Download as pdf or txt
Download as pdf or txt
You are on page 1of 36

ANNUAL

REVIEWS Further
Ann. Rev. Earth Planet. Sci. 1989. 17:439-74 Quick links to online content
Copyright © 1989 by Annual Reviews Inc. All rights reserved

MAGMA CHAMBERS

Bruce D. Marsh
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

Department of Earth and Planetary Sciences,


The Johns Hopkins University, Baltimore, Maryland 21218
Access provided by McMaster University on 01/23/15. For personal use only.

1. INTRODUCTION

Magma chambers are subterranean caldrons, vats, or reservoirs containing


molten silicate fluid. Some are staging areas for volcanic eruptions; others
have no link to the Earth's surface and in cooling solidify to form coarse­
grained plutons. Magma chambers no doubt exist at every level of the
Earth. Although of unusual composition, the greatest terrestrial magma
chamber is the molten outer core; its magma is impounded by its sheer
density. The crust is littered with evidence of past magma chambers-sills,
laccoliths, plutons, layered intrusions; it is a magma chamber repository.
Undoubtedly the mantle too contains numerous magma chambers, but its
constant stirring and inaccessibility precludes direct observation. Even the
mantle itself acts much as a magma chamber.
Growing interest in magma chambers really stems from the generally
held concept of the magmatic process (Figure I) whereby magma is formed,
ascends, occupies a chamber, and eventually erupts. The overall process
is so chemically and physically multifaceted that petrologists have mostly
chosen to expect the chemical identity of the magma to be affected most
strongly at the source and in the chamber itself. A large part of this
viewpoint comes from ease of modeling. The magma is given its chemical
composition at the source with subsequent modification in the chamber.
Magmatic evolution is thus often characterized by deep-seated (source)
and high-level (chamber) processes. There is no good reason for this habit.
There is nothing that prevents the processes of ascent and eruption from
being equally important in determining the end product. It has not always
been this way.
The magma chamber concept is relatively new. Prior to about 25 years
ago, the term was used only in passing (e.g. Grout 1932, p. 177) to denote
a quantity of magma. Batholith, laccolith, and simply intrusion were the
439
0084-6597/89/0515-0439$02.00
440 MARSH

Magmatic Life Cycle


��
'- I

ERUPTION
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org
Access provided by McMaster University on 01/23/15. For personal use only.

EXTRACTION

CO� : �UGH
IO
MELT PRODUCTION
P
� � BOU RY

Figure 1 The magmatic life cycle. Melt produced by convection through a phase boundary
is extracted to form upward-moving dikes, diapirs, or stoping bodies. Magma collects in the
near-surface, forming a magma chamber in which further cooling and crystallization is
thought to give rise to the spectrum of chemical compositions displayed in volcanic
emanations.

more common words. The source area and final pluton were thought to
be more directly linked throughout the magmatic cycle, with the deep
roots of plutons passing imperceptibly into the source rock (Buddington
1 960). This concept emerged in response to the question of the very origin
of igneous rock itself. Is it always a molten fluid? Basalt surely is, but is
granite? As debate waned on this issue, the ever-present problem of the
great chemical diversity of igneous rocks loomed large.
For nearly 50 years, N. L. Bowen persisted in the careful enunciation
of the reaction principle and crystal fractionation as the dominant process
in turning basalt into almost any other more siliceous magma. At the same
time large basaltic plutons, like Bushveld, Stillwater, and Skaergaard, were
found to show undeniably strong effects of such sedimentation under
MAGMA CHAMBERS 441
magmatic conditions. These results, coupled with growing evidence from
experimental phase equilibria that most magmas could be chemically
related, strongly turned petrologists to the belief that sedimentation in
magmatic vats gives magmatic diversity. Today, magmatic differentiation
is synonymous with Bowen's crystal fractionation. This presumed answer
has never been proven. The best examples of fractionating plutons show
relatively little diversity. There is no doubt that separation of liquid and
crystals breeds diversity, but how does it happen? What is the physical and
chemical history of magma chambers?
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

2. EXISTENCE, SIZE, AND SHAPE


Access provided by McMaster University on 01/23/15. For personal use only.

The strongest evidence for the existence of magma chambers actively


undergoing convection and crystal sorting comes from plutons, which are
the most fundamental igneous body. They are virtually everywhere and of
nearly every shape and dimension, ranging from saucerlike bodies to tall
stocks and necks. Some are very small ('" 1 km3), such as in island arcs,
whereas others, such as in batholithic terranes, easily exceed 103 km3• From
the Sierra Nevada, an area containing many plutons, Fyfe (1970) estimated
an average pluton to have a volume equivalent to a sphere of radius 3-7
km ('" 100-1400 km3). A similar average by Gastil et al (1975) of more
than 400 plutons in the Baja California and Sierra Nevada batholiths gave
a radius, corrected for a random depth of emplacement and erosional
level, of about 3 km. The largest plutons seem to be of either ultramafic
or silicic composition. The Bushveld ('" 105 km3) and Stillwater ('" 104
km3) plutons are giant ultramafic bodies, and the many large silicic ash
flows, such as the '" 3000 km3 Fish Canyon flow (Whitney & Stormer
1985), give minimum measures of reservoir size.
Volcanoes themselves, especially those of island arcs, also give minimum
measures of reservoir size because they often represent one (albeit some­
times long) period or phase of activity. The volume of such composite and
stratocone volcanoes is fairly well approximated by a cone of radius equal
to its height, equivalent to a sphere with a radius equal to 0.625 of the
volcano relief. The largest arc volcanoes have heights of about 3-4 km,
while more common heights are 1-2 km, representing overall magmatic
volumes (ignoring porosity) of ",4-60 km3. Although perhaps somewhat
surprising, plutons in such terranes are of similar sizes, which may suggest
that these volcanoes are fair measures of the chamber volume itself. This
would of course further imply that new magma replaces that lost to
building the volcano.
The nearer a body is to the Earth's surface at the time of emplacement,
the more sheetlike it becomes. The leading edges of diapirs either lose their
442 MARSH

density contrast or congeal while the deeper parts continue to rise, which
causes the bodies to spread or balloon laterally (Marsh 1982, 1984).
Ascending dikes also lose buoyancy and spread normal to the least prin­
cipal stress, which is vertical, thus forming laccoliths or sills (Jackson &
Pollard 1 987). The many diabase sills throughout the world are excellent
examples of such bodies (e.g. Walker 1 940, Walker & Poldervaart 1 949,
Smith et aI1 975). In fact, using gelatin as a substitute for magma, Hynd­
man & Alt (1987) have experimentally produced laccoliths from tangential
feeder dikes beneath surface loads. The large ultramafic bodies, no matter
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

what the depth of emplacement, are sheetlike or funnellike (e.g. Wager &
Brown 1 957). There are also bodies that are apparently more equant, and
Access provided by McMaster University on 01/23/15. For personal use only.

even some (especially stocks, volcanic necks, and dikes) that are tall bodies,
but these never seem to reach the volume of sheetlike bodies.

2.1 Detection
The direct detection of magma chambers has not proven easy; a thorough
review is given by Iyer ( 1984). A microearthquake study by Matumoto
( 1 971) of the subsurface of Mt. Katmai in Alaska, where a large ash-flow
eruption occurred in 1912, revealed shear wave shadowing along certain
ray paths. He found perhaps three bodies at depths less than 10 km and
one at 20-30 km. Each zone of attenuation is roughly circular or elliptical
and 1 0-25 km in diameter. The actual sizes of such bodies are uncertain
by at least the wavelength of shear waves ( 3 km). At the other extreme,

teleseismic events have been used to map local crustal attenuation beneath
the young calderas at Yellowstone (e.g. Iyer 1 984) and Long Valley
(Steeples & Iyer 1 976, Dawson et al 1 988). The advantage of teleseismic
studies is that the source events are far from the receivers, which are near
the region of interest, and thus source and deep path irregularities can be
deconvolved from the signal of interest. The steep arrival angle allows an
arbitrarily dense receiver network to map out virtually any region with
minimal refraction effects. By themselves, such mappings show large
regions of attenuation, but there is always ambiguity in deciding where
along the ray path (and of what amplitude) to place the anomalous veloci­
ties. This is helped by crossing rays, but the regions are still somewhat
fuzzy simply due to sampling uncertainties. Augmentation of these data
with gravity surveys and crustal refraction surveys allows much better
definition of the possible magma chamber (Hill 1 976). Long Valley, Cali­
fornia, has been studied extensively and shows good evidence for an
irregular magma chamber at intermediate crustal levels with a vertical
cross-sectional area of 75 km2• Spread over the caldera itself, this value

suggests a volume of 500 km3 or greater. Many geophysically inferred


bodies seem to be either 1 0 km in diameter or 50 km in diameter; the


� �
MAGMA CHAMBERS 443
latter are those bodies associated with large silicic calderas like Yellow­
stone. The splendid depictions by M. P. Ryan and associates (e.g. Ryan
et al 1 983, Ryan 1 987) of the e1asto-mechanical behavior of magmatic
bodies within the Hawaiian system clearly demonstrate the strong influence
of regional fractures and stress fields on the scale of magmatic confinement.
As more and more geophysical data have accumulated, the inferred
magma chambers have shrunk to sizes approaching the volume of erupted
material. The hopes of drilling a short distance beneath a caldera to enter
an active magma chamber have also shrunk. To grasp the difference
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

between a mostly petrologic perspective and a geophysical perspective,


compare the cover illustrations of Eos, Transactions of the American
Access provided by McMaster University on 01/23/15. For personal use only.

Geophysical Union for 30 April 1 985, 27 May 1 986, and 1 9 January


1988. That is, magmatists tend to envision magma chambers as largely
uncomplicated vessels that impose magmatic conditions on their environs.
Geophysical surveys and many postmortem field studies often show a
much more complex picture, one in which although magma is abundantly
present, it fills nooks and crannies and has resided under the strict struc­
tural control of the country rock. It has also become clear that the very
structure of calderas themselves are excellent examples of Daly's (e.g.
1 903) idea of stoping and Richey'S ( 1 927) similar idea of bell-jar intrusion.
The forms, sizes, and depths of magma chambers are exceedingly variable,
but a generic body, for use as a standard-state model, might be sheetlike
at one extreme and dikelike at the other. Their possible thermo-mechanical
regimes are not that different.

