Script

for the lecture Mineral Materials as part of the

study path Materials Engineering

Prof. Dr. rer. nat. Christian Roos

07.08.2021

Preliminary mark:
The script is intended to be used only by students attending the lecture
mentioned above. Any form of distribution, even only of extracts hereof, is not
permitted.
The present work is also valid for some German study paths and is a translation of that
script. However, those lectures are more comprehensive than this lecture. Therefore,
you find more information in this script (especially background information) than
discussed in this respective lecture. However, this should in no way seen as a
hindrance to read through the script and deepen your knowledge about glass
properties, production and forming.
 Page 2

Table of Contents
1. Introduction.................................................................................................................................... 7

1.1. Secondary Literature............................................................................................................... 7

1.2. The concept of glass................................................................................................................ 7

2. Viscosity ........................................................................................................................................ 10

2.1. Viscosity in general ............................................................................................................... 10

2.2. Viscosity and Glass Transition ............................................................................................... 10

2.3. Description of the viscosity-temperature dependence ........................................................ 14

2.3.1. VFT approach .................................................................................................................... 14

2.3.1.1. .................................................................................... Linearization of the VFT-Equation

17

2.3.2. WLF Approach .................................................................................................................. 18

2.3.3. Adam-Gibbs Equation ....................................................................................................... 18

2.3.4. Further Models ................................................................................................................. 20

2.4. Freezing of the structure at sufficiently high viscosity ......................................................... 20

2.5. The Concept of Fictive Temperature .................................................................................... 23

2.6. Angell Plot ............................................................................................................................. 26

2.7. Viscosity Measurement ........................................................................................................ 27

3. Structure of Glasses...................................................................................................................... 30

3.1. Heat Capacity of Glasses and Its Correlation to the Structure.............................................. 32

3.2. Structural Models and Percolation Theory ........................................................................... 37

4. Thermodynamics of glass ............................................................................................................. 40

4.1. Thermodynamics of the Glass Transition.............................................................................. 41

4.2. Glass and the crystalline reference state .............................................................................. 46

4.3. The concept of constitutional phases ................................................................................... 48

4.4. On the entropy of glasses ..................................................................................................... 51

4.5. Simon‘s approximation ......................................................................................................... 52

 Page 3

4.6. Kauzmann paradoxon ........................................................................................................... 53

5. Glass chemistry ............................................................................................................................ 55

5.1. Approach according to Dietzel for cation-anion packings ................................................... 58

5.1.1. Estimations based on homologous compounds ............................................................... 61

5.2. General rules for glass formation ......................................................................................... 61

5.3. Technologically relevant glasses ......................................................................................... 62

5.3.1. Silicate Glasses.................................................................................................................. 63

5.3.2. Borate and Borosilicate Glasses ....................................................................................... 64

5.3.3. Phosphate glasses............................................................................................................. 66

5.4. General aspects of glass formation / summary .................................................................... 68

5.5. Network structures in real systems ...................................................................................... 69

6. Mechanical and thermomechanical properties of glasses ........................................................... 72

6.1. Strength of glasses (theoretical and practical) ..................................................................... 72

6.1.1. Chemical preloading ........................................................................................................ 74

6.2. Structural dependencies and mechanical properties ........................................................... 75

6.3. Thermal conductivity in glasses ............................................................................................ 78

6.4. Thermal expansion coefficient ............................................................................................. 80

6.4.1. Negative thermal expansion coefficient ........................................................................... 82

6.4.2. Thermal stresses ............................................................................................................... 83

6.4.3. Thermal Glass Tempering ................................................................................................. 84

7. Optical properties of glasses ........................................................................................................ 87

7.1. Reflection and refraction ...................................................................................................... 88

7.2. Absorption properties of glasses and melts ......................................................................... 92

7.3. A window and a car… ............................................................................................................ 96

8. Devitrification and crystallization ................................................................................................. 97

8.1. Nucleation ............................................................................................................................. 98

8.1.1. Nucleation rate ............................................................................................................... 101

 Page 4

8.1.2. Heterogeneous nucleation ............................................................................................. 102

8.1.3. Problem of considering the pre-factor ........................................................................... 103

8.2. Crystal growth ..................................................................................................................... 105

8.2.1. Crystal growth rate ......................................................................................................... 107

8.2.2. Characterisation of crystallization in glasses .................................................................. 108

8.2.3. Interfacial and diffusion Controlled Growth ................................................................... 109

8.2.4. Avrami-Erofe’ev kinetics ................................................................................................. 110

8.3. Glass ceramics ..................................................................................................................... 113

8.3.1. LAS glass-ceramics .......................................................................................................... 114

8.4. Metallic glasses ................................................................................................................... 116

9. Industrial (large-scale) glass melting .......................................................................................... 118

9.1. The batch (raw materials and cullet) .................................................................................. 119

9.2. Phenomenological processes in the batch blanket............................................................. 123

9.3. Initial melting ...................................................................................................................... 124

9.4. Sand dissolution .................................................................................................................. 126

9.4.1. Batch-Free-Time (BFT) .................................................................................................... 128

9.5. Redox state of a glass.......................................................................................................... 129

9.6. Fining and refining .............................................................................................................. 130

9.7. Gas analysis ......................................................................................................................... 134

9.8. Specific fining reactions ...................................................................................................... 134

9.8.1. Sulfate fining ................................................................................................................... 135

9.8.2. Antimony / Arsenic Oxide fining ..................................................................................... 137

9.8.3. Halogenide fining ............................................................................................................ 138

9.8.4. SnO2 fining ...................................................................................................................... 138

9.9. Conditioning of the melt ..................................................................................................... 138

9.10. General considerations on the functioning of a glass tank ................................................. 139

9.11. Alternative approaches to CO2 avoidance ......................................................................... 141

 Page 5

10. Energy consideration of the melting process ............................................................................. 141

10.1. Determination of Hex ........................................................................................................... 144

10.2. Determination of Hin ........................................................................................................... 147

10.2.1. Heat capacities of air and exhaust gas ........................................................................... 149

10.3. Determination of the exchanged heat Hexch ........................................................................ 150

10.4. Determination of the heat loss Hw+stack ............................................................................... 151

11. Plant engineering ....................................................................................................................... 155

11.1. Air-gas firing / Oxy-gas firing............................................................................................... 157

11.2. Regenerative and recuperative air preheating ................................................................... 159

11.3. U-flame furnace, regenerative ........................................................................................... 161

11.4. Cross-heated furnace, regenerative .................................................................................. 163

11.5. Special glass furnaces ......................................................................................................... 163

11.6. Electrically heated tanks ..................................................................................................... 164

11.7. General information on the design of furnaces .................................................................. 165

11.8. Corrosion of melting tanks.................................................................................................. 167

12. Glass Shaping .............................................................................................................................. 169

12.1. Glass fiber drawing ............................................................................................................. 169

12.2. Hollow glass production...................................................................................................... 173

12.3. Drawing tubular glass ......................................................................................................... 176

12.4. Pressing glass ...................................................................................................................... 178

12.5. Drawing / rolling of flat glass .............................................................................................. 179

12.6. Float process ....................................................................................................................... 180

13. Incremental systems .................................................................................................................. 183

13.1. Modelling of viscosity ......................................................................................................... 185

13.2. Modelling of the mechanical properties ............................................................................. 186

13.3. Modelling of the thermal expansion................................................................................... 186

13.4. Modelling of the optical properties .................................................................................... 187

Page 6
 Page 7

1. Introduction
1.1. Secondary Literature
The following secondary literature is recommended for closer examination of some topics and
includes books written in German and English language:
• H. Vogel – Glaschemie
• W. Scholze – Glas – Natur, Struktur und Eigenschaften
• H. Schaeffer – Allgemeine Technologie des Glases - Grundlagen des Schmelzens und der
Formgebung
• W. Trier – Glasschmelzöfen
• Jebsen-Marwedel & Brückner - Glastechnische Fabrikationsfehler
• J. Schmelzer & I. Gutzow - Glasses and the Glass Transition
• Mysen & Richet – Silicate glasses and melts
• I.W. Donald – Inorganic Glasses and Glass-Ceramics
• F. Wallenberger – Fiberglass and glass technology
• S. Turns – An introduction to combustion
• A. Paul - Chemistry of Glasses
• M. B. Volf - Chemical approach to glass

1.2. The concept of glass

Ideally, glass is a homogeneous and isotropic material. In everyday use, the term “glass” is
ambiguous and often used unclearly. One may refer to an object (e.g. a wine glass) or a certain
chemical composition (e.g. window glass or lead crystal glass).
In the scientific context, the term “glass” denotes a state of matter which, in analogy to the terms
gas, liquid, crystalline solid, is completely independent of any chemical composition. This state is
reached if crystallization, i.e. the formation of a long-range order, is avoided. In general, this is
achieved through rapid cooling from a melt. However, it should be pointed out that there are also
other ways of producing a glass (e.g. via sol-gel processes or pressure-controlled processes, by
reduction of a solution, precipitation from a liquid or gaseous phase, by a polymerization reaction
or exposure of a crystalline material to shock waves).
The prerequisite is that the respective process takes place so quickly that a structural relaxation
into the equilibrium state (the crystalline state) does not take place. The result is called a “frozen
phase”. However, one thing applies universally: Once you have produced a glass (regardless of the
method used) such a material—when heated—shows a characteristic, abrupt change in heat
capacity, coefficient of thermal expansion and compressibility to the respective value of the
equilibrium liquid well below the liquidus temperature. This behavior is called the glass transition.
 Page 8

Technologically speaking, the term “glass” is usually used to describe a group of oxidic, mostly
silicate, non-crystalline materials. As shown in the overview below, there are also other glasses.

The list below summarizes some properties of glasses in which they are usually superior to other
materials. The fragility of the glass, on the other hand, is a permanent challenge for improvements
to the material.
Note that “beautiful” actually represents a material property that is meant to be serious. It
represents an essential part of the definition of quality for a glass. Historically, glass was mainly
made for its beauty. A property in which glass differs from all other inorganic materials: Glass is a
material without a structure, without internal dislocations or phase boundaries. This is a
prerequisite for optical transparency, but also causes the (often) catastrophic behaviour with
respect to breakage.

Beautiful (aesthetic function)

Beautiful (optical
Transparent (only anfunction)
aesthetic function)

Resistant to chemicals

Extremely stiff

High rigidity

Very smooth surfaces

Very low fracture toughness

Composition highly variable

Continuously producible

Arbitrarily shapeable
 Page 9

Glass is omnipresent in our daily life:

However, partly due to its transparency, glass does not receive the same attention as other
materials because it cannot be seen and perceived “directly”. One example is the cover glass of a
smartphone, which with a thickness of less than 100 µm, its touch-capability and its extreme
strength is a technologically highly developed product, but is hardly noticed.
A common definition for glasses, that is widely used:
§ GLASS is a (predominantly) non-crystalline material (no translation symmetry of the
atomic structure).
§ GLASS is a supercooled frozen liquid („frozen“: a non-equilibrium state with the atomic
degrees of freedom and the mechanical properties of a solid; „liquid“: a condensed phase;
„supercooled“: an supercooled liquid is a metastable liquid equilibrium state that exists
below the melting temperature—the liquidus—Tliq).
§ GLASS shows a glass transition (when a solid glass is heated up, the system does not pass
through a phase transition (melting) into the liquid state, but rather through a so-called
glass transition at Tg < Tliq from the frozen to the metastable supercooled state; this also
applies, if a glass was not made via a melting route)
§ GLASS is a material with a non-crystalline microstructure (no internal phase boundaries;
optically, electrically, mechanically isotropic).
 Page 10

2. Viscosity
2.1. Viscosity in general
Ideal liquids show Newtonian behavior: In this case the shear stress τ is proportional to the shear rate
∆"
!∆#".

∆𝑣
𝜏=𝜂∙
∆𝑦
The constant of proportionality η is the absolute (or dynamic) viscosity ([η] = Poise = dPa·s
= g·cm–1·s–1 = 0.1 kg·m–1·s–1, nowadays often in Pa·s = kg·m–1·s–1).
The reciprocal is the fluidity ϕ. The following applies to the relationship between dynamic viscosity
and kinematic viscosity 𝜈:
𝜂 = 𝜈 ∙ 𝜌 = 𝜙 $% .
The viscosity is often represented as a decadic logarithm due to the wide range of values that it can
exceed. The logarithm of a number has no units. This poses a small problem, since, unless previously
defined, one does not know whether the log relates to e.g. dPa·s or Pa·s. The following can be provided
with any plot or graph:
𝜂
log%& 0 1
𝜂'
Using the previously defined: 𝜂' = 1 dPa·s (or 𝜂' = 1 Pa·s). You can do that—and it is perfectly correct—
but...it’s extremely tedious. Therefore, in the upcoming explanations the expression
log 𝜂 , 𝜂 in dPa·s
is used. For the sake of simplicity, Pa·s or dPa·s is usually written as Pas or dPas in the script. Moreover,
the following simplified notation is also often used:

log 𝜂(𝑇) =L

log 𝜂(𝑇 → ∞) = L∞

log 𝜂8𝑇( 9 = 𝐿(

where 𝐿) can be seen as the high-temperature limit of the viscosity.

2.2. Viscosity and Glass Transition

The viscosity represents the most important kinetic property of a glass melt. We start with the example
of producing a glass from a melt by continuously cooling down the melt.
The kinetics determine whether crystallization or glass formation occurs when a liquid is cooled down.
If the melt is supercooled and this supercooling as well as the further temperature reduction—and
thus viscosity increase—take place very quickly, no crystal nuclei can form or nuclei that have already
 Page 11

formed cannot continue to grow. This is how you can reach the area of the glass transition; the liquid
is no longer in an internal equilibrium and the curvature of the cooling curve changes.
Let us first take a look at an infinitely slow cooling (see Figure below):
If you cool down a one-component melt, it begins to crystallize slightly below the melting temperature.
The system follows the thick red line, enters a regime of critical supercooling needed for the first
nucleation to begin, returns exactly to the melting temperature Tm and remains there until the melt
has completely crystallized.
If the cooling rate is sufficiently high, however, the system no longer returns from the state of
undercooling (thick red line) to the melting temperature, but behaves like indicated by the blue curve.
Without any anomaly in the cooling curve, it gradually reaches the glass state. Only with a very keen
eye one can discover that the course of the cooling curve deviates from that of an exponential function
at a critical point (indicated by the black circle in the right Figure). This is exactly where the glass
transition takes place and indicates the point Tg with the glass melt having a viscosity of 1013 dPas
(strictly speaking, it is a range, as will be explained later).

From a thermodynamic point of view, the melt goes through the states shown below. The solid glass
is then a supercooled melt in which there are no longer any states of equilibrium.
 Page 12

Philip W. Anderson, one of the Nobel Prize Winners for Physics in 1977, made the following statement
in 1995: "The deepest and most important unsolved problem in condensed matter science is probably
the theory of the nature of glass and the glass transition." We are, more than a quarter of a century
later, still unable to give a proper answer to this problem. From a scientific point of view, it is therefore
worthwhile to take a closer look at the phenomenon of the glassy state and the glass transition.
If you want to identify the optimal points for glass production on the basis of a given phase diagram,
the relationships with respect to the viscosity are essential. On the one hand, the melt at Tliq should
have a high viscosity in order to prevent the formation of crystal nuclei (then the crystallization rate,
which is proportional to h–1, is rather low at T < Tliq) and at the same time there should only be a small
temperature difference between Tliq and the glass transition Tg, so that this region, which is critical for
crystallization, can be passed through very quickly, giving nuclei that have already formed no time to
grow any further. After reaching Tg the viscosity is so high that no significant rearrangements and no
ion diffusion can take place, inhibiting any further crystal growth.
In the figure below, the positions of the congruent and incongruent melting compounds are given as
NS = Na2O·SiO2 and NS2 = Na2O·2SiO2 or N3S8 = 3Na2O·8SiO2 (the usual cement notation of the phases
is used) and, in addition to the liquidus curve (bold solid line), also the lines of constant viscosity
(dashed lines) are drawn. The ideal points for glass production are therefore the two eutectics and the
100% SiO2 melt. The dystectics or the melt with a small amount of Na2O, on the other hand, are less
suitable. Tg [dashed line for T(13)] depends only slightly on Na2O over a broad range of compositions
(0.5 to 0.95 Ma.-% SiO2), but changes dramatically in the range of SiO2-rich melts. One can understand
this by assuming that the degree of polymerization of the melt (see below) is already greatly reduced
by adding small amounts of components with predominantly ionic bonds, but with further addition is
only reduced by an insignificant amount.
 Page 13

It can also be seen that in the above-mentioned system, Na2O acts as both a viscosity and a liquidus
flux.
If one looks at the progression of the heat capacity in the region of the glass transition, one recognizes
that the jump from cp at Tg does not have a sharp progression, as is usually associated with a phase
transition. When heating up from lower temperatures, one often finds a characteristic overshooting
of the value prior to reaching the equilibrium value of the liquid (as shown in the figure below). For the
sake of simplicity, however, the course of the cp curve can be approximated by the hatched area (I).
The difference to the true area (II) is calorimetrically insignificant. The picture illustrates two options
for determining the position of Tg using a calorimetric measurement. A third option (giving a value of
530 K, due to the first deviation of the slope from the cp of the ideal solid) is not recommended. The
other two options (551 K, by using a tangent method and 563 K by the turning point of the cp curve)
are typically much closer together than in the example below. The turning point of the cp curve is
generally a good reference point for Tg.
 Page 14

2.3. Description of the viscosity-temperature dependence

2.3.1. VFT approach

Liquids generally show Arrhenius behavior with regard to their viscosity-temperature dependence:
!"
𝜂 = 𝜂& ∙ 𝑒𝑥𝑝*#∙%+ ,

with Eη being an activation energy related to viscosity (referring to the energy that is required to move
the "structural units” [J/mol]), η denoting the dynamic viscosity, η0 referring to a material constant, T
being the temperature and R standing for the universal gas constant.
This approach applies to constant values for Eη and to spherical particles with non-directional bonds.
Unfortunately, this is not the case for (glass) melts in the visco-elastic range, since here Eη clearly
depends on the temperature. Thus, another approach is needed to describe the viscosity-temperature
dependence of glasses in the visco-elastic range. One approach generally used is the empirically
deduced Vogel–Fulcher–Tammann equation (VFT equation). The derivation from the Arrhenius
approach is shown in the Figure below. In the VFT equation a third parameter T0 is thus introduced,
which allows a good description of the viscosity-temperature dependence in the visco-elastic range:
!
" # $
𝜂 = 𝐴! ∙ 𝑒𝑥𝑝 "#"$ , respectively log 𝜂 = 𝐴 + -%&% .
$

In detail, the VFT equation is obtained beginning with the Arrhenius approach following the steps
outlined below:

-"
, 1
𝜂 = 𝜂& ∙ exp .∙0 Arrhenius Equation

-"
, 1 Introduction of a third
𝜂 = 𝜂& ∙ exp .∙(0$0& )
empirical parameter

5' -"
, 1
𝜂=𝐴 ∙ 4
exp 0$0 & Note that 𝐵4 = .
and 𝐴4 = 𝜂&

4
𝐵4 Vogel–Fulcher–Tammann
ln 𝜂 = ln 𝐴 + D G
𝑇 − 𝑇& (VFT)-equation

Rearranged, the final VFT equation is usually given using the decadic logarithm and thus is:
5 6' ,5 '
log 𝜂 = 𝐴 + !0$0 ", with 𝐴, 𝐵 = 89 %&
&

The logarithm has no units. But T0 must have the unit °C or K, otherwise the quotient cannot be
calculated. This follows from the equation above in the second line of the formulas. B (“in the equation
 Page 15

not yet logarithmized”) = Eη/R and thus has the unit °C or K, which ensures that the quotient has no
unit.
In the logarithmic VFT equation, B must also have the unit °C or K. However, it is not a matter of a
“temperature” as in the sense of T0, but rather the composite quantity Eη/R.
Please note: In the logarithmic VFT equation, B, A and T0 refer to the decadic logarithm. This must
therefore be taken into account when calculating back from log η to η.
The parameters A‘ or A and B‘ or B, renamed from the Arrhenius approach, can be interpreted
physically, but do not represent any physical laws. The same applies to T0. T0, A and B are constants
that only depend on the glass chemistry. Still, this interpretation is helpful in understanding the VFT
equation. Following is an at least somewhat meaningful interpretation of the different parameters.
This can be better understood by looking at the Figure below.

The first parameter, A, seems to reflect the viscosity value for T towards infinity (10–4 dPas), but in fact
for most fluids A is usually significantly higher than 10–4 dPas. The highest possible viscosity (according
to VFT) is reached at T0. B is not the activation energy in the sense of the Arrhenius approach. This only
applies, as can be seen from the graph above, in the high-temperature region. Because VFT strives
against the Arrhenius approach, since T minus T0 trends towards T at high enough temperatures. Thus,
in this region it does not matter whether the viscosity-temperature dependency is described with the
VFT or the Arrhenius approach.
The VFT equation is used, for example, by measuring three viscosity-temperature pairs, inserting them
in and solving the VFT equation and then determining the parameters A, B and T0.
Using this viscosity-temperature dependency (VFT curve), some very important characteristic viscosity
points and ranges for glass production can be defined. In general, it should be noted at this point that
glass technologists generally do not speak of a viscosity at a given temperature, but rather, conversely,
of a temperature for one of these characteristic viscosity points. T(13) therefore means the
temperature at which a glass melt of specific composition has a viscosity of 1×1013 dPas.
 Page 16

The range from 4 to 6 dPas (machine shaping), respectively 7.6 dPas (manual shaping), represents the
so-called working range. The temperature interval in which this working range is passed is called the
“length” of the glass. This is the temperature range that is available for machine (4–6 dPas) or manual
(4–7.6 dPas) shaping.

Caution! The “length” of the glass says nothing about the time in which this temperature interval is
passed through. Two chemically almost identical glasses can have very different cooling rates. For
example, if one glass has a brown color by adding a few 100 ppm FeO it has a significantly different
thermal conductivity than a non-colored glass with nearly the same chemical composition. The same
applies if T(4) (i.e. the temperature at which the glass melt has a viscosity of 4 dPas) is at very different
temperatures, as the loss of heat due to radiation is proportional to the fourth power of the absolute
temperature.
The main characteristic viscosities are summarized in the following table:

log !"
Fix Popints / Behaviour / Technical usage
(!"#$"%&'()
Melting range
1.5 Float glass melt at 1450 °C; defibration point
3.0 Droplet temperature; Transfer for shaping
Working range
4.0 Working point, beginning of shape memory
6.0 End of working range for machine shaping
Prestressing range
7.6 Littleton point; macroscopic shape stability
11.3 Dilatrometric softening point; *(x) +T(x) *)
Cooling range
11.3 Dilatrometric softening point; *(x) +T(x) *)
Upper cooling point; ,"@ 1 min
13.0
Dilatometric glass transition.
14.5 Lower cooling point; ,"@ 30 min
*) the stress field begins to connect to the temperature field
 Page 17

2.3.1.1. Linearization of the VFT-Equation

(after R. Conradt)
In certain temperature ranges (directly below Tliq), viscosity measurements cannot be carried out for
certain melts, as they crystallize too quickly. Results are then only available above Tliq and in the region
of Tg, where rearrangement processes slow down and freeze. A linearization of the VFT equation makes
it possible to look behind this “crystallization curtain” and to make reliable predictions regarding the
viscosity in an area where direct measurements are not possible.
For more details, please refer the publication by R. Conradt. Ultimately, the VFT equation is converted
into the linear equation below:
𝑇 𝑇:
01 − 𝑇& 1 ∙ 𝑇 (1 − 𝑐) ∙ 𝑥
𝐵 :
log 𝜂 = 𝐴 + ∙ = 𝛽& + 𝛽% ∙
𝑇: − 𝑇& 𝑇 𝑇: 1−𝑐 ∙𝑥
1 − 𝑇& ∙ 𝑇
:

Y = 𝛽& + 𝛽% ∙ Χ
Y = log 𝜂
𝛽& = 𝛢
5 0&
𝛽% = with 𝑐 =
%$< 0(

(%$<)∙= 0( 0&
𝑋= with 𝑥 = and 𝑐 =
%$<∙= 0 0(

c contains Tg (you have to know this one) and T0 as an unknown. Thus, c in the slope (β1) and c in X
must both be taken into account when performing the regression and adjusted at the same time until
the best fit is achieved (coefficient of determination R2 as close as possible to unity). This can be done,
among other programs, in Excel after a few iteration steps.
The detailed procedure is as explained hereafter:
(%$<)∙=
Tg is determined experimentally and Y = log η is plotted against X !𝑋 = %$<∙=
".

The experimental value for Tg is used to determine the variable x = Tg/T. Then c (= T0/Tg occurring in β1
and X) is iteratively varied until the error is minimal (R2 ➝ 1). In the example below, the best linear fit
is obtained for c = 0.638. The deviation of ±δlog(η) between the straight line and the experimental
values is used as a quality criterion for the fit. For c = 0.638 the minimum is δlog η = 0.02. This
corresponds to an error in the viscosity δη of ±100.02 which is approx. ± 6%. The fit results in a slope
β1 and an axis intercept β0. The VFT constants A, B, and T0 are then calculated from Tg, c, β0, and β1.
 Page 18

In this way, even with only two L/T-pairs, it is possible to generate the viscosity curve or, for example,
to recalculate the VFT parameters from the linearization. In principle, all of the viscosity models
presented here can be linearized in this way.

2.3.2. WLF Approach

The approach named after Williams–Landel and Ferry is based on the "free volume theory": the
viscosity depends on the free volume. The decisive factor here is that v = V/Vref represents a volume
normalized to a reference volume and that the free volume Vfree increases with increasing temperature
and freezes at a value V0 at the glass transition. Hence Vfree = V0 + b·T for large T and Vfree = V0 for
T = Tg.
For the viscosity the following then applies:

"> '('?
log 𝜂 = log 𝜂! − ∙ (WLF-Gleichung)
#.%&% "@ )'('?

It can be shown that the WLF-equation can be converted into the VFT equation. In fact, the equations
are identical. The VFT equation thus also represents a consideration of the free volume.

2.3.3. Adam-Gibbs Equation

(For the following paragraph it may be helpful to deal with the thermodynamics of the glass transition
in more detail beforehand (see later chapter)).
In 1965, Adam and Gibbs theoretically studied the thermodynamic properties of a material with a two-
level hierarchy of structure. (As we will see later, the existence of mid-ranged-ordered (MRO) units is
derived from this. In theory, these were called “cooperatively rearranging units”). As an outstanding
 Page 19

result of the theory, there was a direct proportionality between the logarithm of the macroscopic
viscosity η and the term 1 / T·Sc(T), where Sc(T) represents the so-called configuration entropy:
𝑇:
log 𝜂 = log 𝜂) + 𝐶 ∙
𝑇 ∙ 𝑆A (𝑇)
If the expression for the configuration entropy SC(T)
0 ∆𝑐 𝑇: ∆𝑐F 𝑇:
F
𝑆B (𝑇) = 𝑆 CDE + X 𝑑𝑇 ≈ 𝑆 CDE − ∆𝑐F ∙ ln = 𝑆 CDE ∙ 01 − CDE ∙ ln 1
0( 𝑇 𝑇 𝑆 𝑇

is combined with the approach by Adam and Gibbs mentioned above, an expression for the viscosity-
temperature relation is obtained that no longer contains a fit constant.
𝑇: 1
log 𝜂 = log 𝜂) + 𝐶 ∙ ∙
𝑇 ∙ 𝑆 CDE ∆𝑐F 𝑇:
01 − CDE ∙ ln 𝑇 1
𝑆
The structure parameter a0 = ∆cp/Svit is a dimensionless thermodynamic quantity that characterizes the
glass-forming system. a0 is generally calculated from viscosity data. If one considers that the following
applies to the viscosity at Tg:
1
log 𝜂 (𝑇' ) ≈ log 𝜂( + 𝐶 ∙
𝑆 )*+
If the viscosity log(η) at Tg is assumed to be equal to 13 and the viscosity log(η) at T towards infinity
equals –4, then it again follows
C = 8log 𝜂 (𝑇: ) − log 𝜂) 9∙ 𝑆 CDE ≈ 17 ∙ 𝑆 CDE

The assumption that the viscosity log(η) at T towards infinity is equal to ‒4 can certainly be seen as the
most severe approximation in the consideration. From the last three equations it then follows, with
the simplifications to log(η) explained at the beginning:
𝑇:
17 ∙ 𝑇
𝐿 ≈ 𝐿) +
∆𝑐F 𝑇:
1 − CDE ∙ ln 𝑇
𝑆
We can also rewrite this term to get another form of the the Adam-Gibbs Equation describing the
viscosity (again with the structural parameter a0 = ∆cp/Svit):
𝑇:
𝐿 ≈ 𝐿) + (𝐿: − 𝐿) ) 𝑇
𝑇:
1 − 𝑎& ∙ ln 𝑇

For the first time, this opens up the possibility of determining the viscosity of a solidifying glass melt
(at least in part, because L∞ must ideally be approximated more precisely than just 10–4 dPas) on the
basis of thermodynamically measurable quantities.
For now, let us assume the structure parameter a0 as a constant value (with the knowledge that Svit
depends on the cooling rate and can only be assumed to be constant within limits). We abbreviate Tg/T
with y and approximate ln y with y – 1. Then, combining all values kept constant into the constants b
and c, gives us the following expression for the Adam-Gibbs equation:
 Page 20

𝑏
log 𝜂 → 𝐿) +
(𝑇 − 𝑐)
This equation was first established in 1921 on an empirical basis. It's the well-known VFT equation (see
before).

