U ≈ Zr < REE ≈ Y ≈ Nb ≈ Ta ≈ B ≈ Be <
Mg < Ca < Sr ≈ Cs ≈ Ba < F ≈ Sb ≈ Rb ≈ Cl ≈ CO2 ≈ Ar < K < Na ≈ Li
where REE means trivalent rare earth elements. Eu2 + probably diffuses at a rate similar to
Sr, and Ce4 + probably diffuses at a rate similar to U4 + . U in the above sequence could be
either U4 + or U6 + because the diffusivity difference between the two is small (LaTourrette
and Wasserburg 1997). The above sequence is expected to be applicable to dry silicic (highly
polymerized) melts. The location of the HFSE Nb and Ta in the above sequence (Fig. 73a)
shows that the group of HFSE does not behave the same (Nb and Ta have higher diffusivities
than Hf, Ge, Th and U, even though Nb and Ta also have higher field strength), contrary to the
suggestion of Mungall (2002).
For hydrous rhyolite melts, the main data for multiple elements are on hydrous HR7 with
3.6 wt% H2O by Mungall et al. (1999). By adding H2O, the diffusivity of all components
increases, but those with smaller diffusivities increase more, leading to smaller difference in
diffusivities of different elements as H2O content is increased. The limited data combined with
the alkali element sequence yield the following diffusivity sequence:
Hf < Ti < Zr ≈ Nb ≈ Ta ≈ B < REE ≈ Y ≈ Be < Mg
< Ba < Cs ≈ Ca < Sr < Rb < K < Na
Comparing the diffusivity sequences of alkalis in dry and wet HR7, the largest difference
is that Ba diffusivity does not increase as much as the other alkalis as H2O concentration
increases, so that its diffusivity in wet HR7 is between that of Mg and Ca, rather than similar
to Sr. The positions of Hf, Ti, Nb, Ta, and B also are changed slightly.
For HR7 melt plus 20 wt% Na2O, the estimated diffusivity sequence based on the data of
Mungall et al. (1999) and other information is:
Hf < Zr < Ta ≈ Nb ≈ REE ≈ Y < Ti ≈ W ≈ Be < B < Mg
< Ca < Sr ≈ Cs ≈ Ba < Rb < K < Na
Diffusivity sequence in jadeite melt. In jadeite melt at 0.1 MPa (Dingwell and Scarfe
1985; Roselieb and Jambon 1997, 2002), the diffusivity sequence is:
Cs < Mg < Ba ≈ F < Rb ≈ Sr ≈ Ca < K
which differs from the sequence in HR7 in the divalent cations: in HR7, diffusivity increases
from Mg to Ca to Sr ≈ Ba, whereas in jadeite, diffusivity increases from Mg to Ca and then
decreases from Ca to Sr to Ba.
Combining the above sequence with data of Nakamura and Kushiro (1998), the sequence
in jadeite melt is:
Zr ≈ Th < Nb ≈ U < REE ≈ Y < Cs < Mg < Ba ≈ F < Rb ≈ Sr ≈ Ca < K
Diffusivity sequence in trachyte and phonolite melts. For dry trachyte melt, the diffusivity
sequence based on the data of Behrens and Hahn (2009) is estimated as:
Zr < Hf ≈ Nb ≈ Ta < REE ≈ Y ≈ Sn < Ni ≈ Zn < Ba < Sr < Rb < K < Na
For wet trachyte melt containing about 1.5 wt% H2O, the diffusivity sequence based on the
data of Behrens and Hahn (2009) is estimated to be:
Zr ≈ Hf < Nb ≈ Ta < REE ≈ Y ≈ Cr < Sn < Zn < Ni < Ba < Sr < Rb < K < Na
In dry phonolite melt, the diffusivity sequence is estimated to be (Behrens and Hahn 2009):
Hf < Zr < Nb < REE ≈ Y < Cr < Ni ≈ Zn ≈ Sn < Ba < Sr < Rb < K < Na
In wet phonolite melt containing about 1.7 wt% H2O, the diffusivity sequence is roughly
(Behrens and Hahn 2009):
Hf ≈ Zr < Nb ≈ Ta < REE ≈ Y < Cr ≈ Sn < Ni ≈ Ba ≈ Zn ≈ Sr
Diffusivity sequence in dacite melts. In dry dacite melt, data by Cunningham et al.
(1983), Henderson et al. (1985), Baker (1992a), and Tinker and Lesher (2001) provide the
following constraints:
Si ≈ O < Ga ≈ B < Cs < Co ≈ Mn ≈ Ba ≤ Na < Li
Diffusivity sequence in andesite melts. In dry andesite2 and haploandesite, the following
diffusivity sequence can be obtained based on the data of Lowry et al. (1982) and Koepke and
Behrens (2001):
Zr < Nb < REE ≈ Y ≈ Sc < Cs < Ba ≈ Zn < Rb ≈ Cr ≈ Co ≈ Sr ≈ Mn ≈ Sb < Na
Diffusivity sequence in basalt melts. Self and tracer diffusivity data in dry basalt and
haplobasalt melts are shown in Figure 74. The diffusivity sequence can be roughly determined
as follows:
U ≈ Th ≈ Zr < Si ≈ O ≈ Ti ≈ Te < REE ≈ Y ≈ Sc ≈ Pb ≈ Cs ≈ Tl < Ba <
CO2 ≈ Brim < Sr ≈ Ca ≈ Co ≈ Mn ≈ Cl < Cd ≈ F < Na < Li
Summary on diffusivity sequences. In summary, there is overall consistency in the
diffusivity sequences, with tetravalent cations and P having the smallest diffusivity, then the
pentavalent Nb and Ta, then REE and other trivalent ions, and then the divalent and univalent
ions having the largest diffusivity. In detail, there are differences among various melts. For
example, in dry silicic melts, diffusivity increases from Be to Mg to Ca to Sr ≈ Ba. However,
in hydrous silicic melts, diffusivity increases from Be to Mg to Ca ≈ Ba to Sr. In basalt melts,
diffusivity increases from Ba to Sr to Ca to Mg. Undoubtedly, the explanation lies in the melt
structure (including ionic porosity), and the interplay between bond strength and ionic size
as discussed earlier, but it would be very helpful if a quantitative theory can be developed to
accurately predict the various sequences.
For univalent and divalent ions, the diffusivity depends more strongly on ionic radius or
bond strength (related to ionic radius). For ions with valence of 3 or higher, the diffusivity
does not depend strongly on ionic radius. For example, diffusivities of REE, Sc, Cr, and
B are often within a factor of 2 even though the ionic radius and hence the bond strength