3. COOLING HISTORY
3.1 Historical
The essential scale by which all magmatic processes are measured is set by
the solidification time. It is the maximum allowable time to effect any
processes within a dynamically active magma chamber. (The exact mean­
ing of "dynamically active" is set out in the following section.) The earliest
attempt to calculate the cooling of igneous bodies, if we ignore Lord
Kelvin's estimate of'the Earth age, was apparently by Lane (e.g. 1 902),
who considered conductive cooling of dikes in order to obtain a time scale
for gauging the rate of crystal growth. Most subsequent solutions have
also been analytical. Ingersoll & Zobel ( 1 9 l 3) considered the cooling of a
lava flow, Lovering ( 1 935) and Larsen ( 1 945) considered lavas, dikes, and
stocks, Ingersoll et al ( 1 9 54) considered spherical bodies, and Jaeger (e.g.
1 9 57, 1968) considered many geometries and discussed the effects of latent
heat and convection. All of these studies, with the exception of a few
specific examples by Jaeger, treat both the body and the wall rock as
444 MARSH

purely conductive with constant and uniform physical properties. The most
geologically thoughtful solutions are those of Larsen and Jaeger. The
critical consideration in such calculations is the form of the boundary
condition at the contact. Any fixed or even time-varying choice of tem­
perature or heat flux strongly influences the solution. Alternatively, La­
place-type solutions, where only the initial temperature field is prescribed
and then allowed to decay, give more realistic results. In all such cases the
cooling time t comes from scaling the heat conduction equation (e.g.
Marsh 1 984) to give
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org
Access provided by McMaster University on 01/23/15. For personal use only.

(1)

where L is the characteristic length scale of the body (e.g. its radius or
half-width), K is the thermal diifusivity, and C is a constant (of order
unity) that is dependent on the initial conditions. For virtually any problem
of pure conduction, a solution can be found in Carslaw & Jaeger ( 1959).

3.2 Contemporary Models


These investigations are basically of two types: those that consider the wall
rock to dominate heat transfer, and those that consider the heat to bc
driven from the body by magmatic convection. Of the first type, Irvine's
( 1970) consideration of a thermally well-mixed body cooled by conduction
through the wall rock treats a variety of wall-rock conditions, including
metamorphism, depth of burial, and possible roof melting. Somewhat
complementary studies by Cathles ( 1977) and Norton & Taylor ( 1979)
consider cooling of plutons with inclusion of the effects of hydrothermal
circulation, the efficiency of which is subsequently considered. The
opposite extreme, where cooling is dominated by convection within the
body itself, has been explored by Bartlett ( 1969), Spera ( 1980), Spera et al
( 1982), Lister ( 1 983), Martin et al (1987), and Huppert & Sparks ( 1 988)
among others.
The central issue here of the role of each medium in cooling is of critical
importance to all magmatic problems. An appreciation of this issue can
be gained by considering some simple models.

3.2.1 CONDUCTION ONLY The rate of cooling for the case in which both
the magma (assumed stagnant) and the wall rock cool only by conduction
is well known and can be taken as a basic or standard state whose total
flux or rate of heat transfer from the body is QCd(t). The efficiency of all
other cooling models having a.net or total outward heat flux of QT can be
compared with this state by forming the ratio
MAGMA CHAMBERS 445

QT (2)
�u=---,
Qed

which is defined to be the �usselt number. A dimensionally free, pure


number, �u is a measure of the rate of heat transfer measured in multiples
of that due purely to conduction. The well-known wind chill factor is an
everyday example of �u. It is important here to recognize that for mag­
matic problems, Qed(t) represents time-dependent conductive cooling of a
body identical in all respects to the convecting body except that it is
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

stagnant. It is important to emphasize that all aspects of magmatic cooling,


be it by conduction or by convection, are time dependent, and that such
Access provided by McMaster University on 01/23/15. For personal use only.

a definition of �u conforms exactly with other similar problems (e.g.


Homsy 1973, Ihaveri & Homsy 1982).

3.2.2 MAXIMUM Nu If the wall rock is taken to be conductive, the


maximum rate of cooling of an enclosed, deeply buried body occurs when
the body is kept perfectly mixed thermally-that is, the magma tem­
perature is everywhere uniform at all times, and heat loss to the walls is
uniformly and instantly felt throughout the body. In this (albeit unrealistic)
way the highest possible temperature is maintained at the magma-wall­
rock contact, and the maximum temperature gradient is maintained in the
wall rock. In a physical sense, this model is tantamount to assuming
a magma chamber of, for example, water kept turbulently stirred with
externally driven propellers whose energy needs are supplied elsewhere;
viscosity changes due to cooling are minimal, and the fluid always remains
highly mobile.
For any well-mixed body enclosed by conductive material, regardless of
shape, the maximum heat transfer is twice that of conduction alone (Marsh
& Maxey 1985, Marsh 1988a), i.e.
m 2.
�u ax =
(3)

This limit is physically sensible and easily comprehended. For conduction


only, the temperature at the contact immediately after emplacement, if we
assume identical thermal properties, is the average of the initial tem­
perature of the wall rock (Twr) and of the magma (Tm); the temperature
difference driving heat flow is thus (Tm+Twr)/2-Twr (Tm-Twr)/2. On
=

the other hand, for the well-mixed body the eady contact temperature is
not the average of the two media but is exactly the initial magma tem­
perature Tm; thus the heat flux is governed by the temperature difference
Tm-Twr. Because the characteristic conduction length scale ["-'(Kt)lj2] in
the wall rock is the same in both extremes, we have
446 MARSH

(4)

No amount of magmatic convection, however vigorous, can increase


the rate of heat transfer, and the great convenience in this choice of Qcd(t)
in defining Nu is now clear: Nu S 2. The only way to increase Nu further
is to make the wall rock more transparent to heat transfer. Metamorphic
reactions may help somewhat, but in the upper crust, hydrothermal cir­
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

culation and shallow depth of burial can increase Nu severalfold or more.


Such effects are considered later. The greatest importance of this simple
Access provided by McMaster University on 01/23/15. For personal use only.

and highly revealing model is in understanding other models and in placing


their physical bases in perspective.

3.2.3 CONVECTION DOMINATED A great deal of effort has gone into


understanding heat transfer from convective bodies (e.g. Turner 1 973,
Chap. 7) such as the Earth's mantle [see O'Connell & Hager (1980) for an
excellent review], and it is principally these results that have been applied
to magma chambers. The usual physical setup is as described already­
namely, an enclosed body deeply embedded within conductive wall rock.
All dynamics and heat transfer in convective fluids are related to the
Rayleigh number

IY.g!J.TL3
Ra= --­ (5)
vK '

where IY. is thermal expansion, 9 is gravity, v is kinematic viscosity, K is


thermal diffusivity, and !J. T and L are the temperature difference and the
length scale driving convection, respectively. Ra, a pure number like Nu,
is a measure of the importance of buoyancy relative to viscous drag and
conductive cooling effects. For high-viscosity, highly conductive material,
convection is unlikely unless !J.T and L can be made sufficiently large. Ra
is thus a general measure of convective vigor and of the rate of heat
transfer.
Almost all intuition into convection comes from consideration of a layer
of fluid heated from below and cooled from above. The layer of thickness
L is conductive, supporting a temperature gradient of !J.TjL (see Figure
2a), until !J.Tbecomes sufficiently large, which defines a critical Ra (i.e. Rae)
beyond which convection commences. This is called a stability problem
because convection only occurs if Ra > Rae. Stability is not an issue for
magma chambers with large side-wall cooling relative to floor and roof
cooling, for in the presence of horizontal temperature gradients, convection
occurs for all Ra.
MAGMA CHAMBERS 447
T� 2.00 r-------...,
a
b
® _1.75
�:J

1i 1.50
E
:J
Z
'ii 1.25

*
z
§ 1.00 r\-----=:::===�
E
'i,i
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

Wall Rock ::;;


.75
Access provided by McMaster University on 01/23/15. For personal use only.

.50 10..._...._
... ....__
. ...._..1..
. _.-1
0.0 1.0 2.0 3.0 4.0 5.0
Dimensionless time
"2
11�(4ktlL�

Figure 2 (a) The two possible basic thermal states commonly used in defining heat transfer
from a magmatic layer of thickness L. The steep gradient A is associated with time-dependent
conductive cooling, whereas the much less strong gradient B is associated with time-invariant
conductive cooling. It is the ratio AlB that is estimated by Equation (12). (b) The variation
of Numax with time (dimensionless). This is the rate of heat transfer from a thermally well­
mixed, buried sheet relative to the same stagnant sheet. For all times greater than 0.5, the
stagnant body cools more rapidly.

For large Ra (i.e. Ra/Rae > '" 1 03), we know from both experiment and
theory that
(6)
where CR is a constant usually in the range 0.2-0.7, and b is a constant in
the range 1/3-1/5, both depending on boundary conditions and possibly
on the Prandtl number (Pr == v/K). [Common values used for sheetlike
magma chambers are CR � 0.3 and b 1 /3, whereas typical values for
=

wall flows are CR � 0.5 and b 1 /4 (fixed T) or b 1 /5 (fixed flux).] This


= =

Nusse1t number (Nuo), although of the same general definition as (2), is


markedly different from (3) in that it utilizes for a basic or standard
state a steady-state (i.e. linear) temperature gradient across the body (i.e.
Qed'" I'lT/L, as shown by Figure 2a).
The velocity of convection is given by

(7)

and from this a Reynolds number (Re) can be formed by multiplying


throughout by L/v, i.e.
448 MARSH

VL K
-:= Re � -Ra 2b, (8)
V V

or

(9)
Using (6), we obtain

Re � Pr- 1 NU6. (10)


Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

Sinee Re is a more familiar measure of flow intensity, this provides a key


Access provided by McMaster University on 01/23/15. For personal use only.

insight into the role of the outward heat flux Nu in deciding the vigor of
convection. The value of Pr for magmas is '" 104 for ultramafic com­
positions and '" 108 or more for silicic compositions.
It would be convenient if Numax from (3) could be used in (10) to bound
Re, but this cannot be done because (6)-(10) are based on a much different
basic conductive state [Qed ¥ Qed(t)]. Although similar relationships must
hold, the constants may differ markedly. Some appreciation of the strong
effect this simple choice of basic state has in deciding Ra, for example, can
be gained by noting that from (6), the thermal boundary-layer thickness
d within the chamber will be of order
(11)

For the perfectly well-mixed body with no internal thermal boundary layer
(d 0) Ra must necessarily become very large (i.e. approach infinite
=

magnitude) in order to satisfy this limit. On the Numax scale, then, as Numax
goes from 1.0 to 2.0, Ra tends from 0 to 00 when measured on the NUo
scale. In terms of the cooling history of the body, both schemes give the
same result. But it is in assigning a physical meaning to the large values
of Ra and NUo that misunderstandings could arise.
Ifwe consider a typical basaltic magma chamber, for example, Ra could
easily be 1015, whence, from (9) and using b = 1(3, we have Re � 1 06 (and
Nuo � 105). Such a value, if actually measured in a flow, would certainly
represent a state of extreme turbulence. That this physical interpretation
may not be meaningful for sudden magmatic cooling from an initially
isothermal state can be seen by finding how large Ra and Nuo must be just
to meet the requirement that Qcd(t) = Qed for a stagnant magma. That is,
for Numax 1.0, what are the equivalent or virtual magnitudes of NUo and
=

Ra? Put yet another way, how many multiples of the weak basic state Qed
are necessary to achieve Qed(t )? Hence for a stagnant, sill-like, and initially
isothermal body instantaneously emplaced in cool wall rock, where the
temperature contrast is I1T, in order to maintain purely conductive heat
MAGMA CHAMBERS 449
flow from the upper boundary into an infinite half-space, it is necessary
that

Qed(t) '" (f1T)/(4Kt) 1/2


Nu(t)ed '" I] -I
_
( 1 2)
'"
Qed '" (f1T/2)/(L/2) - ,
where I] 2(Kt/L2)1/2. The associated virtual values of Ra needed to
=

achieve these rates of heat transfer are, from (6),

Ra(t)ed = [N��)CdTb = (CRI])-I/b. (13)


Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org
Access provided by McMaster University on 01/23/15. For personal use only.