2.3.4. Further Models

There are other models for describing the viscosity-temperature dependency. The approach by
Avramov is based on a so-called stretched exponential function (or "Kohlrausch–Williams–Watts-
function"). Here the chemical properties of the glass, namely the proportion of network converters
(modifiers) α, are linked to the viscosity:

𝑇+ ,
𝐿 = 𝐿* + +𝐿+ − 𝐿* , ∙ - / 𝛼 ≈ 1.2 + 6 ∙ 𝜒-./ ≈ 1.2 + 6 ∙ (1 − 𝜒01.@ )
𝑇
α thus acts as a measure of the fragility (see later) of the melt/glass. Strong glasses have an α of
approx. 1. The fragility increases with α.
Another approach was used by Mauro et al. (MYEGA) and, like the Adam-Gibbs equation, is based on
the approach of configuration entropy:

𝑇+ 𝑚 𝑇+
𝐿 = 𝐿* + +𝐿+ − 𝐿* , ∙ ∙ exp <= − 1? ∙ - − 1/@
𝑇 𝐿+ − 𝐿* 𝑇
These approaches will not be discussed in detail here. Each approach has its strengths and weaknesses.
For example, the MYEGA approach describes the course of the viscosity around Tg very well, but clearly
shows weaknesses in the area of L → L∞. Here again, for example, the VFT approach is superior. It is
crucial, however, that all approaches are models that describe reality only more or less well, depending
on certain circumstances and that we are not dealing with physical laws here.

2.4. Freezing of the structure at sufficiently high viscosity

How can we interpret the freezing of a structure with a correspondingly high viscosity? A hierarchical
structure type helps to understand the occurrence of a relatively sudden transition (the “solidification
of the glass”), which results from a gradual change in an external parameter (viscosity) but does not
represent a phase transition. The idea sketched in the picture below was first formulated by Tammann
in 1933. He drew an analogy between the transition from plastic to brittle in drying clay masses and
the glass transition. In the picture below, the small squares have vibrational, rotational and
configurational degrees of freedom like in a liquid. A gradual reduction in size of the box leads to a
sudden loss of degrees of freedom, first the configurational and finally the rotational degrees of
freedom. A material that has a similarly structured hierarchical structure would freeze within a narrow
parameter interval and only retain the solid-state vibration degrees of freedom without, however,
achieving a crystalline order.
 Page 21

Another analog for hierarchical structures is provided by densifying traffic on a motorway:

With gradually increasing traffic density, degrees of freedom (independent choice of speed and lane)
are lost. Cooperative clusters are formed that move like a single, larger object. Freezing into a rigid
state (traffic jam) happens suddenly (hence the risk of accidents at the end of the traffic jam!). A solid-
state-like status is achieved with no recognizable topological order. The terms "flowing traffic",
"viscous traffic" intuitively lean towards viscosity.
 Page 22

Both the traffic jam and the parking lot, shown in the picture above, feature a "solid" with no flow with
the fixed mean distances between all elements and the number of degrees of freedom being "solid-
like". However, only the parking lot shows a “translational symmetry”.
With regard to what has been said above, the activation energy EA (in the sense of an activation energy
for the viscosity) is a function of the temperature in the case of glass and it is not identical to the
binding energies. For example, EA(η) is 310 kJ/mol for B2O3- and 710 kJ/mol for SiO2-melts, but the
binding energies are very similar (B-O: 460 kJ/mol, Si-O: 444 kJ/mol). Ergo: no bonds are "broken" with
viscous flow. There is rather an assumption of a layer movement.
However, EA(η) and EA(diffusion) are similar. So, there are similar basic mechanisms behind both
processes. Hence the connections via the well-known Stokes-Einstein relationship with
λ = hydrodynamic radius of the particles. In the case below, it is still linked to the movement of the
particles within a time t (relaxation time). The diffusion coefficient D is then the quotient of the mean
square displacement 〈𝑟(𝑡)G 〉 and the time t:
〈"($)! 〉 ( )
"
𝐷= '$
= '*+,

However, since viscous flow is a cooperative phenomenon, the Stokes-Einstein relationship only
applies directly in the limiting case of high T or large free volume.
The thought experiment below illustrates the dramatic effect of viscosity increase on relaxation time.
Relaxation time τ [s] is the time that a molecule needs to travel the length of its diameter.

Temperature Viscosity Relaxation

T [°C] η [dPa s] time τ [s]
SLS glass 300 1019 506
450 1015 5400
550 1013 (*) 50
1500 100 10–1
water 20 10–2 10–12
 Page 23

2.5. The Concept of Fictive Temperature

We have already learned that the speed of changing a suitable parameter (most frequently the cooling
rate q) can determine whether, for example, a melt solidifies glassy or crystalline. The faster a melt is
cooled down, the better the crystallization can be suppressed (the viscosity at Tliq of course also plays
a role, as we have already seen).
If you now cool a melt and come into the area of glass transition, you leave the (metastable)
thermodynamic equilibrium and enter the area of supercooled, solidified melt in which there is no
longer any thermodynamic equilibrium. The structural units can relax up to Tg (1013 dPas). Below Tg
this is no longer the case in realistic times. Tg also shifts with q. The higher q, the higher Tg. The
relationship between cooling rate and glass transition is quantitatively represented by the Bartenev-
Ritland equation, with A and B as constants:
1 d𝑇
= 𝐴 − 𝐵 log 𝑞, with 𝑞 = −
𝑇( d𝑡

Thus, the time needed for passing the range Tg also determines which structural state is frozen in the
glass. The molecular mobility of the building blocks decreases with cooling. From a certain temperature
onwards, the time it takes for the molecular building blocks to reconfigure is as high as the cooling rate
that was applied. The configuration changes slow down and ultimately a glass is created.
This structural state in solid glass is not an equilibrium state; so it does not reflect the temperature of
the glass at a later, colder point in time. This is a crucial difference to a crystal: In the crystal the current
temperature corresponds to the "temperature" of the structure. The crystal is always in structural
equilibrium; the glass is not.
However, since this structure is decisive for many properties of the glass, it is interesting to “know”
this structure or the “temperature of the structure” at which these relaxation processes were frozen
in the glass. The temperature at which the relaxation state was frozen is called the fictive temperature
Tf. Like Tg, it is depending on the cooling rate q (see figure below) and determined during heating;
namely by the fact that the glass then falls back (relaxes) into its equilibrium structure due to the
regained structural "mobility" (hence the term "structural relaxation").
 Page 24

If Tg is passed through slowly at a low cooling rate, the units have more time to arrange themselves.
The structure comes “a little closer” to the crystalline state and a smaller molar volume is frozen, Tf is
lower. Conversely, if Tg is passed through at a high cooling rate, a larger molar volume is frozen and Tf
is higher. The fictitious temperature describes the current state of relaxation of the glass. This
temperature Tf thus contains information about the history of the glassy material and its structure. Tf
is defined thermodynamically as the point of intersection of the enthalpy curves of the glass and the
liquid state.
But what has been explained above also means that the energy differences between glass and iso-
chemical crystal depend on Tf. The higher Tf, the higher is Hvit (= enthalpy of vitrification, that is the
notation for the energy difference between glass and iso-chemical crystal) and consequently of course
also ∆cp. But since, as mentioned before, Tg and Hvit only vary noticeably with extreme variation in the
cooling rate (see picture below), one can nevertheless introduce "the" glass state as a defined term if
one moves within the "normal" technical limits, where then Tf and Tg are very close to one another. At
high cooling rates (e.g. several 100 K/s), however, Tf and Tg can be 100 K or more apart.

In other words: Tf is the temperature at which the equilibrium structure of the melt is identical to the
structure that the glass is currently showing, which is equivalent to the temperature at which the
structure of the glass was effectively frozen. The fictive temperature Tf is a measure of the molecular
order and the structural equilibrium of a glass (structural relaxation) as a function of the cooling history
and is the temperature at which the system falls out of internal equilibrium during cooling.
An example: If you hold a glass in the glass transition area for a certain time and then quench the glass,
the structure is frozen, that is characterized, by the holding temperature and time at which the glass
was previously held: Tf.
The fictive temperature Tf is not the same as Tg. The difference between Tf and Tg depends on the
cooling rate and increases with increasing cooling rate. Tg is also influenced by the measurement
method and the heating rate used.
A Tg determined by means of DSC at a heating rate of 20 K/min will differ from a Tg determined by
means of a dilatometer at a heating rate of 4 K/min. And none of these values are the fictive
 Page 25

temperature. The fictive temperature Tf is almost always lower than (at most equal to) Tg. As an
example, investigations by Badrinarayanan et. al. (J. Non. Cryst. Solids 2007) on Polystyrene showed:

Dependence of Tg and Tf on the cooling rate (q) from capillary dilatometry and DSC
q [K/min] Tg [°C] Tf [°C] Tg – Tf [°C]
Capillary dilatometry 0.2 94.6 ± 0.2 - -
0.1 93.8 92.4 1.4
0.03 92.9 ± 0.2 91.4 ± 0.1 1.5 ± 0.3
0.01 91.5 90.6 1.1
0.003 90.2 89.5 0.7

DSC 30 101.5 ± 0.3 100.2 ± 0.1 1.3 ± 0.4

20 101.1 ±0.3 99.7 ± 0.1 1.4 ± 0.4
15 100.6 ± 0.1 99.5 ± 0.1 1.1 ± 0.2
10 99.9 ± 0.2 99.1 ± 0.2 0.8 ± 0.4
4 98.9 ± 0.2 98.4 ± 0.3 0.5 ± 0.5
3 - 98.0 ± 0.1 -
2 98.3 ± 0.2 97.5 ± 0.1 0.8 ± 0.3
1 - 96.9 ± 0.2 -
0.3 - 95.6 ± 0.1 -
0.1 - 94.5 ± 0.1 -
0.03 - 93.8 ± 0.2 -
0.01 - 92.7 ± 0.2 -

As already mentioned, the structure and the macroscopic properties of the glass depend on Tƒ. The
fictive temperature influences e.g. the strength or refractive index of a glass. This is easy to understand
when you reconsider that a larger or smaller molar volume is being frozen. The larger molar volume
results in a lower density, lower refractive index and lower strength or hardness as is shown below:
 Page 26

2.6. Angell Plot

From a scientific point of view, the viscosity of a glass can be plotted as a function of a universal,
dimensionless temperature scale "Tg/T" (so-called "ANGELL plot", according to A. Angell). The point
Tg/T = 0 corresponds to the hypothetical liquid state at infinitely high temperatures. Glass melts reach
a high-temperature viscosity limit at L0 = log η ≈ –4 (η in dPa·s), valid for this hypothetical, liquid state
at T = ∞ (condensed matter cannot become thinner). At Tg/T = 1 the glass transition is reached; here
approximately Lg = log η(Tg) = 13 applies. The course of the curve between these two boundary layers
depends on the degree of crosslinking (degree of polymerization) of the melt. An infinitely extensive
polymer network reaches the glass state through interdependent steric hindrance of large structural
units. The cp jump of such a melt at Tg disappears (there is no configurational or rotational contribution
to cp!).
For melts that essentially freeze due to the loss of configuration degrees of freedom within individual
(as will be introduced later: MRO) areas, there is a deviation in the viscosity curve towards lower values
and a steeper increase in viscosity in the range T → Tg. A clear cp jump is observed during the glass
transition of these melts.

If the slope m of the curve with which T approaches Tg is increasing (becoming steeper), then it is called
a fragile glass. The glass then shows great deviations from the behavior of ideally polymerized melts.
m is thus a measure of the degree of polymerization (degree of crosslinking) of the structure.
If the slope of the curve with which T approaches Tg is constant, then it is a strong glass. The glass
shows no deviation from the behavior of ideally polymerized melts. The fragility also shows how much
the temperature/viscosity dependency deviates from the Arrhenius behavior.
The concept of fragility has nothing to do with the strength of a glass. It goes back to the terms “fragile
vs. strong glass formers” originated by A. Angell. With a “strong” glass former, the topology and
structure change only slightly with temperature. The viscosity shows approximately Arrhenius
behavior. In the case of a “fragile” glass former, the topology and structure change very strongly with
temperature (mostly from very small units at high temperatures to large clusters at low temperatures).
The fragility of a glass former is given by the gradient of the viscosity-temperature function in the
vicinity of Tg and the fragility of a glass former is determined by the height of the cp jump and the
amount of entropy frozen at Tg. The quotient ∆cp/Svit thus determines how steep the course of the
viscosity-temperature function is for T → Tg; ∆cp/Svit is also called a structure parameter.
 Page 27

These relations can be clearly seen in the Angell plot, as demonstrated for polystyrene melts in the
figure below.

With a higher degree of polymerization M, the ability to rearrange coordination for the purpose of
relaxation decreases. Relaxation (and increase in viscosity) is only achieved by relaxation
(displacement) of the (fully cross-linked) polymer units with respect to one another.

2.7. Viscosity Measurement

In glass production, the viscosity varies over several decimal powers and is generally represented
logarithmically. An extreme example is o-terphenyl, the viscosity of which increases by around 15
decades during the glass transition. Such a large, continuous change in a measured variable is unique
in physics.
The methods for measuring the viscosity must be adapted accordingly. There is no measuring method
with which all viscosity ranges can be recorded. For individual measurement methods, please refer to
the lecture slides. However, a few particularly relevant methods should be repeated here.
The rotational viscometer can be used from T = 1000 to about 1450 °C, corresponding to about log h
= 3.0 to 1.0 and represents a direct viscosity measurement. A suitable cylinder is immersed in a melt
and the torque required to maintain the rotational speed is measured at a constant temperature. The
measurement corresponds to a parallel plate viscometer in which the two plates are exchanged for
cylinders.
The dilatometer only allows an indirect viscosity measurement, but is well suited to determine two
characteristic points in glass technology: Tg and TD. A rod with a defined length is heated up. The
characteristic change in the coefficient of thermal expansion is evident in the glass transition Tg. And
at the dilatometric softening point TD, the sample loses its shape stability and can no longer be
measured in the dilatometer:
 Page 28

Effects of the fictive temperature can also be clearly observed in the dilatometer. If a glass is cooled
down quickly and then heated up again at a lower heating rate, the glass relaxes in the area of the
glass transition into the structure with the lower temperature:

Another method, although only approximate, but very quick and therefore practical, is the heating
microscope. It is suitable for quick approximation determinations, especially for semi-crystalline
systems, e.g. glass solders. One observes the shadow cast by a glass sample in an oven. Depending on
the temperature, the sample shows changes in its general shape:
 Page 29

At the softening point (SP) the corners of the sample are rounded off. In the case of a soda-lime silicate
glass, this point corresponds to a certain viscosity. At the hemisphere point (HP) the height of the
sample is half of the base area and at the flow point (FP) the height of the sample is 1/3 of the initial
height, which in turn corresponds to a respective viscosity. There are different approaches to
determining the viscosity, with some significant differences. It is therefore advisable to find out about
the model that you want to use before using the method. Some values of the models are shown below
(from Pascual, et. Al., "A new method for determining fixed viscosity points of glasses", 2005).
Scholze Pascual et al This work
Viscosity points
logη±s (P) logη±s (P) logη±s (P)
First shrinkage 10.0 ± 0.3 8.9 ± 0.25 9.1 ± 0.1
Maximum shrinkage 8.2 ± 0.5 7.9 ± 0.2 7.8 ± 0.1
Deformation 6.1 ± 0.2 6.6 ± 0.1 6.3 ± 0.1
Sphere - - 5.4 ± 0.1
Half ball 4.6 ± 0.1 4.5 ± 0.1 4.1 ± 0.1
Flow 4.1 ± 0.1 3.1 ± 0.15 3.4 ± 0.1

In general, with this method, several viscosity-temperature pairs can be obtained very fast and used
e.g. for a VFT description.
 Page 30

3. Structure of Glasses
Unlike crystalline solids, glasses show no long-range order. This becomes visible, amongst other, in X-
ray diffraction experiments, where no discrete intensity maxima (diffraction peaks) can be observed
for a glass sample. Instead, a broad and not very discreet rise in the background (see below on the
right) is obtained, which, however, still contains some information. It is interesting to note, that the
position of the hump’s maximum is in good agreement with the main peaks of the crystalline reference
phases —even in the case of multicomponent glasses. The left part of the figure below shows a Debye–
Scherrer pattern of a glass and a crystalline solid.
Discrete diffraction rings, which represent the peaks in the one-dimensional diffractogram, are clearly
visible in case of the crystalline sample. The intensity of those rings in the left pattern corresponds to
the peak height in the right one (information on texture or internal stresses is of course lost in the one-
dimensional display, or has to be brought to light again through suitable measurements). For the glass
sample, just a diffuse intensity distribution can be observed corresponding to the broad hump in the
pattern on the right.

However, if we now evaluate the half width B (FWHM = Full Width at Half Maximum) of such a "glass
hill", according to:

Peak broadening b calculated via FWHM*:

1 𝜋
𝑏= ∙ (2𝜃234 − 2𝜃215 ) ∙
2 180°
(22±4)° → b = 0.0698

we obtain a possible correlation length b or, alternatively, a fluctuation ε around a mean translation
symmetry. The interpretation of both approaches is as follows:
Interpretation I: Half width B (of a certain reflection at a defined 2θ) translates into a correlation length
b, within which translation symmetry still applies. This would therefore indicate nano-crystallinity in
 Page 31

the glass. For example, in this "crystallite hypothesis", b would correspond to just 7 to 8 structural
repetitions per spatial direction for a -Si-O-Si-distance of 0.26 nm
Interpretation II: Peak width B and the corresponding correlation length b can be interpreted as a
HI
fluctuation ε around a mean translation symmetry, 𝜀 = of the interplanar spacing. In a quartz glass,
I
the fluctuation would correspond to a variation of the mean atomic distance of approx. 9 %. This would
indicate that a glass is an (amorphous) form of a corresponding crystal with fluctuations in the mean
spacing around a translational order so large that discrete diffraction peaks can no longer form.
Lebedew as well as Zachariasen & Warren both presented their own theories on the structure of
glasses. Lebedew’s theory was clearly disproved with the development of increasingly better
transmission electron microscopes. No evidence for nano-crystallinity in the glass structure could be
found in the propagated size range. Thus, over time Zachariasen & Warren‘s random network theory
prevailed. Glass is generally referred to as an "infinitely" extended amorphous network with
fluctuations in the mean distance around a translation order. Typical illustrations can be found in
almost all relevant textbooks:

However, this representation is not completely satisfactory, at least if you want to dive more deeply
into the structure of glass. For example, effects like ion conduction in glasses (see below) cannot be
explained with such a simple picture.
In crystalline solids, lattice defects of zero dimensionality (point defects) determine the ionic
conduction. In glasses, however such defects do not exist. So how should e.g. a Na-ion (shown in the
structure below) be able to migrate through the network? We'll get to that later.
 Page 32

3.1. Heat Capacity of Glasses and Its Correlation to the Structure

We now want to take a closer look at the structure of glasses in order to get a deeper insight.
Therefore, we first try to compare the heat capacities of a glass and its iso-chemical crystal.
Let us consider every atom (every “particle”) of a structure to be a discrete oscillator. Each particle has
a kinetic energy of Ekin = ½·kB·T per degree of freedom and a potential energy equal to the mean kinetic
energy. This results in a total energy Etotal of 2·½·kB·T = kB·T for each degree of freedom. For simply
structured, crystalline solids (e.g. mono-atomic metals) and sufficiently high temperatures (RT), the
law of Dulong and Petit applies as a good approximation for the molar heat capacity: cn = 25 J/(mol K):

As there are 3 degrees of freedom in the solid state:

𝐸EJEK8 = 3 ∙ 𝑘L ∙ 𝑇 per atom
𝐸EJEK8 = 3 ∙ 𝑘L ∙ 𝑇 ∙ 𝑁M = 3 ∙ 𝑅 ∙ 𝑇 per mole

If a solid has a complex structure (composed of more complicated molecules), the three degrees of
freedom of the lattice vibration per particle are added to the degrees of freedom of the molecular
vibration. cn can then be significantly higher than predicted by the rule of Dulong–Petit.
The law of Dulong and Petit deviates from the experimental results at low temperatures. The heat
capacity of solids decreases at low temperatures and tends towards zero for T → 0 (third law of
thermodynamics).
For better predicitons, Debye developed a model which takes into account elastic vibrations of the
whole body instead of independent, individual vibrations of the individual atoms:
 Page 33

However, these have different frequencies depending on the wavelength and are therefore
temperature-dependent.

The Debye model describes cn as a function of temperature. From these considerations, the well-
known course of the heat capacities of solids above and below their Debye temperature follows:

The Dulong-Petit Limit of 3·R = 24.9 J·mol–1·K–1 applies if all possible phonon frequencies are occupied
in a (simple) solid. This case is valid above the so-called Debye temperature θD (material-dependent,
 Page 34

all phonon frequencies occupied) with θD = h·νD/kB (with h = Planck’s constant), νD = Debye cut-off
frequency and kB = Boltzmann's constant).
3·R is reached the sooner, the smaller θD and the lower the cut-off frequency νD. θD on the other hand
is lower, if the bond coupling of the vibrating atoms is "softer" (the "softer" the solid is). Or to put it
another way: A low θD means that all vibrational modes of the crystalline lattice are excited even at a
low T.
As can be seen from the picture above, cv lies on a “master curve” at the transition to low
temperatures, which tends towards 0 in proportion to T3. In the case of electron conductors, the
contribution of the electron gas to the heat capacity results in a linear trend of cP towards T = 0. At
extremely low temperatures, the T3-law therefore changes to a T1-law. In the case of isolators, the T3
proportionality also applies there.
Being monoatomic solids, metals are mostly in good agreement with the Dulong-Petit law. Due to the
metallic bond, one could initially expect otherwise, since the atoms release electrons from their outer
electron shell, which can move freely through the crystal (electron gas). Each electron would have to
contribute 3 degrees of translational freedom. Assuming that if each atom releases an electron, the
molar heat capacity would have to be 3·R+3·R/2 = 9/2 R.
However, since all states below the Fermi distribution are already occupied in the electron gas, most
electrons cannot achieve a state of higher energy and therefore also cannot contribute to the heat
capacity.
Let us compare the heat capacities of a glass and it’s iso-chemical crystal, bearing in mind the facts
discussed above. It can be seen that crystal and glass differ directly below Tg, since the mentioned
relaxation effects based on the fictive temperature come into play here. However, this does not change
the fact that the course of cp a little further below Tg is identical for both, glass and iso-chemical crystal
(see below):
 Page 35

This necessarily leads to the conclusion that the short-range order (SRO) in the glass is identical to that
of the iso-chemical crystal since cP (crystal) → cP (glass). We have already learned about this in the
chapter on thermodynamics of the glass transition. So far, so good.
The picture changes when considering low temperatures. More complex crystal structures then always
deviate slightly from the T3 law explained above. Glasses, however, at low temperatures fundamentally
deviate from the T3-law by a significant degree:

(Note: In the image shown above, it should be noted that the area around Tg is explicitly not a "digital
jump". The jump from cp at Tg does not have a sharp course (blue curve) as it is found in a phase
transition of the first order (red curve). The upper picture further neglects the structural relaxation and
the exact description of the losses of configuration and rotational degrees of freedom (see later
chapter), being an idealization in order to show the quasi-identical cp curves below Tg.)
From a plot of the ratio cp/T3 over T, one can estimate the position of the maximum deviation from the
T3-law. The corresponding Temperature Tmax can be interpreted as a frequency or, if the speed of sound
is known, as a wavelength, which is just 2–3 nm.
 Page 36

This implies, that at the lowest temperatures structural units of 2–3 nm in size must vibrate against
each other before neighboring atoms can vibrate against each other in any distribution.
This further hints at a structure in which atoms are relatively firmly linked to one another in areas of 2
nm, with a lower degree of linkage being realized between the individual areas. This consideration is
an essential modification of the network hypothesis of Warren and Zachariasen. Indeed, nowadays it
is assumed that glasses have a short-range order (SRO) as well as a medium-range order (MRO) in the
range of a few nm. Crystals, on the other hand, have a SRO and a long-range order (LRO)
The existence of a medium-range order is supported by two further observations: By carrying out
extended X-ray diffraction experiments on glasses, one obtains a distribution function (nothing else is
an X-ray diffraction diagram) that is similar to that of a liquid:

Accordingly, at certain distances there are greater accumulations of atoms around a central atom than
at other distances. A clear proof that a glass is not amorphous (i.e. shapeless), but that there is an
order beyond the direct polyhedral coordination.
Finally, drawing a randomly disturbed, computer-simulated, network with e = 0.09 (derived from our
b-value via the previous FWHM investigation), one receives further confirmation:
Let us take an undisturbed, for the sake of simplicity, 2-dimensional lattice and distort it statistically
with ε = 0.09. Each line connecting two points having a distance above/below a certain limit (here
arbitrarily ±𝜀 ∙ √2) is removed, since from a certain distortion on, a bond can no longer exist. Thus,
only bonds with 81 − 𝜀 ∙ √29 < 𝑑 < 81 + 𝜀 ∙ √29 are accepted as valid bonds (assuming a bond length
of 1 in the undistorted lattice). We get an image (see below) that shows more and less strongly linked
areas. After coloring these areas two networks become apparent: a coherent network (so-called
"bond-percolation"/"bond percolation"), which is responsible for the strength, and a network of open
channels (so-called "void percolation"/"diffusion path percolation”) with which properties such as the
aforementioned ionic conduction in glasses can be easily explained: Ions can migrate through these
channels inside the network at an appropriate high level of excitation.
 Page 37

At this point, let us summarize our excursion into the glass structure:
The short-range structure (SRO) of a glass is very similar to that of the iso-chemical crystalline state.
Additionally, glass is not "disorganized". There are two hierarchies of order: the short-range order
(SRO) and the medium-range order (MRO). Local energetic compensation is provided by the size of the
MRO units.
As an analogy: The crystal also has two hierarchies of order: the short-range order (SRO) and the long-
range order (LRO). Local energetic compensation takes place here through defects (displacements,
vacancies, ...)
The glass transition is caused by the loss of (liquid-like) configurational degrees of freedom within the
MRO units and the viscous flow is caused by the relative movement of the MRO units. As in the
crystalline state, the defects and dislocations determine the mechanical properties and diffusion, so
do the clusters, their percolation and the percolation of bond-free areas in an irregular network (and
these thus explain the previously described ionic conductivity).

3.2. Structural Models and Percolation Theory

(from the Latin word „percolare“ = to filter, to trickle through)
We have already seen that glass is a material without a crystalline microstructure. So how can we
precisely tailor the properties of a such a material? This can only be done by changing the structure
and topology at the atomic level, the key word being „topological engineering” or „chemical bond
engineering”. Therefore, material science of glass in this area is much more similar to that of organic
polymers than that of metallic or ceramic materials. It thus becomes evident, that the inclusion of glass
into the basic group "ceramics", as is often done in classification of materials, is clumsy and only refers
to the material systems, but does not reflect the above-mentioned aspects.
This gives rise to the perfectly valid question of how a medium-ranged aggregation (clusters) can arise
through a random (i.e. unordered) process. Let us perform a thought experiment:
Let p be the probability that a point (or an edge) in the picture below is occupied. Networking occurs
at pc, the critical probability. If p is the probability that a field is occupied, larger clusters are formed as
p increases (see below).
The so-called percolation threshold pc is then defined as the value of p at which at least one cluster
reaches a size that extends through the entire system, i.e. extends on the grid from the right to the left
 Page 38

and from the upper to the lower side. In this case the cluster percolates through the system. There are
thus two possible growth mechanisms:
a) Adding a bond
b) existing clusters are connected to each other through a bond (degree of cross-linking increases
much more strongly)
In the figure below, the percolation threshold is reached by occupying only 3 additional points in each
direction, although p increases by only 1% from figure c) to d). pc depends on the size of the network,
but also on the type of network (hexagonal, triangular, cubic, ...). The percolation limit of a structure
with corner-linked tetrahedra (as found in many glasses) is pc = 0,43.