change significantly. Similarly, diffusivities of Si, Ge, Ti, Zr, Hf, U, and Th are also not hugely
different. One possible explanation is that for the highly charged cations, the diffusing species
are different from the simple cations, and the size of the diffusing species does not change
significantly from one ion to another.
In dry silicic melts at relatively low temperatures such as 1200 K, the diffusivity difference
between rapidly and slowly diffusing elements is large, about 8 orders of magnitude from Si to
Li and Na. As the melt becomes more mafic, as the temperature increases, and as more H2O is
added to the melt, the diffusivity difference becomes smaller, only a few orders of magnitude.
For example, Figure 74b shows that the difference between Li and U diffusivities in dry basalt
melt at 1640 K is only 2.4 orders of magnitude.
Even though the diffusivity sequence is not universal but depends on melt composition
and other conditions, a rough sequence for a given type of melt (such as haplorhyolite melt)
at some temperature can be useful for at least two purposes: (i) the sequence gives a rough
idea of which element diffuses more rapidly, which is useful in forming a semi-quantitative
understanding of diffusion rate in natural melts, and in predicting diffusive elemental
fractionation; and (ii) the change in the diffusivity sequence may be used to probe the melt
structure.
402 Zhang, Ni, Chen
figure 74. Self and tracer diffusivities of divalent cations in dry basalt and haplobasalt melts.
Fig. 74. Self and tracer diffusivities of divalent cations in dry basalt and haplobasalt melts.
CoNCluDING reMarKS
An impressive amount of diffusion data on natural melts and analogs of natural melts has
been accumulated in the last 30 years. Among the naturally occurring elements, no diffusion
data are available for In, N, As, Bi, Se, I, V, Mo, Cu, Ru, Rh, Pd, Ag, Os, Ir, Pt, Au and Hg
in natural silicate melts. For some other elements, such as Tl, Te, Ta, W, Zn, Cd, and Re,
few studies have been made. On the other hand, for major elements such as O, H, Si, Al, Fe,
Mg, and Ca, numerous data are available, and the data can be 74used in quantitative modelings
(such as bubble growth, crystal dissolution, etc). However, even for these elements, it is still
impossible to accurately predict their diffusivities as a function of temperature, pressure
and melt composition (including concentration of H2O) due to limited data coverage and/or
discrepancies among the data. Much experimental work is still needed in the future.
It is more important to carry out well-planned systematic diffusion studies rather than
numerous random studies (often with other goals, and extraction of diffusivities being only side
products). For example, a dozen data points through well-designed experiments by Tinker and
Lesher (2001) are sufficient to pin down the temperature and pressure dependence of Si and O
diffusivity at ≤ 4 GPa in a haplodacite melt. Behrens and Zhang (2009) and Wang et al (2009) also
showed that with coordination, only a small number of experiments are necessary to quantify the
complicated behavior of H2O diffusion as a function of temperature, pressure and H2O content.
In general, the compositional effect on diffusivity is much more difficult to constrain
compared to temperature and pressure effects because (i) compositional variation of natural
melts is complicated and (ii) there is no theory on what compositional parameters are the most
important and how the diffusivity should be related to the composition. Furthermore, limited
and random experimental data often do not resolve the compositional trends well. Even though
in this review we made some ad hoc attempts to derive empirical relations between diffusivity
and melt composition, the success is limited. (Trying to use a couple of consistent compositional
parameters would not work well, and experimental data are not enough to constrain the fits if
many compositional parameters, e.g., fractions of all major cations, are used.) It appears that
the compositional dependence of diffusivity for rapidly diffusing (network modifiers) species
differs from that for slowly diffusing (network formers) species.
Among the effect of various components on diffusivities in silicate melts, dissolved H2O
content often has the largest effect, often increasing the diffusivity by orders of magnitude
(e.g., Watson 1979a). However, this effect has not been quantified for most elements. Because
most natural melts contain significant H2O, it is critical to make major efforts to quantify the
effect of H2O for accurate prediction of diffusion rates in natural melts.
In the literature, the Eyring diffusivity estimated from viscosity is often associated with
oxygen diffusivity. Recent experimental data show that oxygen diffusivity in the presence of
H2O is orders of magnitude greater than the Eyring diffusivity (Behrens et al. 2007). Because
natural melts often contain a significant amount of H2O, oxygen diffusivity and Eyring
diffusivity are rarely similar in nature (Zhang and Ni 2010, this volume). On the other hand,
silicon diffusivities are usually similar to or smaller than oxygen diffusivities in silicate melts.
Furthermore, the addition of H2O is not expected to affect silicon diffusivity as drastically