Immediately after emplacement, Ra(t) cd -+ <Xl; and aftcr significant cool­


ing, '1 = O. l and, for CR 0.2 and b = 1/3, Ra(t)"d � 10 5• During all this
=

time, however, the body is, by definition, stagnant and only conductive.
These apparent Rayleigh numbers thus should not necessarily be con­
strued as indicators of convective vigor unless they are normalized to
Qed(t) or otherwise measured relative the magmatic basic state of sudden
cooling.
In this regard, it is useful to note that a common criterion used in
investigating the stability of a fluid layer during transient heating or cooling
is the value of Nu relative to the basic state Qeit). Homsy and associates
(e.g. Homsy 1 973, Ihaveri & Homsy 1 980, 1 982) take the point of insta­
bility, which gives Rae, to be when Nu(t) is 1 % or so above unity relative
to the time-dependent basic state QedCt). Magmatic cooling is similarly
transient, inasmuch as both the initial and final thermal states are iso­
thermal and cooling merely connects these two stable states through a
series of possibly unstable or convective states. Because the very early ('1 --->
0) conductive state is highly efficient in transferring heat, convection is
unnecessary and the fluid is stable. Only with increasing time does insta­
bility commence when convection becomes more efficient than conduction.
In terms of stability theory, this feature appears in the general result (see
Marsh 1 988a,b)
Ra tJ!2 = constant = Ct, ( 1 4)
where Ra is based on the layer thickness and the initial temperature
contrast, and tL is the dimensionless time (Kt/L2) for the onset of convec­
tion. From this result, we obtain
agf1T[(Kt) 1/2]3
----- = Rat = Ct. (1 5)
vK
Equation ( 1 5) shows that the characteristic length scale for initiation of
convection is (Kt) 1/2 ( == d), which is a thermal boundary-layer thickness,
450 MARSH

and that convection sets in once d grows conductive1y to some critical


thickness. (This fundamental feature later crops up again when considering
solidification and convection.) Once the layer becomes unstable, the vigor
of convection can increase to any level of vigor, provided the flux of
buoyancy to the system (i.e. the rate of cooling) is sufficient.
On these grounds it is interesting to follow Numax [i.e. Q perfectly mixed
relative to Qcd(t)], which is shown in Figure 2b (Marsh 1 988b). From an
initial value of 2, Numax decreases sharply-nearly linearly-and becomes
less than unity for IJ � 0.5. Evidently, the fluid is stable for all times greater
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

than IJ 0.5, which regardless of solidification would disallow convection


=

in the later stages of cooling (more on this below), and from ( 1 5) the fluid
Access provided by McMaster University on 01/23/15. For personal use only.

is also stable for IJ O. Convection is thus only possible for 0 < IJ < 0.5,
=

and inclusion of solidification (see below) further restricts vigorous thermal


convection to states of superheat.

3 .2.4 HYDROTHERMAL EFFECTS SO far, we have only considered con­


duction in the wall rock, but hydrothermal flows may also be important.
There are two features of this flow important to consider: (a) the transient
time necessary to set up the hydrothermal flow field relative to the time
for the magma to become congested owing to crystallization, and (b) the
magnitude of enhancement of cooling relative to conduction (i.e. how
much Numax is increased).

3.2.4. 1 Transient time Consider the time taken to establish fully the
flow due to porous media convection upward along a tall magma chamber.
Just after emplacement, heat from the body diffuses outward, increases
the temperature of the wall rock, and gives any pore water buoyancy,
which encourages upward flow. A strong upward boundary-layer flow will
eventually develop, and the central question here is the time necessary for
establishment of this flow. (In the true situation, the rock nearest the body
will be strongly scaled from heating, forming a purely conductive envelope
through which heat must diffuse before reaching the hydrous medium. For
the present, the time needed for this process is ignored). For a vertical
heated plate embedded in a porous medium, Cheng & Pop ( 1984) find that
the flow becomes established in a time

( 16)

where y is the ratio of heat capacity of the saturated porous medium to


that of the fluid, Kp is the permeability, and L is the height of the chamber
MAGMA CHAMBERS 45 1
wall; the other symbols are as before but pertain, where necessary, to the
hydrothermal solution.
Taking y �1 , v �3 X 10- 3 cm2 S I , Kp = 1 0 - 12 cm 2 (Norton & Taylor
-

1 979), 9 = 103 cm S- 2, ex = 5 x 1O-4oC-I, AT= 200°C , and L = 1 05 cm,


we find that

ftr �200,000 yr. ( 1 7)

And because it is only the outermost edge of the thermal front that initiates
the flow, the temperature contrast could be smaller by a factor of 4. In
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

addition, the permeability could easily be smaller for crustal metamorphic


rocks (Norton & Taylor 1 979), such that Itr could be of order 10 6 yr. It is
Access provided by McMaster University on 01/23/15. For personal use only.

interesting to compare this time with the conduction time ( 1 ) for cooling
by, say, 1 0% of a body of characteristic size L, for which C �1 0 - 1 and
the body would be crystallized to the point of a mush and be nondynamic
(see below). The ratio of these two times is

Itr 1. 707Kyv
�= �50' ( 1 8)
ted KpgexATCL
where K = 10- 2 cm2 S-I and all else is as above. The large value of this
ratio, which in reality may be even larger, clearly suggests that the magma
will be congested by crystals and have passed its dynamic stage long before
the hydrothermal flow becomes established. Adding also the time taken
for the initial heat to cross the sealed, conductive region nearest the body
and to reach the hydrothermal medium further increases this measure.
Although porous flow early on is ineffective in enhancing cooling, once
established there is no doubt that cooling is greatly increased. These
transient effects and the role of hydrothermal circulation in possibly
quenching magma at the contact have been considered as problems in
conjugate (i.e. hydrothermal-magma) convection by Bergantz & Lowell
(1 987).
In a paper deserving more appreciation, Lister ( 1974; see also Carrigan
1 986) has considered the heat transfer from a hydrothermal flow associated
with a propagating fracture front moving with the thermal anomaly. He
explicitly includes the effect of a conductive envelope inside the fracture
front but does not include the transient effect of establishing the hydro­
thermal flow itself, which is taken from the start to be a fully established
boundary-layer flow. Even under these conditions, the study shows that
the main thermal resistance is still in the wall rock. When the fracture
front moves inward with time, however, it is tantamount to a flow associ­
ated with an inwardly moving heated wall, which may never escape the
452 MARSH

transient period because the start-up period is renewed at every new posi­
tion of the wall. Motion of the wall or fracture front stabilizes the flow
against establishment of normal boundary-layer flows (Bergantz 1 988).
Upon emplacement of the pluton, there is an early period of wall-rock
heating when the thermal anomaly propagates outward and the wall rocks
expand owing to thermal stresses and are thus sealed against hydrothermal
flow (e.g. Lister 1 974, Norton & Knight 1 977, Burnham 1979, Marsh
1 982). At some point the thermal regime (e.g. the solidus isotherm) begins
to collapse back in upon the body. This is when the fracture front advances
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

inward, eventually fully traversing the body and allowing hydrothermal


fluid to circulate freely throughout the body. This is the subsolidus stage
Access provided by McMaster University on 01/23/15. For personal use only.

so well delineated by H. P. Taylor and associates using oxygen isotopes


(e.g. Taylor & Forester 1 979).

3.2.4.2 Enhanced heat transfer The overall effect of hydrothermal cir­


culation on the thermal history of magma chambers and plutons has been
assessed in several numerical studies (e.g. Cathles 1 977, Norton & Knight
1 977, Norton & Taylor 1 979). Norton & Taylor's ( 1979) study of the
Skaergaard intrusion used a detailed distribution of wall-rock permeability
and found generally that "conductive transfer of energy from the pluton
occurred at a similar rate in all three numerical models . . . for the initial
300,000 years. . . . Even though convective heat transfer was taking place in
[only two of the three models], it removed a minor part of the energy from
the plutons. The factor of 10 difference in permeability between [two of the
models] made only a small difference in the cumulative heat transfer from the
pluton." (They also found that even for a body aspect ratio of only about
2, the conductive cooling rate was almost exactly the same as that given by
an anaJytic mo�el of a horizontal infinite sheet.) These findings are also
consistent with the purely conductive numerical study of Alae lava lake (Peck
et al 1 977), where the effect of heavy Hawaiian rainfall seems to have had
little influence on the overall solidification time. The rainfall does of course
strongly affect the subsolidus temperature distribution at later times.
The moral of the hydrothermal story seems to be that such effects are
essentially unimportant during the earliest times of magmatic cooling,
when the body is most dynamic in terms of magmatic processes. By the
time that hydrothermal convection becomes important, the magma has
crystallized to the point that it is a mush and, dynamically speaking,
relatively dead. These conclusions hold for the extreme case of fully satur­
ated, fully permeable wall rock. With increasing depth in the crust, the
degree of saturation and the permeability both decrease, especially with
increasing metamorphic grade and geologic complexity, and these effects
become even less important.
MAGMA CHAMBERS 453

4. RHEOLOGY AND DYNAMICS OF


SOLIDIFICATION FRONTS
4.1 Crystallinity and Rheology
Perhaps the single most important physical feature of magma is its monu­
mental change-by a factor of about 1 0 16-in viscosity during solidi­
fication. Basalt, for example, at its liquidus temperature has a viscosity of
about 1 02 poise, whereas at its solidus it is about 1 0 18 poise [see the useful
reviews by McBirney & Murase ( 1 984) and Ryan & Blevins (1 987)], and
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

this change occurs over a variation in temperature of only about 200"C.