The question now arises as to how such an extended network (no LRO!) can be linked to a medium-
range order (MRO).
In the following consideration, we assume the classic 3D hard-shell model. We have learned that melts
that lie on high slope curves in the ANGELL plot form glasses due to structural sphere packing
frustration (“fragile glasses”). If spheres of the same size are packed tightly, then there is a conflict
between locally closest and most densely packed methods. The tetrahedron is optimally packed
locally. Tetrahedra, however, cannot be arranged in a space-filling manner. A frustration with a 7.4%
gap remains; no translation symmetry is possible. Densely packed crystalline structures, hcp or fcc,
avoid this by compromising the formation of an S.R.O with additional octahedron gaps:
 Page 39

The next optimum is the icosahedron, which is made up of slightly distorted tetrahedra. F.C. Frank
(1952) therefore proposed the icosahedral packing for glasses. But even icosahedra cannot be packed
to fill a room. Here, too, “gaps” remain and the icosahedron connection changes from chains to rings
to an open network with greater expansion. If one considers the potential energy within the structure
for such an open network (for such clusters), one finds energy minima for different cluster sizes
depending on the selected energy model:

It is noticeable that in both cases there is an energy minimum for cluster sizes of N = 78 atoms. With
the different potential types, these clusters correspond to a network with gaps or one with a chain
structure:

Both structures are found in glasses (e.g. soda-lime-silicate or phosphate glasses), which show a
tetrahedral or distorted icosahedral short-range order. This is a further indication that corresponding
glasses can form densely packed areas that are statistically less densely connected to one another.
 Page 40

4. Thermodynamics of glass
As an introduction, some changes in properties will be explained based on the temperature
dependence of the volume of crystal and glass when heated. In the case of a crystal (or the crystalline
reference system), the volume increases sharply during heating when TM is reached due to the phase
transition from solid to melt. In contrast to this, the volume-temperature curve of the glass when it is
heated, due to the same short-range order but higher enthalpy and entropy, runs approximately
parallel to that of its crystalline reference system up to Tg. Its slope changes at Tg (to that of the liquid
melt). Further qualitative changes in a system with temperature and their derivation up to the
occurrence of discontinuity can—for crystalline phases—in thermodynamic equilibrium be classified
according to Ehrenfest using the systematics of phase transformation. A glass transition, however, is
not a transformation of the nth order, in the Ehrenfest sense. The n+1th-derivative of the cp curve never
runs towards infinity.

In addition, the diagram above shows the difference between enthalpy and entropy between glass and
iso-chemical crystal (the so-called crystalline reference system), given by the so-called vitrification
enthalpy Hvit and entropy Svit. These variables are discussed in more detail below. For the moment we
just want to state that Hvit and Svit describe some kind of enthalpic and entropic difference between a
glass and a crystal of the same chemistry.
 Page 41

When assessing the figures above, a few points should be pointed out: The course of the heat capacity
is calculated at Tg according to Dulong-Petit. The course of the enthalpy-temperature-curve below Tg
is approximately described by a polynomial:
)
1 1 1 )
/ 01 2
Δ𝐻 = 𝐻(𝑇) − 𝐻(𝑇- ) = * 𝑐. ∙ 𝑑𝑇 = .[𝐴 ∙ 𝑇 + 𝐵 ∙ 𝑇 − 𝐶 ∙ 𝑇 + 𝐷 ∙ 𝑇 + 2𝐸 ∙ 𝑇 / 8
)# 2 3 )#

Above Tg, the heat capacity is described by a straight line with an approximately constant value:

𝑐.,456 = 9 𝑐.,( (𝑙𝑖𝑞) ∙ 𝑛(

(

Therefore, by differentiating the polynomial for the course of the enthalpy-temperature curve below
Tg, the polynomial for calculating the heat capacity below Tg can be obtained:
1
𝑐. = 𝐴 + 𝐵 ∙ 𝑇 + 𝐶 ∙ 𝑇 0/ + 𝐷 ∙ 𝑇 / + 𝐸 ∙ 𝑇 0/
cp,liq can be obtained from the slope of the enthalpy-temperature curve above the transformation
temperature Tg in the same way. Finally, the jump in heat capacity at Tg can be calculated using

∆𝑐. = 𝑐.,$%& − 𝑐.,'($)*+,-./%/

Using the approaches mentioned above, it is possible to determine the cp curve of a glass. Between
298 K and Tg, the heat capacities of glasses and their crystalline reference system show—at leat
theoretically—identical curves, which can be approximated by the above-mentioned polynomial. At
lower temperatures, deviations occur, which can be attributed to a medium-range order (MRO) in the
glass.

4.1. Thermodynamics of the Glass Transition

Due to the previously explained loss of degrees of freedom in the glass transition (configuration and
rotation), there is also, for example, a decrease in the heat capacity and the thermal expansion
coefficient:
 Page 42

At this point it is advised to introduce some fundamental terms that characterize the glass, the iso-
chemical crystal and the corresponding melt. Let us consider the following aspects:
The solid, iso-chemical, crystalline state of equilibrium corresponding to a respective glass (or glass
component) should, as introduced above, be called the “crystalline reference state”. The left part of
the figure below shows that a glass is described relative to its crystalline reference state by the position
of the glass transition temperature Tg and the magnitude of the glass formation enthalpy Hvit, as we
G %
have already seen above. A rule of thumb states Tg = ·TM (on the Kelvin scale) and Hvit = ·Hm. The
N G
precise position of Tg and the size of Hvit depend on the cooling rate q. In this respect, Tg and Hvit can
be seen as the two additional parameters that are needed besides the state variables in order to
quantitatively characterize a glass state.
On the other hand, we can specify a cooling rate (for example: 2 K·min–1 at p = 1 bar) and thus define
a very specific glass state. Since Tg and Hvit vary noticeably only when extreme variations in the cooling
rate occur, one can introduce “the” glass state as a term (see chapter on the fictitious temperature).
The right picture shows the enthalpy and entropy difference between glass and the crystalline
reference state. Below Tg, constant values for Hvit and Svit are retained down to T = 0 K. Between Tg and
Tliq the values rise to Hm and Sm, respectively. At Tg the heat capacity (= slope of the enthalpy curve)
increases by the value Δcp = cp,liq – cp,gl. As a rule of thumb, we can assume:
%
Svit = N ·Sm. Tm = Tliq (please be aware that Tm and Tliq are used synonymously in this script), Hm, Sm
represent melting temperature, enthalpy and entropy of the system.
 Page 43

If one now looks at the states of melt, glass and iso-chemical crystal, further important observations
can be made:

(Note: Strictly speaking, Hvit and Svit have to be added to the above approaches in the picture for HC
and SC! The above therefore only applies if you come from the melt and determine HC(T) and SC(T)
down to Tg. The entire course of HC(T) and SC(T) is determined starting from 0 K: When determining
the entropy difference between glass/supercooled melt vs. crystal, integration is always carried out
 Page 44

from T = 0 on. Thus, Hvit and Svit become an integral part of HC(T) and SC(T). For the entire integral of
SC(T) we have:
0
𝑐8DP (𝑇) − 𝑐Q (𝑇) 𝑇
𝑆O (𝑇) = 𝑆 CDE + X d𝑇 ≅ 𝑆 CDE + ∆𝑐 ∙ ln
𝑇 𝑇:
0(

Where Svit describes the part below Tg and the integral describes the part Tg up to a maximum of TM.

The course of the heat capacity of the glass or the crystalline reference phase up to the melting point
Tm determines the relationship between the glass-related quantities Hvit, Δcp, Tg and the equilibrium
quantities Hm, Tm. In the figure above it is shown that the same applies to Svit, Δcp, Tg on the one hand
and Sm, Tm on the other. (For later transfer to multicomponent systems, from here on the melting point
Tm is replaced by the liquidus temperature Tliq and the heat of fusion Hm and melting entropy Sm by the
quantities Hfus and Sfus assigned to Tliq, respectively. Tliq marks the upper limit of the melting range. Tliq,
Hfus and Sfus are equilibrium properties (just like Tm, Hm, Sm); only two of the three quantities are
R )*+ R/
independent of each other in each case: Sfus = 0,-.
and Sm = 0 . In the same way, only two of the four
,-.

glass-related variables Tg, Δcp, Hvit, Svit are independent of each other. Thus, there are two additional
parameters that are needed to describe the state of the glass. The quantity Hc(T) describes the gradual
increase of the shaded area in the image before, when T increases from Tg to Tliq.
In the same way as described above, Sc(T) describes the increase of the shaded area with temperature
from Tg to Tliq in the diagram shown above. The area is bounded by the course of the cp/T curves of
glass and crystalline reference state. Sc(T) is called configurational entropy. Sc(T) takes the value Svit
between 0 K and Tg and increases to the value Sfus between Tg and Tliq. Sc(T) quantitatively summarizes
the structural differences between glass and the crystalline reference state—whatever the structure
may look like in the spatial image. This may not satisfy scientific curiosity, but it is invaluable for the
quantitative description of processes involving the glass and its melt. We summarize:
 Page 45

0,-. 0
Hvit = Hfus – ΔcP (Tliq – Tg), Svit = Sfus – ΔcP ·ln 0(
= Sfus + ΔcP ·ln 0 ( .
,-.

In the picture below the difference of the observation from 0 K to Tg and above Tg is explained again.
The entropic differences between glass and crystal from 0 K to Tg are summarized here as Sexcess. Later,
we will almost always use the previously introduced term of vitrification entropy, Svit, for this. Svit is a
simplification and approximation of Sexcess:
'S
𝑐+9388 (𝑇) − 𝑐7:;8<39 (𝑇)
𝑆 647688 = H d𝑇
𝑇
&

If one compares the well-known states of gaseous, liquid and solid with the glassy state, it is evident
that the liquid state and the glassy state are very similar with regard to their degrees of freedom. To
this end, a lot of what we don't understand about the glassy state is due to our lack of understanding
of the liquid state.
The gaseous state is described by the kinetic gas theory and for an ideal gas with three degrees of
freedom (N) and A particles, the internal energy is described by:
3
𝑈= ∙ 𝐴 ∙ 𝑘L ∙ 𝑇
2

For a solid body, the phonon theory applies and with this assumption of phonon oscillations

𝑈 = 3 ∙ 𝐴 ∙ 𝑘L ∙ 𝑇

is obtained. However, neither for a liquid nor for a glass does such a unifying theory exist. However,
many liquids obey
 Page 46

𝑈 = 𝑁 ∙ 𝐴 ∙ 𝑘L ∙ 𝑇

with 3.0 < N < 4.0, as can also be seen in the picture below.

4.2. Glass and the crystalline reference state

In the images shown previously, it was assumed that the cp curves of the solid glass and the crystalline
reference state have approximately the same values shortly below Tg. In real systems, however, this is
by no means the case, as is shown by the example below. We therefore have to modify the concept of
the crystalline reference state a little bit: The crystalline reference state is that solid equilibrium state
which would be stable as a crystal phase at the temperature Tg of the associated glass. If the definition
is made in this way, then the assumption made for cp is correct.
So, for the NAS6 glass shown below, the high-temperature modification of albite is the crystalline
reference state. For a pure SiO2 glass, it is cristobalite, not quartz.
In general, the following applies: The glass transition takes place as soon as cp exceeds a value of 3·R
per gram atom (= number of atoms in the stoichiometric formula, Dulong-Petit-Limit). With cp = kB per
degree of freedom in condensed matter and with three degrees of freedom per atom, one obtains
cp (T = Tg) = 648.5 J·mol–1·K–1 for the example shown below (R = 8.314 J·mol–1·K–1).
 Page 47

The relationships described above lead to the curves of the density and the heat content of a glass
shown above. In industrial practice, the density of a glass product varies at most on the 4th decimal
(dρ < 0.0003 g/cm3); the refractive index of optical glasses even varies only on the 5th decimal. It is
therefore possible to set "the" state of the glass very precisely.
To clarify:
H0 = standard enthalpy (298 K, 1 bar)
for the crystal stable at T = Tg
S0 = standard entropy (see above)
cp(T) = A+B·T+C/T2
Hfus = enthalpy of fusion
Sfus = entropy of fusion
Tliq = liquidus temperature
Hvit = vitrification enthalpy
Svit = vitrification entropy (zero Kelvin entropy)
Tg = glass transition temperature
Δcp = jump of heat capacity (glass) at Tg

To distinguish between glass and crystal, let us summarize once again:

ΔHf = Hfluid – Hcrystal = Hfus
ΔHg = Hglass - Hcrystal = Hvit
ΔSf = Sfluid – Scrystal = Sfus
ΔSg = Sglass – Scrystal = Svit
 Page 48

4.3. The concept of constitutional phases

Note in advance: The term crystalline reference phases is used here synonymously with the term
constitutional phases. Additional information about the topics covered in this chapter can be found in
chapter 10.1.
We have seen that the local structures present in glasses and in the iso-chemical crystals are identical.
Below Tg, glass and the iso-chemical crystal differ (only) by the corresponding glass-determining
variables of the vitrification entropy Svit and the vitrification enthalpy Hvit. These—on the other hand—
depend on the MRO/LRO (but not the SRO). If we were able to determine these local structures (which
refer to the corresponding crystalline solid), we would open up the possibility of determining glass-
relevant data, such as the standard enthalpy of formation H0Glass, as long as we know the corresponding
enthalpy Hvit. In this case we could simply calculate H0Glass according to H0Glass = H0Cry+ Hvit.
The corresponding glass-determining variable can then be written as the amount of substance
weighted sum of the proportions of the constitutional phases nk (nk being either the molar or the
weight percentage, depending on the tabulated values), e.g.:
&
𝐻TUVW = r 𝑛X 8𝐻X& + 𝐻X"YZ 9

Mixtures are considered ideal mixtures, the mixture entropy Smix is negligible, since only SRO with few
atoms is relevant:
_
[D\
𝑁M ∙ 𝑘L
𝑆 = ∙ r 𝑛^ ∙ ln𝑥^
𝑛KEJ[]
^`%

For the case of a pure quartz glass, the procedure is simple. The local structural unit in the glass, as in
the crystal, is the [SiO4]4–-tetrahedron. To determine the standard enthalpy of formation of glass H0Glass,
we only have to look up the standard enthalpy of formation of cristobalite as well as Hvit (vitrification
enthalpies and entropies are nowadays often included in the corresponding tables, see also later
chapter) and we can calculate the standard enthalpy of formation of a quartz glass.
Note: The crystalline reference state is the solid equilibrium state that is stable at the temperature Tg
of the respective glass! This is usually the low-density polymorph. In the case of SiO2, this is therefore
cristobalite, not quartz. If the state is not known, one can try to determine it by theoretical
considerations.
So far, so good. But what if we are dealing with a multicomponent glass and the structural units of the
crystalline reference phases, which are present as SRO in the glass, cannot be determined so easily?
As an example, a glass from the system CaO-Al2O3-SiO2 is shown below:
 Page 49

The blue dot indicates the composition of the glass. The reference phase space, given by the immediate
phase triangle, is enclosed at the corners by quartz, anorthite and pseudowollastonite. In these
crystalline reference structures (or constitutional phases) the SROs are known and the corresponding
values for H0Cry or S0Cry are tabulated. Thus, we are now optimally prepared to calculate the standard
enthalpy of formation for the above-mentioned glass, provided that Hvit of the constitutional phases is
also known. If this is not the case, a first approximation for "normal" cooling rates q between 10 and
100 K·s–1 can be used:
∆R 0-1 ∆a 0-1
∆R /
≈ 1/2 and ∆a /
≈ 1/3.

If one tries to change the compositional range of a glass in a multi-component system, one has to rely
on appropriate thermodynamic software, such as FactSage™, which calculates the stable phases at Tliq
based on the minimum Gibbs energy.
In advance of the chapter on industrial melting of glasses, one important aspect has to be pointed out:
assuming a reference phase composition as shown in the example above, SiO2 is present as a reference
phase in the glass. This SiO2 then bears the name "constitutional silica" or "residual silica" because it
represents SiO2 that is not bound in one of the other phases by reaction (and thus cannot melt in a
eutectic, for example). This residual SiO2 must dissolve (slowly) in the melt.
What is needed now is the proportion of crystalline reference phases in the glass. For this purpose, the
reference phase space (in the ternary system composed of the phases A, B and C) is taken as a basis
and, in the example of the ternary phase diagram, the lever law for binary systems is extended to
ternary systems with the oxides a, b and c. If X represents the respective quantity of the component,
the following applies:
 Page 50

8𝑋Y6 − 𝑋Y 9𝑋 6 + 8𝑋Y5 − 𝑋Y 9𝑋 5 + 8𝑋YB − 𝑋Y 9𝑋 B = 0

with i = a, b or c

Boundary condition: 𝑋 6 + 𝑋5 + 𝑋B = 1

For a phase diagram, which is generally given in wt.-%, it must be converted into the amount of
substance. The number of moles of an oxide component a, b or c in the respective constitutional
phases A, B and C gives the total amount of this oxide in glass G. Generally formulated:

𝑛=? + 𝑛=@ + 𝑛=" = 𝑛=A

For a ternary system:

𝑛V6 + 𝑛V5 + 𝑛VB = 𝑛VT

𝑛b6 + 𝑛b5 + 𝑛bB = 𝑛<T

𝑛<6 + 𝑛<5 + 𝑛<c = 𝑛bT

Matrix of variables Target matrix

The procedure: One establishes the number of moles of the basic components i (in the example
i = a, b and c) of the system in the reference phases R (in the example R = A, B and C) (example: one
mole NC3S6 contains: 1 mole Na2O, 3 moles CaO and 6 moles SiO2). The number of moles of the
reference phase NC3S6 is unknown (n). However, we know that the sum of a basic component i (e.g.
first line above for a) of the system over all reference phases must be equal to the total amount of this
oxide in the glass G (and we know this because of the position in the phase diagram, or because of the
chemical composition). In other words: The sum of the respective number of moles of the basic
components i of the system in the reference phases R must yield the total number of moles of the
basic component i in the glass G.
This gives three equations with three variables for a ternary system. The linear system of equations is
solved either by releasing the variables and inserting them or with the help of matrix calculation.
As a basic aid to estimating the quantity ratios, it should be noted here: The closer the point of the
glass composition lies to one of the corners, the greater the associated phase content in the reference
phase. If the point of the glass composition lies on a tie-line, the quantity content of the opposite phase
is zero. The point of the glass composition must lie on the tie-line, i.e. if one side of the tie-line triangle
passes through the point of the glass composition, then the phase content and the composition of the
glass will no longer change with further cooling.
 Page 51

4.4. On the entropy of glasses

According to the second law of thermodynamics, the entropy S never decreases during a spontaneous
∆R
change of state. The entropy change ΔS is defined as ΔS = . The fact that the entropy tends towards
0
zero when the temperature approaches absolute zero (third law) applies to a perfect crystalline
substance, but not to glass.
The heat capacity of the crystal increases slightly until TΘ is reached while cgl remains constant. At Tm,
cx = cgl applies. We have already learned about this relationship before.

In entropic terms, the difference in S (or also in cp) between supercooled liquid and crystal decreases
and eventually disappears (extrapolated, dashed red curve below) when a liquid is supercooled further
below Tg:

If a liquid were to be cooled further than TC and then actually had an S or cp smaller than the crystal
(dashed red curve), this would be paradoxical. This means that the glass transition is not only a kinetic
effect (rapid cooling), but that there is also some kind of thermodynamics behind it, i.e. there is a
 Page 52

thermodynamic glass transition. The solutions to this problem is known as the Kauzmann paradox (see
also later section ''Kauzmann paradox''):
1. before the above takes place either a phase transition must occur
2. or the decrease of S/cp over T becomes smaller
3. or a 1st order liq-liq phase transition takes place
In reality, the system avoids the physically unreasonable state Ssolid > Sliquid through the glass transition.
However, this causes another "problem", which is contained in the Third Law of Thermodynamics: In
the pure crystal, the heat capacity cp ≈ cv approaches zero when approaching absolute zero according
to Debye's T3 law. The reaction entropy (the difference in entropies between products and reactants)
becomes smaller and smaller (Nernst theorem). Or, in other words, since no more particles can
oscillate in a perfect crystal at absolute zero, there can be no more entropy changes:
UYe
0→&∆𝑆 =0
By this law, however, one cannot say the absolute entropy value at zero, only that the change is equal
to zero and all substances have the same entropy at absolute zero. Planck then defined, that the
entropy of an ideally crystallized, pure solid at absolute zero is zero: ST = 0 = 0.
According to the above, however, glass would have an entropy higher than zero at absolute zero. This
controversy has been and is still being hotly debated. Computer simulations of the glass structure at
absolute zero still cast doubt on the thesis ST = 0 (glass) ≠ 0. However, the devil is also in the detail here
and the majority of glass scientists agree that glasses do indeed have a zero-point entropy: We cannot
conceive glass as an ideal solid (and certainly not ideally crystallized). In the case of glasses, as
supercooled melts, a large number of different configurations are possible. It is also conceivable that
at T → 0 the specific heat capacity cp ≠ 0. Glass has a lower density than the iso-chemical crystal
modification. This leads one to conclude that there are cavities in the glass, which give the molecules
mobility and could thus make many different configurations possible. In silica glass, the void accounts
for 21 % of the volume when compared to quartz.

4.5. Simon‘s approximation

We have seen that there is a zero-point entropy of the glass. This leads to the curve of S over T shown
below on the left side (and also schematically before) with SG(0) as the zero point entropy of the glass.
The difference between SG and SX is shown below on the right. The sigmoidal progression of cp < Tg and
the glass transition as a sigmoidal progression can be approximately ignored and the glass transition is
instead assumed to be a cp jump. In other words, the deviations between glass and crystal at close to
0 K and the (very small but existing) deviations between glass and crystal in the range of the Dulong-
Petit limit result in a non-constant difference between glass and crystal (course of the red line in the
right image below, below Tg). However, this difference can also be approximated as a straight line. This
gives a constant Svit (analogue Hvit) for the entropic (and enthalpic) difference between glass and iso-
chemical crystal. This is the so-called Simon's approximation:
 Page 53

The following then applies approximately:

∆Cp = 0 for T < Tg and ∆S(T) = ∆S(Tg) = const. ≠ 0 for T < Tg
The entropy of a glass is thus described as almost constant for T < Tg and greater than zero at T = 0.
Simon's approximation represents a simplification that greatly facilitates the treatment of the
thermodynamics of a glass from room temperature to high temperatures. It also opens up the actual
justification for Hvit and Svit. Simon's approximation is valid for T < Tg, i.e. of course also at and around
0 K. This follows implicitly from what has been described above.
The deviations between glass and iso-chemical crystal at the Dulong-Petit limit mentioned at the
beginning can be explained to some extent with longer-wave temperature oscillations, which are
always subject to the maximum temperature oscillation, and which make a further minimal
contribution to cp due to the glass structure compared to the crystal structure.

4.6. Kauzmann paradoxon

When a liquid is supercooled, the entropic difference between solid and liquid phase decreases. If one
extrapolates the course of the entropy under Tg, one obtains a temperature at which the difference
between solid and liquid phase becomes zero (intersection of the dashed red curve with the blue
curve):
 Page 54

This is the so-called Kauzmann temperature. The paradox is that if you then cool a liquid below its
Kauzmann temperature, you theoretically get a phase that has a lower entropy than the corresponding
solid, crystalline phase. This would of course contradict all thermodynamic principles.
One solution to the Kauzmann paradox is to postulate a phase transition at a temperature before the
liquid reaches the Kauzmann temperature. This temperature is called the calorimetric ideal glass
transition temperature T0c. The interesting thing is that with this model, the glass transition is no longer
just a kinetic effect, but that there is actually a thermodynamic basis behind it:
d𝑇
𝑇, → 𝑇-. as → 0.
d𝑡
In the Kauzmann temperature range, the Gibbs-DiMarzio model predicts that the configurational
entropy of a supercooled liquid vanishes near the Kauzmann temperature and the microstructure of
the liquid then corresponds to that of the iso-chemical crystal and the curves intersect in a second-
order phase transformation. This has not yet been demonstrated experimentally, as the very slow
cooling rates required for detection are not achievable due to the unwanted crystallization.
There are other solutions to the Kauzmann Paradox. For example, the heat capacity of a supercooled
melt could slowly fall to a lower value. Similarly, before the Kauzmann temperature is reached, a first-
order phase transformation could take place to another liquid state, in which case the heat capacity of
the new liquid phase is lower than the extrapolated one. Kauzmann himself solved the paradox by
postulating that all supercooled liquids must always crystallize before reaching the Kauzmann
temperature.
 Page 55

5. Glass chemistry
For the sake of simplicity, in the following section we refer in most cases to a glass with SiO2 as the
glass-forming component. However, all considerations can be applied analogously to other oxidic
glasses. For a silicate-based glass, the [SiO4]4–-tetrahedra are linked by corners:

For a glass with four bridging oxygen-bonds per Si-atom the degree of polymerization amounts to
4:1 = 4. Each ºSi-O-Siº bond is referred to as a „bridging oxygen“ (BO). Each ºSi-O–···M+ bond (with e.g.
M = Na+ or K+) is called „Non-bridging Oxygen” (NBO). Thus, considering just the coordination we
always end up with a [SiO4]4–-tetrahedron. When we additionally consider the bonds, for example with
a former BO that was broken up and has been converted into an NBO we arrive at the notations
[SiO3O]– or [SiO5/2]–.
To abbreviate the notation, so-called QX-groups are defined, 'x' being the number of bridging oxygen
atoms in a single tetrahedron. The structure of the fully polymerized silica glass consists of Q4
tetrahedra (four bridging oxygen, no NBO's). Network converters generate lower Q groups by
exchanging BO´s for NBO´s:
Q4: tetrahedron: 4 BO, 0 NBO per silicon
Q3: tetrahedron: 3 BO, 1 NBO per silicon
Q2: tetrahedron: 2 BO, 2 NBO per silicon
Q1: tetrahedron: 1 BO, 3 NBO per silicon
Q0: tetrahedron: 0 BO, 4 NBO per silicon
This nomenclature is not restricted to glasses with [SiO4]4–-tetrahedra as the basic structural unit.
Bonds and the acid-base concept in glasses:
Most of the other materials that the students deal with have a structure. Their properties are primarily
determined by their structure, and only to a lesser extent by their chemical composition. We therefore
have to deal intensively with the structure of the chemical bond in the material class of glass, which
leads us to the next section.
Since glass is a material without a (micro-)structure, there is (almost) no other way of influencing its
properties than through the chemistry of its bonds, the structure and topology at the atomic level. The
basics of chemical bonds, orbital theory and hybridization are assumed to be known.
 Page 56

In the simplest assumption, four types of bonding can be distinguished: metallic bonding, ionic
bonding, covalent bonding and dipolar bonding. In chemistry as well as in materials technology, mixed
forms of these bond types often occur. Therefore, the illustration shown below should be seen as a
depiction of "extrema", so to speak.