as oxygen diffusivity because H2O carries oxygen but not silicon. Based on limited data, we
hypothesize that the Eyring diffusivity is more likely associated with silicon diffusivity, and the
similarity may even apply to hydrous silicate melts. If this hypothesis is verified, the extensive
viscosity data for silicate melts may be used to estimated silicon diffusivity, and vice versa.
As expected, the SEBD values are less consistent than self and tracer diffusivities.
The dependence of SEBD values on the specific conditions (such as a specific mineral
dissolving in a specific melt) also means that it is more difficult to quantify SEBD values.
Eventually, quantification of chemical diffusion in multicomponent melts will require the use
of more complicated diffusion models, either a diffusion matrix approach or more advanced
approximate methods (Liang 2010; Zhang 2010, this volume).
Almost 30% of all the experimental diffusion data (Online Supplementary Table 1) by
geochemists are on silicate melts of simple synthetic systems such as jadeite, albite, anorthite,
orthoclase and diopside melts. These data turn out to be less useful in quantitative applications
to geological problems. Furthermore, diffusion in these systems is not necessarily simpler in
terms of mathematical treatment, or in terms of inferring the diffusion mechanism. We suggest
that focusing future diffusion studies on natural melts and their analogs (e.g., well-chosen
haplorhyolite and haplobasalt melts) would be more productive.
404 Zhang, Ni, Chen
aCKNoWleDGMeNtS
This research is supported by NSF grants EAR-0711050 and EAR-0838127. H. Ni
acknowledges financial support by the visitors program of the Bayerisches Geoinstitut.
Charles Lesher and James Mungall provided formal reviews. We thank Daniele Cherniak for
editorial handling.
refereNCeS
Alletti M, Baker DR, Freda C (2007) Halogen diffusion in a basaltic melt. Geochim Cosmochim Acta 71:3570-
3580
Bai TB, Koster van Groos AF (1994) Diffusion of chlorine in granitic melts. Geochim Cosmochim Acta 58:113-
123
Baker DR (1989) Tracer versus trace element diffusion: diffusional decoupling of Sr concentration from Sr
isotope composition. Geochim Cosmochim Acta 53:3015-3023
Baker DR (1990) Chemical interdiffusion of dacite and rhyolite: anhydrous measurements at 1 atm and 10 kbar,
application of transition state theory, and diffusion in zoned magma chambers. Contrib Mineral Petrol
104:407-423
Baker DR (1991) Interdiffusion of hydrous dacitic and rhyolitic melts and the efficacy of rhyolite contamination
of dacitic enclaves. Contrib Mineral Petrol 106:462-473
Baker DR (1992a) Tracer diffusion of network formers and multicomponent diffusion in dacitic and rhyolitic
Baker DR (1992b) Estimation of diffusion coefficients during interdiffusion of geologic melts: application of
transition state theory. Chem Geol 98:11-21
Baker DR (1993) The effect of F and Cl on the interdiffusion of peralkaline intermediate and silicic melts. Am
Mineral 78:316-324
Baker DR (1995) Diffusion of silicon and gallium (as an analogue for aluminum) network-forming cations and
their relationship to viscosity in albite melt. Geochim Cosmochim Acta 59:3561-3571
Baker DR (2010) Erratum to: Chemical interdiffusion of dacite and rhyolite: anhydrous measurements at 1
atm and10 kbar, application of transition state theory, and diffusion in zoned magma chambers. Contrib
Mineral Pertrol 159:597
Baker DR, Balcone-Boissard H (2009) Halogen diffusion in magmatic systems: our current state of knowledge.
Chem Geol 263:82-88
Baker DR, Bossanyi H (1994) The combined effect of F and H2O on interdiffusion between peralkaline dacitic
and rhyolitic melts. Contrib Mineral Petrol 117:203-214
Baker DR, Conte AM, Freda C, Ottolini L (2002) The effect of halogens on Zr diffusion and zircon dissolution
in hydrous metaluminous granitic melts. Contrib Mineral Petrol 142:666-678
Baker DR, Freda C, Brooker RA, Scarlato P (2005) Volatile diffusion in silicate melts and its effects on melt
inclusions. Ann Geophys 48:699-717
Baker DR, Watson EB (1988) Diffusion of major and trace elements in compositionally complex Cl- and
F-bearing silicate melts. J Non-Cryst Solids 102:62-70
Baker LL, Rutherford MJ (1996) Sulfur diffusion in rhyolite melts. Contrib Mineral Petrol 123:335-344
Balcone-Boissard H, Baker DR, Villemant B, Boudon G (2009) F and Cl diffusion in phonolitic melts: influence
of the Na/K ratio. Chem Geol 263:89-98
Behrens H (1992) Na and Ca tracer diffusion in plagioclase glasses and supercooled melts. Chem Geol 96:267-
275
Behrens H, Hahn M (2009) Trace element diffusion and viscous flow in potassium-rich trachytic and phonolitic
melts. Chem Geol 259:63-77
Pertrol 157:765-780
Blank JG (1993) An experimental investigation of the behavior of carbon dioxide in rhyolitic melt. Ph. D.
Dissertation, California Institute of Technology, Pasadena
Bowen NL (1921) Diffusion in silicate melts. J Geol 29:295-317
Brearley M, Scarfe CM (1986) Dissolution rates of upper mantle minerals in an alkali basalt melt at high
pressure: an experimental study and implication for ultramafic xenolith survival. J Petrol 27:1157-1182
Chakraborty S, Dingwell D, Chaussidon M (1993) Chemical diffusivity of boron in melts of haplogranitic