Magma viscosity is a strong function of the degree of crystallinity and
Access provided by McMaster University on 01/23/15. For personal use only.

interstitial liquid composition. Maximum packing of solids for most


basaltic magmas occurs near a crystallinity N of about 0.55 (volume
fraction), which separates a dilatant material (N) 0.55) from a highly
viscous, albeit certainly non-Newtonian, mush (N < 0.55). The crystal­
rich material is dilatant because it expands under shear. In drilling the
crust over solidifying Hawaiian lava lakes, for example, until the point of
critical crystallinity (N > 0.55) is encountered, the crust acts as a stiff, rigid
solid (e.g. Wright & Okamura 1 977, Helz 1 987). Beyond this point, the
drill stem can be pushed ahead by hand. The exact viscosity at this point
is not known, but it is certainly many orders of magnitude above that at
the liquidUS. A striking example of this sudden change in rheological
behavior is the observation that lavas never carry more than about 55%
phenocrysts, and the more viscous the interstitial liquid, the lower this
maximum observed crystallinity (Marsh 1 98 1 ).
Convection within fluids whose viscosity varies strongly with tem­
perature has been found by experiment and theory to be confined to a
region in which the viscosity varies by no more than a factor of about 1 0
(Stengel e t al 1 982, Richter e t a l 1 983, Morris & Canright 1 984, Jaupart
& Parsons 1 985, Smith 1 988). All cooler fluid effectivcly becomes a part of
the container holding the fluid. This is a critical finding for understanding
magmatic convection, for it delineates the freely convectable portion of
the magma chamber-that is, it partitions the thermal boundary layer or
solidification front into a part that is locked within a rigid (in terms of
convection) crust and a part that is able to participate in convection. For
tholeiitic basalt this factor of 10 increase in viscosity occurs after about
20-25% crystallization (i.e. N � 0.20-0.25).
The solidification front can thus be classified into three parts (Figure 3):
1 . rigid crust, 1 .0 � N � 0.55,
2. mush, 0.55 � N � 0.25,
3. suspension, 0.25 � N � 0,
454 MARSH
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

....
10 110
Crystal Free Magma �
Access provided by McMaster University on 01/23/15. For personal use only.

�..
c
C;

Figure 3 The rheological regimes about the margins of a magma chamber. The viscosity
(fl) variation is shown with flo as the liquidus viscosity and IOf1o as thc viscosity at the
capture front.

where all crystallinities are approximate (as they depend on the specific
magmatic composition and phase equilibria). The boundary between sus­
pension and mush is that found from variable-viscosity convection studies,
and it further separates magma into two gross regions: an inner region
(N ::;; 0.25) where all physical processes are dominated by inherently fluid­
like behavior; and an outer, cooler, much more viscous region dominated
by a crystalline mesh or network of solids (Marsh 1 988b). A similar
division has been suggested by Peterson ( 1987). In terms of convection in
the usual sense, all material outward of (i.e. cooler than) the suspension is
rigid; fluid flow is, however, possible, as in porous media.

4.2 Crystal Capture


Crystals within the suspension can move as a result of gravity and possibly
fluid flow, but motion within the mush becomes increasingly difficult owing
to strongly increasing viscosity. Crystals can escape entrapment by the
advancing mush by becoming sufficiently large within the suspension such
that their settling velocity is larger than the rate of advance of the leading
edge of the mush. Thus thc boundary between suspension and mush is
called the capturefront.
In an initially crystal-free body, crystals nucleate at the inwardly prop­
agating liquidus and grow through the suspension zone. The time avail­
able for growth is set by the time taken for the capture front to advance
through the suspension thickness. Because cooling slows with time, this
MAGMA CHAMBERS 455
time becomes progressively larger, and the mean crystal size near the
capture front accordingly increases. Some crystals eventually escape the
capture front and settle deeper into the magma (Marsh 1 988c). At greater
depths, however, the magma is hotter and these crystals suffer resorption.
Mangan & Marsh ( 1988) find that only after about 60% solidification of
the whole body (assumed sheetlike) can these single crystals actually reach
the upward-growing lower crust and form a cumulate, which because of
absorption is actually a microcumulate. The surprising result of this study
is that initially crystal-free magmas cannot undergo strong chemical
differentiation by fractional crystallization of singly settling crystals. The
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

study by MacLeod ( 1 98 1 ) of an initially crystal-free gabbro sill seems to


Access provided by McMaster University on 01/23/15. For personal use only.

support this conclusion.


Single crystals settle slowly and are highly vulnerable to resorption, but
if crystal-laden plumes form in the upper suspension zone, they may travel
much faster. Such plumes have been suggested by Brandeis & Jaupart
( 1986) based on results of their experiment on sudden cooling of uniform­
viscosity fluids (Jaupart et al 1 984, Jaupart & Brandeis 1 986). Stemming
from the ideas and experiments of Bradley ( 1 965) on lake sedimentation,
Marsh (1 988c) has suggested a similar process of transport and sorting of
crystals.
In initially crystal-rich intrusions, only those phenocrysts originally near
the roof can be captured by the downward-growing crust (Figure 4).
Eventually all phenocrysts deeper than some critical initial depth-pro­
vided that they are heavy-will escape capture and go on to form a basal
cumulate within the upward-growing lower crust. An initially homo­
geneous distribution of, for example, bronzite phenocrysts will be increas­
ingly depleted downward from the roof and concentrated upward from
the floor. This process was apparently first recognized by Osborne &
Roberts (193 1 ) in studying the Shonkin Sag laccolith, which has both
basal and roofward (leucite) cumulates. Many others have subsequently
recognized the importance of this process that gives rise to broadly S­
shaped profiles of modal variations within sills and lava lakes (e.g. Evans
& Moore 1 967, Jaeger 1 968, Wright & Okamura 1977), and Gray & Crain
(1 969) and Marsh ( 1 988c) have examined some of the mechanics of the
process.
Plume formation would seem to hasten settling of these initial pheno­
crysts and encourage stronger concentrations of basal cumulates. What
seems to be more common, however, is less concentration than what would
be expected even by settling of single crystals. Why this is so is not entirely
clear, but in the Kilauea Iki lava lake, modal olivine near the roof is highly
correlated with vesicle content (Mangan & Helz 1 985, Helz 1 987), which
suggests that either hindered settling or actual buoying up of individual
456 MARSH
o.o�--�-� ,--___--, �--�-_

5
! 0.5


1 .0 ...l-__.L-_.......J '--_----'-__--' L__-L-_.......J

0/0 Phenocrysts % Phenocrysts 0;., Phenocrysts


- - -
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

(A) (9) (C)

Prospect Intrusion Shonk]n Sag Prehistoric Makoapuhi Tasmanian Dolerite


Access provided by McMaster University on 01/23/15. For personal use only.


1i
.!./
5 0.5
�0 II

�"
I
1.0
10 20 20 40 60 10 20 30 2.8 2.9 3.0
% olivine % mafics % olivine density (g/cm 3)

Figure 4 (top) The possible final modal distribution of phenocrysts in a layer due to a rate
of solidification that is (A) infinite, (B) zero, and (C) time varying. (bottom) Observed
phenocryst distribution in some well-known sills.

crystals by bubble attachment is occurring. In large intrusions, it is remark­


able to realize that capture is still only possible near the roof and that an
initially 2-km-thick magma containing, say, 1 5% phenocrystic olivine
would, considering porosity, form a basal cumulate of some 600 m before
much upper crust could grow at all. Adding to this the usual history of
upper- and lower-crust growth, later field studies would show a body
that did not produce much upper crust until late-as measured by the
cumulates-in its cooling history. Through careful understanding of such
processes in large sills and their subsequent scaling to large intrusions, we
can add much to our understanding of magmatic processes.

4.3 Differentiation
The strong role of solidification fronts in controlling magmatic evolution
may be evident, for example, in Hawaiian lavas. A compilation using
Wright's ( 1 971) data for summit lavas erupted from Kilauea volcano
(Figure 5) shows a consistent truncation in compositional evolution at
about 7 wt% MgO and 5 1 -52 wt% Si02, a result that has been also noted
MAGMA CHAMBERS 457
53 -r-------,

52
----------------------------------------------..-------- '-
A

51
i�
Lavas ... :�.
, ....
.00 D O ·
O
l1li OO
Jt
°o: w.
50 l1li
o .0
�l1li:
0

(II
l
••
a
o
en 49
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

4B ... .

.
Access provided by McMaster University on 01/23/15. For personal use only.

47

46 -r----.---_.--�,_--�.�
25 20 15 10 .... ." 5
.·rvtgO

0 ��-r--_r--_.�--6r_--�--_r--_;3

10

..
CD


t:
iii
20�_____,,_--�_,,_------_r-L--�
400 BOO 1200

Temperature °c

Figure 5 Truncation of Kilauea lava composition near MgO � 7.0 wt% and Si02 � 5 1-52
wt% (top) in relation to the interstitial melt composition observed in Kilauea Iki lava lake
(bottom).

independently by Murata & Richter (1 966). The variation in MgO up to


this truncation point has been shown by Murata & Richter ( 1966) to
correlate with eruptive output: the larger the eruptive flux, the larger the
MgO content. They liken this to the similar process of sediment transport
in rivers, which suggests that it is the nature of the transport of magma
458 MARSH

that determines its phenocryst content at the moment of emplacement or


eruption. The magical truncation composition, however, is another matter.
Also shown in Figure 5 are the magma compositions encountered in the
eruptions leading to the formation of Kilauea Iki lava lake and the liquid
compositions found in subsequent drilling of the lake itself (Helz 1987).
As near as can be ascertained, the capture front of Kilauea Iki lies exactly
at this truncation composition. Much more evolved liquids reside outward,
within the mush and rigid crust, but were this an active subterranean
chamber they would be inaccessible to concentration and eruption. The
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

chemical evolution of the system is thus restricted to that which can be


effected by the removal of the initial phenocrysts. In some systems, these
Access provided by McMaster University on 01/23/15. For personal use only.

late liquids might be denser than the underlying magma (i.e. suspension)
and escape, being replaced by deeper, less evolved liquids. In Kilauea Iki
this process may be occurring in the lower crust, where small diapirs of the
siliceous interstitial liquid seem to have migrated upward and to have been
incorporated into the upper crust (Helz 1 988). In larger bodies compaction
of the crystalline network of the lower mush and rigid crust may expel this
evolved liquid much more efficiently (Sparks et a1 1 985, Kerr & Tait 1 986).
Thus, the so-called liquid line of descent of magmas may be controlled by
two processes: (a) Transport by way of phenocryst carrying capacity gives
the breadth of chemical evolution, and (b) the capture front sets the
truncation composition beyond which further evolution is by processes
within the solidification fronts themselves. The volumes of refined liquid
are apt to be disproportionately smaller for mush and crust processes
(Marsh 1 988a), and they are much less accessible to eruption.