If you start from a pure hard-shell model based on the cation-anion packings, you can easily determine
the resulting coordination polyhedra. The following limits can be deduced from the polyhedral
geometries of trigonal planar (3-fold), tetrahedral (4-fold), octahedral (6-fold) and cubic (8-fold)
coordination:

f231 G N
f34
= N
√3 − 1, vG − 1, √2 − 1, √3 − 1

The cation must always at least fill the corresponding gap, so it must not "rattle" in the coordination
polyhedron. Even if this model is very simple, it helps to get an initial estimate for the coordination of
a cation in an anion compound.
However, in most cases bond formation is not due to Coulomb forces alone, but rather includes the
energetically preferred formation of electron pairs known as covalent bonding. These electron pairs
and covalent bonds are directed and are based on the participating orbitals. The overlapping orbitals
shorten the interatomic distances compared to the simple spherical model, as is shown in the Figure
below.
 Page 57

Which bonding type is prevails depends on the electronegativity difference of the partners and on the
mean electronegativity. The picture below may serve as a guide:

As the next step, we introduce the term ionicity, which is defined according to:
Ionicity = 1 − exp (−0.18 ∆𝜒 G )
Ionicity (i.e. the proportion of ions in a bond in relation to the covalent proportion) can be interpreted
as a measure of anion-cation dissociation. As a convention, bonds with more than 50% ionic content
are not marked with a hyphen, but for example as Ca2 +O2–. The hyphen is reserved for predominantly
covalent bonds such as Si–O or S–O.
So, on the left side of the picture below there is a very strong covalent bond with low ionicity. Such
strong covalent bonds as encountered in S–O form (complex) oxo-anions. The unit [SO4]4– is not able
to polymerize via bridging oxygen, but behaves like an isolated ion: SO3 + 3H2O ➝ 2 H3O+ + SO42–.
 Page 58

At the upper part of the triangle (shown above) are strongly ionic bonds (with correspondingly high
ionicity). The strongly ionic Na···O bond leads to the formation of isolated Na+ ions and, in turn, no
polymerization occurs: Na2O + H2O ➝ 2 Na+ + 2 OH–
Strongly covalently bound oxides form strong (strongly dissociated) acids, strongly ionically bound
oxides form strong bases.

Mixed covalent-ionic bonds such as Si-O lead to monomeric silica, but this is able to polymerize to
larger units (colloids) via –O– bridges: SiO2 + 2 H2O ➝ H4SiO4
In silicate glasses, the ratio of the bridging (right) to the separation point oxygen (left) can be
understood as an acid-base equilibrium:
≡ Si − O$ + O$ − Si ≡ ↔ ≡ Si − O − Si ≡ + OG$
The isolated O2– ion plays a role similar to that of the isolated H+ in aqueous chemistry. Of course, there
are no isolated protons in the water. Nor does one have to speculate about the “true” existence of
isolated O2 in the glass structure. H+ simply denotes a region with an excess proton. In the same way,
O2– denotes an oxygen atom with two predominantly localized electrons. The concept can be extended
to any oxide system.
In aqueous solutions, the H+ concentration is a measure of the acidity (pH value). In the oxidic system,
basicity can be determined by an O2 concentration (i.e. p(O2) value): the higher the p(O2) (i.e. the O2
concentration), the more basic.

5.1. Approach according to Dietzel for cation-anion packings

On the basis of what has been explained above, it is generally not sufficient for glasses to only use the
classic hard-shell model if one wants to describe the coordination and bonding relationships and,
accordingly, the role of an individual oxide component in a glass.
Dietzel therefore chose a different approach: evaluating various minerals with regard to their actual
cation-anion distances a0 as encountered in the real substance (not based on the ionic radii!). The
covalent part of a bond is always implicitly taken into account. The cation-anion interaction in the
 Page 59

vicinity of a common anion is then only determined by the different charge numbers of the cations and
their characteristic distances (for example with oxygen as the common atom). In this respect, Dietzel's
model does not represent a primitive spherical model despite the simple shape for the force F between
two ions:

1 𝑧<VZ ∙ 𝑧V_ ∙ 𝑒 G 𝑧V_ ∙ 𝑒 G 𝑧<VZ

𝐹< = ∙ = ∙ G , 𝑎 ≤ 𝑟<VZ + 𝑟V_
4𝜋𝜀& 𝑎G 4𝜋𝜀& 𝑎

𝑎 = distance between the charge centres [m] Dietzel's "field strength" F

e = elementary charge [As] ! avoid the term "field strength" alone! A field
z = valence cation/anion strength has the unit [N/C] = [V/m].
ε& = Vacuum permittivity [As/Vm]

(with N/A·s = N·m/A·s·m = J/A·s·m = W/A·m = V/m)

With oxygen being the common anion for most materials used in glass production, the approach
reduces to the charge of the cation and the cation-anion distance:
𝑒 G 𝑧<VZ
𝐹< =
2𝜋𝑒& 𝑎G
With this Dietzel model, there is an extension of the ionicity model. The coordination of the cation is
also taken into account (which is not the case when considering the ionicity) and leads to the following:

Bond character of oxides in a mass glass

Me-O bond % ionicity coordination Dietzel’s field
number strength
S-O 13 4 2.60
Si-O 41 4 1.56
B-O 35 3; 4 1.62; 1.45
Al-O 51 4; 6 0.97; 0.84
Ca-O 67 6 0.35
Ba-O 70 8 0.24
Pb-O 47 6 0.34
Zn-O 51 6 0.52
Na-O 70 6 0.19
K-O 73 8 0.13

It can be seen that the Dietzel’s field strength does not change with the ionicity, which is solely based
on the electronegativity. It is also noteworthy that compounds such as K2O and Na2O still have a
covalent portion of > 25% of the M-O bond. The role of various elements in the formation of oxide
glasses has now been classified using the Dietzel’s field strength F. The following applies:
 Page 60

§ Polymerization and glass formation via oxygen ions is most likely to be realized for medium-
high Dietzel’s field strengths with 1.4 < F < 2.1.
§ Too high Dietzel’s field strengths (e.g. S, Cl) indicate the formation of strong oxo anions (acid
residues of strong mineral acids): no glass formation
§ For F < 0.4 the ionic bond dominates; the oxide dissociates in the melt as a metal cation and
O2–.
In this way, relevant elements in the periodic table can be classified in terms of their effect on glass
formation and glass structure. It should be borne in mind that the groups of conditional glass formers
and “imperfect” glass formers cannot be clearly separated. The classification shown below is therefore
only a guide. The important thing is the classification into network modifiers (1st and 2nd type),
structurally active elements and network formers.

Dietzel's field strength can also be used to predict phase separations and compounds within a system.
The table below shows the difference in Dietzel’s field strength ΔF of two oxides (SiO2 as a constant
component) that form a binary oxide system:

F F ΔF compounds
K2O 0.13 SiO2 1.56 1.43 KS4, KS2, KS, K2S
Na2O 0.19 1.37 NS2, NS, N3S2, N2S
PbO 0.34 1.22 PS, P2S, P3S, P4S
CaO 0.35 1.21 liq/liq phase separation;
CS, C2S, C3S2, C2S, C3S
MgO 0.45 1.11 liq/liq phase separation;
MS, M2S
ZnO 0.52 0.97 liq/liq phase separation;
Z2S
TiO2 1.25 0.31 liq/liq phase separation
 Page 61

Large field strength differences mean a great tendency for the formation of compounds on the side of
the oxide with the high field strength. As the difference decreases, the tendency towards formation
shifts to the side of the low-field strength oxide. On the side of the oxide with a high field strength,
liquid/liquid phase separations occurs. With a smaller difference, the miscibility gap increases and the
number of compounds decreases.

5.1.1. Estimations based on homologous compounds

Dietzel's field strength and other dependencies shown in earlier parts of this script can also be helpful
in other areas. For example, if you do not know the behavior (in the example below the formation
entropy, S0) of an element (e.g. barium chromate) in a glass structure, but the influence of so-called
homologous compounds, you can search for the data of these homologous compounds (caution with
very light elements like Li and Be! They are out of the ordinary in almost every homologous series) and
plot these as a function of one of the values given in the table (see lecture) z/a, z/a2, χ, ionicity, △Hf,
△Hbond.
In at least one of these orders, you will generally discover a very good linear dependency that allows a
reliable estimate of the unknown quantity. In the book “Chemical Approach to Glass” by M. B. Volf,
more examples can be found.

5.2. General rules for glass formation

To derive rules for glass formation, we fall back on the well-known Angell plot: A structure that is
already highly cross-linked, as in a SiO2-melt, draws its resistance to arrange in a crystalline structure
precisely from this high degree of cross-linking. In this case we speak of "Frustration due to large
polymer units". To repeat this again, there is then no or only a small cp-jump at Tg, since no degrees of
freedom are lost. These are so-called "strong" glasses. As a sidenote: It is also not possible to crystallize
an SiO2 melt without suitable nucleating agents.
 Page 62

If the glass transition occurs due to the loss of configuration degrees of freedom within individual MRO-
areas, then there is a large jump in cp at Tg, which means that we have a so-called "weak" (or “fragile”)
glass. In this case we are talking about "Ball packing frustration". Although these glasses could
rearrange themselves into a crystalline structure due to (still) existing degrees of freedom, they do not
manage to rearrange themselves into a long-range ordered crystal in the context of the (rapid) cooling
due to the different structural units.

For glasses with geometrical frustration due to large polymer units (actually, these represent the basic
type, which can then be converted from the fully polymerized state with only BO's to the partially
polymerized state with corresponding NBO's by means of appropriate elements) the following applies:
(1) Cations M has a small coordination number: [3] or [4]
(2) Oxygen polyhedra share corners, not edges or faces (maximize cation-cation distance)
(3) Anions X should not be linked to more than 2 cations M, like M - X – M
So, the only stoichiometries for which this is possible are M2X3, MX2, M2X5. Corresponding examples of
such glasses are B2O3, As2S3, BeF2, SiO2, GeO2, P2O5, As2S5,…

5.3. Technologically relevant glasses

In principle, a distinction is made between one-component glasses such as silica SiO2, borate B2O3,
Germanate GeO2 or phosphate glasses P2O5 and multi-component glasses. The latter include e.g.
silicate glasses, alkali-silicate (R2O-SiO2), soda-lime-silicate and potassium-lime-silicate (R2O-RO-SiO2),
borosilicate (R2O-RO-B2O3-SiO2) or alumino-/alumosilicate (Al2O3-SiO2) glasses. For details on the
respective properties and applications, please refer to the lecture slides. Some systems are discussed
in more detail below.
 Page 63

5.3.1. Silicate Glasses

The table below lists typical, average compositions for various technologically relevant silicate glasses:

display Insulation wool Textile fibre

Oxide Mass*) Lamp Lab**) AF Glass Rock® Stone E ECR

SiO2 72 67 80 65 63 41 43 53 60
Al2O3 1 1 3 14 2 21 17 14 14
B2O3 - - 10 1 5 - - 6 -
Fe2O3 - - - - - - 7 - -
FeO - - - . - 7 - - -
MgO 4 2 1 8 4 8 13 1 3
CaO 9 12 1 5 8 22 18 25 23
BaO - - - 3 - - - - -
Na2O 14 18 5 - 18 1 2 - -

*) float, container, tableware

**) with α = 3.3 or 4.0 ppm/K

We have already discussed the structure of silicate glasses. The fully polymerized network of SiO2 is
modified by various network modifiers such as Na2O, CaO, MgO or K2O. Intermediate oxides such as
Al2O3 cause an increase in viscosity and glass length, an improvement in mechanical strength, an
increase in scratch hardness and an improvement in chemical resistance. Further oxides can be added
e.g. for nucleation, to change the chemical resistance or for coloring. In the following, however, some
peculiarities of the above-mentioned silicate systems and the function of the intermediate oxides will
be discussed in more detail.
If an alkali-silicate glass is present, the existing NBO’s will weaken the structure, which in turn leads to
a high coefficient of thermal expansion, a low chemical resistance as well as low Tg and Tliq, with the
reduction of the latter one allowing for the fusibility of the first glasses.
In some cases, however, these alkali oxides also cause undesirable properties, such as very low
chemical resistance.
If, however, Al2O3 is introduced into alkali (and/or alkaline-earth) containing silicate glass at the
expense of SiO2, an increase in chemical resistance and a decrease in the coefficient of thermal
expansion is observed. This is due to the Al3+-ions being located in charge deficit tetrahedra.
 Page 64

Charge deficit tetrahedron

The Al3+ prefers the tetrahedral coordination in the glass, the same applies to B3+. It is built into the
tetrahedron like a Si4+, but has too little positive charge to ensure the charge balance of the four
surrounding oxygen atoms (hence the name "charge deficit tetrahedron"). The charge equalization is
then ensured by an alkali- or alkaline-earth-ion. However, these cations are no longer available for
charge compensation at an non-bridging oxygen, thus either preventing the formation of NBO’s or
forcing previously existing NBO’s to be closed and build a BO. As a result, the degree of polymerization
of the structure increases, as does the chemical resistance and the coefficient of thermal expansion.
This is of course only valid as long as alkali or alkaline earth ions are present in order to ensure the
charge equalization. If there are no more alkali or alkaline earth ions available, Al3+ can also act as a
network modifier and is incorporated in a [6]-coordinated manner (octahedral surrounding).
The calculation below illustrates these dependencies with the following composition:

oxide wt. % mol / kg

SiO2 80.3 13.38
Al2O3 2.8 0.28
B2O3 12.2 1.75
CaO 0.3 0.05
Na2O 4.0 0.64
K2O 0.3 0.03

Together, B2O3 and Al2O3 can form up to 2·(0.28 + 1.75) = 4.06 mol charge deficit tetrahedra. CaO, Na2O
and K2O are only able to compensate 2·(0.05 + 0.64 + 0.03) = 1.44 mol of these tetrahedra. Therefor,
some of the Al3+ (to a minor fraction also B3+), have to be used for charge compensation and can no
longer form a charge deficit tetrahedron.
A technologically extremely relevant example are borosilicate glasses such as DURAN®, PYREX®, so-
called "3.3 glasses" (α = 3.3·10–6 K–1), which are commonly encountered in chemical laboratories. These
glasses have a low coefficient of thermal expansion α, but are still significantly easier to melt
(Tg = 525 °C, η = 104 dPas at 1260 °C) compared to pure quartz glass (Tg= 1200 °C and η > 104 dPas
at 2500 °C).

5.3.2. Borate and Borosilicate Glasses

Borate glasses consist of [BO3]-triangles and are therefore coordinated in a planar manner. The [BO3]-
triangle is a very stable structure with bond energies for B–O of around Ebond = 540 ± 5 kJ/mol which
 Page 65

are even higher than that of the Si-O bond (444 ± 8 kJ/mol). Due to the more covalent character of the
bond, the B-O-B bond is not as symmetrical as the Si-O-Si bond.
Therefore, in pure B2O3 and in systems rich in B2O3, the [BO3]-triangles are arranged in a hexagonal
superstructure forming a so called [B3O9/2] „Boroxol-ring”. These Boroxol-rings form the monomeric
subunit in the network. However, there are only 3 bridging oxygen atoms per 3 B atoms = 1 bridge per
1 B atom. The degree of polymerization is thus 1:1 = 1.

This leads to a weak structure of pure B2O3 glass despite the high bond energy. These glasses therefore
have a high thermal expansion α, low Tliq and low Tg.
With the presence of R+ and R2+ (R = alkali or alkaline earth ions), however, completely cross-linked
[BO4]-tetrahedra are formed. Basic oxides therefore initially lead to a higher degree of crosslinking of
the complete network.
Only if there is an excess of R separating oxygens—like B–O–···R+ or B–O–···R2+···–O-B—are formed
again. Thus, if the amount of RO/R2O is continuously increased, the borate glass structure first
becomes stronger and from a certain point on weak again (see next picture). Accordingly, α first
decreases, but later on increases again. This behavior is also known as the „boron oxide anomaly“.

So how can we interpret this behavior? In the range of high alkaline amounts there are enough NBO
oxygen atoms so that three oxygen atoms are sufficient for shielding a single boron atom.
 Page 66

With decreasing R2O-percentage, the number of available O atoms also decreases and the B3+ atoms
can no longer be adequately shielded. This causes the change towards [BO4]-tetrahedra-groups
balanced by a single R1+ per tetrahedron. With further decrease of R, the polyhedra can only continue
to share corners if [BO3]-structures (aka the already known Boroxol-rings) form again.

On introducing alkali or alkaline earth ions into a borosilicate glass, tetrahedral structures with charge
deficits start to form (the same applies to Al3+ in soda lime silicate glass). In the case of triple-
coordinated boron, the monovalent or divalent cation ensures that boron can adopt a tetrahedral
coordination through additional charge compensation and thus a Q4- instead of a lower-Q-structure is
present, which in turn is stronger and more stable.

Charge deficit tetrahedron

If you have been paying attention, you know from the previous explanations that B3+ [4] has an ionic
f231
radius of 27 pm, leading to a ration f34
of 0.19. It thus should NOT go into the tetrahedron (Limit =
0.225, it "rattles"). Here we reach the limit of our model considerations. We only have the hard-shell
model in mind (Coulomb). With the charge-balancing cation and due to the electron structure,
however, the B3+ does go into [4]-coordination. A sp3-hybrid is formed with an unoccupied sp3-hybrid-
orbital. So, we once more become aware, that none of these considerations is always 100 % correct.
Once the maximum proportion of alkalis or alkaline earths has been reached and all further boron
atoms are accommodated in the charge deficit tetrahedron, any further increase in alkalis or alkaline
earths leads to the formation of NBOs and thus again to a weakening of the structure. In the overall
interpretation, however, it is made even more difficult by the fact that NBOs may only be linked to B3+
in [3]-coordination and that certain rules are applicable for the neighborhood of B3+ in [4]-coordination
and B3+ in [3]-coordination.
Many studies have been made on the effect of boron in silicate glasses. For example, Ba can be
compensated by increasing B, Al and Mg to increase durability and strength. On the other hand, this
leads to undesirable boron evaporation in the melting tank and a stronger corrosive attack. Above a
certain B2O3 content in the glass, the evaporation of boron components (e.g. NaBO2 and KBO2)
increases strongly. For details and further properties and interactions of boron in (silicate) glasses,
please refer to the relevant technical literature.

5.3.3. Phosphate glasses

Phosphorus—as an element of the fifth main group—generally features a sp3-hybridization with one
of the sp3-orbitals being doubly occupied. The [PO4] unit thus has a tetrahedral structure like [SiO4],
 Page 67

which is stabilized by the sp3-hybridization. As an element of the 5th main group, P has an "excess" of
one electron. As a result, one sp3-suborbital is doubly occupied and is not available for the formation
of further P-O single bonds. Instead, a double bond P=O is formed, which does not further participate
in the polymerization of the network. The consequence is again a lower degree of cross-linking (only 3
instead of 4 corners). The tetrahedra either form isolated P4O10 molecules or they form chains that
(only) occasionally branch and cross-link. Again, a weakly cross-linked structure is formed, which can
be seen from the relatively low Tliq and Tg, as well as the high α-values.

Pure phosphate glasses are strongly hygroscopic and therefore only of academic interest. As an
additive oxide, P2O5 is responsible for a number of (abnormal) properties and is used in silicate, borate,
borosilicate glasses, etc... (laser-glasses, waveguides, electronic conductivity, semiconducting glasses,
etc...).
The table shown below once again compares the properties of SiO2, phosphate and borate glasses:

Tliq Tg α
[°C] [°C] [10 K-1]
-6

SiO2 1723 1220 0.6

B2O3 450 257 15.1

P2O5 570 380 13.1

 Page 68

5.4. General aspects of glass formation / summary

The following pictures give a brief summary of the information on glass formation as presented in
the previous chapters:

OH and F– also depolymerize a silicate network. There are also BeF glasses. The large S2– ion can occupy
oxygen positions to a certain degree.
The oxo-anion SO42–, on the other hand, is not able to participate in the polymerization of the network
due to its very strong sulfur-oxygen bond, despite its tetrahedral structure. Polymerization via oxygen-
atoms by sigma bonds generally requires that the covalent bond Me-O is not too strong. None of the
strong mineral acid anions is therefore a good glass former.
Below, the network-forming groups, the network-modifying groups and the charge-deficit tetrahedra
are briefly shown again:
 Page 69

5.5. Network structures in real systems

The near-order structures in the crystalline mineral and in the glass are the same. In albite and albite
glass (chemistry: Na2O-Al2O3-6SiO2 = 2NaAlSi3O8), Al is present in a tetrahedral environment (alumino-
silicate, charge deficit tetrahedron). In contrast to the Na2O-SiO2 system, Na2O does not cause any
depolymerization in the albite system. The structure that forms thus results in:

According to this procedure, one can determine the respective coordination polyhedra for a glass with
different oxidic components. For example, the phase diagram of a binary sodium silicate glass is shown
below:
 Page 70

system Na2O-SiO2 (Na2O = N, SiO2= S)

1800

1600
T(1.0) T(2.0)
1400

1200
T(3.0)
T in °C

1000
T(4.0)
800
T(13.0)
600 NS NS2 N3S8 S

400
0.6 0.8 1.0
wt. fraction of SiO2

We now assume a pure quartz glass and move successively into the range of increasing sodium
content.

At the point of 100 % SiO2:

Degree of
Composition Structural unit(s)
polymerization

SiO2 = [SiO4] 4

In the area of the following first eutectic, we encounter coordination polyhedron:

Degree of
Composition Structural unit(s)
polymerization

3Na2O·8SiO2 =
Na6Si8O19 =
3.25
Na3/4SiO19/8 =
[SiO3.25O0.75] –Na+0.75
 Page 71

In the region of the first dystectic, the following coordination polyhedron exist:

Degree of
Composition Structural unit(s)
polymerization

Na2O·2SiO2 =
Na2Si2O5 =
3
NaSiO5/2 =
[SiO3O] –Na+

In the area of the second dystectic the coordination polyhedron is given by:

Degree of
Composition Structural unit(s)
polymerization

Na2O·SiO2 =
Na2SiO3 = 2
[SiO2O2]2 –2Na+

One can clearly see, that the degree of polymerization decreases with increasing sodium content.
 Page 72

6. Mechanical and thermomechanical properties of glasses

Thermodynamic, thermomechanical or mechanical properties in glasses (and ceramics) are often (but
not always!) determined by the basic structural units, see previous chapter on crystalline reference
phases / constitutional phases.

6.1. Strength of glasses (theoretical and practical)

From a mechanical point of view, glasses are brittle materials. All theories on the fracture mechanics
of brittle materials are based on a fracture-inducing defect, which is the center of the fracture
propagation. The break-causing defect can be localized in the volume or on the surface. In the case of
glass as a homogeneous material without a structure, primarily the surface defects play a role. Volume
defects are only to be considered in connection with inclusions in the bulk of the glass, i.e. with quality
defects such as bubbles or inclusions.
A glass surface can have a number of different surface defects. Fluctuations in flatness extend over
2 – 3 cm (surface waves in the float process), waviness can be caused by cooling corrugations in
moldings (0.1 – 1 mm). Defects in flatness and waviness do not play a role in terms of strength, but
roughness does. However, even an average well-produced surface is not rougher than 10 nm (“fire-
polished surface”). With everyday use, however, some (individual) defects in the range of 10 – 20 μm
will soon appear. They are much more critical for the practical strength of a glass object than the
roughness.
In the standard theories of fracture mechanics of brittle materials, as is known, three different loading
cases of a surface defect that has once existed are dealt with. Mode I, in which tensile stresses occur
perpendicular to the defect, is the most important; it is dealt with exclusively in the following. The
critical tensile stress under load type I is denoted by σIc or sIc. It is synonymous with the strength of a
glass.

First, we want to consider the theoretically maximum possible strength in a glass: The creation and the
continuation of a crack create two new surfaces. The total work of fracture Wfrac is proportional to the
surface tension σ of the glass:
 Page 73

And the maximum stress sIc (derived from the well-known Griffith equation) that the glass can
withstand without breaking defines its strength and depends on the defect size (e.g. crack length) b:

2∙𝐸∙𝜎
𝑠gA ≅ 𝑓:hJ ∙ Š
𝑏

The geometric factor is determined by the type of fracture-inducing defect. For a short, linear volume
%
defect (e.g. pore), 𝑓:hJ = (this case applies more to a ceramic than to a glass). A flat surface defect
√j
√j
has 𝑓:hJ = . In a "perfect" glass, no surface defects exist and structural features, such as grain
k
boundaries or crystallite size distributions, do not exist in a glass. The maximum stress therefore sets
in immediately before the crack tip at the minimum feasible distance between the bonds, with
0.1 nm < a0 < 0.3 nm. Thus, the defect size would correspond to the bond distance. If fmax then exceeds
the binding force of a ºSi-O-Siº bond, the crack will migrate further. In other words, a "perfect" silicate
glass would assume a strength corresponding to that of the ºSi-O-Siº bond. Let us now take a look at
this:
We use the heat of formation of SiO2 from the elements as the bond strength. This is 908 kJ/mol or
1.5×10–18 J per Si atom. The molar volume, i.e. the quotient of molar mass (60.084 g/mol) and density
(2.2 g/cm3) of SiO2 glass, is 27.3 cm3/mol or 4.5×10–29 m3 per Si atom. The binding energy and the
volume then result in a strength of approx. 33 300 MPa (!) (less strongly cross-linked glasses approx.
20 000 MPa, alloyed high-strength steels 1 300 MPa). Glass thus shows an extremely high strength,
provided (and this is the crucial point!) no extrinsic defects are present.
The table below shows how much the actual strength of glass (the service strength) decreases with
increasing damage, starting from a soda-lime glass with a theoretical strength of 20 GPa):
 Page 74

property s in MPa limiting factor

theoretical strength 20,000 bonds, glass structure
strength of glass fibres 5,000-10,000 micro flaws, glass
surface
pristine glass, 500-1,000 micro flaws, glass
acid polished glass surface
safety glass, 100-200 cracks, glass surface
thermally strengthened
standard glass products 20-50 cracks, glass surface
damaged glass <10 cracks, glass surface

6.1.1. Chemical preloading

At this point, it makes sense to anticipate measures to increase strength (see also later chapter on
thermal glass tempering). If compressive stresses can be introduced into the surface of a (glass)
product, this can increase the fracture strength. In soda lime silicate glass, this is possible (in addition
to thermal toughening, see later) by means of so-called chemical toughening. In a chemical salt bath
small Na+ ions are exchanged for larger K+ at a sufficiently high temperature, thus inducing compressive
stresses of up to 300 MPa. Gorilla-Glass™ developed by Corning is an example of such a chemically
toughened glass used for smartphones:

Due to steric hindrance the larger K+-ion cannot occupy octahedral sites [NaO6]. The large chemical
potential difference ∆µ, however, forces the ion exchange, thus introducing mechanical compressive
stresses.

The exchange depths are in the range of 10–30 µm. The advantages of the process are the high
compressive stresses achieved and the fact that the formation of high tensile stresses inside is avoided
 Page 75

(see picture above). However, chemical tempering in the salt bath below Tg causes problems in the
duration of the process and potentially in the surface appearance.
Alternatively, Na+ is exchanged for Li+ and a subsequent surface crystallization takes place.
Compressive stresses of up to 500 MPa can be generated here (CHEMCOR ™ process).