composition. Geochim Cosmochim Acta 57:1741-1751
Cherniak DJ, Hervig R, Koepke J, Zhang Y, Zhao D (2010) Analytical methods in diffusion studies. Rev
Cooper AR (1968) The use and limitations of the concept of an effective binary diffusion coefficient for multi-
component diffusion. In: Mass Transport in Oxides, vol 296. Wachman JB, Franklin AD (eds) National
Bureau Standards Spec Publ, pp 79-84
Cooper AR, Kingery WD (1964) Dissolution in ceramic systems: I. molecular diffusion, natural convection,
and forced convection studies of sapphire dissolution in calcium aluminum silicate. J Am Ceramic Soc
47:37-43
Crank J (1975) The Mathematics of Diffusion, Clarendon Press, Oxford
Cunningham GJ, Henderson P, Lowry RK, Nolan J, Reed SJB, Long JVP (1983) Lithium diffusion in silicate
melts. Earth Planet Sci Lett 65:203-205
DeGroot SR, Mazur P (1962) Non-Equilibrium Thermodynamics. Interscience, New York
Dingwell DB, Scarfe CM (1984) Chemical diffusion of fluorine in jadeite melt at high pressure. Geochim
Dingwell DB, Scarfe CM (1985) Chemical diffusion of fluorine in melts in the system Na2O-Al2O3-SiO2. Earth
Dunn T, Ratliffe WA (1990) Chemical diffusion of ferrous iron in a peraluminous sodium aluminosilicate melt:
0.1 MPa to 2.0 GPa. J Geophys Res 95:15665-15673
Einstein A (1905) The motion of small particles suspended in static liquids required by the molecular kinetic
theory of heat. Ann Phys 17:549-560
Finnila AB, Hess PC, Rutherford MJ (1994) Assimilation by lunar mare basalts: melting of crustal material and
dissolution of anorthite. J Geophys Res 99:14677-14690
Frank RC, Swets DE, Lee RW (1961) Diffusion of Ne isotopes in fused quartz. J Chem Phys 35:1451-1459
Freda C, Baker DR (1998) Na-K interdiffusion in alkali feldspar melts. Geochim Cosmochim Acta 62:2997-
3007
Freda C, Baker DR, Scarlato P (2005) Sulfur diffusion in basaltic melts. Geochim Cosmochim Acta 69:5061-
5069
Gabitov RI, Price JD, Watson EB (2005) Diffusion of Ca and F in haplogranitic melt from dissolving fluorite
crystals at 900°-1000°C and 100 MPa. Geochem Geophys Geosys 6:doi:10.1029/2004GC000832
Glasstone S, Laider KJ, Eyring H (1941) The Theory of Rate Processes. McGraw-Hill, New York
Harrison TM, Watson EB (1983) Kinetics of zircon dissolution and zirconium diffusion in granitic melts of
variable water content. Contrib Mineral Petrol 84:66-72
Hayden LA, Watson EB (2007) Rutile saturation in hydrous siliceous melts and its bearing on Ti-thermometry
of quartz and zircon. Earth Planet Sci Lett 258:561-568
Henderson P, Nolan J, Cunningham GC, Lowry RK (1985) Structural controls and mechanisms of diffusion in
natural silicate melts. Contrib Mineral Pertrol 89:263-272
Hess KU, Dingwell DB, Webb SL (1995) The influence of excess alkalis on the viscosity of a haplogranitic melt.
Am Mineral 80:297-304
Hofmann AW (1980) Diffusion in silicate melts: a critical review. In: Hargraves RB (ed) Physics of magmatic
processes. Princeton Univ Press, Princeton, p 385-417
Hofmann AW, Magaritz M (1977) Diffusion of Ca, Sr, Ba, and Co in a basaltic melt: implications for the
geochemistry of the mantle. J Geophys Res 82:5432-5440
Hui H, Zhang Y (2007) Toward a general viscosity equation for natural anhydrous and hydrous silicate melts.
Hui H, Zhang Y, Xu, Z., Del Gaudio P, Behrens H (2009) Pressure dependence of viscosity of rhyolitic melts.
International Commission on Radiation Units and Measurements (1984) ICRU Report 37: Stopping Power for
Electrons and Positrons. ICRU, Bethesda, MD
Jambon A (1982) Tracer diffusion in granitic melts: experimental results for Na, Rb, Cs, Ca, Sr, Ba, Ce, Eu to
1300°C and a model of calculation. J Geophys Res 87:10797-10810
Jambon A, Carron JP (1976) Diffusion of Na, K, Rb and Cs in glasses of albite and orthoclase composition.
Jambon A, Semet MP (1978) Lithium diffusion in silicate glasses of albite, orthoclase, and obsidian compositions:
an ion-microprobe determination. Earth Planet Sci Lett 37:445-450
Jambon A, Shelby JE (1980) Helium diffusion and solubility in obsidians and basaltic glass in the range 200-
300°C. Earth Planet Sci Lett 51:206-214
406 Zhang, Ni, Chen
Jambon A, Zhang Y, Stolper EM (1992) Experimental dehydration of natural obsidian and estimation of DH2O at