5. CONVECTION WITH SOLIDIFICATION


5.1 No Solidification
A layer of fluid of uniform viscosity strongly cooled from the top and
bottom, as shown by Jaupart & Brandeis ( 1 986, Jaupart et al 1 984), first
forms upper and lower conductive boundary layers, the upper one of which
soon becomes unstable and gives forth cascades of plumes. At first the
plumes are small and penetrate only a part of the layer, but with time they
strengthen and reach the floor, joining other cool fluid already there
from basal cooling to form a basal stagnant or stable zone of fluid. The
continuous, haphazard return flow to the roof promotes systematic cooling
throughout the thermally well-mixed layer (Figure 6). The central tem­
perature decreases systematically with time, and convection weakens in
response to the falling temperature contrast between the boundaries and
the center of the layer. If a magma chamber should convect and cool in
this fashion, it is important to realize that crystallization would take place
MAGMA CHAMBERS 459

MEAN HORIZONTAL TEMPERATURE


0.0 0.5 1.0 0.5 1.0
0.0 0.0

UPPER BOUNDARY LAYER


(UNSTABLE)
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

J: WELL-MIXED LAYER
l-
Access provided by McMaster University on 01/23/15. For personal use only.

ll. 0.5 0.5


W 77
. 4.5

ZONE OF PENETRATION

1.0

Figure 6 The thermal history measured experimentally for a fluid layer of constant viscosity
cooled from both above and below (after Jaupart et al 1984). Note the strong variation in
central temperature with time.

systematically everywhere within the body, although it would be stronger


at the margins. The cooling time itself is rapid; the constant in Equation
(1), for example, was found experimentally to be C = 0.077, which is much
less than that for the time required to cool the same, but stagnant, layer
by conduction (C '" 1 .0). If this constant-viscosity layer was sandwiched
between infinitely thick solids, cooling would of course take longer in both
cases because of the progressive thermal saturation of the bounding solids
with time.
The vigor of this convection must be measured by an appropriate Ray­
leigh number, but it is not clear that the governing length scale is that due
to the full layer thickness. Using such a length scale would imply that the
source of buoyancy driving the flow is always distributed across this length,
either as internal heat sources or as a thermal gradient whereupon cold
and hot boundary layers are coupled across this length to drive the flow.
In addition, this is a transient, inherently unsteady convective state con­
necting an initially stable (i.e. isothermal) state with a final stable state;
the convection never reaches steady state. That is, the time to reach steady
convection in a layer heated from below, for example, is on the order of
460 MARSH

the conduction time through half the layer thickness (i.e. Kt/L2 0.25), �

when here the layer is once again stable in a time Kt/L 2 0.077. On these

grounds the usual parametric relations [e.g. Equations (6) and (7)] may
not be exactly applicable except in a time-averaged fashion. Nevertheless,
in the analysis of these experiments, Jaupart & Brandeis ( 1986) find that
the layer generally cools almost as fast as would be predicted using (6)
with an exponent b 1 /3 . The fit is better if the slow cooling during the
=

start-up is ignored, but the overall rate of cooling is still broadly predictable,
although presumably this could be made better through a more appro­
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

priate choice of the exponent b in (6); it is commonly found in experiments


that b is not exactly 1/3 (e.g. Peltier 1 985, Marsh 1 988b), which makes the
Access provided by McMaster University on 01/23/15. For personal use only.

convective heat transfer dependent on layer thickness.


The key feature in promoting rapid convective cooling here is the insta­
bility of the entire cool thermal boundary layer. In plunging downward,
it allows cold fluid to exchange heat with the deeper and hotter inner fluid
while at the same time bringing hot fluid to the cool boundaries.

5.2 Solidification
Solidification at the boundaries of a layer stabilizes the coldest, densest
fluid, preventing it from falling back into the fluid; convection and cooling
become less vigorous. Consider, for example, a layer consisting of a molten,
single-component fluid (say, Si02) initially at a uniform temperature. If its
initial temperature is also its liquidus temperature, any amount of cooling,
however small, produces a solid. Both boundary layers are always solid,
and there is no cool fluid anywhere in the system that is able to become
unstable and promote convection and cooling throughout the layer. There
is thus no convection, the layer is stagnant, and the central temperature
remains constant until the solidification fronts meet. The layer cools by
conduction.

5.2. 1 SUPERHEAT If the layer is initially superheated, on the other hand,


convection is possible, but instability now depends on the rate of solidi­
fication. The inward-propagating thermal boundary layers are no longer
completely solid. A certain thickness of the outermost part is cool fluid
that is unstable and can cause convection much as in the Jaupart &
Brandeis ( 1 986) experiments. The central temperature will thus decrease
with time as convection uniformly cools the inner region of fluid. Hot fluid
arriving near the roof will retard advancement of the upper crust, whereas
the cool, stagnant basal zone will hasten growth of the lower crust. The
ever-thickening upper crust and the diminishing inner temperature pro­
gressively reduce the outward heat flux, causing convection to wane. As
soon as the superheat has been convected away, the fluid is once again at
MAGMA CHAMBERS 461
its liquidus and convection ceases. There is no temperature gradient across
the inner fluid, both top and bottom temperatures are constrained by the
liquidus temperature, and convection and solidification occur sym­
pathetically. As long as the contact temperature is below that of the
liquidus, solidification will proceed, for the layer is cooling and all dynamics
can be thought of as a reflection of this advance. Thus, no advance of
solidification means no cooling. Solidification could be artificially halted
by supplying a heat flux at the base equal to that outward at the roof.
Convection cannot reverse or stop solidification unless such a heat source
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

is present, but of course magmas typically cool from all sides.


The role of solidification in suppressing convection was shown ana­
Access provided by McMaster University on 01/23/15. For personal use only.

lytically by Smith ( 1 988), who studied sudden cooling of a very thick layer
of fluid whose viscosity depends strongly on temperature. He found for a
single-component fluid, superheated by an amount AT, that the governing
Rayleigh number for instability is (for large Pr)
rxgI1 T(Kj V)3
Ra v = vK � 1 4, ( 1 9)

where V is the velocity of advance of the solidification front, and the


reference viscosity is that at the initial fluid temperature. The governing
length scale for initiation of convection is not the full layer thickness L,
but is instead Kj V, which represents the essential competition between
thermal instability and stabilization by solidification. If V is sufficiently
large, Rav is small, and no convection is possible. The thermal instabilities
are overcome by solidification before maturation to finite-amplitude
motions. With slowing of the advance of solidification, convection sets in
as a series of vertically elongated cells that resemble plumes with a shallow
penetration depth (Figure 7). For a basaltic magma with v = 103 cm2 s- I ,
the critical velocity is

(20)

which shows the inherent insensitivity to the temperture contrast (which


in a strict sense is unknown). For a granitic magma, with v ::::: 1 06 cm2 S - l ,
V is only a factor of 1 0 smaller. These velocities are rather large ( "-' 1 0 km
yc 1 ) and unlikely to be exceeded except in the earliest stages of cooling
after emplacement, during which time chill zones are formed. (The chill
zone itself does not of course necessarily reflect convective stability, but it
is a kinetic effect of crystallization.) The approximate thickness of the chill
zone when convection commences can be estimated by noting that the
thermal boundary layer (d) grows as d E(Kt) 1/2, where E is a constant
=
462 MARSH

III
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org
Access provided by McMaster University on 01/23/15. For personal use only.

Fi,qure 7 Onset of convection (i.e. streamlines) in a superheated layer solidifying by cooling


from above (after Smith 1988): (left) constant viscosity, where solid-liquid interface is
across the top margin; (right) variable viscosity, showing the stagnant high-viscosity lid and
confinement of the flow to a regime where the viscosity varies by a factor of 3 .

(i.e. 2, 3, n, etc.) that depends on the detailed geometry of the boundary


layer. Thus the critical velocity is related to this by d = E2K/2 V. For the
velocity given by (20) with t::.. T = 1°C and E � 2, we have d � 10 cm.
However, since t::.. T could be much smaller (i.e. t::.. T « 1), convection could
begin much later, at much smaller values of V, and the corresponding
crust thickness would be much larger (see below).
The stability criterion ( 1 9) and the nature of the flow only apply to the
onset of convection. With time, especially as solidification slows, con­
vection will grow stronger and, providing that some superheat remains,
plumes may reach into the deeper parts of the layer, cooling the entire
layer (as in Figure 6), whereupon convection is characterized by a much
larger length scale.
5.2.2 NO SUPERHEAT An almost ubiquitous feature of lavas is the pres­
ence of phenocrysts, however small, which suggests that truly superheated
magmas are rare. The general case of concern to petrologists is thus a
magma emplaced at its liquidus temperature, with or without phenocrysts.
Consider, then, a layer of magma of uniform liquidus temperature that
suddenly begins cooling from top and bottom. In all respects except one
(possibly two), this is exactly the same situation as for solidification of a
single-component melt. The exception is the presence now of a liquidus
and solidus separated by the crystal-liquid rheological transitions of Figure
3. (The other possible exception is the density of the residual liquid, which
for magmas can be either less than or greater than the solids.) The critical
question here is, What is the temperature of the convective liquidus? That
is, how much of the crystallizing boundary layer is able to participate in
MAGMA CHAMBERS 463
convection? As explained before, if liquidus and solidus were one and the
same, there would be no convection at all.
As mentioned earlier, the results of experiments on layers heated from
below suggest that the flow is contained in a region where the viscosity
varies by no more than a factor of about 1 0, but this is for a vigorous
flow driven by heating. For more passive flows this possible variation in
viscosity may be less. There is some suggestion of this in the results of
Smith ( 1 988), who found for the weak flows at initial instability that the
flow had essentially ceased when the viscosity had risen by a factor of ,...., 3
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

over that of the deep (superheated) interior.


To examine this possible effect, Brandeis & Marsh ( 1 988a,b) solidified
Access provided by McMaster University on 01/23/15. For personal use only.

superheated paraffin to measure the position of the convective liquidus


relative to the true liquidus. Two types of fluid were used: (a) high-purity
n-nonadecane with a melting point at 3 1 .9°C and a small crystallization
interval ( ,...., I-2°C) due to sluggish crystallization kinetics and impurities,
and (b) a paraffin mixture with a crystallization interval of 4-5°C and
a liquidus at ,...., 22°e. The otherwise insulated, isothermal, and initially
superheated box of fluid (20 x 20 x 20 cm) was suddenly cooled from
above by chilled water circulating through a copper plate. A crust of solid
immediately formed, whereas small convective plumes commenced a few
minutes later and eventually grew to an intensity proportional to the
amount of initial superheat. Small (I-2°C) amounts of superheat produced
convective flows confined approximately to the upper half of the tank
(Figure 8).
In every instance, the convective liquidus was found, as near as can be
measured (± 0.1 0c), to be at exactly the same temperature as the liquidus of
the material itself. Once this temperature was reached, convective motion
became all but undetectable, the only motion apparently being due to
slight sidewall cooling by the ambient room temperature. The entire region
inward of the crust remained isothermal, crystal free, and highly fluid until
being encountered by the crust growing by conduction alone. Relative to
the time for complete solidification ( ,...., 8 days), the initial superheat was
lost in a matter of a few hours. An actual magmatic system could thus be
expected to spend an exceedingly brief, essentially unrecognizable time in
the highly convective stage I.
It is clear from these results that although significant convective heat
transfer did not extend to temperatures below the liquidus, the convective
liquidus in magmas may be more difficult to determine and may possibly
vary significantly relative to the usual liquidus with bulk composition. The
only possible natural information on this issue comes from the thermal
history of the Hawaiian lava lakes, which have central temperatures that
vary in time almost exactly as do those seen in the experiments of Brandeis
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org
Access provided by McMaster University on 01/23/15. For personal use only.