6.2. Structural dependencies and mechanical properties

In the chapter on glass chemistry, we have already talked about the strength of a glass and its
dependence on several factors:

• The strength of the atomic bond

• The degree of polymerization (number of bonds per formula unit)
• The atomic packing density (number of formula units per unit of space)

• The type of linkage of the assemblies

The chemical composition obviously also determines the mechanical properties. One would now
assume that less strongly cross-linked glasses with many NBO's would have a lower strength than
highly cross-linked glasses.
But, comparing an anorthite glass with a diopside glass, one sees that the answer is not as simple as
one would hope: the anorthite glass has a degree of polymerization of 100 %, the diopside glass of 50
%. One would therefore spontaneously expect the anorthite glass to have a higher strength. But if you
compare the Young’s moduli, the diopside glass is clearly ahead:

If you calculate the atomic packing density, you get 68,3 % for the diopside glass, but only 58 % for the
anorthite glass. So here the packing density is the decisive parameter and not the degree of
polymerization. A similar example can be given for the structurally iso-typical compounds diamond
and cubic SiC: The bond enthalpies are ∆HDiss = 447 KJ/mol for SiC and ∆HDiss = 348 KJ/mol for diamond.
However, the hardness of diamond with HKnoop > 70 GPa is significantly higher than that of cubic SiC
with HKnoop = 30 GPa. If we now look at the (packing) density, we see that diamond with ρ = 3.52 g/cm3
/ a = 3.567 Å is significantly more densely packed than SiC with ρ = 3.21 g/cm3 / a = 4.3596 Å.
For silicate glasses, it is often true that the packing density is very decisive for the mechanical
properties. A higher degree of polymerization here means easier "foldability" of the structure and thus
lower elastic properties. The balancing cation (if present) in the charge deficit tetrahedron (e.g. in the
feldspar glasses such as anorthite, Q4) is incorporated into the structure differently than a network
transformer cation. The latter then makes it more difficult to "fold together" in lower cross-linked
silicate glasses (e.g. diopside, Q2), resulting in better mechanical properties.
 Page 76

Another example for the reasoning given above is shown below. The mixed series of jadeite and
diopside glasses is evaluated in terms of packing density and degree of polymerization versus Young's
modulus (Jindal, et. al., Ceramics Int. 37, 2011):

Series of diopside jadeite glasses, DI-JD, normally MgCaSi2O6 – NaAlSi2O6

System DI-JD, x(JD) = 0.0 0.2 0.4 0.6 0.8 1.0

ρ in g/cm³; ± 0.003 2.847 2.765 2.655 2.607 2.510 2.442
VM in cm³/g-atom 7.388 7.505 7.708 7.740 7.926 8.029
Vpack in cm³/g-atom 4.555 4.528 4.502 4.475 4.448 4.387
Cpack 0.617 0.603 0.584 0.578 0.561 0.554
degree of polymerization 0.5 0.6 0.7 0.8 0.9 1.0
E in GPa; ± 0.2 102.0 94.4 86.1 84.1 75.1 72.5
µ; < ±0.03 0.28 0.26 0.25 0.24 0.22

4𝜋 N N
𝑉FV<X ∙ 𝑁6 ∙ ∑(𝜒V_ ∙ 𝑟V_ + 𝑥<VZ ∙ 𝑟<VZ )
𝐶FV<X = = 3
𝑉 l 1
∙ ∑(𝑥V_ ∙ 𝑀V_ + 𝑥<VZ ∙ 𝑀<VZ )
𝑝
It can also be seen here that the jadeite glasses with a lower packing density and higher degree of
polymerization have a lower strength:

In other cases, what is gained in binding energy is lost in packing density (possibly to a greater extent,
see the example of SiC and diamond).
If we now compare the jadeite/diopside glass and the iso-chemical crystals again, we notice that other
conditions apply to the iso-chemical crystals. In the crystal, the higher cross-linking dominates, which
is why the jadeite is stronger than the diopside. In the glass, however, the packing density dominates.
As seen above, the packing density decreases significantly more from the crystal to the glass in jadeite
glass (25 %) than in diopside glass (approx. 14 %):
 Page 77

phase composition Young’s modulus E in GPa

low-T crystal glass
cristobalite SiO2 65 73
enstatite MgSiO3 184 107
wollastonite CaSiO3 100 93
diopside • CaMgSi2O6 163 ← 37 % loss → 102
anorthite CaAlSi2O8 103 98
albite NaAlSi3O8 74 70
Jadeite • NaAlSi2O6 213 ← 66 % loss → 73

However, one should be careful with generalizations here as well. The mechanical properties of a glass,
as well as of a crystalline material, when the structure recedes into the background (e.g. in a single
crystal), are always determined by a combination of the strength of the atomic bond, the degree of
polymerization, the atomic packing density and the linkage of the structural units. If one of these
quantities falls completely out of line, another—normally dominant—quantity can no longer
compensate.
For example, when O is substituted for N in the so-called oxy-nitride glasses (or for C in carbon glasses),
there are yet other dependencies. By replacing the two-bond oxygen with the three-bond nitrogen (or
four-bond carbon), the degree of polymerization is greatly increased (in the case of C insertion, the
packing density is also additionally increased due to the significantly smaller carbon). This also causes
a massive increase in strength:

As a kind of "cream cake or black forest cake" (depending on what one prefers) of these considerations,
one can now try the jadeite-diopside mixing series again and plot the packing density e.g. against the
 Page 78

bulk modulus K. If one extrapolates the curve to K = 0, one ends up with a value of the packing density
of approx. 0.45:

But what does it mean if there is a compression modulus of 0? And what would be the correlation with
the packing density? The value 0.45 is in excellent agreement with the percolation limit of a structure
with corner-linked tetrahedra (pC = 0.43, see chapter Structural Models and Percolation Theory). Below
a packing density of approx. 0.45 to 0.43, the network collapses, there are no more strength-relevant
cluster cross-links and consequently the compressibility of the material disappears. A very nice
confirmation of the percolation theory in the context of glass.

6.3. Thermal conductivity in glasses

Heat conduction is primarily responsible for heat transport in glass and ceramic materials. At higher
temperatures, heat radiation must additionally be taken into account. In glass, heat is transferred by
phonon oscillations of the lattice building blocks. A simple relationship applies here for the thermal
conductivity λ:

% c = specific heat capacity

λ]J8DH = N 𝑐𝑣F 𝑙F with
vp = speed of the phonones

lp = mean free path length of the phonons

The thermal conduction in glass is thus determined by lp (is in the order of magnitude of the atomic or
molecular spacing (SRO/MRO)). Due to the lack of long-range order, λamorph < λkrist applies to iso-
chemistry. The temperature dependence of the thermal conductivity is given as a function of the
change of the heat capacity c with temperature: At temperature T = 0 K, c = 0 and increases
proportionally with T3 up to the Debye temperature θD = hν/kb.
 Page 79

We then get the following picture for the dependence of λ on T:

The increase in thermal conductivity at low temperatures
is proportional to T3 for glasses (T << θD, green arrow).
When c = 3R is reached in the Dulong-Petit limit, the
thermal conductivity tends towards a constant value of 25
J·mol–1·K–1 and thus has no influence on λ (blue lines).
For most oxides, in the range at around 1000 °C, the
specific heat becomes almost independent of the
temperature.
In the range in between, the density of the phonons
increases and lp decreases. λ then becomes proportional
to 1/T (red arrow).

The situation is different when there are pores in the glass. Pores are virtually a second phase within
the material. Porous glasses can therefore be regarded as "two-phase materials". This second phase
can be used to optimize the material properties with respect to thermal conductivity. The thermal
conductivity decreases linearly with the pore size. When dpore < lgas (at RT about 50 nm), the thermal
conductivity decreases abruptly because the shock transfer is hindered (characteristic here is the so-
called Knudsen number Kn = f(lgas/dpore)).
If the pores are large enough so that the mean free path of the photons is smaller than the pore size,
the heat transfer in porous materials increases with increasing temperature because the heat radiation
in the gases of the pores increases.
The following then applies to the thermal conductivity in a pore:

λ] = 4 γ 𝑑 σ ε 𝑇 N d = pore diameter

γ = form factor

σ = Stefan - Boltzmann constant

ε = total emissivity
 Page 80

6.4. Coefficient of thermal expansion

The thermal expansion is attributed to the non-harmonic oscillations (change of the interatomic
potential) of the atoms with increasing temperature and can be described with a special case of the
Mie potential, the Lennard–Jones potential LJP. The LJP describes the entire intermolecular potential
as a hard-sphere potential:

In a solid, every single atom oscillates around an equilibrium point. If these were harmonic oscillations,
the distance between the atoms would have to remain equal to the equilibrium distance on average,
because the atoms would then oscillate to the same extent in the direction of a neighboring atom as
in the opposite direction. But the potential energy increases more when two atoms approach each
other than when they move away from each other. With increasing temperature, the oscillation of the
atoms thus increases. But since there is a shift in the energetic minimum with increasing distance to
the neighbor (namely 2r for identical atoms), the oscillation is non-harmonic. Thus, there is an increase
in volume, which we call thermal expansion of the material.
For moderate temperatures, the slope 𝜕r/𝜕T (or 𝜕V/𝜕T) is approximately constant. Thus a ∼ const. is
also valid. For T ➝ 0 also trends towards zero (a ➝ 0). ATTENTION: With this concept, a < 0 cannot be
explained (this will be dealt with later). The coefficient of volume expansion is defined as the change
in volume with changing temperature at constant pressure:

1 𝜕𝑉
𝑎" = ∙0 1
𝑉& 𝜕𝑇 F

The isotropic change in volume of a cuboid is given by:

∆𝑉 = 𝐿G 𝐿N ∆𝐿% + 𝐿% 𝐿N ∆𝐿G + 𝐿% 𝐿G ∆𝐿N

% ∆m % ∆ n5 % ∆ n6 % ∆ n7
𝑎" = m&
∙ ∆0
= n5 ∆ 0
+ n6 ∆ 0
+n
7∆ 0

𝑎" = 3 ∙ 𝑎U
 Page 81

The coefficient of thermal expansion is generally measured with a dilatometer. At this point it seems
appropriate to refer back to the previous considerations on Tg and Tf: A dilatometer measurement
should always be carried out according to the standard ISO 7991. To justify this approach, one must
remember the nature of the glass transition as a path-dependent process: at high cooling rates, the
glass transition occurs at slightly higher temperatures than at low cooling rates. In the first case, a glass
structure with a slightly lower density is frozen in (which has absolutely nothing to do with the
phenomenon of thermal stresses!). If a rapidly cooled sample is slowly heated, then the glass relaxes
into a denser state near Tg. The relaxation of the assumedly high temperature structure in a (non-
relaxed) sample creates a dip in the dilatometer curve as the glass structure relaxes to the then
prevailing structural temperature in the vicinity of the glass transition. Again, this is by no means a
residual stress relaxation, but rather a structural relaxation between two different glass states!
A dilatometer measurement is therefore a good way to test the degree of structural relaxation of a
sample. But be careful: If a rapidly cooled sample is slowly heated, then the glass near Tg relaxes to a
denser state and the reading of Tg or α becomes doubtful.
The coefficient of thermal expansion is generally given for a temperature range and suggests linearity.
This is only true if no phase transformations take place in the respective temperature range and ∆l/l is
linear in this range (the latter is not correct in some cases, e.g. for quartz glass).
The linear expansion coefficients of inorganic glasses cover a wide range from almost 0 to
35·10–6 K–1. In the SiO2-TiO2 system, glasses with negative α can be found (so-called ULE glasses, ultra-
low expansion). For amorphous materials it is true that with lower polymerization (more non-bridging
oxygen atoms, NBO's) the bond asymmetry and thus α are increased. In networks with a high free
volume (e.g. quartz glass), thermal energy can be absorbed by rotation and bending, which counteracts
the expansion. Therefore, pure quartz glass has a very low α of approximately 0.5 ·10–6 K–1.
As mentioned earlier, there is an influence of RmOn-addition (depolymerization!) on α. In the figure
below this is shown for a Na-Si glass (already with 18 wt% Na2O), in which SiO2 is exchanged for RmOn:

The effect of boron by forming charge deficit tetrahedra can be clearly seen in this example. A
borosilicate glass for laboratory glass applications therefore also has a quite low α. Borofloat 33, for
example, has an α of 3.3 ·10–6 K–1 (nomen est omen).
 Page 82

There is a fundamental relationship ("iron rule") between Tm or Tg, and α: the higher Tm or Tg is, the
smaller α will be.

6.4.1. Negative thermal expansion coefficient

A negative thermal expansion coefficient cannot be explained by a simple interatomic potential, but is
rather based on the secondary linkage of the structural elements.
A tilting of coordination polyhedra relative to each other with a simultaneous increase in the distance
Me-O (which symbolizes the primary thermal expansion). causes a contraction of the overall structure
brought about by this primary expansion.

An example might be an accordion that is pulled apart in one direction but contracts in cross-section.
Technologically, this effect is very relevant, for example, in glass ceramics, a composite material made
of glass and embedded crystalline areas. Herein, the glass has a positive α whereas the crystal has a
negative α. The thermal expansion is additive, which means that with clever choice of the respective
α and clever adjustment of the quantity ratios, a material with zero expansion can be produced. The
proportion of the crystalline phase in relation to the residual glass phase generally determines the
thermal expansion of the material more significantly, since the α of the crystalline phase can only be
varied within limits. Details are dealt with in the corresponding (upcoming) chapter.
Examples of such zero-expansion materials are, for example, the well-known ceramic glass cooktops
or viewing panels for fireplaces. But these materials are also indispensable in the field of
photolithography and astronomy (see pictures below).
 Page 83

6.4.2. Thermal stresses

Thermal stresses can occur during cooling or heating processes when the inside of a material heats up
or cools down more slowly than the outside. If these stresses lead to failure of the material, this is
called thermal shock.
A distinction is made between 3 cases according to the severity of the thermal shock:
Case 1: For infinitely large heat transfer (the thermal diffusivity α is irrelevant), the body on the surface
increases the temperature instantaneously:
𝐸∙𝛼
𝑠(𝑥) = ∙ (𝑇% − 𝑇& )
1−𝜇
Case 2: If there is a constant heat transfer, there is generally no immediate failure, but there is
weakening and internal cracking, e.g. during alternating cycles (classic example: day/night,
winter/summer):
𝐸∙α
𝑠(𝑥) = ∙ (𝑇% − 𝑇& )
(1 − 𝜇) ∙ λ
Case 3: The material is loaded with a constant heating or cooling rate at the surface and the heat input
occurs at a constant temperature rate. If the thermal diffusivity (as f(ρ, λ, cp)) is now large enough,
"nothing" happens even at a high α:
𝐸 ∙ a ∙ 𝑐F ∙ 𝑝
𝑠(𝑥) = ∙ (𝑇% − 𝑇& )
(1 − µ) ∙ λ
According to the above, the thermal shock properties can be determined for a glass in the same way
as for a ceramic. The dependencies result directly from the above formulae.
For glass, however, the thermal stresses play a further, a double role. First of all, thermal stresses that
can lead to fracture of the glass body must be avoided. Among other things, these stresses build up
when a glass product comes out of a mold (e.g. a hollow glass) and cools down. The cooling is very
inhomogeneous between the outside and the inside as well as over the height of the glass product due
to the different masses of glass at the different spots. This leads to thermal stresses unless the glass
has a coefficient of expansion equal to zero. These strains must then be relaxed in a second step; in a
so-called "annealing oven". Here the glass is heated to a temperature near Tg and kept at this
temperature or slowly cooled so that the stresses vanish and no new stresses can build up due to the
slow cooling rate. The design of such a cooling program to relieve the thermal permanent stresses to
values below a certain threshold is part of an exercise (please do not confuse this with temporary
 Page 84

stresses, such as those that occur when a glass plate is bent or when a glass article is heated without
allowing relaxation to occur (ex: hot tea into a glass)).
The cooling rate dependence of the developed stresses s from upper cooling point (η = 1013 dPa·s) to
lower cooling point (η = 1014.5 dPa·s) can be calculated according to:

V∙- o∙<8 V∙-

𝜎=𝑓∙ ∙ ∙ ℎ ∙ 𝑑G from 𝜎 = 𝑓 ∙ (%$") ∙ (𝑇b − 𝑇V )
(%$") p

The maximum permitted temperature difference then follows with:

1 𝜎 ∙ (1 − 𝑣)
Δ𝑇eV= = 𝑇eV= − 𝑇eY_ = ∙
𝑓 𝑎∙𝐸

h = cooling rate [K·s–1] d = dimension key figure

γ = thermal conductivity [W·m–1K–1] = Thickness with one-sided cooled plate
𝜈 = Poisson number = 1/2 Thickness with double-sided cooled
plate
ρ = density [kg·m–3]
= radius for spheres
E = Young’s modulus [MPa]
α = coefficient of thermal expansion [K–1] = √𝑑 ∙ 𝐿 for bottles with d = wall thickness and
L = bottom thickness
cp = heat capacity [J·kg–1K–1]
f = form factor
= 0.336 for flat substrates
= 0.126 for massive cylinder
= 0.066 for spheres
= 0.3 for hollow glass

6.4.3. Thermal Glass Tempering

The second role of thermal stresses is thermal toughening. The principle of forced cooling is used here
technically to increase the strength of a glass product by introducing permanent compressive stresses
into the surface. For this purpose, finished shaped objects (such as a pane that has already been bent
into the desired shape) are slowly and evenly heated to a temperature below the macroscopic shape
stability limit TL = T(7.6) ≈ 720 – 750 °C. Then the rapid cooling to T < Tg takes place by blowing air over
the surfaces. The process takes less than 30 s, in some cases even less than 5 s in the case of newly
developed processes. The stress profile is then directly correlated with the temperature profile
according to the above-mentioned relationships (see also figures below). The fracture pattern of such
glass products shows many small glass fragments due to the high internal stresses:
 Page 85

In detail, the process can be understood as follows: The homogeneously heated glass is quickly cooled
on the surface. This means that the surface contracts (assuming a positive α). The hot interior can at
least partially follow this movement. There are low tensile stresses on the surface and compressive
stresses inside. Then after a certain time, depending on the conduction of heat, the interior solidifies
and contracts. However, the surface is already cold and solid and opposes the contraction with tensile
stress and is itself placed under (desired) compressive stress. The stress profile at high temperature is
thus reversed when it cools down completely. Stresses of up to ‒150 MPa arise in the surface
(compressive stresses are given a negative sign).
The effect of reinforcement through thermal toughening is demonstrated in comparison to a well-
cooled (annealed) float glass. The samples are subjected to a type of three-point bending test (a). In
order to exclude the influence of edge influences, the test is carried out two-dimensionally by means
of a ring-shaped support and a central pressure stamp (b) (see below):
 Page 86

Pictures (c) and (d) show the stress distribution over the thickness of the glass pane, once with the
application of a load that creates a compressive stress ‒sappl on the upper side and an equal tensile
stress + sappl on the lower side. The breaking stress is sIc. While sappl becomes greater than sIc in the non-
toughened glass—which leads to the destruction of the pane—in the toughened glass only tensile
stresses of the size sappl - sESG < sIc occur; the pane doesn’t break in this case. Here, sESG (ESG = toughened
safety glass) represents the permanently applied stress in the surface of the toughened glass.
Due to the strong increase in compressive stress in the surface of the glass the bending force in the
test is not critical. For construction purposes, civil engineers can expect a strength of 20 MPa for non-
toughened glass; with ESG it is enhanced to about 100 MPa. ESG has one disadvantage: it can no longer
be processed (cutting, drilling, etc.) after it has been toughened.
The following picture clearly summarizes the interrelationships of thermal tempering:

If a crack (green bar) is made in the glass, it is prevented from advancing by the compressive stresses
(depending on its crack tip energy). However, if the crack (red bar) is so deep that it extends into the
tensile stress zone, this leads to breakage and immediate failure of the component.
The Prince Rupert Drops are a clear example of a highly tempered glass (tensile and compressive
prestressing).
 Page 87

7. Optical properties of glasses

At the beginning of this chapter, a few general remarks should be done. The terms "spectral and optical
properties" refer to the interaction between matter and an electromagnetic field. The spectrum of
electromagnetic waves covers several orders of magnitude. The frequencies range from technical
alternating currents and those common in telecommunications to those responsible for thermal
radiation and finally to the highly energetic γ-rays. The visible range (light perceived by the humans
eye) represents a very thin band whose wavelengths range from approx. 380 to 780 nm. The different
frequency ranges can be assigned to characteristic atomic, electronic or nuclear processes.
Frequencies up to 300 GHz are technically generated by oscillations of an electron gas in conductors,
semiconductors and plasmas. The IR range is related to the vibrations and rotations of molecules. The
visible (VIS) and ultraviolet (UV) range is caused by electron transitions in the outer electron shell of
atoms, whereas X-rays are obtained by the same phenomenon in the inner shell. Hard γ-rays are
associated with nuclear processes:

λ unit ν unit
AC net 6000 km 50 Hz
AC 30 km 10 kHz
LW 1 km 150 – 300 kHz
MW/AM 100 m < 3000 kHz
SW 10 m 3 – 30 MHz
FM, TV 1 m 30 – 300 MHz
mobiles, microwaves 10 cm 0.3-3 GHz
radar 1 mm 0.3-300 GHz
IR, 48 K 300 µm 1.0E+12 Hz
IR, 25 °C 17 µm 1.8E+13 Hz
IR, 1600 °C 2.7 µm 1.1E+14 Hz
IR 6000 °C 830 nm 3.6E+14 Hz
VIS, red 780 nm 3.8E+14 Hz
VIS, violet 380 nm 7.9E+14 Hz
UV 10 nm 1.0E+18 Hz
X-ray, γ-ray < 1 nm >1.0E+18 Hz

We will concentrate on the VIS, UV and IR range when considering glasses. For this purpose, we will
look at the band structures of dielectrics (e.g. glass), semiconductors and conductors (metals):
Dielectrics absorb IR radiation by means of molecular vibrations and UV radiation in electron
transitions of the outer shell. In between there is a narrow band with little interaction. Thus, dielectrics
are transparent as long as there are no multiple reflections at phase boundaries.
The following applies to semiconductors: If the energy of an incident photon EPhoton is greater than the
band gap, electrons of the valence band can be raised to unoccupied states in the conduction band.
Semiconductors thus allow frequencies up to the cut-off frequency of their band gap to pass. (1 eV
 Page 88

corresponds to 1240 nm, 3 eV corresponds to 413 nm). They are therefore opaque in the VIS or lower-
energy wavelength range (they nevertheless show reflection, absorption, refraction, transmission).
Conductors (metals) have free electrons in the conduction band, which are raised to unoccupied states
by VIS radiation. There is no band gap, therefore metals are (generally) opaque and only show
reflection and absorption. At extremely high frequencies of the electromagnetic wave, the electron-
gas of the metals is no longer able to follow the frequency; as a result, the interaction decreases and
the material becomes transparent. The cut-off frequency is called plasma frequency nP with nP2 =
e0·(e/m)·(e·ce); the plasma wavelength λP belonging to nP for different metals in nm — Al: 79, Cu: 82,
Ag: 141, Au: 80. For very thin metal layers, λP shifts towards long wavelengths — layer thickness for Ag
[in nm]: 8, 11, 14 ➝ limit T > 80 %: λ [in nm] < 940, 730, 605 nm. This effect is exploited in coating of
glass.

Various optical effects can occur in a glass: Scattering, refraction, reflection and
absorption/transmission. Here we will only refer to refraction, reflection and absorption/transmission,
omitting scattering due to the lack of scattering centers in glassy materials.

7.1. Reflection and refraction

The known optical laws apply to reflection and refraction: The sine values of the angles of incidence
and refraction are in the same ratio to each other as the velocities c1 and c2 that the light can reach in
the respective materials. If the light initially propagates in air or a vacuum and strikes a transparent
material, the ratio of the angular quantities results in a constant value that depends on the material
and is referred to as the (absolute) refractive index n. After refraction and reflection, the wave that
continues to travel has a different direction and a different speed:
 Page 89

In a vacuum, c0 is the same for all wavelengths. This is no longer true in a medium. In glass, the speed
of light is slowed down by interaction with ions and n becomes a function of λ. Dispersion occurs.
Short-wave light is slower and is refracted more strongly when it is not incident perpendicularly.
The following applies to reflection: In the case of waves, the wavelength must be considerably greater
than the distances between the scattering centers (for example atoms). Otherwise, several "reflection
rays" may be formed, for example in the case of X-rays reflected from a crystal (note: the refractive
index n for X-rays is slightly smaller than 1 for all materials, so total reflection is possible at the
transition from "optically" thin to "optically" dense). In the VIS range, the refractive index n for all
materials is greater than 1. Total reflection is therefore possible at the transition from "optically" dense
to "optically" thin (see picture above).
In all transparent materials, part of the incident light is reflected. The reflection is due to the abrupt
change of the refractive index at the interface of two media. With a vertically incident light beam,
approx. 4 – 5 % of the light is reflected at each air/glass interface (also at the glass/air transition!).
In glass, as in all transparent materials, there is always a (minimum) proportion of reflection R due to
n1 ≠ nglass:
_ $% G
𝑅 = !_9 '%" (as a function of the angle of incidence α)
9

With perpendicular incidence of light, a soda-lime glass with a typical refractive index of n = 1.5 has a
reflected portion of 4 %. Reflections (at each interface) may then lead to efficiency losses, e.g. in
photovoltaics of about 8 %.
 Page 90

The refractive power of an optical material can be represented as a molar quantity, the so-called molar
refraction r*, i.e. the specific refraction related to the mean molar mass M. r*, is given in [m3], since n
is unitless and the molar volume is given in m3. The molar refraction r* is an additive property and is
proportional to the polarizability α:
𝑛G − 1 𝑀 4𝜋𝑁6
𝑟∗ = ∙ ~ ∙ 𝛼
𝑛G + 2 𝑝 3
n is therefore a function of the density and the polarizability of a material. The greater the polarizability
α of an ion, the greater its contribution to light refraction. Understandably, anions have a greater
polarizability than cations. The refractive index determines, for example, the use of a particular glass
as an optical lens. For optical applications, however, not only the strength of the refractive index n is
important, but also its dependence on the wavelength (i.e. the dispersion). This is given by the Abbe
number:
_: $%
𝑣r = _
;' $_<

With ne = green Hg-line (546.08 nm), nF‘ = blue Cd-line (479.99 nm) and nC = red Cd-line (643.85 nm).
It describes the ratio of the mean diffraction angle to the "fanning angle" of the respective spectrum.
Small Abbe numbers mean strong dispersion.
The figure below shows an earlier problem with the production of highly refractive glasses featuring a
small dispersion (after all, one wants a true-color image): the struggle for the independent adjustment
of refractive index and dispersion. Known as the earlier "iron rule", an increase in the refractive index
n also leads to a stronger splitting of the spectrum, resulting in the so-called "iron straight line", which
was a result of this supposed proportionality.
 Page 91

Only the development of glasses that show independent adjustment of n and ve (by Otto Schott and
Ernst Abbe) made it possible to construct color-corrected or color-true optics, the so-called achromats.
This is also relevant for smaller wavelengths in the UV range. A modern lithograph lens has a diameter
of > 50 cm and costs several million euros.
Another remark on refraction in glasses: Since in inorganic glasses the anions occupy by far the largest
space and are more polarizable, substantial changes in optical properties (i.e. e.g. refractive power)
are often achieved by varying the anions rather than the cations. The table below clearly illustrates
this partition of space:

species wt. % mol % r in nm vol. %

oxides elements
Si4+ 73.0 25.0 0.041 1.0
Al3+ 1.2 0.5 0.050 0.1
Mg2+ 3.9 2.0 0.065 0.3
Ca2+ 6.8 2.5 0.099 1.4
Na+ 15.1 10.0 0.095 4.8
O2- 60.0 0.140 92.4

Fluoride lenses have a significantly lower, telluride lenses a significantly higher refractive power than
oxide lenses. This is related to the respective polarizability α:
α: F– < OH– < Cl– < O2– < S2– < Se2– < Te2–
Quartz glass has a Q4 structure, hence the low polarizability, which results in the relatively low
refractive index of quartz glass with n = 1.4589. As the NBO content increases, the polarizability and
refractive index increase. This is due to the higher polarizable oxide fraction, but also to the more easily
polarizable alkali ions. However, one would expect n to increase with the polarizability, i.e. in the order
Li < Na < K. But, besides α, the molar volume also plays a role. And since Li-Si glasses have a lower
molar volume, they are ultimately above Na-Si and K-Si glasses.
 Page 92

Of course, the refractive index n also depends on the temperature, since the density of the material is
a function of the absolute temperature. Likewise, n depends on Tf (structural relaxation), as the fictive
temperature also has an influence on the density.

7.2. Absorption properties of glasses and melts

The color of a glass is one of its most important product characteristics and can have different functions
(technical, aesthetic, etc.).
A brief background: The valence shells are responsible for the chemical bonds in molecules and solids.
In the molecule, the valence orbitals split into new energy levels. In solids, the splitting leads to energy
bands that are separated from each other by band gaps. By absorbing photons, electrons can be
excited into higher energy states, whereby the following rule, E2 ‒ E1 = h⋅ν, has to be fulfilled. In this
process, characteristic wavelengths are generally absorbed from the incident light. The excited
electrons usually fall back to their initial level by releasing thermal energy. Recombination can take
place with the emission of a photon or the release of an electron (chemical reaction), etc..... Thus, for
this coloring, electrons must be available in the valence band, which can be excited at the
corresponding energy of the photons. Sodium has the electron configuration [Ne] 3s1, so in the excited
Na-vapor it can come to the recombination 3p ® 3s and thus to an emission at 589 / 590 nm (the p-
orbitals have slightly different energy levels). This is the typical yellow color known from spectroscopic
investigations. However, with Na+ (electron configuration [Ne]) in the glass, there is no 3s electron.
Therefore, Na+ acts as a non-coloring agent in the glass.
Relevant for this ion coloration are therefore cations with free electrons in the valence orbitals. The
transition metals and the lanthanides with their d- and f-orbitals are therefore chromophores of this
type. As an example, the intrinsic transition for Ti(III) is briefly explained: The valence electron of Ti3+
is lifted from the ground state into an unoccupied energy level. The excitation occurs mainly in the VIS
and NIR range and a faint yellow color is produced.