low water contents. Geochim Cosmochim Acta 56:2931-2935
Karsten JL, Holloway JR, Delaney JR (1982) Ion microprobe studies of water in silicate melts: temperature-
dependent water diffusion in obsidian. Earth Planet Sci Lett 59:420-428
Koepke J, Behrens H (2001) Trace element diffusion in andesitic melts: an application of synchrontron X-ray
fluorescence analysis. Geochim Cosmochim Acta 65:1481-1498
Koyaguchi T (1989) Chemical gradient at diffusive interfaces in magma chambers. Contrib Mineral Pertrol
103:143-152
Kubicki JD, Muncill GE, Lasaga AC (1990) Chemical diffusion in melts on the CaMgSi2O6-CaAl2Si2O8 join
under high pressures. Geochim Cosmochim Acta 54:2709-2715
Lasaga AC (1998) Kinetic Theory in the Earth Sciences, Princeton University Press, Princeton, NJ, p 811
LaTourrette T, Wasserburg GJ (1997) Self diffusion of europium, neodymium, thorium, and uranium in
haplobasaltic melt: The effect of oxygen fugacity and the relationship to melt structure. Geochim
LaTourrette T, Wasserburg GJ, Fahey AJ (1996) Self diffusion of Mg, Ca, Ba, Nd, Yb, Ti, Zr, and U in
haplobasaltic melt. Geochim Cosmochim Acta 60:1329-1340
Lesher CE, Walker D (1986) Solution properties of silicate liquids from thermal diffusion experiments. Geochim
Lesher CE, Walker D (1991) Thermal diffusion in petrology. In: Diffusion, Atomic Ordering, and Mass
Transport, vol 8. Ganguly J (ed) Springer-Verlag, New York, p 396-451
Liang Y, Richter FM, Davis AM, Watson EB (1996a) Diffusion in silicate melts, I: self diffusion in CaO-Al2O3-
SiO2 at 1500°C and 1 GPa. Geochim Cosmochim Acta 60:4353-4367
Liang Y, Richter FM, Watson EB (1996b) Diffusion in silicate melts, II: multicomponent diffusion in CaO-
Al2O3-SiO2 at 1500°C and 1 GPa. Geochim Cosmochim Acta 60:5021-5035
Linnen RL, Pichavant M, Holtz F (1996) The combined effects of fO2 and melt composition on SnO2 solubility
and tin diffusivity in haplogranitic melts. Geochim Cosmochim Acta 60:4965-4976
Linnen RL, Pichavant M, Holtz F, Burgess S (1995) The effect of fO2 on the solubility, diffusion, and speciation
of tin in haplogranitic melt at 850°C and 2 kbar. Geochim Cosmochim Acta 59:1579-1588
Lowry RK, Henderson P, Nolan J (1982) Tracer diffusions of some alkali, alkaline-earth and transition element
ions in a basaltic and andesitic melt, and the implications concerning melt structure. Contrib Mineral Petrol
80:254-261
Lowry RK, Reed SJB, Nolan J, Henderson P, Long JVP (1981) Lithium tracer-diffusion in an alkali-basaltic
melt - an ion-microprobe determination. Earth Planet Sci Lett 53:36-40
Lundstrom CC (2003) An experimental investigation of the diffusive infiltration of alkalis into partially
molten peridotite: implications for mantle melting processes. Geochem Geophys Geosys
4:doi:10.1029/2001GC000224
Lundstrom CC, Hoernle K, Gill J (2003) U-series disequilibria in volcanic rocks from the Canary Islands: Plume
versus lithospheric melting. Geochim Cosmochim Acta 67:4153-4177
MacKenzie JM, Canil D (2006) Experimental constraints on the mobility of rhenium in silicate liquids. Geochim
MacKenzie JM, Canil D (2008) Volatile heavy metal mobility in silicate liquids: implications for volcanic
degassing and eruption prediction. Earth Planet Sci Lett 269:488-496
Magaritz M, Hofmann AW (1978a) Diffusion of Sr, Ba and Na in obsidian. Geochim Cosmochim Acta 42:595-
605
Magaritz M, Hofmann AW (1978b) Diffusion of Eu and Gd in basalt and obsidian. Geochim Cosmochim Acta
42:847-858
Medford GA (1973) Calcium diffusion in a mugearite melt. Can J Earth Sci 10:394-402
Morgan Z, Liang Y, Hess PC (2006) An experimental study of anorthosite dissolution in lunar picritic magmas:
implications for crustal assimilation processes. Geochim Cosmochim Acta 70:3477-3491
Mungall JE (2002) Empirical models relating viscosity and tracer diffusion in magmatic silicate melts. Geochim
Mungall JE, Dingwell DB (1997) Actinide diffusion in a haplogranitic melt: Effects of pressure, water content,
and pressure. Geochim Cosmochim Acta 61:2237-2246
Mungall JE, Dingwell DB, Chaussidon M (1999) Chemical diffusivities of 18 trace elements in granitoid melts.
Nakamura E, Kushiro I (1998) Trace element diffusion in jadeite and diopside melts at high pressures and its