464
MARSH

«
a:
w

a:

I-
w
:2
(L
w
I-
::::>
.e:......
0
a

30
35
40

20
25
stage
11 -

0.1
0.2
0.3

D I M ENS ION LESS TIME [ K t/(4 d .d)]


MAGMA CHAMBERS 465
& Marsh. That is, once a crust forms over the lake, the interior temperature
seems to be essentially constant with time until being encountered by the
thermal regimes associated with the growing crusts themselves (Figure 9).
This thermal history is in striking contrast to the early thermal history
calculated for vigorous convection of ponded komatiite lava by Turner et
al ( 1 986), which is allowed to convect through its liquidus until it reaches
a significantly large crystallinity, and also to that found experimentally in
the constant-viscosity experiments of Jaupart et al ( 1984). This evidence
may suggest that the convective liquidus for these tholeiitic basalt magmas
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

is also very nearly the actual liquidus itself. A change in viscosity by a


factor of 3, if one considers the inherent uncertainties in magma rheology
Access provided by McMaster University on 01/23/15. For personal use only.

(McBirney & Murase 1 984, Ryerson et al 1 988, Spera et al 1 988), is so


small that such a variation is essentially at the rheological noise level. And
ordinarily small effects, such as yield strength and crystal nucleation, may
in the limit of very slow flow have a major influence on the convective
liquidus. Thus, strictly speaking, if magmas are emplaced already at their
convective liquidus, the effective !1T driving convection is arbitrarily small,
and unknown, and therefore Ra is always small (e.g. 0 to 1 00) regardless
of the size of the chamber. The critical time for initiation of convection,
as modulated by solidification, may extend over a large part of the full
solidification time.

5.3 Wall Flows


Convection in tall bodies is dominated by flow along steeply inclined or
vertical walls. These flows may be important in forming double-diffusive
systems (see reviews by Huppert & Sparks 1 984, Turner 1 985), in con­
trolling heat and mass transfer (e.g. Spera et al 1 984, Nilson et al 1 985),
and in controlling the distribution of solids near walls (Irvine 1980, Thomp­
son & McBirney 1 985). The standard method of analysis is to use bound­
ary-layer methods, which show that the flow is strongly dependent on
the height of the wall. This is so because for steady, well-developed flows

Figure 8 Convection and cooling during solidification of n-nonadecane. (a) Variation of


temperature with time (dimensionless) at various depths ( 1 .95, 5.35, 8.9 cm) in the tank from
an initial temperature of 36°C and an upper-plate temperature of7 AOC. The loss of superheat
is rapid (stage I) relative to the overall solidification time (stage II). (b) Streamlines of
convection in the paraffin mixture as depicted by time-lapse photos and aluminum flakes.
Notice the downward-growing crust at the top of each picture, which attains approximately
2 cm in thickness. As the temperature decreases to the liquidus, convection wanes and is
increasingly confined to the upper part of the tank. This series of pictures spans a period of
about 4 hr, beyond which the entire tank is essentially stagnant except for the flow due to
unavoidable heat loss through the walls (from Brandeis & Marsh 1988b).
466 MARSH

0 . 0 0 2 5 0 . 0 1 0 0 . 0 2 2 5 0 . 04 0 . 0 6 2 5 0 . 0 9 0 . 1 2 2 5 0 . 1 6

0.9

Centra l
0 8
Temperature .
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

T •T 0.7
Access provided by McMaster University on 01/23/15. For personal use only.

TI =�
c T · T
m W 0.6

0.5 well mixed &


J BA overburden

O . 4 ..__________�__��__________���
o 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

TJ =
J4Kt
L

Figure 9 The variation i n central temperature (dimensionless) with time (dimensionless) for
various bodies under the indicated conditions. Curve JBA ( 1 984) represents the experimental
results of Figure 6; curve THS (1 986) is the calculation of Turner et al ( 1 986) for cooling of
a 1 00-m-thick komatiite lava lake with a convective liquidus well below the usual liquidus.
The letters near the top (i.e. M and A) represent actual measurements from Hawaiian lava
lakes (see Marsh 1988b). Notice the close agreement between the experimental results of
Figure 8 (i.e. temperature with time) and the temperature history of Hawaiian lava lakes.
It is also possible, nevertheless, that the lake temperatures may decrease slowly with time.
The uncertainty in the present data does not rule this out for later times.

the thermal boundary layers thicken with distance down the wall. This is
generally a good approximation because, as discussed by Spera et al (1982),
the transient stage leading to boundary-layer development is geologically
rapid (10-100 yr). During this time an initially purely conductive, cool
border zone of constant thickness progressively sloughs from the wall and
is replaced by a thickening downward thermal boundary layer. When there
is solidification along this wall, this transition time may be much longer.
Solidification acts in essence as the continual formation of a new wall
along which the transient state must again propagate. At either a
sufficiently large solidification rate or small Ra, the transition point divid-
MAGMA CHAMBERS 467
ing a true convection-dominated boundary layer from a conduction-domi­
nated thermal front may stay at the top of the wall and the system may
cool mainly by conduction. This general situation has been considered by
Bergantz & Marsh ( 1 988), who show that through a transformation of
coordinates centered on the moving solidification front the effect of solidi­
fication becomes completely analogous (mathematically) to fluid flow or
suction through the front itself. They find that the sidewall flow partitions
itself into two flow regimes: a convection-dominated one, where the bound­
ary layers open outward and downward; and a suction-dominated one,
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

where the boundary layers are of constant thickness. This same feature
appears in the solidification of horizontal layers (e.g. Smith 1988). With
Access provided by McMaster University on 01/23/15. For personal use only.

this transformation, the energy and momentum equations become

(21 )

(22)

where Vo is the velocity of the solidification front, y is the coordinate


normal to the front, U is the fluid velocity along the wall, and Ts is the
solidification temperature.
Because the far-field temperature is above that of the liquidus, a thermal
front develops in the fluid. Its length scale is, from (21),

(23)

and the governing Rayleigh number becomes the same as ( 1 9), which also
has the length scale KI Vo. Thus Rav here again becomes independent of
the height of the system, which is a direct reflection of the ability of
solidification to keep the system essentially always in the transient stage.
The position Yc along the wall where the flow changes from the usual
boundary layer to solidification dominated is found by equating the thick­
ness of the thermal boundary layer to that of the thermal front:
K
Yc � Ra v· (24)
Vo

The larger the solidification rate, the less important will be the usual
thermal boundary layers.
The possible importance of solidification in the cooling of magma cham­
bers may be evaluated in the field by studying the temporal progression
of the thickness of the sidewall crust. Separation of the effects of crystal-
468 MARSH

debris flows may show whether or not the crust grows with relatively
uniform thickness, which would be the ideal case of solidification domi­
nation. The effect of nonvertical and irregular walls must also be evaluated.
Once again, however, it is likely that no strong wall flows will develop for
magma emplaced either at or near its convective liquidus because of the
strong stabilizing effects of solidification.

5.4 Mantle Convection


There is a close similarity between downward solidification of sheetlike
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

magma chambers and small-scale convection beneath downward-growing


lithosphere. Buck & Parmentier ( 1986) showed in a numerical study an
Access provided by McMaster University on 01/23/15. For personal use only.

eventual flow field similar to that of Smith (1988; see Figure 7). Using
(20) and Ra � 1 06, convection would be expected to commence once
lithosphere growth slowed to - 0.35 cm yr- 1 , which is when the thermal
thickness of the lithosphere is about 55 km. Sublithospheric convection
would not be expected near the ridges, but in the Pacific, for example, it
would commence certainly by the time that the plate had reached Hawaii.
This small-scale convection would be strongest at initiation, would wane
strongly with time, and at late times would be augmented by heating from
below by the underlying, deeper mantle.

6. ROOF MELTING

Mafic sills and plutons occasionally show clear evidence of either melting
or slow solidification at the roof. This evidence is often in the form of
patches and layers of granophyre that may have come from melting of
stoped blocks or from melting in place. These features are important to
understand because of the possible light they may shed on the intrusive
cooling history and rate of solidification.
Huppert & Sparks ( 1988) present an analytical and experimental study
of this problem by evaluating the dynamics of the melting of a wax layer
overlying a hot aqueous solution. The critical feature in promoting melting
is to keep the upper contact temperature as high as possible, which can be
done in this case through vigorous convection in the aqueous solution.
From their analytical treatment of the experiments, Huppert & Sparks
then extend their results to the geological problem of granite overlying a
layer of basalt. They find that a basalt layer of SOO-m thickness undergoing
turbulent convection from an initial temperature of 1 200°C to a non­
covectable state of mush at 1 091 °C (i.e. they place the convective liquidus
at a crystallinity of 55%) will melt about 300 m of granite if its initial
'"

temperature is everywhere 500°C. After 7 days the granitic layer of melt,


MAGMA CHAMBERS 469
which is then only a few meters thick, begins to convect and thereafter is
itself assumed to be in turbulent convection.
Interestingly, even if both layers (i.e. wax and water) were solid, the wax
would show some melting. This is because the contact temperature of two
instantaneously juxtaposed solids of initial temperatures TJ and T2, if we
ignore long-term conduction effects, is

(25)
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

where p is density and Cp is specific heat. For identical material properties


Access provided by McMaster University on 01/23/15. For personal use only.