However, there is another relevant mechanism here, the charge-transfer reaction. An electron is not
lifted to another energy level within the cation, but a transfer to a ligand (generally oxygen) takes
place. This extrinsic transfer requires significantly more energy. A charge transfer occurs between the
oxygen ligand and the central ion (or vice versa). This charge-transfer transition (mainly in the UV) is
several orders of magnitude more intense than that of the intrinsic transition and therefore causes a
significantly stronger absorption.
 Page 93

For charge-transfer transitions, the ligand field theory (see in relevant textbooks) becomes important,
since the d-orbitals are no longer energetically equivalent depending on the coordination. This leads
to the ligands coming "close" to the orbitals of the central cation at different lengths. Different charge-
transfer transitions are distinguished (acceptor-donor):
§ Transition ligand to metal: Electron of the ligand is transferred to the metal atom (p-
orbitals of the ligand to d- or s-orbitals of the metal, typical for highly charged cations).
Example: potassium permanganate
§ Transition metal to ligand: metal is donor and ligand is acceptor. (from the d-orbital of the
metal into empty orbitals of the ligand.) Example: organic complexes.
§ Metal to metal transition: When a metallic cation is present in different oxidation states in
a compound (ex: Fe(II) to Fe(III)).
§ Ligand to ligand transition: Electron transfers between different ligands (less frequent).
The colors of polyvalent ions with empty or incompletely occupied d-shells (3d, Ti to Cu) are thus
determined by the valence of the central ion and the coordination number of the oxygen complex (see
ligand field theory). These depend on the redox state of the melt and on its basicity (i.e. simplified: the
ratio of the network formers to the network transducers as well as their field strength).
In general, all 3d elements show slight changes in valence and thus a different absorption behavior. It
is very difficult to obtain only one valence in a glass or glass melt as—generally—valence equilibria
occur. If p(O2), temperature and chemistry are fixed at a given time of the melt, then this equilibrium
is also fixed. Therefore, the p(O2) in a glass melt is of enormous importance:
𝑦 𝑦
M\' + OG ↔ M(='#) + OG$
4 2
This equation is particularly important for iron. The Fe2+/Fe3+-ratio is a decisive parameter for the redox
state of a melt:
4FeG' + 𝑂G ⟷ 4FeN' + 20G$
The table below shows how the Fe2+/ Fe3+-ratio and the redox state (here represented by the redox
number, for details see under redox state of a glass) are related to the color of a glass.
Fe2+/Fetot Redox number Color
0.10 - 0.40 + 20 ↔ 0 White / Slightly yellow
0.40 - 0.60 0 ↔ -15 Green
0.60 - 0.75 -15 ↔ -25 Yellow / Green (“feuille morte”)
0.75 - 0.90 -25 ↔ -30 Brown (Amber)
 Page 94

The picture below shows the consequences of the change in valency of iron on the color impression.
Fe3+ absorbs only weakly in the blue range, while Fe2+ absorbs strongly in the yellow and red ranges. A
glass with an Fe2+ content of 20 % of the total iron results in the spectrum given by the thick dotted
line. Interestingly, it is exactly complementary to the spectrum of Mn3+. A superposition of both spectra
would result in a neutral color. This effect was used in the past. If the saturation of the color to be
compensated is not too high, then a color-neutral glass can be melted by adding MnO to a glass
containing Fe2O3 (oxides are always given as a lump sum in the composition, not broken down by
valency). The effect is called "physical decolorization". Before this, however, it makes sense to reduce
the saturation of the color to be compensated. This is done by an oxidizing melting process that keeps
the Fe2+ content low. This procedure is called "chemical decolorization".

In general, the color of a glass—once melted—does not change. Valence and coordination are frozen.
However, some ions can change their valence when irradiated by UV light (sunlight), provided that
another polyvalent ion is available as a partner. This very undesirable process is called solarization. It
leads to color defects of already installed glasses.
In general, the optical properties of the glass melt (in the VIS and especially IR range) are also very
important. The energy transfer from the energy carrier to the glass melt in the industrial manufacturing
process or the heating and cooling during the forming process depend essentially on the spectral
properties of the glass melt in the IR range.
The picture below shows the spectral distribution of the radiation intensity of a flame (assumed to be
a black body according to Planck) at different temperatures. Transmission curves of coloring oxides
present in the melt are also shown.
 Page 95

A flame is not a black body unless it contains a lot of soot. A flame radiates in discrete bands of H2O
and CO2 oscillations and rotations. However, the curve of the blackbody radiator forms an envelope
and thus provides a good reference point. It turns out that neither the Fe3+ nor the intensely coloring
Cr3+ significantly affect the transmission in the near IR range. In contrast, Fe2+ makes the glass melt
impermeable precisely in the range that is important for heat transfer in the melting furnace.
Therefore, precise control of the Fe2+ content in the melt is particularly important. Due to the very high
decadic extinction coefficient of Fe2+, this already applies to small amounts in the range of 0.001 to
0.01 Wt.-% (across-the-board Fe2O3). The effects on the glass melting process are considerable.
For example, a furnace for flint glass is built deeper than a furnace for amber glass. For the latter there
is an intrinsic danger that most of the heat radiation is already absorbed in the upper layers [by the
Fe(II)] and the glass near the bottom will not remain hot enough and freeze. This can be counteracted
by melting electrodes. Conversely, in a flint-glass tank there is a risk that the bottom will overheat.
The effect of electrical boosting to compensate for the absorption of radiation by Fe (II) is shown
below:
 Page 96

7.3. A window and a car…

Based on what we have discussed before, some everyday observations can now be explained.
For example, why is a window transparent? This can be seen in the transmission spectrum of a flint
(white) soda-lime-silicate glass shown below:

In addition to the absence of phase boundaries that would cause scattering, it is important that there
are no substances coloring agents absorbing in the VIS that would disturb the transparency.
Furthermore, we have seen that glass is a dielectric. There are no free electrons as in a metal and the
band gap is too large for photons in the visible range to raise electrons. The most easily excitable
electrons are the valence electrons of oxygen. However, they only interact at higher energy excitation
in the UV range. So above UV to VIS there are no electronic interactions with light. The atomic groups
only interact in the IR range (OH at 3.5 µm, the framework oxygens from 4 µm). Thus, in principle,
there is an optical window between UV and VIS. Of course, one always loses the reflection part of
approx. 4 % per interface.
This also explains why a car parked in the sun heats up. The short-wave heat radiation of the sun
(T > 1400 K, 700 – 3000 nm) is transmitted by the glass to a large extent (more than 50 %). The interior
heats up to, let's say 50 °C, and radiates according to its temperature at 16 to 17 μm in the clearly
longer-wavelength IR range. But this is where the interactions of the atomic groups and the framework
oxygens are located. This radiation is therefore almost completely absorbed and reflected by the glass
and the heat remains trapped (greenhouse effect). The only way to compensate for this is through
(slow) heat conduction. The result is that the interior heats up strongly.
By the way: As can be seen above, the glass is largely impermeable in the UV range, the integral
proportion of the radiation passing through in the UV range below 400 nm is very much reduced. It is
therefore extremely unlikely to get a sunburn inside the car. Unless you remain seated for an extremely
long time, but then the radiation intensity per time is probably no longer sufficient to damage the cells
(or you have collapsed in the meantime due to heat stroke).
 Page 97

8. Devitrification and crystallization

We have learned how to prevent the crystallization of a melt and thus obtain a glass. The undesired
crystallization of a melt is called "devitrification". However, if one manages to let crystallization take
place partially and in a controlled manner, one can obtain a material consisting of a glass phase and a
crystalline phase with fundamentally new properties. These glass ceramics are materials that are
produced from glass melts through controlled crystallization. Glass ceramics thus consist of a glassy
and one or more crystalline phases. Controlled crystallization leads to a composite material with
generally fundamentally different properties than the corresponding starting glass. A characteristic of
glass ceramics compared to glasses of the same chemistry is the often higher thermal and mechanical
resistance due to a more rigid crystalline structure. Glass-ceramics differ from ceramic materials of
comparable composition by being free of pores when produced in a suitable manner. To a certain
extent, it is therefore possible to combine the advantages of glass (manufacturing method and
shaping) and a conventional ceramic (thermal and mechanical resistance).
The properties of glass ceramics are defined by both the type and content of the crystal phase(s) they
contain and the residual glass phase. There are therefore several degrees of freedom in setting the
properties of a glass ceramic. Glass-ceramics (like glasses) can exhibit intrinsic and extrinsically
imposed gradients. Glass ceramics (in contrast to glasses) can exhibit anisotropy if the crystal phase is
aligned in a preferred direction (thermal conductivity, electrical and optical properties).
Thus, according to the above, a distinction is made between uncontrolled crystallization
("devitrification") and controlled crystallization (desired for production of glass-ceramics). Every
crystallization (or condensation), whether desired or undesired, is preceded by a nucleation of some
kind. Depending on whether melt, nucleus and crystal have the same chemistry or not, one generally
speaks of homogeneous or heterogeneous nucleation. Attention, the terms homogeneous or
heterogeneous nucleation are described and used vaguely even in many renowned textbooks. Details
are discussed in the next chapter under heterogeneous nucleation.
Nucleation can occur through atomic or molecular precipitation or segregation in the melt, but
secondary particles or interfaces can also act as nuclei. Segregation in the melt can occur, for example,
through density fluctuations or molecular movements by providing kinetic energy of some kind (see
next chapter).
 Page 98

8.1. Nucleation
Let us now take another look at the supercooled melt. The melting temperature is the temperature at
which a substance melts, i.e. changes from a solid to a liquid state. The melting temperature depends
on the substance but in contrast to the boiling temperature, it depends very little on the pressure
(melting pressure). Melting temperature and pressure together are called melting point, describing
the state of a pure substance and being part of the melting curve in the phase diagram of the
substance. Some substances cannot melt because they chemically decompose, and others can only
sublime under normal conditions.
For pure chemical elements, the melting point is identical to the freezing point and remains constant
throughout the melting process. Impurities or mixtures usually lower the melting temperature [melting
point depression, (cryoscopy)], and with some substances the temperature can also rise during the
melting process, which results in a melting range.
If a non-monoatomic melt is supercooled, spontaneous crystallization does not occur in the entire
volume below Tliq as Nuclei of a certain size must first be present.

In contrast to chemical elements, there can be deviations between the melting point and freezing point
of pure chemical compounds. If the freezing point temperature is below the melting point
temperature, this is called thermal hysteresis. For example, in the case of pure water: Without the
presence of nuclei and under a pressure of 1 bar, water freezes at approx. –40 °C and melts at approx.
0 °C.
Without supercooling, therefore, no nuclei form and no crystallization can take place. Nucleation
always initiates a first-order phase transition and takes place when the free enthalpy of formation ΔG
becomes negative (free enthalpy of the melt is greater than the free enthalpy of the nucleus): ΔG =
Gnucleus – Gmelt.
Let us now take a closer look at this fact with the help of the picture below. Local density and
compositional fluctuations form a subcritical crystal nucleus, which initially is less stable than the
surrounding melt. A solid nucleus is always less stable than the surrounding melt due to the
competition of volume and surface energy. Up to a critical size, given by the critical radius rc, a nucleus
therefore assumes a more stable state through resolution. If the fluctuation leads to the formation of
 Page 99

a nucleus that is larger than the required critical radius rc, then the nucleus strives for a more stable
state through further growth.

As soon as a size of r > 1.5·rc is reached, the crystallite is absolutely stable with respect to the
environment. The size of rc and the height of the energy barrier Gc depend on the degree of
supercooling T/Tliq of the system, the interfacial energy (later on expressed by the surface tension σ)
between the nucleus and the melt, and the enthalpy of fusion Hm (as will be derived later on). The
theory does not make any statement about the time dependence of the process. This is due in
particular to the fact that the viscosity of the melt is not taken into account at all. The thermodynamic
description for nucleation thus results in:

∆𝐺 = 𝐺 _s<UrsW − 𝐺 erUZ ΔGv = Change in free volume enthalpy during phase

transition
∆𝐺 = (𝐺m_ − 𝐺me ) + 𝐺& = ∆𝐺m + ∆𝐺&
Δgv = Change in free volume enthalpy during phase
4 N transition per volume unit
∆𝐺 = 𝜋𝑟 ∆𝑔" + 4𝜋𝑟 G ∆𝑔&
3 ΔG0 = Change in free Interfacial enthalpy during
crystal growth
Δg0 = Change in free Interfacial enthalpy during
crystal growth per volume unit
σ = surface tension
r = nucleus radius

For ΔGv, as the nucleus is more ordered than the surrounding melt, the free energy of the system is
decreased. The formation of a new interface (between nucleus and melt) leads to an increase in the
free energy due to the additional interfacial energy ΔG0. Simplified, G0, i.e. the energy that the nucleus
has to expend to grow against the surface tension, can be set equal to the prevailing surface tension
between the nucleus and the melt. It thus follows:
4 N
∆𝐺 = 𝜋𝑟 ∆𝑔" + 4𝜋𝑟 G 𝜎
3
In the above consideration, special attention should be given to the signs as here the textbooks start
from different definitions in the sense of ΔG = Gnucleus - Gmelt. The picture below illustrates how the two
 Page 100

energies (ΔGv, the energy released during crystal nucleation and G0, the energy the nucleus has to
expend to grow against interfacial energy/tension) change depending on the radius of the nucleus:

Of course, the maximum enthalpy ∆G* that must be expended for nucleation (and thus also the critical
nucleation radius rc) is of decisive interest. For ∆gv the following applies

∆𝑇
∆𝑔" ≈ −∆𝐻me
𝑇UYt
with Hmv as the heat of fusion related to the unit volume and the supercooling
being ΔT = Tliq – T. Thus, from derivation and zeroing of the equation for ∆G the following is valid:
16 𝜎 N 16 N 1
∆𝐺 ∗ = 𝜋 G= 𝜋𝜎 G
3 ∆𝑔" 3 ∆𝐻e 𝑇
D− 01 − 1G
𝑉l 𝑇UYt

For the determination of σ there are approaches based on the enthalpy of fusion:
σ = β·ΔHs / (NA1/3·V2/3) with β for simple silicates having values of about 0.45.
The example of SiO2 melt shows that above Tliq no formation of a supercritical nucleus is possible at all.
However, subcritical nuclei will also form in the melt above Tliq and immedialtely resolve again. If the
melt can now be cooled quickly to below Tg, a non-crystalline state is frozen. However, the viscosity is
not an explicit component of this theory, which therefore makes a prediction of the time-dependent
crystallization probability impossible. The Tg of a SiO2 melt is between 1200 and 1400 °C, depending on
the degree of purity. Subcritical crystal nuclei r > 2 nm or r < 4 nm can therefore be present in a glass
frozen in this way. The presence of clusters with a few nm in size is readily compatible with the concept
of a non-crystalline solid. This is another reason why the hierarchical SRO – MRO structural concept is
surprisingly supported.
 Page 101

We conclude: With increasing temperature T, i.e. decreasing supercooling, the necessary critical
nucleation radius rc and the maximum nucleation work ΔGmax become larger. At Tm, the values for rc
and ΔGmax become infinite.
Criteria for an effective nucleating agent are: The interfacial energy between nucleating agent
(interface or nucleus) and primary crystal phase must be small and ideally the difference in lattice plane
spacing is < 15 % to facilitate epitaxial growth. Possible substances are the noble metals (colloidally
dissolved), TiO2, ZrO2, P2O5 as well as various sulfides and fluorides. Additionally, liq-liq phase
segregations can occur.

8.1.1. Nucleation rate

The nucleation rate is always significantly higher than the crystal formation rate. We have just defined
the change in free volume enthalpy per volume Δgv during nucleation to be:
∆0 ∆R @ 0
∆𝑔" = ∆𝐻me 0 = mA
01 − 0 1
=>? =>?

78
We have determined the maximum nucleation work with
7"
= 0 to be

16 𝜎 N 16 N 1
∆𝐺 ∗ = 𝜋 G= 𝜋𝜎 G
3 ∆𝑔m 3 ∆𝐻e 𝑇
D− 01 − 1G
𝑉l 𝑇UYt

For the kinetic description we choose an Arrhenius approach with the activation energy 𝐸𝐴 being equal
to ∆𝐺∗. From the thermodynamic consideration above, with c = (nucleation) constant, the nucleation
rate IKB (nuclei per sec and cm3) can be deduced as:
∆𝐺 ∗
𝐼u5 = 𝑐 ∙ 𝑒𝑥𝑝 0− 1
𝑘𝑇
 Page 102

Additionally, considering the diffusion of atoms/elements to the nucleus (for further growth) we have
to add the diffusivity of the respective particles and the activation energy of diffusion. We thus arrive
at (homogeneneous or heterogeneous nucleation):
∆T∗'-B v(&)∙∆T ∗ ' -B
𝐼u5 = 𝐶& ∙ 𝑒𝑥𝑝 !− " or 𝐼u5 = 𝑐& ∙ 𝑒𝑥𝑝 !− "
X0 X0

c0 is then defined as c0 = (n/VM )· f , i.e. the number of molecules per unit volume nV times the frequency
f with which particles are attached to a single nucleus. The Arrhenius equation describes the case of
monomolecular reactions. The diffusion coefficient in solids is also described by an Arrhenius equation.
It then follows with the help of the Stokes-Einstein relation

and after conversion to the natural logarithm:

w $xT ∗ '-B %
In 𝐼u5 = In 𝑐 4 + ∙
0 X 0

By measuring two rate constants and two temperatures of the same reaction, the activation energy
can then be calculated by setting up the Arrhenius equation for both measurements.
The above describes the "steady-state" condition. For the time up to this steady-state state, the
following still applies (with τ as induction period, negligible if t > 5τ):
D
𝐼u5 (𝑡) = 𝐼u5 ∙ 𝑒 $E

8.1.2. Heterogeneous nucleation

Generally, there are foreign surfaces in a melt (bubbles, walls, impurities, etc...). On such foreign
surfaces, crystal formation takes place preferentially, since the formation of a new interface here
requires less energy than the formation of a new interface between nucleus and melt. Foreign particles
or foreign interfaces thus act as an "initial nuclei" and crystal growth preferentially takes place at their
surface. However, this can unfortunately also lead to uncontrolled crystallization.
ΔG*heterogen = f (θ) · ΔG*homogen
with f between 0 and 1 (dimensionless factor, depending on contact angle θ). The whole consideration
is then broken down to a question of wetting (see figure below). For θ = 180°, f(θ) = 1 and
homogeneous nucleation occurs, for 0° < θ < 180°, 1 >> f(θ) > 0 and heterogeneous nucleation occurs.

If the nucleus and the crystal phase precipitated on it are not iso-chemical to each other, we are talking
about heterogeneous nucleation. If a uniformly distributed segregation (so-called nucleating agent
segregation) is produced in the matrix by suitable process control, controlled crystallization takes
 Page 103

place. Growth of the crystals onto the (non-iso-chemical) nucleus is (presumably) favored in an
epitaxial way.
There is a problem in this definition that melt, seed and crystal are not iso-chemical. For example, if a
nucleus grows on an interface (say the wall of a crucible) in the same chemistry of the melt and a
crystal of the same composition crystallizes on this nucleus, is it a homogeneous or heterogeneous
nucleation? Is this wall by definition a nucleus or does the actual nucleus only grow on this wall? At
what point does one define an aggregation of atoms or molecules as a nucleus? 3, 5 or 10 atoms? At
what point is it (still) a segregation? The interested reader will realize here, that the above-mentioned
and often used definition falls short. Let us therefore agree for all considerations here that
heterogeneous nucleation always exists when the actual maximum nucleation enthalpy ∆G* is reduced
by one of the above-mentioned aspects. By this aspect it also becomes apparent, that heterogeneous
nucleation requires less energy than homogeneous nucleation.

8.1.3. Problem of considering the pre-factor

When determining the nucleation rate (steady state), the pre-exponential factor c0 (or also referred to
as A in the following) is usually given as nv multiplied by a frequency term, which is described e.g. via
an Arrhenius approach or a viscosity approach (Stokes–Eyring) (nv = number of available building units
per unit volume, i.e. the "particle density", this value is in the order of approx. 1028 m–3).
The pre-exponential factor thus reaches very high values in the range 1041 and thus "compensates" for
the high values of ∆G for ED. The pre-exponential factor A or c0 is also often calculated as:
%
𝑘X ∙ 𝑇 𝑎G ∙ 𝜎 G
𝐴 = 2 ∙ 𝑛" ∙ D G
ℎ 𝑘b ∙ 𝑇
%
𝐷 𝜎 G
𝐴 = 𝑛" ∙ ∙ 0 1
𝑎 𝑘b ∙ 𝑇
1 %
𝐴 = 𝑛" ∙ G
∙ (𝜎 ∙ 𝑘b ∙ 𝑇)G
𝜂∙𝑎

With a = mean size of the accumulating particles, nv ~ 1/a3, thus a measure for the "diffusion paths".
The corresponding factors next to nv describe the frequency term. The nV (particle density) establishes
a connection between the elementary step of the single particle and the nucleation rate Io. But that's
where a problem begins: how many of the particles present are actively participating in the process?
This must have something to do with a number related to the surface of the new phase to be formed.
Analogy: When dissolving a glass—a problem quite related to the formation of a new phase—the
number of H+ or OH– adsorbed on the surface of the glass is used, summarized as the surface occupancy
𝜃. Then one obtains the dissolution rate r0 as a function of
$%
𝑘b ∙ 𝑇 ∆𝐺
𝑟& = 0 1 ∙ 𝜃 ∙ §𝑒𝑥𝑝 0− 1¨
ℎ 𝑘b ∙ 𝑇

However, there is still a "permeation factor" 0 < ß < 1 missing, which describes the proportion of H+ or
OH– present to those actually taking part in the reaction. In the case of glass corrosion, the value of
 Page 104

about 0.22 is quite well known. What is the point of the comparison? It becomes obvious that the
knowledge of nV alone does not say anything.
In the overall equation for IKB
∆T∗'-B
𝐼u5 = 𝐶& ∙ 𝑒𝑥𝑝 !− "
X0

this is "ironed out" by the factor exp(–ΔG*/kT), which also effectively also defines a kind of probability
of passage. The problem here is knowing the value of ΔG*. Often, there are uncertain assumptions
behind this or ΔG* is derived by comparison with a known IKB, which is tantamount to circular
reasoning. One can determine ΔG* via correspondingly more complex barrier calculations. For further
details, please refer to more in-depth lectures and literature.
Furthermore, a geometry factor must always be taken into account in the case of new phase formation
(dendrites grow differently than spherulites or platelets); to regard nuclei a priori as spherulites is an
unsubstantiated assumption.
In the above formula for A, the diffusion coefficient D is used instead of 𝜂. According to the theory of
multi-particle diffusion in charged particles, this should be the value of the second-fastest species,
because this is primarily responsible for establishing the local charge neutrality, without which any
transport would immediately come to a standstill. Diffusion coefficients of ions in condensed matter
hardly reach values above 5·10–5 cm²/s equivalent to 5·10–9 m²/s (e.g. H+ in water). In silicate systems
with alkalis and alkaline earths, a value around 10–10 m²/s is already a very high value. Further on, for
dry melts, one can then use the approach of Eyring
X
F 0 _yG %/N XF 0
𝐷 = (By'%) ∙ ! " or 𝐷≅
w mA wV

to link diffusion coefficient and viscosity for determining values for A. However, it should be noted that
D is not always proportional to 𝜂. Close to Tg, the crystal growth can become independent of 𝜂. The
approach of Eyring is then no longer valid in this form. D is then generally larger than predicted.
The fact that the pre-exponential factor A has something to do with the surface is reflected here by
σ (surface tension) in the formula for A. But what exactly is σ? Most likely the interfacial tension
between the liquid and the developing phase. This may be easy to estimate for precipitation of salts
from aqueous solution, but it is difficult to obtain reasonable values for a LAS crystal from a melt, for
example. Although the corresponding physics are in principle correct, the respective values are not
readily available. σ is then certainly smaller than the tabulated surface tension of the melt, for example.
One could estimate it’s value by
6
% a/ 7
𝜎 ≈𝑘∙𝑇∙ !X ∙ m / "

For NS2, one then obtains e.g. σ/T = 1.37· 10–4 N/(m·K), at 1200 K correspondingly σ = 0.16 N/m. With
pure SiO2 it would only be 0.09 N/m. But since σ always occurs with power 1/2, the effect is rather
small. It reduces A by a factor of 2 at maximum.

But what is a in this consideration? It is "the distance to bring about a positional change", a measure
of the "diffusion path". This is identical to an atomic diameter only in simple ionic systems. In silicate
systems, the distance is much smaller (roughly corresponding to the distance one has to overcome to
get from tetrahedral to primitive cubic packing of the oxygen atoms; because this is the only way they
 Page 105

can change their position. In a nutshell, one could now say: the physics is right again, but one does not
know a.

In a nutshell, when it comes to getting reliable values for A, a realistic possibility is an approach of the
form A = ƒ(T)–(1/𝜂), where 𝜂 is the actual viscosity of the melt and ƒ(T) is derived from experiments.
Otherwise, the uncertainties and assumptions about ΔG*, σ or a accumulate to such a large number
of "adjusting screws" that the correct A may be obtained in individual cases, but this does not provide
a basis for reliably predicting it.

8.2. Crystal growth

If a melt is kept for a sufficiently long time at a temperature T with Tliq > T > Tg , a crystallization of the
corresponding phases (controlled or uncontrolled) can take place. For controlled crystallization, high
crystal densities of 1010 to 1015 crystals per mm3 are required. Therefore, nucleating agents are used.
Once the nucleus is formed, crystal growth can take place. The crystallizing crystallites are initially iso-
chemical with the surrounding melt. In two or more component melts, zoning occurs according to the
phase diagram. However, the central questions in crystallization are: How fast can an "amorphous"
glass structure be converted into a periodic crystal structure (crystal growth rate U)? And how quickly
can the enthalpy of crystallization Hcry be dissipated (heat flow away from the crystal)?
In the picture below, the areas of nucleation and crystal growth are shown. With decreasing
temperature, the nucleation enthalpy Hkb can be dissipated better, therefore the nucleation rate IKB
increases. However, the viscosity η increases with decreasing T, which in turn counteracts the
nucleation rate. There is therefore an ideal temperature for the maximum nucleation rate (and also
for crystallization).

In the Ostwald–Miers range, no nucleation takes place, i.e. only nuclei formed at lower temperatures
can grow. Here the melt is thermodynamically metastable; it would therefore crystallize when seed
crystals are added, for example. This is the desirable range for crystal growth, since no new nuclei are
formed during crystal growth, on which crystals would then grow again with a time delay. This in turn
would lead to a heterogeneous structure. It also follows that for well-controlled crystallization, the
overlap Tcrymin – Tkbmax should be minimal. If nucleation and crystal growth curves are separated, the
corresponding systems are very good glass formers.
 Page 106

This also explains the quasi-non-vitreous producibility of the pure metals. Here, nucleation and crystal
growth curves overlap almost completely. As soon as nuclei are formed during cooling, crystals can
instantaneously crystallize on them (as the shown in the picture below).
The position of the nucleation curve relative to the crystallization curve thus determines by which
effort crystallization can be avoided when cooling a melt. In summary, once again: In systems that can
be easily supercooled and easily form glasses, the maxima of both curves are clearly separated. Thus,
during cooling, there are no initiators for crystal formation despite increasing driving forces. Systems
where both maxima overlap require very high cooling rates for glass formation.