geochemical implication. Geochim Cosmochim Acta 62:3151-3160
Perez WA, Dunn T (1996) Diffusivity of strontium, neodymium, and lead in natural rhyolite melt at 1.0 GPa.
Pichavant M, Montel JM, Richard LR (1992) Apatite solubility in peraluminous liquids: experimental data and
an extension of the Harrison-Watson model. Geochim Cosmochim Acta 56:3855-3861
Poe BT, McMillan PF, Rubie DC, Chakraborty S, Yarger J, Diefenbacher J (1997) Silicon and oxygen self-
diffusivities in silicate liquids measured to 15 GPa and 2800 K. Science 276:1245-1248
Rapp RP, Watson EB (1986) Monazite solubility and dissolution kinetics: implications for the thorium and light
rare earth chemistry of felsic magmas. Contrib Mineral Petrol 94:304-316
Reid JE, Peo BT, Rubie DC, Zotov N, Wiedenbeck M (2001) The self-diffusion of silicon and oxygen in diopside
(CaMgSi2O6) liquid up to 15 GPa. Chem Geol 174:77-86
Roselieb J, Chaussidon M, Mangin D, Jambon A (1998) Lithium diffusion in vitreous jadeite (NaAlSi2O6): an
ion microprobe investigation. N Jb Miner Abh 172:245-257
Roselieb K, Jambon A (1997) Tracer diffusion of patassium, rubidium, and cesium in a supercooled jadeite melt.
Roselieb K, Jambon A (2002) Tracer diffusion of Mg, Ca, Sr, and Ba in Na-aluminosilicate melts. Geochim
Roselieb K, Rammensee W, Buttner H, Rosenhauer M (1995) Diffusion of noble gases in melts of the system
SiO2-NaAlSi2O6. Chem Geol 120:1-13
Ruessel C, Wiedenroth A (2004) The effect of glass composition on the thermodynamics of the Fe2 + /Fe3 +
equilibrium and the iron diffusivity in Na2O/MgO/CaO/Al2O3/SiO2 melts. Chem Geol 213:125-135
Saal AE, Hauri EH, Cascio ML, Van Orman JA, Rutherford MJ, Cooper RF (2008) Volatile content of lunar
volcanic glasses and the presence of water in the Moon’s interior. Nature 454:192-196
chalcogenides. Acta Crystallogr A32:751-767
Shaw CSJ (2000) The effect of experiment geometry on the mechanism and rate of dissolution of quartz in
basanite at 0.5 GPa and 1350 °C. Contrib Mineral Petrol 139:509-525
Sheng YJ, Wasserburg GJ, Hutcheon ID (1992) Self-diffusion of magnesium in spinel and in equilibrium melts:
Constraints on flash heating of silicates. Geochim Cosmochim Acta 56:2535-2546
Shimizu N, Kushiro I (1984) Diffusivity of oxygen in jadeite and diopside melts at high pressures. Geochim
Shimizu N, Kushiro I (1991) The mobility of Mg, Ca, and Si in diopside-jadeite liquids at high pressures. In:
Physical Chemistry of Magmas. Perchuk LL, Kushiro I (eds) Springer-Verlag, pp 192-212
Tinker D, Lesher CE, Baxter GM, Uchida T, Wang Y (2004) High-pressure viscometry of polymerized silicate
melts and limitations of the Eyring equation. Am Mineral 89:1701-1708
Tinker D, Lesher CE, Hucheon ID (2003) Self-diffusion of Si and O in diopside-anorthite melt at high pressures.
van der Laan SR, Wyllie PJ (1993) Experimental interaction of granite and basaltic magmas and implications
for mafic enclaves. J Petrol 34:491-517
van der Laan SR, Zhang Y, Kennedy A, Wylie PJ (1994) Comparison of element and isotope diffusion of K and
Van Orman JA, Grove TL (2000) Origin of lunar high-titanium ultramafic glasses: constraints from phase
relations and dissolution kinetics of clinopyroxene-ilmenite cumulates. Meteor Planet Sci 35:783-794
Wang H, Xu Z, Behrens H, Zhang Y (2009) Water diffusion in Mount Changbai peralkaline rhyolitic melt.
Contrib Mineral Pertrol 158:471-484
Watson EB (1979a) Diffusion of cesium ions in H2O-saturated granitic melt. Science 205:1259-1260
Watson EB (1979b) Calcium diffusion in a simple silicate melt to 30 kbar. Geochim Cosmochim Acta 43:313-
322
Watson EB (1981) Diffusion in magmas at depth in the Earth: the effects of pressure and dissolved H2O. Earth
Watson EB (1982) Basalt contamination by continental crust: some experiments and models. Contrib Mineral
Petrol 80:73-87
Watson EB (1991a) Diffusion of dissolved CO2 and Cl in hydrous silicic to intermediate magmas. Geochim
Watson EB (1991b) Diffusion in fluid-bearing and slightly-melted rocks: experimental and numerical approaches
illustrated by iron transport in dunite. Contrib Mineral Petrol 107:417-434
408 Zhang, Ni, Chen
Watson EB (1994) Diffusion in volatile-bearing magmas. Rev Mineral 30:371-411