T is given by (TJ + T2)/2; for these experiments, we have T = (65°C +


1 5°C)/2 � 40°C, which is slightly above the wax melting point. And if
we consider the large heat capacity of water (i.e. p J CP 1 � 2p 2Cp),
T is found to be even larger ( � 48°C); including the latent heat of the wax
will lower these temperatures slightly. Any water temperature less
than about 55°C probably would not allow melting. Sustained melting,
however, is only possible through convection in the underlying layer.
In light of the earlier discussions on solidification and superheat, it is
clear that these results are most applicable when the basaltic magma is
initially superheated. This heat is vigorously driven off by convection and,
providing the roof rock is initially hot enough, could well cause melting.
Once the basalt reaches its convective liquidus, which may be near its
actual liquidus, convection is exceedingly weak and cooling is essentially
everywhere by conduction. The critical issues of interest in combining
these models are to incorporate solidification and little superheat into the
Cambridge (e.g. Huppert & Sparks 1 988) aqueous magma models and to
evaluate the conditions of roof crust instability into the Johns Hopkins
(e.g. Marsh 1 988b,c) viscous solidification models. It would also be of
much interest to repeat the roof-melting experiments with an initial fluid
layer of molten wax. In order to melt the roof material, the contact
temperature, just after emplacement, must be above the melting tempera­
ture, but at the same time the underlying molten material must not be
quenched too strongly against the roof. The best means of doing this is to
have the highest possible initial temperatures for both layers, which
suggests that the roof rock should be preheated or reside deep in the crust
(Marsh 1 988b). However, typical rocks in the deep crust are dehydrated
and of intermediate composition, and thus they are much more difficult
to melt extensively by basalts at normal temperatures.
In passing, it should be noted that the Cambridge and Johns Hopkins
models of magma chambers differ strongly on one, perhaps testable point:
The Cambridge model is dominated by large-Ra convection that causes
470 MARSH

rapid uniform cooling and crystallization of the whole chamber to sig­


nificantly subliquidus temperatures, and convection apparently stops only
when the overall crystallinity approaches 50% (v); the Johns Hopkins
model is characterized by sluggish, small-Ra convection driven by cooling
and crystallization at the leading edges of inward-moving solidification
fronts around the margin of the chamber. The chamber interior remains
essentially crystal free (having lost the initial phenocrysts) and isothermal
at a near-liquidus temperature until encroachment by the solidification
fronts. This latter view seems also to have been envisioned independently
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

by Morse (e.g. 1 988).


Access provided by McMaster University on 01/23/15. For personal use only.

7. QUESTIONS FOR FURTHER RESEARCH

The present state and direction of research on the physics of magma can
be summed up by the following series of questions:
1 . What are the shapes, sizes, and positions of magma bodies beneath
active volcanic systems, both large and small? How do these dimensions
relate to the observed volumes of erupted materials?
2. Is it possible to invert the observed modal and chemical variations
within sills, plutons, and sequences of lavas to give the solidification
and convective history of the body?
3. Is vigorous thermal convection only possible in superheated magmas?
If so, because of the rapid rates of heat transfer, is magma ever likely
to become superheated?
4. What is the role of solidification fronts in modulating convection, in
truncating the liquid line of descent, and in controlling crystal
nucleation and growth? (Can liquid lines of descent be used to infer
magma dynamics?)
5. What is the dependence of the convective liquidus on bulk composition,
phase equilibria, and rheology of the common magmatic compositions?
6. What is the dynamic relationship between solidification fronts along
steep walls and the development of debris flows and crystal and liquid
fractionation?
7. What are the fundamental controls on roof melting?
8. What is the exact evolution in time of the central temperature of lava
lakes?

ACKNOWLEDGMENTS

Reviews and discussions by G. Brandeis, G. Bergantz (who pointed out


the paper by Cheng & Pop), M . Smith, and M . Mangan are appreciated.
Stimulating debates with H. Huppert have been beneficial, as have dis-
MAGMA CHAMBERS 471
cussions with F. Spera and D. Yuen. G. Bergantz and G. Brandeis kindly
helped with the figures. This work is supported by a grant (EAR88 - l 7394)
from the National Science Foundation and grant NAG5-32 MASTER
from the National Aeronautics and Space Administration.

Literature Cited

Bartlett, R. W. 1969. Magma convection, cooling of intrusives by groundwater con­


temperature distribution, and differ­ vection which includes boiling. Econ.
entiation. Am. J. Sci. 267: 1067-82 Geol. 72: 804-26
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

Bergantz, G. W. 1988. Convection and solidi­ Cheng, P., Pop, I. 1 984. Transient free con­
ficat ion in tall magma chambers. PhD vection about a vertical flat plate em­
Access provided by McMaster University on 01/23/15. For personal use only.

thesis. Johns Hopkins Dniv., Baltimore, bedded in a porous medium. Int. J. Eng.
Md. 127 pp. Sci. 22: 253-64
Bergantz, G. W., Lowell, R. P. 1 987. The Daly, R. A. 1903. The mechanics of igneous
role of conjugate convection in magmatic intrusion. Am. J. Sci. 16: 107-26
heat and mass transfer. In Structure and Dawson, P. B., Iyer, H. M., Evans, J. R.
Dynamics of Partially Solidified Systems, 1988. Magma under Long VaHey caldera.
ed. D. E. Loper, pp. 369-82. Boston: Eos, Trans. Am. Geophys. Union I March
Martinus Nijhoff. 506 pp. 1988: 1 24 (Lett.)
Bergantz, G. W., Marsh, B. D. 1988. Solidi­ Evans, B. W., Moore, J. G. 1 967. Olivine
fication as suction in convection along tall in prehistoric Makaopuhi tholeiitic lava
walls. Submitted for publication lake, Hawaii. Contrib. Mineral. Petrol. 15:
Bradley, W. H. 1 965. Vertical density 202-23
currents. Science 1 50: 1 423-28 Fyfe, W. S. 1 970. Some thoughts on granitic
Brandeis, G., Jaupart, C. 1 986. On the inter­ magmas. In Mechanism of Igneous
action between convection and cry­ Intrusion, ed. G. Newall, N. Rast, pp. 201-
stallization in cooling magma chambers. 1 6. Liverpool: Gallery Press. 380 pp.
Earth Planet. Sci. Lett. 77: 345--6 1 Gastil, R. G., Phillips, R. P., Allison, E. C.
Brandeis, G., Marsh, B. D. 1988a. Con­ 1 975. Reconnaissance geology of the state
vection and crystallization of sheet-like of Baja California. Geol. Soc. Am. Mem.
magma bodies: II. Fluid dynamic experi­ 140. 170 pp.
ments. Eos, Trans. Am. Geophys. Union Gray, N. H., Crain, I. K. 1969. Crystal set­
69: 526 (Abstr.) tling in silIs-a model for crystal set­
Brandeis, G., Marsh, B. D. 1988b. The con­ tling. Can. J. Earth Sci. 6: 1 2 1 1-16
vective liquidus in solidifying magma Grout, F. F. 1 932. Petrography and
chambers: a fluid dynamic investigation. Petrology. New York: McGraw-Hill. 454
Submitted for publication pp.
Buck, W. R . , Parmentier, M. 1986. Con­ Helz, R. T . 1987. Differentiation behavior of
vection beneath young oceanic litho­ Kilauea Iki lava lake, Kilauea Volcano,
sphere: implications for thermal structure Hawaii: an overview of past and current
and gravity. J. Geophys. Res. 9 1 : 1961-74 work. In Magmatic Processes: Physico­
Buddington, A. F. 1 960. Granite emplace­ chemical Principles, ed. B. Mysen, pp . 241-
ment with special reference to North 58. University Park, Pa: Geochem. Soc.
America. Geol. Soc. Am. Bull. 70: 671-747 Am. 500 pp.
Burnham, C. W. 1 979. Magmas and hydro­ Helz, R. T. 1988. Diapiric transfer of melt in
thermal fluids. In Geochemistry of Hydro­ Kilauea Iki lava lake: a quick, efficient
thermal Ore Deposits, ed. H. L. Barnes, process of igneous differentiation. Geol.
pp. 71-1 36. New York: Wiley-Inter­ Soc. Am. Bull. In press
science. 798 pp. 2nd ed. Hill, D. P. 1 976. Structure of Long Valley
Carrigan, C. R. 1 986. A two-phase hydro­ caldera, California, from a seismic refrac­
thermal cooling model for shallow tion experiment. J. Geophys. Res. 8 1 : 745-
intrusions. J. Volcanol. Geotherm. Res. 28: 53
175-92 Homsy, G. M. 1 973. Global stability of time­
Carslaw, H. S., Jaeger, J. C. 1 959. Con­ dependent flows: impulsively heated or
duction of Heat in Solids. New York! cooled fluid layers. J. Fluid Mech. 60: 1 29-
London: Oxford Univ. Press. 5 1 0 pp. 2nd 39
ed. Huppert, H. E., Sparks, R. S. J. 1984.
Cathles, L. M. 1 977. An analysis of the Double-diffusive convection due to crys-
472 MARSH

tallization in magmas. Ann. Rev. Earth igneous intrusions. J. Geophys. Res. 9 1 :


Planet. Sci. 1 2: 1 1-37 3591-3608
Huppert, H. E., Sparks, R. S. J. 1 988. Melt­ Lane, A. C. 1902. Studies of the grain of
ing the roof of a chamber containing a igneous intrusions. Geol. Soc. Am. Bull.
hot turbulently convecting fluid. J. Fluid 14: 369-84
Mech. 1 88: 107-3 1 Larsen, E. S. 1945. Time required for the
Hyndman, D. W., Alt, D. 1987. Radial dikes, crystallization of the great batholith of
laccoliths, and gelatin models. J. Geol. 95: southern and lower California. Am. J. Sci.
763-74 243A: 399-4 1 6
Ingersoll, L. R., Zobel, O. J. 1 9 1 3 . Mathe­ Lister, C R . B . 1 974. O n the penetration of
matical Theory of Heat Conduction. Bos­ water into hot rock. Geophys. J. R. A stron .
ton: Ginn & Co. 2 1 2 pp. Soc. 39: 465-5 1 0
Ingersoll, L. R., Zobel, O. J., Ingersoll, A. C. Lister, C. R. B. 1 9 8 3 . On the intermittency
1954. Heat Conduction With Engineer ing , and crystallization mechanisms of sub­
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

Geological, and Other Applications. Madi­ seafloor magma chambers. Geophys. J. R.


son: Univ. Wisc. Press. 325 pp. Astron. Soc. 73: 3 5 1-66
Access provided by McMaster University on 01/23/15. For personal use only.