The further growth of the stable crystals is proportional to 1/η and the thermodynamic driving force is
given by Gm = Hm(1 – T/Tliq), as shown below. The crystallization rate is proportional to Hm (the enthalpy
of fusion) and becomes zero at Tliq, passes through a maximum at approx. 50 to 100 K below Tliq and
approaches zero asymptotically at Tg.
 Page 107

8.2.1. Crystal growth rate

For the crystal growth rate U, two energetic contributions must be considered. The energy required
for an atom to overcome the glass-crystal interface, ΔG0, and the energy gained by incorporating
elements into the crystal (free volume energy of crystallization, ΔGv):

$∆T& ∆TH a0 = interatomic distance of the particles

𝑈 = 𝑎& ∙ 𝑓 ∙ 𝑒𝑥𝑝 ∙ (1 − 𝑒𝑥𝑝 )
X0 X0
f = frequency at which particles impinge on the
crystal-glass interface

Furthermore, it must be taken into account that only a part p of the available spots is available for
growth (given energy):
$∆T& ∆TH ∆0
𝑈 = 𝑝 ∙ 𝑎& ∙ 𝑓 ∙ 𝑒𝑥𝑝 X0
∙ !1 − 𝑒𝑥𝑝 X0
" with 𝑝 = G{0
,-.

p is small for a low subcooling, but increases for strong subcooling. Taking diffusion into account, it
then follows again:
∆𝐺
!1 − 𝑒𝑥𝑝 " 𝑓
𝑈 = 𝑝 ∙ 𝑓4 ∙ 𝑘 ∙ 𝑇 ∙ 𝑘𝑇 , with 𝑓‘ =
3𝜋𝜂𝑎&G ω
From this it becomes apparent, that crystal growth rates can thus be adjusted by control of the
viscosity. Under certain circumstances, the heat dissipation Q during crystallization can influence the
crystal growth rate as well.
𝑈𝐴𝑝∆𝐻u
𝑄=
𝑀
Here A is the heat transporting surface, M the molar mass of the crystal and ΔHK the molar enthalpy
of crystallization. A small thought experiment: Let's assume a crystal growth rate U of 1 µm/sec results
in a ΔT of about 3 K. This is a reasonable value for many silicate systems and would lead to U being
only slightly dependent on HK and heat dissipation processes. However, as soon as the crystal growth
rate increases, for example to 10 µm/sec, a ΔT of about 30 K would result. Such high crystal growth
rates can occur in fast growing glass ceramics. Under these conditions, there is an influence of HK on
the growth process.
If one now wants to determine the total crystallizing volume α (interfacially controlled—more on this
later—and time-dependent), then the following applies:
−𝜋 N
𝛼 = 1 − exp ! 𝑈 𝐼u5 ∙ 𝑡 k "
3
Respectively
𝜋 N
−𝐼𝑛(1 − 𝛼) = ∙ 𝑈 ∙ 𝐼u5 ∙ 𝑡 k ≈ 𝛼 𝑓ü𝑟 𝛼 ≪ 1
3
with the corresponding equations for IKB and U given above.
If one knows the crystallization rate U and the nucleation rate IKB as a function of temperature, one
can now choose a fixed value for the degree of conversion α (as a kind of conversion criterion) and
 Page 108

create a so-called TTT-plot (Time-Temperature-Transformation, TTT). One can now define a certain α-
value as a "quality criterion", for example a very low α of 1·10–6 as a just acceptable limit for a defect-
free glass or alternatively a high value of α = 0.99 for a highly (through-)crystallized glass-ceramic.
From a diagram such as the one shown below, tN can then also be determined. This is the minimum
time required to reach a value of α = 1·10–6, which by definition is understood as the devitrification
limit, i.e. the limit of crystalline fraction up to which "a glass is still a glass" [Ullmann 1972]. This
definition then again provides surprising parallels to Lebedev's crystallite hypothesis (see chapter
Structure of glasses). A glass is therefore, at least from this point of view, a matter of interpretation.

8.2.2. Characterization of crystallization in glasses

The process of controlled crystallization is by no means applicable to all glasses, but presupposes that
the nucleation and crystallization curves overlap so little that no crystals form spontaneously during
processing of the glass and no disturbing crystal growth takes place in the area of high nucleation rates.
Only then is a homogeneous, finely-grained structure achieved, which is a prerequisite for favorable
material properties. If the first nuclei form in the area of high crystal growth rate, they quickly grow
into larger crystals and a coarse crystalline structure is obtained. If the nuclei initially form in an area
of low crystal growth, many nuclei are already present before crystal growth begins. The result is a
much finer-grained structure.
The crystallization behavior of glasses is usually investigated in a gradient furnace. If necessary,
samples can be kept at temperatures in the range T > Tg for long periods. The type and size of the
crystalline phases formed are afterwards documented for the treated samples. For glasses with a
somewhat stronger tendency to crystallize, thermo-analytical methods are also suitable. The picture
below shows the result of a DTA (differential thermal analysis) applied to a crystalline basalt sample.
The temperature curve and the thermal effects are shown as a function of temperature. The large
endothermic peak (at 110 min) characterizes the melting of the main part of the crystalline phases, the
exothermic peak (at 180 min) the crystallization during cooling. The application of the thermal effects
 Page 109

as a function of temperature shows that the temperature of melting (1161 °C) is significantly higher
than that of recrystallization (1027 °C). A certain amount of supercooling is therefore always required
for crystallization.

8.2.3. Interfacial and diffusion controlled growth

Crystal growth can be diffusion or interfacial controlled or both as a function of time. If the crystal and
melt are iso-chemical or if the crystallizing component is present in quasi "infinite" excess, crystal
growth will always be interfacially controlled. Thus, the growth rate is solely dependent on how fast a
building block (atom or molecule) can be incorporated into the crystal. A homogeneous crystal without
gradients is formed.
In the opposite case, if a building block must first diffuse towards the growing crystal, the diffusion
coefficient of this element or molecule determines the growth rate. The growth is diffusion controlled.
Usually, a compatible element is incorporated into the crystal at the beginning. This element then
depletes around the crystal over time and has to diffuse towards the crystal at an ever greater expense
of time. In this case, often less compatible elements, but closer to the crystal, can also be incorporated.
The consequence is that a chemical zoning, i.e. a chemical gradient across the crystal, is created (which
may be balanced by diffusion within the crystal).

Gradient within crystal

 Page 110

In general, if there is no single-component melt, growth will often start in an interfacial-controlled

manner and then change to a diffusion-controlled growth. The kinetic dependencies will be described
in the next chapter using Avrami-Erofe'ev kinetics.

8.2.4. Avrami-Erofe’ev kinetics

For a detailed derivation see the lecture slides and/or corresponding literature.
The volume V*(t) of a single crystal grows according to a time law that depends on the geometry of
the growth and the local mechanism (interface controlled = linear time law or diffusion controlled =
square-root of time). From the above and the geometry considerations, the following dependencies
result:

the (4·D·t)0.5 dependence is the so-called Boltzmann substitution, with which one solves the differential
H< c∙I 6 ∙<
equation HZ
= HZ 6
. It corresponds (like a standard deviation) to the mean range of the diffusion
profile according to time t. The standard kinetics of the JANDER equation etc. also use this approach
accordingly. Whether one uses (4·D·t)0.5 or (D·t)0.5 here, however, is of rather marginal importance. In
the end, all that matters is the determination of the exponent n in the sum n+m in the term
exp[–k·t(n+m)], where m comes from the nucleation and k is a temperature-dependent constant that is
not determined in detail via this approach. The exponent is always the same. In this respect, it does
not matter whether D or 4·D is used in the root. But 4·D·t seems to be more suitable in front of the
Boltzmann background as a counterpart to v·t.
Depending on the geometry and local mechanism, time laws V*(t) = B×tn with exponents n between ½
and 3 result.
 Page 111

n is therefore a function of the geometry and the growth law.

Using the same line of though as for the crystallization growth, the nucleation density NV can also be
either constant in time, according to a root law or increase linearly in time (exponents m = 0, ½, 1). The
nucleation density NV, with C as the initial nucleation (rate) (resp. IKB) can be described by the following
equation:
1 𝐶
𝑁" = 𝑡 e , 𝑚 = 0, , 1 ⟹ 𝑑𝑁" = 0, , 𝐶 ∙ d𝑡
2 2 ∙ √𝑡

For the mass balance then follows:

𝑑𝑉== = 𝑉 ∗ (𝑡) ∙ (𝑉& − 𝑉== ) ∙ 𝑑𝑁m
The temporal increase in volume dVxx of total crystallized matter is thus determined by the growth law
of the single crystal V*(t) = B×tn multiplied by the number of places where crystallization takes place,
dNv (only the not yet crystallized residual volume Vo - Vxx is available for further crystal growth). From
the above, the change in crystalline volume then follows by integrating the above equation for dVxx,
with Vcryst/Vo = α, i.e. the degree of transformation related to the initial volume Vo, 0 ≤ α ≤ 1,

d𝑉== d𝛼
X =X = X 𝑉 ∗ (𝑡) ∙ 𝑑𝑁m
𝑉& − 𝑉== 1−𝛼

− ln(1 − 𝛼) = 𝑉 ∗ (𝑡) ∙ 𝑁m = 𝐵 ∙ 𝐶 ∙ 𝑡 _'e = 𝑘 ∙ 𝑡 |

And finally, the actual Avrami-Erofe'ev kinetics:

𝛼 = 1 − 𝑒𝑥𝑝(−𝑘 ∙ 𝑡 | ), 𝑋 = 𝑛 + 𝑚
 Page 112

X is the Avrami-Koeffizient. The sum X of the exponents n (for the crystal growth type) and m (for the
nucleation type) is decisive. k is the Avrami kinetics coefficient. Remember: k contains B (with D, resp.
U0) and the initial nucleation rate C (resp. I0). The Avrami-Erofe'ev kinetic thus describes interfacial and
diffusion-controlled nucleation and crystal growth as a function of time. The general solution is:
𝛼 = 1 − exp8−𝑘 ∙ 𝑡 } 9 , 𝑋 = 𝑛 + 𝑚 ⟹ −𝐼𝑛(1 − 𝛼) = 𝑘 ∙ 𝑡 _'e

We can now use this equation to represent the degree of transformation α as a function of time t
(shown below as a function of a normalized time in order to have comparability of the curves). The
course of the curves is then characteristic for the geometry of the crystal (sphere, needle-like or
platelet), type of nucleation (nucleation density linear growing, diffusion-controlled growing or
constant) and the crystal growth (interfacial controlled or diffusion-controlled).

Therefore, if the rate of crystallization of a sample (reaction turnover) can be determined via an
independent experiment (e.g., XRD, heating stage microscope, ...), the mechanism of nucleation and
crystal growth can be determined:

crystal crystal growth nucleation kinetics

geometry kinetics const.: diff.: linear:
m=0 m=0.5 m=1
N=2 linear: n = 1.0 1.0 1.5 2.0
diff.: n = 0.5 0.5 1.0 1.5
N=1 linear: n = 2.0 2.0 2.5 3.0
diff.: n = 1.0 1.0 1.5 2.0
N=3 linear: n = 3.0 3.0 3.5 4.0
diff.: n = 1.5 1.5 2.0 2.5
 Page 113

8.3. Glass ceramics

The aim of controlled crystallization is to produce a glass-ceramic. The manufacturing process is initially
identical to that of a glass. A glass of suitable chemistry is melted and cooled. Through a controlled
process (so-called ceramization), the glass is partially crystallized by an additional heat treatment step.
As a reminder, it is not possible to obtain the desired crystal phase through controlled cooling, as
nucleation is necessary before crystal growth. Nucleation is carried out first, then the temperature is
increased and crystallization is initiated.

The production route described and illustrated in the figure above is the "classical" glass route.
However, there are other ways to produce a glass ceramic. For example, a glass powder can be
subjected to a temperature treatment without nucleating agents. In most cases, interfacial
crystallization occurs during the sintering of the powder. This results in a glass ceramic with
0.5 – 3.0 % residual pore content. If a glass powder with nucleating agents is used, volume
crystallization can be achieved. This sintering route is interesting for glass-ceramics made from systems
with a low tendency to glass formation. If necessary, these systems can be made glassy via fritting, as
higher cooling rates can be achieved here. Sinter crystallization would then take place. Another route
is the "indirect" glass ceramic. Here, a glass powder is mixed with a crystalline powder and fused
without crystallization.
There are various glass-ceramic systems that are characterized by their main crystalline phase,
including:
§ LAS glass ceramics → low or zero-expansion (Low CTE*) glass ceramics for cooking surfaces,
mirror telescopes, photolithography, laser gyroscopes, ....
§ MAS glass ceramics → high strength applications (cordiertite as main crystal phase) for HDD
substrates, high E-module applications, dental applications
§ Apatite-woolastonite glass-ceramics → medical technology, bone replacement, bio-
"ceramics".
§ Disilicate → dental applications
§ Mica glass-ceramics → medical technology, machine-machinable GC
§ Leucite glass-ceramics → medical technology, dental technology
 Page 114

8.3.1. LAS glass-ceramics

One of the most important three-material systems for the production of glass-ceramics is the
Li2O×Al2O3×SiO2 system (LAS system) and its modifications. Of the many different glass-ceramics, only
glass-ceramics of the LAS system will be presented here due to the sake of brevity. Hummel and Smoke
have done fundamental work on the main crystal phases found in this system, namely high quartz solid
solution (HQSS) and keatite solid solution (KSS).

These crystal phases show a very low to even negative coefficient of thermal expansion (CTE).
Therefore, with these phases as the main component, glass ceramics with excellent thermal shock
properties and also very good mechanical strength can be realized. Applications are manifold as
material for laser gyroscopes, as telescope mirror substrates or as protective glasses with high thermal
shock resistance, as well as in the household sector as hobs and cookware. Well-known product names
here include CERAN® for hobs and ZERODUR® for telescope mirror substrates.
Glass-ceramics with HQSS as the main crystalline phase have a very good thermal shock resistance due
to their low thermal expansion coefficient α20-700 of about 0,1·10–6 K–1. If the chemical composition of
the pure Li2O×Al2O3×n SiO2 system is n ≥ 3.5, the HQSS transforms into keatite solid solution at
temperatures above 950 °C. The phase transformation is irreversible and reconstructive, i.e. coupled
with bond breaking. Nevertheless, the two crystal phases are very similar in their structures. The LAS
phases are derivatives of silica structures and are formed by replacing Si4+[4] with Al3+[4] and charge
balancing via "stuffed" cations (H+, Li+, Zn2+, Mg2+, Fe2+, Mn2+, Co2+, …).
 Page 115

After the conversion, the CTE of the glass ceramic increases to about 1·10–6 K–1 in the composite due
to the higher CTE of the KSS. This causes higher thermally induced stresses in case of thermal shock
and thus a higher risk of fracture of the material under thermal load. Note: The negative coefficient of
thermal expansion of the HQSS results from a structural feature, see chapter "Negative coefficient of
thermal expansion". A negative coefficient of thermal expansion cannot be explained by an interatomic
potential.
The initial glass (usually a flat glass) is rolled, cut and stored. There is subsequent post-processing
(decoration, cutting, drilling, bending, etc...), then ceramization in roller kilns is achieved in times less
than 2 hours. Temperature-time control is very crucial, as crystallization generally does not start
homogeneously throughout the volume at the same time. Crystallization is associated with large
volume changes (up to 2%), therefore dimensional changes / distortions may occur.
The microstructures of the HQSS and KSS glass-ceramics are very different, as can be seen in SEM
images below. The glass ceramic with HQSS as the main crystalline phase typically has a residual glass
phase content of 40 – 60 vol-% and a CTE of 0.1·10–6 K–1. The glass ceramic with KSS as the main
crystalline phase typically has a residual glass phase content of < 10 vol-% and a CTE of 1·10–6 K–1.

The properties of such a glass ceramic can be adjusted by the type of crystal phase but also (and in the
case of CTE much more strongly) by the proportion of residual glass phase to crystal phase. In such
glass ceramics, which consist of many components, secondary crystal phases always form, see TEM
 Page 116

image below left. Likewise, they often show a layered structure due to the ceramization, see SEM
image below right.

The table below shows typical compositions and modes of action of the individual components for
industrial LAS glass ceramics.

Oxide Range [w.-%] Function / Impact

Al2O3 18 – 25 HQSS / KSS-former
Li2O 3–6 HQSS / KSS-former
SiO2 60 – 75 HQSS / KSS-former
MgO 0–2 HQSS / KSS-former
ZnO 0–2 HQSS / KSS-former
P2O5 0–5 HQSS / KSS-former
TiO2 1–6 nucleation agent
ZrO2 0–4 nucleation agent
Na2O 0–2 melting behaviour
K2O 0–2 melting behaviour
BaO 0–3 melting behaviour
As2O3 0–2 refining agent
Sb2O3 0–2 refining agent
Transition- and 0 – 2 (each) coloration
RE-metal oxides

8.4. Metallic glasses

A short note on metallic glasses. If the viscosities of metal melts are analyzed in detail, it becomes
apparent that metal melts (and also salt melts) show just as steep increases in viscosity with falling
temperature as silicate melts—albeit starting from much lower values. This observation has long
challenged scientists to represent such systems as glasses as well. Metallic glasses—metals without a
structure(!)—have outstanding mechanical properties: they combine the well-known high tensile
 Page 117

strengths of metals with the high yield strengths known from wood or polymers (i.e. deformation
without permanent plastic components). In the beginning, the production of metallic glasses required
extremely high cooling rates Rc. The metallic glasses could only be produced as very thin foil fragments
or later as very thin strips. Soon, however, it was understood that the packing frustration (see chapter
on glass structure) of the glass formation could be overcome by mixing metals of very different atomic
radii. It was even possible to make a glassy NaCl sample. H2O also forms a glass; its Tg is at 136 K or at
169 K (the exact value is disputed). A very successful realization of glass formation in volume was
achieved for the system Zr-Ti-Cu-Ni-Be (so-called "bulk metallic glasses"). It can be seen that the
viscosity-temperature curve, with the exception of the very low values at T > 1400°C, is similar to the
curve of a float glass.

The TTT curve below shows that crystallization is already avoided at very comfortable cooling rates <
10 K/s.
 Page 118

9. Industrial (large-scale) glass melting

Glassmaking processes are among the oldest technologies of mankind. At the same time, they are still
one of the pillars of our industrial society and the glass industry is an important branch of the economy
in Germany (for details, please refer to the lecture slides). All glassmaking processes are subject to
continuous development. In addition to the quality of the products, the efficient use of resources and
the protection of the environment are becoming more and more important. In light of the worldwide
shortage of raw materials and energy sources and the increasingly noticeable effects of climate
change, the glass industry—like all high-energy industries—is facing enormous challenges.
The first step in large-scale glass production is the melting process, which transforms the raw materials
into a processable melt. Please note here: The glass is melted from the raw materials. Only in rare
cases is an already produced glass melted again. The melting process itself basically takes place in four
steps (batch input excluded):

The batch is placed in the furnace and dehydration of the raw materials, water evaporation (up to
700°C if clay is used) and solid-state reactions between the individual batch components occur. Further
evaporating species can be: SO2, HBO2 (if borates are used as raw materials), HCl, NaCl, KOH, PbO, Pb,
NaOH, etc...
Either the primary (eutectic) melts are formed (depending on the composition) at 700°C – 900°C, or
the initial melts originate from the low-melting raw materials. These are often the carbonate melts
with a very low viscosity and a high aggressiveness towards the other batch components (ideally
towards the sand). CO2 is emitted when the carbonates decompose (500°C – 1000°C) or some salts
melt directly when Ts is reached. The phase of the rough melt follows, in which the respective solid-
state and melting reactions take place according to the corresponding phase diagram. The sand reacts
between 750°C – 1000°C with the sodium silicates or soda and forms liquid sodium silicates and
reaction foam.
In the end, all the melt phases are present in this rough melt, with the exception of the batch
components that cannot (or no longer) be transferred into the melt phases by some kind of reaction.
In many bulk glasses the last undissolved phase is sand (SiO2), but there can also be other refractory
 Page 119

phases such as corundum. These components must then be dissolved in the melt phase. They do not
melt (any more). Sand has a melting point of 1700 – 1800°C and these temperatures are typically not
reached in a melting tank. Melts of soda-lime-silica glasses generally have temperatures of up to
1450°C. The initial melting phase is thus followed by a phase where the remaining sand is dissolved
(quartz dissolution). Finally, there will be refining and conditioning steps (homogenizing; refining). The
individual steps differ in terms of mass conversion, temperature level, heat conversion and time
required. It is noticeable that the greatest heat conversion is offset by a very short dwell time.
However, the individual stages of the above block diagram are not run through strictly one after the
other (see structure of tanks). At least the first three take place—spatially more or less well
separated—in a single reaction volume. The respective phases will be discussed in detail later.

9.1. The batch (raw materials and cullet)

A distinction is made between primary and secondary raw materials. Primary raw materials are further
subdivided into natural (sand, limestone, dolomite, feldspars, ...) and synthetic raw materials (soda,
sulfates, coloring oxides,...). Secondary raw materials are, for example, cullet (see below) and filter
dusts from evaporation reactions or from dusting, which are returned to the melting process from the
waste gas purification.
The raw materials have to be prepared, weighed and mixed. The mixture of raw materials forms the
actual batch.
Suitable raw materials for classic SLS glasses are sand, soda, limestone, dolomite, feldspars. In general,
mineral raw materials are always more suitable than synthetic raw materials due to the melt
thermodynamics. For example, it is always more favorable for the melt to use feldspar as an aluminum
carrier instead of corundum, which would be one of the refractory components of a melt due to its
extremely high melting temperature.
The advantages and disadvantages of mineral and synthetic raw materials are summarized below:

natural raw materials synthetic raw materials

§ cheaper § more expensive

§ Often contaminated § high purity
§ variable composition § constant composition
§ only limited process control § good process control
§ multiple samples for tests § simple samples for tests
§ not always certified § certified

In particular, each raw material has its own special characteristics that need to be taken into account.
In the case of sand, for example, not only is the impurity, grain size and grain size distribution
important, but also the grain morphology, which can have an influence on the melting-in behavior.
Soda ash is one of the most expensive raw materials in an industrial glass batch, as it is generally used
at more than 10 Wt-% and—as a synthetic raw material—has to be produced in a very energy-intensive
way (approx. 3200 kWh/t).
 Page 120

Alkaline earth carbonates, if introduced as dolomite, have another peculiarity that must be taken into
account. As can be seen from the schematic phase diagram below, pure calcite can dissolve larger
amounts of Mg and Fe. The area of existence of the single-phase, Mg-rich calcite is relatively large. A
pure dolomite can dissolve almost no Ca at all, but a lot of Fe, whereby the possible Ca content in the
dolomite increases slightly with increasing Fe content):

This means, however, that a natural, Fe-poor dolomite will practically always be present in two phases,
a dolomite phase with some Fe and Ca dissolved and an Mg-rich calcite phase. In the structure of
dolomite, one can see that the dolomite phase can be present in lamellae in the Mg-rich calcite phase.
As soon as dolomite is no longer in a single phase, a complex structure is formed. It consists of dolomite
grains (with dissolved Ca) and calcite grains (with dissolved Mg), whereby structures (twin lamellae;
also lamellae of the respective other phase) are present within the grains.
 Page 121

The Mg-rich calcite and the dolomite, however, show fundamentally different decomposition
behavior. Selective decomposition of the dolomite phase above 680 °C occurs. This leads to dramatic
changes in the entire structure, the dolomite "decripitizes": It virtually explodes and becomes dusty.

In a pure CO2 atmosphere, the first stage of decomposition is at 680 °C, the second at 900 °C. The mass
losses correspond exactly to the decomposition of MgCO3 to MgO or CaCO3 to CaO. At partial pressures
pCO2 below 16 mbar, only one decomposition stage below 680°C occurs. Above pCO2 = 16 mbar, the first
stage remains fixed at 680 °C, while for the second stage the respective temperature increases with
increasing pCO2. This leads to an explosion of the dolomite grains. The resulting particles and
evaporation condensates can cause massive problems, especially in the regenerator, e.g. clogging due
to condensation or also due to so-called "carry over" of particles.
The reaction of the first decomposition stage is diffusion-controlled, thus dependent on the grain size
of the dolomite used: a decomposition temperature of 680 °C was only found with finely pulverized
dolomite; qualities used in the glass industry with grain size distributions of up to approx. 1 mm
decomposed at approx. 720 °C. The reaction of the second decomposition stage is controlled by
diffusion. The reaction of the second decomposition stage is a first-order reaction occurring in the
volume. With the exception of heat conduction effects (i.e., at moderate heating rates), it is therefore
independent of grain size.
Cullet is another very essential and important "raw material" for many SLS glasses (especially in the
container glass industry). On the one hand, the cullet is already a thermally converted raw material (so
no more energy is needed to form a glass from it; moreover, the CO2 has already been expelled).
Secondly, depending on their size, cullet acts as a kind of window for the heat radiation that is
transferred from the flame into the batch. In this way, cullet facilitates the melting process. If the
shards become too small, they react quickly with the soda, leaving the sand without a solid-state
reaction partner. Small cullet is also too compact and behaves similarly to batch. However, with even
 Page 122

smaller cullet size, again the interfacial reactions are greatly enhanced. This leads to the diagram below
of the dependence of the meltdown on the size of the cullet:

It should also be noted that the overall thermal properties of the batch significantly influence the
melting rate. A batch without cullet reaches the filled pore stage relatively late. The finer the shard
fraction, the sooner the very shard-rich batch reaches this stage. Extremely fine cullet (< 63 μm, not
shown in the picture below) initiates the phase of closed porosity particularly early (from 750 °C), but
traps the batch gases all the longer, which can lead to problems with degassing and then to foam
formation.

The material's own cullet or foreign cullet (from the well-known "glass collection containers") can be
added. The whole batch can be conditioned, e.g., as pellets (with binder) or as briquettes (only
mechanically pressed together).
 Page 123

9.2. Phenomenological processes in the batch blanket

Before going into detail about the individual melting processes, a general overview of the melting
process is given here. The melting of the batch carpet is a very crucial step. The reaction kinetics and
diffusion rates depend strongly on the temperature. Therefore, the heat transfer into the batch carpet
is very important for the melting rate. The turnover rate of the melting therefore strongly depends,
among other things, on the mobility of the atoms in the melt [T, η of the initial melt, defect structure
of the solids (solid state reactions)] and the distance and contact area of the partners to each other (as
a function of wettability, spatial arrangement, grain size, amount of initial melt, ...). The energy
demand of the melting process is, as seen at the beginning, very high (typically 0.5 MWh/t glass at 50
% cullet) and the time demand is less than 60 min. The melting-in of the batch is therefore an area of
very high energy demand.
The batch is heated from two sides, the combustion space (here mainly by heat radiation) and the melt
(here mainly by convection). The heat input into the batch depends on the amount of heat that can be
"delivered" to the batch, the heat transfer at the interface (how "fast" the heat can be transferred,
characterized by the heat transfer coefficient α) and the thermal diffusivity of the batch a = λ/(ρ·cp).
However, this also means that a high heat flux q of flame and melt does not necessarily guarantee a
high melting rate. What is also decisive is how quickly and efficiently the energy can be brought into
the batch. And this heating rate within the batch depends on the thermal conductivity and the
(endothermic) chemical reactions in the batch.

Melting (batch-to-melt conversion) is completed at the liquidus temperature (Tliq) of the system. Below
Tliq, crystalline components (e.g. non-dissolved quartz) remain.
The picture below illustrates, how the batch melting proceeds at a batch heap or a batch blanket of
height h0. In a first stage (1), the batch only absorbs heat, but remains in the state of a granular bulk
material. The heat input occurs, as previously described, both from the top furnace and from the hot
backflow of the molten glass emanating from the "hot spot". The melt fronts move from the outside
towards the centerline of the batch. Again, two stages can be distinguished: (2) the occurrence of a
spatially extensive initial melt in an essentially still granular batch and (3) the formation of a rough
melt, i.e. essentially a liquid phase that still contains many bubbles and crystalline relicts. The quite
precisely obtainable transition temperatures T1: (1) → (2), T2: (2) → (3) can be assigned to the stages.
 Page 124

9.3. Initial melting

The initial melting generally takes place through the carbonate reactions. Soda melts at about 820 °C.
Limestone decomposes at about 830 °C unless there is a sufficiently high CO2 partial pressure in the
atmosphere. In principle, this decomposition of limestone is seen as unfavorable, since it leaves a
heavy-melting oxide (CaO, Ts = 2580 °C!), unless all CaO can be transferred into melting phases by
earlier reactions. Two reactions compete with each other in the melt:

• The so-called carbonate route, which is present at high heating rates (>100 K/min) and which
drives the reactive dissolution of SiO2 in a binary melt phase of Na2CO3 and CaCO3:
Na2CO3 + CaCO3 ⇌ Na2Ca(CO3)2 (conversion via solid state reaction)
Na2Ca(CO3)2 melts at about 820 °C and becomes very reactive towards the sand. A
Na2O·CaO·SiO2 melt plus CO2 (g) is formed:
Na2Ca(CO3)2 + 2SiO2 ⇌ Na2SiO3 / CaSiO3 + 2CO2 (g) (> 820 °C)

• And the so-called silica / silicate route, which dominates at low heating rates (10 K/min) and
in which the eutectic mixture is formed from the solid-state reaction between sand and soda:
Na2CO3 + 2SiO2 ⇌ Na2Si2O5+ CO2 (g) (700 °C - 860 °C)
Na2CO3 + SiO2 ⇌ Na2SiO3+ CO2 (g)
Another eutectic melt of SiO2 and Na2Si2O5 forms at 799 °C, forming a silicate-rich melt.
The ratio of these reactions to each other is also depending on grain size of the sand (and depending
on the grain size distribution).
Some melting point and further eutectics are listed below:
Melting points:
 Page 125

Ts (Na2SiO3) = 1089 °C
Ts (Na2Si2O5) = 874 °C
Eutectics:
Teu (Na2O × 2SiO2 + SiO2 ) = 793 °C
Teu (Na2O × 2SiO2 + Na2O × SiO2 ) = 846 °C
Teu (2Na2O × SiO2 + Na2O × SiO2 ) = 1020 °C
In combination with CaO, an even lower (ternary) eutectic can be achieved:
Teu (Na2O × 3CaO × 6SiO2 + SiO2+ Na2O × 2SiO2 ) = 725 °C
During this stage of a granular bulk material with open pores, a free exchange between batch pores
and furnace atmosphere is possible. Organic components burn off; emission peaks of HF and SO2 are
also observed. Upon reaching the stage of the so-called rough melt, the pore spaces of the batch are
separated from the furnace atmosphere. The primary emissions are thus immediately suppressed. The
first melting stages provide a large surface area for the enclosed pores ("reaction foam"). During this
short phase, which lasts about 30 min, the redox-active additives develop their effect (see later) and a
very specific Fe2+/Fetotal ratio is established, which cannot be changed independently of this initial value
during the entire further course of the process (beware of the influence of Fe2+ on the absorption
properties of the melt!). At the end of this whole melting step / melting sequence, the rough melt is
present. The picture below illustrates once again in detail the mechanisms described above. The
solubility of the total sulfur in the melt is determined as a very important consequence of the earlier
determined redox state (or Fe2+/Fetotal ratio). This typical curve, known as the "Budd curve" (see
chapter on refining), shows that melts of glasses of different colors have very different sulfur contents.
A color change from white to brown in a tank should only be carried out if the amount of Na2SO4 in the
batch is drastically reduced beforehand (and only adjusted again after the change has been made).
Otherwise, the melt will foam up.
 Page 126

It is also important to note that the speed of the chemical reactions taking place in the mixture are a
function of the atomic mobility (which is reflected in the diffusion coefficients or the electrical
conductivity), the distance between the reaction partners and their contact area. The wetting behavior
of the melting phase plays an important role and electrical conductivity measurements can be used to
characterize such a melting phase.
The figure below shows the time at which the initial melt has formed by the sudden increase in
electrical conductivity and compares this with the thermal conductivity. The sharp increase in thermal
conductivity follows that of the electrical conductivity at a distance of 100 K.