Watson EB, Bender JF (1980) Diffusion of cesium, samarium, strontium, and chlorine in molten silicate at high
temperatures and pressures. Geol Soc America Abstracts with Program 12:545
Watson EB, Dohmen R (2010) Non-traditional and emerging methods for characterizing diffusion in minerals
and mineral aggregates. Rev Mineral Geochem 72:61-105
Watson EB, Jurewicz SR (1984) Behavior of alkalis during diffusive interaction of granitic xenoliths with
basaltic magma. J Geol 92:121-131
Watson EB, Sneeringer MA, Ross A (1982) Diffusion of dissolved carbonate in magmas: experimental results
and applications. Earth Planet Sci Lett 61:346-358
Winther KT, Watson EB, Korenowski GM (1998) Magmatic sulfur compounds and sulfur diffusion in albite
melt at 1 GPa and 1300-1500°C. Am Mineral 83:1141-1151
Wolf MB, London D (1994) Apatite dissolution into peraluminous haplogranitic melts: an experimental study of
solubilities and mechanisms. Geochim Cosmochim Acta 58:4127-4145
Zhang Y (2008) Geochemical Kinetics, Princeton University Press, Princeton, NJ
225
Acta 55:441-456
Zhang Y, Walker D, Lesher CE (1989) Diffusive crystal dissolution. Contrib Mineral Petrol 102:492-513
Zhang Y, Xu Z (1995) Atomic radii of noble gas elements in condensed phases. Am Mineral 80:670-675
Zhang Y, Xu Z, Zhu M, Wang H (2007) Silicate melt properties and volcanic eruptions. Rev Geophys 45:RG4004,
doi:4010.1029/2006RG000216