Irvine, T. N. 1 970. Heat transfer during Lovering, T. S. 1 935. Theory of heat con­
solidification of layered intrusions. l. duction applied to geological problems.
Sheets and sills. Can . J. Earth Sci. 7: 103 1 - Geol. Soc. Am. Bull. 48: 69-94
61 MacLeod, N. S. 198 1 . Differentiation of a
Irvine, T . N. 1980. Magmatic density cur­ gabbro sill in the Oregon eoast range by
rents and cumulus processes. Am. J. Sci. crystallization-zone settling. US Geol.
280-A: 1-58 Surv. Prof Pap. 1165. 22 pp.
Iyer, H. M. 1 984. Geophysical evidence for Mangan, M . , Helz, R. T. 1 985. Vesicle and
the locations, shapes and sizes, and inter­ phenocryst distribution in Kilauea Iki lava
nal structures of magma chambers lake, Hawaii. Eos, Trans. Am. Geophys.
beneath regions of Quaternary volcanism. Union 66: 1 1 33 (Abstr.)
Philos. Trans. R. Soc. London Ser. A 3 1 0 : Mangan, M . , Marsh. B. D. 1 988. Convection
473-5 1 0 and crystallization of sheet-like magma
Jackson, M . D . , Pollard, D. D . 1987. The bodies: III. Crystal capture and basal
laccolith stock controversy: new results cumulates. Eos, Trans. Am. Geophys.
from the southern Henry Mountains, Union 69: 526 (Abstr.)
Utah. Geol. Soc. Am. Bull. 100: 1 1 7-39 Marsh, B. D. 1 98 1 . On the crystallinity,
Jaeger, J. C. 1957. The temperature in the probability of occurrence and rheology of
neighborhood of a cooling intrusive sheet. lava and magma. Contrib. Mineral. Petrol.
Am. J. Sci. 255: 306- 1 8 78: 85-98
Jaeger, J. C. 1 9 6 8 . Cooling and solidification Marsh, B. D. 1 982 .. On the mechanics of
of igneous rocks. In Basalts: The Pol­ igneous diapirism, stoping, and zone melt­
dervaart Treatise on Rocks of Basaltic ing. Am. J. Sci. 282: 808-55
Composition, ed. H. H. Hess, A. Pol­ Marsh, B. D. 1 984. Mechanics and energetics
dervaart, pp. 503-36. New York: Inter­ of magma formation and ascension. In
science. 862 pp. 2 vols. Studies in Geophysics, ed. F. R. Boyd, Jr.,
Jaupart, C., Brandeis, G. 1986. The stagnant pp. 67-83. Washington, DC: NatL Acad.
bottom layer of convecting magma cham­ Press
bers. Earth Planet. Sci. Lett. 80: 1 83-99 Marsh, B. D. 1 988a. Causes of magmatic
Jaupart, C, Parsons, B. 1985. Convective diversity. Nature 333: 47
instabilities in a variable viscosity fluid Marsh, B. D. 1 988b. On convective style and
cooled from above. Phys. Earth. Planet. vigor in sheet-like magma chambers. 1.
Inter. 39: 1 4-32 Petrol. In press
Jaupart, c., Brandeis, G., Allegre, c. J. 1 984. Marsh, B. D. 1 988c. Crystal capture, sorting,
Stagnant layers at the bottom of con­ and retention in convecting magma. Geol.
vecting magma chambers. Nature 308: Soc. Am. Bull. In press
535-38 Marsh, B. D., Maxey, M. R. 1985. On the
Jhaveri, B. S., Homsy, G. M. 1 980. Ran­ distribution and separation of crystals in
domly forced Rayleigh-Benard convec­ convecting magma. J. Volcanol. Geo­
tion. J. Fluid Mech. 98: 329-48 therm o Res. 24: 95-150
Jhaveri, B. S., Homsy, G. M. 1 982. The onset Martin, D., Griffiths, R. W., Campbell, L H .
ofconvection in fluid layers heated rapidly 1 987. Compositional and thermal con­
in a time-dependent manner. J. Fluid vection in magma chambers. Contrib.
Mech. 1 14: 251-60 Mineral. Petrol. 96: 465-75
Kerr, R. C., Tait, S. R. 1 986. Crystallization Matumoto, T. 1 97 1 . Seismic body waves
and compositional convection in a porous observed in the vicinity of Mount Katmai,
medium with application to layered Alaska, and evidence for the existence of
MAGMA CHAMBERS 473
molten chambers. Geol. Soc. Am. Bull. 82: systems. In Magmatic Processes: Physico­
2905�20 chemical Principles, ed. B. O. Mysen, pp.
McBirney, A. R., Murase, T. 1 984. Rheo­ 259�87. University Park, Pa: Geochem.
logical properties of magmas. Ann. Rev. Soc. Am. 500 pp.
Earth Planet. Sci. 1 2: 337�57 Ryan, M. P., Blevins, J. Y. K. 1987. The
Morris, S., Canright, D. 1984. A boundary viscosity of synthetic and natural silicate
layer analysis of Benard convection in a melts and glasses at high temperatures and
fluid of strongly temperature-dependent I bar ( l OS pascals) pressure and at higher
viscosity. Phys. Earth Planet. Inter. 36: pressures. US Geol. Suru. Bull. 1 764. 563
355 73 pp. .
Morse, S. A. 1 9R8. Motion of crystals, Ryan, M. P., Blevins, J. Y. K., Okamura,
solute, and heat in layered intrusions. Can. A. T., Koyangi, R. Y. 1983. Magma reser­
Mineral. 26: 209�24 voir subsidence mechanics: theoretical
Murata, K. J., Richter, D. H. 1 966. The set­ summary and application to Kilauea vol­
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

tling of olivine in Kilauean magma as cano, Hawaii. J. Geophys. Res. 88: 4 1 47�
shown by lavas of the 1959 eruption. Am. 81
J. Sci. 264: 1 94-203 Ryerson, F . J., Weed, H . c., Piwinskii, A . J.
Access provided by McMaster University on 01/23/15. For personal use only.

Nilson, R. H . , McBirney, A. R., Baker, B. H . 1988. Rheology of subliquidus magmas 1 .


1 985. Liquid fractionation. Part I I : Fluid Picritic compositions. J. Geophys. Res. 93:
dynamics and quantitative implications 3421�36
for magmatic systems. J. Volcanol. Smith, M. K. 1988. Thermal convection dur­
Geotherm. Res. 24:25�54 ing the directional solidification of a pure
Norton, D., Knight, J. E. 1977. Transport liquid with variable viscosity. J. Fluid
.
phenomena in hydrothermal systems: Mech. 1 88 : 547�70
cooling plutons. Am. J. Sci. 277: 937�8 1 Smith, R. C. II, Rose, A. W., Lanning,
Norton, D., Taylor, H. P. Jr . 1 979. Quan­ R. M. 1975. Geology and geochcmistry
titative simulation of the hydrothermal of Triassic diabase in Pennsylvania. Geo!.
systems of crystallizing magmas on the Soc. Am. Bull. 86: 943�55
basis of transport theory and oxygen iso­ Sparks, R. S. J., Huppert, H. E., Kerr,
tope data: an analysis of the Skaergaard R. c., McKenzie, D., Tait, S. R. 1 985.
intrusion. J. Petrol. 20: 421�86 Postcumulus processes in layered
O'Connell, R. J., Hager, B. H. 1980. On the intrusions. Ceol. Mag. 122: 555�68
thermal state of the Earth. In Physics of Spera, F. J. 1980. Thermal evolution of plu­
the Earth 's Interior, ed. A. Dziewonski, E. tons: a parameterized approach. Science
Boschi, pp. 270-3 1 7. Bologna: Soc. Ital. 207: 299�301
Fis. Spera, F . J., Yuen, D. A., Kirschvink, S. J.
Orborne, F. F., Roberts, E. J . 193 1 . Differ­ 1982. Thermal boundary layer eonvcetion
entiation in the Shonkin Sag laccolith. in silicic magma chambers: effects of tem­
Am. J. Soc. 22: 33 1�53 perature-dependent rheology and impli­
Peck, D. L., Hamilton, M. S., Shaw, H . R. cations for thermogravitational chemical
1977. Numerical analysis of lava lake fractionation. J. Geophys. Res. 87: 8755�
cooling models: Part II. Application to 67
Alae lava lake, Hawaii. Am. J. Sci. 277: Spera, F. J., Yuen, D. A., Kemp, D. V. 1 984.
4 l 5�37 Mass transfer rates along vertical walls in
Peltier, W. R. 1 985. Mantle convection and magma chambers and marginal upwelling.
viscoelasticity. Ann. Rev. Fluid Mech. 1 7 : Nature 3 1 0: 764-67
561�608 Spera, F. J., Borgia, A., Strimple, J., Fei­
Peterson, J. S. 1987. Solidification con­ genson, M. 1988. Rheology of melts and
traction: another approach to cumulus magmatic suspensions. Part I. Design,
processes and the origin of igneous layer­ calibration, and operation of concentric
ing. In Origins of Igneous Layering, ed. I. cylinder viscometer and application to
Parsons, pp. 505�26. Boston: D . Reidel. rhyolitic magmas. J. Geophys. Res. 93:
666 pp. 10,273�94
Richey, J. E. 1927. The structural relations Steeples, D. W., Iyer, H. M . 1976. Low ve­
of the Mourne Granites (Northern Ire­ locity zone under Long Valley as deter­
land). Q. J. Geol. Soc. London 83: 653�88 mined from teleseismic events. J. Geophys.
Richter, F. M . , Natif, H. c., Daly, S. F. 1 983. Res. 8 1 : 849-60
Heat transfer and horizontally averaged Stengel, K. C., Oliver, D. S., Booker, J. R.
temperature of convection with large vis­ 1982. Onset of convection in a variable­
cosity variations. J. Fluid Mech. 1 29: 1 73� viscosity fluid. J. Fluid Mech. 1 20: 41 1�33
92 Taylor, H . P. Jr., Forester, R. W. 1 979. An
Ryan, M. P. 1987. Neutral buoyancy and the oxygen and hydrothermal isotope study of
mechanical evolution of magmatic the Skaergaard intrusion and its country
474 MARSH
rocks: a description of a 55-my-old fossil Bull. 68: 107 1-74
hydrothermal system. J. Petrol. 20: 355- Walker, F. 1 940. The differentiation of the
419 Palisade diabase, New Jersey. Geol. Soc.
Thompson, M . E., McBirney, A . R. 1985. Am. Bull. 5 1 : 1059-1 1 06
Redistribution of phenocrysts by con­ Walker, F., Poldervaart, A. 1949. Karroo
vectivc flow in a viscous boundary layer. dolerites of the Union of South Africa.
J. Volcano!. Geotherm. Res. 24: 83-94 Geo!. Soc. Am. Bull. 60: 59\-706
Turner, J. S. 1973. Buoyancy Effects in Fluids. Whitney, J. A., Stormer, 1. C. Jr. 1985. Min­
New York: Cambridge Univ. Press. 367 eralogy, petrology, and magmatic con­
pp. ditions from the Fish Canyon tuff, central
Turner, J. S. 1 98 5 . Multicomponent con­ San Juan volcanic field, Colorado. J.
vection. Ann. Rev. Fluid Mech. 1 7 : 1 1-44 Petrol. 26: 726-62
Turner, J. S., Huppert, H . E., Sparks, R. S. Wright, T. L. 1 9 7 1 . Chemistry of Kilauea
J. 1986. Komatiites II. Experimental and and Mauna Loa lava in space and time.
Annu. Rev. Earth Planet. Sci. 1989.17:439-472. Downloaded from www.annualreviews.org

theoretical investigations of post -emplace­ US Geol. Surv. Prof Pap. 735. 40 pp.
ment cooling and crystallization. J. Petrol. Wright, T. L., Okamura, R. T. 1977. Cooling
Access provided by McMaster University on 01/23/15. For personal use only.

27: 397-438 and crystallization of tholeiitic basalt,


Wager, L. R., Brown, G. M. 1957. Funnel­ 1 965 Makaopuhi lava lake, Hawaii. US
shaped layered intrusions. Geol. Soc. Am. Geol. Surv. Prof Pap. 1004. 78 pp.

You might also like