Reviews in Mineralogy - Geochemistry Volume 72 - Diffusion in Minerals - Melts

Uploaded by

Copyright:

Available Formats

You might also like

Reviews in Mineralogy - Geochemistry Volume 72 - Diffusion in Minerals - Melts

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Reviews in Mineralogy - Geochemistry Volume 72 - Diffusion in Minerals - Melts

Uploaded by

Copyright:

Available Formats

Reviews in Mineralogy & Geochemistry

Vol. 72 pp. 1-4, 2010 1

Diffusion in Minerals and Melts: Introduction

INTRODUCTION: RATIONALE FOR THIS VOLUME

SCOPE AND CONTENT OF THIS VOLUME

Diffusion in Minerals and Melts:

Theoretical Background of Diffusion in Minerals and Melts 7

Figure 3. Sketch of fluxes into and out of an

Figure 4. Concentration profile C versus x, and the 4. Concentration profile∂C2C/∂x

interdiffusion or mutual diffusion), multispecies diffusion, multicomponent diffusion, and

where 40Ar and 40K are atomic concentrations (such as mol/m3).

Diffusion in three dimensions (isotropic media)

SOLUTIONS TO BINARY AND ISOTROPIC DIFFUSION PROBLEMS

Theoretical Background of Diffusion in Minerals and Melts 21

pressure; Behrens and Zhang 2001).

Figure 8. Setup of an olivine dissolution

Figure 9. Homogenization of an oscillatorily zoned

Calculated profiles are shown in Figure 10b.

Figure 11. Closure temperature

corresponds to the time when the mineral temperature was Tc.

DIFFUSION IN ANISOTROPIC MEDIA

 Jx   D11 D12 D13   ∂C / ∂x 

Based on mass balance (Eqn. 7), ∂C/∂t can be written as:

The above equation can be simplified using coordinate transformation to become:

The transformed diffusion equation becomes

changed the shape of a crystal, from

 J1   D11 D12 ... D1n   ∇C1 

Figure 14. MgO diffusion

Modified effective binary approach (activity-based effective binary approach)

activity gradient of the component alone (Eqn. 65):

Figure 15. Fitting an uphill

Diffusivity matrix approach

Youxue Zhang (Ch 2) Page 17

Equation (118) is often referred to as the Darken-Hartley-Crank equation. Cooper (1965),

although the two equations may look similar; and (2)Page

 D11 D12 D13 

Diffusion of H, C, and O Components

1529-6466/10/0072-0005$10.00 DOI: 10.2138/rmg.2010.72.5

DIffUSION Of THe H2O COMpONeNT

Diffusion of H, C, O Components in Silicate Melts 175

Diffusion of H, C, O Components in Silicate Melts 177

Other expressions of equilibrium

H2O diffusion literature

figure 4. Experimental H2Ot diffu-

figure 5. Experimental H2Ot dif-

where K is the equilibrium constant

184 Zhang & Ni

Diffusion of H, C, O Components in Silicate Melts 185

186 Zhang & Ni

data (Ni et al. 2009b).

Diffusion of H, C, O Components in Silicate Melts 187

figure 11. Comparison of H2Ot

cases. H2Ot diffusivity is higher

large enough to be distinguished

∂Ctotal H ∂  2∂CH2 2∂CH2 Om 

The equilibrium constant (KH2-H2O) can be written as:

figure 13. (a) Fugacity ratio of H2/H2O as

Fig. 13. (a) Fugacity ratio of H2/H2O as a function of temperature at oxygen