9.4. Sand dissolution

With the rough melt as the starting point, a phase of sand dissolution begins (if sand is the refractory
element, depending on the corresponding constitutional phases → see earlier chapter). In a glass melt,
SiO2 can therefore be present as a reference phase. This SiO2 is then, as explained earlier, called
"constitutional silica" or also "residual silica", because it represents SiO2 that is not bound in one of the
other phases by reaction (and can thereby melt e.g. in an eutectic). This remaining (hence "residual")
SiO2 has subsequently to be (slowly) dissolved in the melt.
Based on the normative phase content, which represents the amount of crystalline phases that would
result if the glass was completely crystallized, the amount of residual or constitutional SiO2 can be
determined. This is therefore the amount of SiO2 that is not consumed by chemical reactions in the
other normative phases, but must be (slowly) chemically dissolved by diffusion processes. This amount
can be quite considerable, as the example below illustrates with an amount of just under 23 %, and
can vary very significantly with the oxide composition (see calculation examples in the lecture slides
and exercises).
 Page 127

raw material i kg/t oxide j kg/t phase k kg/t

sand 666.96 SiO2 720.00 SiO2 227.05
feldspar 77.15 Al2O3 15.00 Na2O·Al2O3·6SiO2 77.15
dolomite 182.98 MgO 40 .00 MgO·SiO2 99.62
limestone 34.54 CaO 75.00 Na2O·3CaO·6SiO2 263.34
soda ash 240.91 Na2O 150.00 Na2O·2SiO2 332.84
CO2 202 .54
sum 1202.54 1202.54 1000.00

Sand dissolution is enhanced by a smaller grain size and a narrow grain size distribution (fine sand with
a few large grains increases the dissolution time, as a SiO2-rich melt forms quickly and the large grains
remain in the highly viscous melt for a long time). Additionally, increasing SiO2 diffusion (higher
temperatures, aggressive melts), increased convection (e.g. stirring, bubbling) or increased SiO2
solubility through adapted composition helps with sand dissolution. Especially the last point is, of
course, clearly limited by the final product requirements. The picture below shows a glass melt with
bubbles and not yet dissolved constitutional sand (white circles).

The kinetics of the sand dissolution process should also be discussed. The dissolution is diffusion
controlled, meaning that the SiO2 that dissolves at the interface must somehow be transported away
from the interface into the melt by diffusion. The force for this is the concentration gradient between
the grain-melt interface and the surrounding melt.
 Page 128

A grain of sand consists of 100 wt.-% SiO2. The surrounding melt has a concentration of perhaps 50
wt.-% at the beginning. This quite steep gradient—as a driver for diffusion—decays quite quickly and
the melt then already contains about 70 wt.-% SiO2. A diffusion layer is formed around the sand grain,
which consists of about 85 wt.-% SiO2. This reduces the driving gradient to just 15 % and the process
becomes very slow without support.
If a large grain of sand dissolves, there is a lowering of the basicity of the melt in the vicinity of the
grain. This leads to a lowering of the CO2 solubility in the melt and thus to an (undesired) release of
CO2 bubbles, see picture below (©GlassService). Due to this, it is important to ensure a homogeneous
grain size distribution of the sand without large grains.

9.4.1. Batch-Free-Time (BFT)

In this context, one quantity becomes important: the batch free time (BFT). It refers to the time
required at a certain temperature to obtain a completely clear melt without non-dissolved
components. Bubbles are not taken into account here. The definition is thus quite simple, but when
did the last grain disappear? This is where it gets interpretative.
The BFT is a function of temperature, heating rate, composition, specific batch constituents, batch
homogeneity, grain size, grain size distribution, shards, etc... Unfortunately, there is no uniformly
 Page 129

accepted standard for a BFT, so each laboratory has its own BFT definition. This should be taken into
account when making quantitative comparisons.

9.5. Redox state of a glass

The redox state of a glass describes the ratio between the oxidized and reduced species (polyvalent
ions) in the system. It determines the color and IR absorption of a glass (due to different valence states
of the ions, see also chapter "Optical properties").
Oxidizing components can be: sulfates, nitrates, polyvalent ions in the highest oxidation state (Fe3+,
Sb5+, As5+, Sn4+, Ce4+, etc.).
Reducing components: Organics, carbon, reduced form of polyvalent ions (Fe2+, Sb3+, As3+, Sn2+, Ce3+,
S2–, etc.) or the pure metals (!).
If a melt contains many oxidizing components (sulfates / nitrates), a lot of O2 will also be dissolved in
the melt as O2– (i.e. chemically). This influences the redox state of the glass and leads to the well-known
color changes in the presence of iron, for example:

Thus, the oxygen partial pressure p(O2) in a glass melt is of enormous importance, as by means of
measuring the p(O2) the redox state of the melt is determined.
The redox state of the glass batch is characterized by the so-called batch redox number R, in which
each batch component is assigned a redox factor Ri (tabulated values) (the more negative, the more
reducing) and the redox factor of the batch can then be calculated by ∑ 𝑅Y ∙ 𝑚Y . Alternatively, the redox
state can be determined via the COD (Chemical-Oxygen-Demand).
Finally, the redox state of the final glass is determined by the ratio Fe(II) to F(III), which is normally
expressed as the ratio of Fe2+ to the total iron content (left column) in the melt/glass:

Fe2+/Fetot Redox number Color

0.10 – 0.40 + 20 ó 0 White / Slightly yellow
0.40 – 0.60 0 ó -15 Green
0.60 – 0.75 -15 ó -25 Yellow / Green (“Feuille morte”)
0.75 – 0.90 -25 ó -30 Brown (Amber)
 Page 130

Changes in the redox state (for example by adjusted refining, see the following paragraph) become
critical when colored glasses are to be melted. For example, a shift of the redox state of a flint glass
into the reducing regime necessitates either a compensation of the Fe(II) content (since more Fe(II)
would result from the total Fe due to the reducing conditions) or the total iron content has to be
reduced. Otherwise, a greenish coloring could occur in the flint glass. In fact, one would need raw
materials with a low iron content. Here, sand, limestone and dolomite can be particularly critical.
Of course, one can carry out laboratory experiments and measure the Fe(II) content or the light
absorption of the produced glass at 1050 nm to determine the color. With visual inspection, there is a
dependence on the thickness of the glass product as a faint greenish haze can be hard to detect. Since
float glass production even desires an increased Fe(II) content for reasons of better energy absorption
in the melt, flat glass panes look slightly greenish when viewed from the side. Lower iron values are
used in container glass production, especially for “white” glass. Of course, for colored glasses
additional amounts of coloring agents are needed.

9.6. Fining and refining

After residual quartz dissolution, the melt still contains very large quantities of small bubbles. The
extremely time-consuming process by which these bubbles are expelled is called refining. Only the
larger bubbles are able to leave the melt by buoyancy. Therefore, a chemical reaction or a component
with a sufficiently high vapor pressure at a given temperature is added to create a swarm of relatively
large bubbles that absorb the smaller bubbles (fining). After fining, the remaining small bubbles are
ideally reabsorbed during settling at falling temperatures (refining):
Mechanisms of purification:

Fining: Refining:

• Bubble buoyancy • resorption of small bubbles at falling

o∙(∙f 6
temperature (solubility of SO2 has
𝑣& = N∙w
; 𝑣~•g€ = (1 − 𝑋m )G ∙ 𝑣& negative T coefficient)
• coalescence (large bubbles „eat“ small
bubbles)
G∙•
𝑃D9 = 𝑃& + f
≫ 𝑃&

• Strategy: production of a swarm of

bubbles by chemical reaction

Na2SO4 ⇌ Na2O + SO2 + ½ O2

The amount of bubbles that have to be removed from a glass after melting is typically 105 bubbles
per kg (diameter approx. 50 to 400 µm).
Thus, during the refining step, gases dissolved in the glass are expelled. In the equilibrium for O2, CO2,
N2 and noble gases, the concentration of the gases absorbed in the melt is a function of the partial
 Page 131

pressure of the respective gas in the bubble (or the furnace atmosphere). A further distinction must
be made here between physically dissolved gases in the glass—interstitially in open areas of the
network and without chemical bonding to the network—and chemically dissolved gases. The latter are
absorbed in the melt by a chemical reaction and their solubility (as a function of composition and
temperature) is orders of magnitude higher than that of physically dissolved gases.
N2 is always physically dissolved, except in very strongly reduced melts (then as nitrides, N3–). O2, CO2,
water vapor and SO2 can be either chemically (predominant) or physically dissolved. Ar and He are
mostly physically dissolved. There is a local equilibrium between chemically and physically dissolved
CO2 in the form:
CO2 (phys. dissolved) + O2– (in glass) ⇌ CO32– (chem. dissolved)

The sulfur dioxide solubility depends on the degree of oxidation of the melt (as sulfate or sulfide):

%
SO2 + G O2 + O2– ⇌ SO42– (oxidizing melt conditions)

N
SO2 + O2– ⇌ S2– + G O2 (reducing melt conditions),

By analyzing the gas composition of a bubble in the glass, one can obtain valuable information about
its formation history and thus about possible melting problems. Bubble analysis is an important part
of melt process optimization.
If bubble release is excessive or if the viscosity is too high, the bubbles cannot escape from the glass
and foaming can occur. This foam formation depends on the amount of gas volume per m2 of the melt
surface (organic contaminations play a major role here) and on the stability of the foam bubbles (this
is a function of furnace atmosphere, temperature, glass composition, surface tension, bubble
diameter, etc...). A distinction is made between primary foam, caused by mixture reactions (carbonate
decomposition, carbonate-sulfate reactions, H2O release), and secondary foam caused by fining
reactions.

If the bubbles rise in relation to their size and reach the density of a random closed packing (RCP, 64
%), the rising stops (relative velocity between bubble and medium, vSLIP = 0) and a stationary foam
 Page 132

forms (see figure above). Exactly the same phenomenon occurs with a beer in a glass when a foam
crown forms on the beer.
In contrast to the foam crown on a beer, however, a foam carpet in a glass tank is very critical because
the heat input from the top furnace into the glass is almost completely prevented. The formation of
large amounts of lamellar foam can even make it necessary to shut down a tank.
The rate of ascent vasc of a bubble is a function of the viscosity of the melt and the radius of the bubble
(squared!):
ρ = density of the glass melt [kg·m−3]
𝑐 ∙ 𝜌 ∙ 𝑔 ∙ 𝑟G η = viscosity of the melt [Pa·s]
𝑣K]A =
𝜂 r = bubble radius [m]

g = acceleration of gravity [m·s−2]

!
c = factor (in most cases c = )
"

The density of the bubble is negligible here, but it would be correct to use ρ = ρmelt – ρgas. For the factor
G
c, c = ‚ applies in most cases, especially if the surface of the bubble can be considered as being stiff. If
the surface of the bubble would be completely mobile, it would rise less due to deformations within
%
the limits of the surrounding viscosity, thus reducing c to c = N.

The picture below shows the rate of ascent for bubbles of different sizes. It becomes obvious, that
bubbles between 100 µm and 200 µm in diameter would need between 1 to 5 days to ascent through
one meter of a glass melt, even at temperatures above 1450 °C. Hence, in general, bubble growth is
necessary for effective fining.

!
This growth is achieved by a suitable component (fining agent) decomposing (or evaporating) at high
temperatures (and thus at low viscosity, at least 100 dPa·s). This leads to bubble growth with
subsequent gas "stripping" from the melt and leads to a higher rate of ascent. This is shown
schematically in the figure below. The release of the fining gases occurs when pgas, melt > pbubble.
 Page 133

In stage 1, the refining agent forms bubbles at high temperatures (>1400 °C). In stage 2, bubble
agglomeration and bubble growth occur. Dissolved gases diffuse from the melt into the bubbles, the
bubbles rise (slowly). Further gases diffuse into the bubble in stage 3, the volume increases strongly
and finally the bubble is large enough to rise rapidly to the surface of the melt ("onset of fining"). Since
a fining reaction is generally associated with a change in the valency of the fining component (see
specific fining reactions), it is recognized that the fining temperature can be influenced by the redox
state of the melt. The control of this complex equilibrium is decisive for stable process control in a
glass-melting tank.
The above-mentioned "stripping" also changes the composition of the bubble. Below is an example of
the change in composition of a bubble with temperature during O2 fining. A large increase in bubble
size is typical for fining with As or Sb (see specific fining reactions). The gas content of a bubble allows
to draw conclusions about the history of its formation; as mentioned, this is used for process analysis.

A word on secondary fining: This "re-fining" or "re-absorption of chemically dissolved gases" happens
during the controlled cooling of the melt. This is because the fining reactions can then be reversed and
gases in bubbles are (re-)absorbed by the melt again. Ideally, the bubble shrinks and completely
disappears. This re-fining is particularly important for small bubbles < 100 µm, which are difficult to
get to rise. However, re-fining is only effective for gases that can be dissolved in cooling melts (CO2,
O2, SO2 + O2). Also, the glass melt should already be low in the amount of dissolved gases. For re-fining,
 Page 134

there is a rather narrow temperature range (e.g. SLS float glass: approx. 1350 – 1275 °C) due to the
rapidly increasing viscosity during cooling, which of course counteracts re-absorption.
When switching from air to oxygen as oxidant, one generally observes a slightly lower viscosity of the
glass melt, but also more foaming and lower fining temperatures (approx. 20 – 40°C lower). This is
partly because water dissolved in the melt supports the fining process (during primary fining, gas
bubbles grow in the melt due to SO2 evolution from sulfate decomposition and water vapor released
from the melt). The glass melt in an oxygen-fired furnace can contain up to 70 % more dissolved water
than the glass melt in an air-fired furnace.
For a highly oxidized SLS glass melt in air-fired furnaces, the onset of fining (standard sulfate fining) is
estimated to be at about 1440 –1450 °C, whereas in oxygen-fired furnaces it is between 1400 °C and
1420 °C. The H2O content is approx. 18 vol.-% in natural gas-air-fired furnace atmospheres and approx.
55 vol.-% in natural gas-oxygen-fired glass furnaces.

9.7. Gas analysis

It has already been mentioned several times that valuable information about the formation history of
a bubble can be drawn from the analysis of its gas composition. However, it is also possible to analyze
the integral amount of gas released during non-isothermal heating of the batch. This is often done via
an Evolved Gas Analysis (EGA). In this process, the evolved gases are fed via a carrier gas (N2) into a gas
chromatograph, FTIR and/or oxygen analyzer.
The gas emission of the batch is obtained as a function of the temperature and corresponding
conclusions can be drawn, for example, about the suitability of a fining agent or a certain gas-releasing
raw material component. An example is shown in the figure below.

In the diagram above, for example, one can see that carbonate decomposition only begins at relatively
high temperatures, which may indicate slower melting kinetics. The release of the fining agent (here
sulfate refining) takes place at just sufficiently high temperatures to ensure a low viscosity of the melt.

9.8. Specific fining reactions

In this chapter, some specific fining reactions are discussed. It should always be noted that a release
of gas by the fining agent takes place at a suitable viscosity of the melt (< 102 dPa·s). If the viscosity is
too high, even larger bubbles cannot develop a sufficient fining effect. The redox dependencies in the
melt must also be taken into account. The redox state of a glass describes the equilibrium between the
 Page 135

oxidized and reduced species (polyvalent ions) in the system. So, if a melt contains many oxidizing
(leaching) components (sulfates/nitrates) there will also be a lot of O2 dissolved in the melt in the form
of O2– (i.e. chemically). This in turn influences the redox state of the glass. The partial pressure of
oxygen p(O2) in a glass melt is therefore of enormous importance
𝑦 𝑦
𝑀 =' + 𝑂G ↔ 𝑀(='#) + 𝑂G$
4 2
as it has a corresponding effect on the color and IR absorption of the glass via the Fe(II)/Fe(III) ratio:

Oxidizing components are: Sulfates, nitrates, polyvalent ions in the highest oxidation state (Fe3+, Sb5+,
As5+, Sn4+, Ce4+, etc…)
Reducing components are: Organic components, carbon, reduced form of polyvalent ions (Fe2+, Sb3+,
As3+, Sn2+, Ce3+, S2–, etc…) or the pure metals (!).

9.8.1. Sulfate fining

Sulfate fining (often via Na2SO4) leads from the chemically to physically dissolved state. Oxygen (g) and
SO2 (g) can form sulfates with each other or with the oxygen in the melted glass. The dissolved sulfates
contain chemically dissolved oxygen and SO2. The equation for the equilibrium reaction is:
O2 (g) + SO2 (g) ⇌ O2 (phys. diss.) + SO2 (phys. diss.) ⇌ SO42– (chem. diss.)
Reading from left to right, the temperature decreases. This therefore describes the "re-fining". The
actual fining reaction is then rather from right to left (with rising temperature), with O2 and SO2 as
fining gases. These gases occur when the sum of the partial pressures of the physically dissolved gases
reaches approximately 1 – 1.3 bar and strong bubble formation and growth occurs ("fining onset",
approx. 1430 °C – 1480 °C).
A special feature of fining with sulfate is the interaction with the polyvalent iron, which together with
the sulfur is responsible for the brown coloring (amber chromophores). A charge-transfer reaction
 Page 136

takes place between Fe3+ and S2–, which leads to the formation of the actual chromophore S2–
(supersulfide ion):
3 Fe3+ + 3 e– ⇌ 3 Fe2+
2 S– ⇌ S2– + 3 e–
3 Fe3+ + 2 S– ⇌ 3 Fe2+ + S2–
N N
Or in oxide notation: G Fe2O3 + 2 Na2S → 3 FeO + G Na2O + NaS2

The relationship between redox state, sulfur content and color is shown in the curve below. During the
transition from flint to amber (oxidizing to reducing), a melt is generated that can dissolve less SO3.
Thus, if you want to change the redox state of your melt, you always have to go through the minimum
of SO3 solubility and at least some of the SO3 will leave the melt; this, in turn, leads to a lot of foam and
blister in the transition from flint to amber. This typical curve, known as the "Budd curve", highlights
that melts of glasses of different colors have very different sulfur contents. To conduct a color change
in a tank from white to brown, the amount of Na2SO4 in the batch needs to be drastically reduced
beforehand (and only adjusted again after the change has been made). Otherwise, the melt will foam
up.

The table below shows the relationship between the Fe(II)/Fe(III) ratio and the resulting glass color.

Fe2+/Fetot Glass colour

0.10 – 0.40 White / slightly yellow

0.20 – 0.60 Green

0.60 – 0.75 Olive green (“feuille morte”)

0.75 – 0.90 Amber (brown)

 Page 137

Depending on the furnace atmosphere (water vapor content) and Na2SO4 addition, very high
temperatures (well above 1430 °C), are needed for a strongly oxidized sulfate fining. In order to achieve
a sufficient glass quality, there are therefore two possibilities: Either reduce the redox (which,
however, can lead to color problems → see explanation before) or actually increase the fining
temperature. If you reduce the redox, you also have to consider other coloring mechanisms, for
example if you want to produce a bluish color based on Se–Co. The main key in such adjustments is to
match the desired color with a reduced redox and to manage the fining and foaming. One can surmise
that there are fewer problems with foaming in more oxidized conditions and thus higher temperatures.
However, even with highly oxidized melts and sulfate fining at high temperatures, foaming can occur
if the addition of fining agents (e.g. Na2SO4 or CaSO4) is (too) high. It is therefore important to control
the addition of fining agent.
Sulfate fining also naturally increases the SO2 content in the flue gases (environmental problem), but
can be partially removed by dust separation of flue gas, which produces sulfur-containing filter dust.
In the past, many companies have switched from antimony (plus nitrate) to sulfate refining for batch
cost and environmental reasons. Generally, sulfate fining takes place under reduced conditions in
contrast to antimony (+ nitrate) fining. Therefore, the concentration ratio of Fe(II)/Fe(III) in the glass
may increase and the glass becomes greener. Sulfate fining, as mentioned, then requires very high
temperatures under very oxidized conditions, which are higher than those of antimony nitrate fining.
Therefore, more reduced batch conditions (using carbon, slags or other reducing—but not coloring—
sulfate leaching agents) are usually preferred for sulfate leaching. However, this can lead to some
green/blue coloration (at higher iron oxide levels).
When switching to sulfate fining (with Co–Se as a decolorizing pair), more foaming may occur after
sulfate addition (see Budd curve). In general, however, new industrial mixture formulations are usually
found with good results.

9.8.2. Antimony / Arsenic Oxide fining

Arsenic fining (and analogously antimony fining) takes place under strongly oxidizing conditions at
1400 - 1500 °C. The fining agents are As2O3 or arsenates [As(V)]. In glass melts, As2O3 or arsenates in
combination with oxidants (e.g. NaNO3) are used as fining agents. The nitrates are then used to keep
the As in its valence state of As(V) or to oxidize As(III) into As(V) at an early stage.
5 As2O3 + 4 NaNO3 ⇌ 5 As2O5 + 2 Na2O + 2 N2
At high temperatures, the reduction of As(V) then forms O2 and As(III):
As2O5 ⇌ As2O3 + O2 (g) (O2 is the fining gas)
It is evident that only arsenic as As(V) is a fining agent. Due to health and environmental considerations,
attempts are made to avoid antimony / arsenic oxide fining. However, this is not always possible with
certain glasses or intended glass colors. The active oxidation state of antimony for fining is therefore
Sb5+, so nitrates are added to the batch (as oxidants) to form as much Sb5+ as possible out of Sb3+ in
order to obtain the corresponding "reverse reaction" at temperatures 1300 – 1400 °C: Sb2O5 ⇌ Sb2O3
+ O2 (g). Some applications directly use antimonates as antimony carriers (here antimony is already
present in the oxidation state +5).
 Page 138

9.8.3. Halogenide fining

A halide fining is suitable for high melting temperatures (e.g. borosilicate glasses), when Na2SO4 is not
yet effective enough, due to the still too high viscosity at fining onset. In this case, halide salts are often
used as fining agents (NaCl, NaBr, NaI, KCl, KBr, KI). For example, the NaCl vapor pressure of melts with
an excess of NaCl (greater than the solubility level → approx. 1.4 mol-% for SLS glass) is almost the
same as for pure NaCl. Above 1.4 mol-%, separated halide-rich melt phases are formed. Thus, the fining
reaction here is not a dissociation reaction, but a pure halide evaporation from the melt (T ≥ 1480 °C
and pNaCl > 1 bar).
The disadvantage is the formation of halogenic acids:
%
2 NaCl + SO2+ H2O + G O2 ⇌ Na2SO4 (particulates) + 2 HCl (g)

The graph below shows the halide fining as a function of the chloride concentration in the glass (Fining
onset: pNaCl = 1 bar [D. Koepsel, 2000]):

9.8.4. SnO2 fining

Tin oxide fining is used for alkali-free glasses (e.g. TFT display glass). These glasses have high melting
temperatures and high viscosities, so that the fining reaction can only be successful at high
temperatures. Nitrates are added to the melt to keep Sn in the valence state (IV). The fining
temperature is 1550 - 1600 °C for alkali-free glasses, which then reach a viscosity of approx. 102 dPas.
The fining gas is once more O2 released in the reaction mentioned below:
%
SnO2 ⇌ SnO + G O2 (gas) (O2 as fining gas)

9.9. Conditioning of the melt

Apart from the fining step an additional conditioning step is necessary to adjust the temperature over
the depth, length and width of the glass stream. This is necessary, because a uniform viscosity and thus
a homogeneous temperature distribution in the glass is required for the later shaping process.
Note: In addition to the temperature, the viscosity of the glass is, of course, strongly depending on the
chemistry. A homogeneous temperature distribution in the glass may therefore pretend a good
viscosity distribution although, for example, due to impurities or refractory corrosion, a different
 Page 139

chemistry—and thus viscosity—is found for the glass at the bottom of the tank compared to the
interior of the glass stream.
For conditioning, see also working tank and forehearth in the chapter on plant technology.

9.10. General considerations on the mode of operation of a glass tank

Without going into detail about tank designs here (see the chapter on plant technology), some general
considerations for fossil-heated tanks should be made: The tank can be regarded as a kind of chemical
reactor. A cold flow (the batch) is introduced into the tank and heated. This mass flow then leaves the
tank (the molten glass) at an increased temperature. In turn, a hot stream is introduced into the
furnace (the fuel gases), which heats up the “cold” stream, is itself cooled down and leaves the furnace
at a lower temperature (the waste gases). In addition to the heating by the fuel gases, part of the
molten glass circulates in the tank and thus supplies energy to the batch from below, thereby
supporting the melting process. The convection currents along the length of the tank are shown in the
figure below:

The flow profile has another important task: it divides the complete reaction space into relatively well-
separated subspaces. The aim is to form a twofold roll. The hot spot or swelling point is the area of
maximum temperature of the melt. The individual processes are assigned to the flow pattern. With
the exception of the furnaces used in the flat glass industry, area 5, the so-called working furnace or
refiner, is thermally and fluidically separated from the melting furnace by a small opening of only
approx. 0.5 m2 cross-section, the throat. However, these convection rolls also cause certain problems.
A large part of the glass does not leave the furnace directly, but enters this convection. Thus, there is
no piston-like flow ("plug flow") in a glass tank, but rather an (ideal) mixer. The differences are
showcased in the picture below:
 Page 140

input output, plug flow

time constant t ,
concentration

t = tank capacity/daily production

output,
ideale mixer

0 10 20 30 40
t in h

This leads to a residence time distribution of the glass in the tank according to the profile given in the
figure below (determined by tracer tests), in this case further broken down into three different feeder
channels:

The result is a dwell time distribution with glass fractions that circulate for several days (!). This should
be into consideration if, for example, a color change has to be carried out in a tank. The following is
also important: The glass must have a certain minimum residence time (tcrit in the picture above) in the
tank to ensure sand dissolution and to successfully perform the fining process. If this residence time is
not given for certain glass particles, quality problems will occur. This can happen if, for example, the
tank is constructed incorrectly or if the throughput is too high.
Another note on convection currents: By its very nature, the molten glass heated from above (from
the furnace) will be hotter than the molten glass at the bottom of the furnace. Unfortunately, this is
not advantageous for convection, as a stationary state would be established and the cold, dense glass
melt at the bottom would show no ambition to rise to the surface and thus set convection into motion.
This is counteracted by forced convection. This convection is induced at a suitable place, e.g. by
electrodes in the bottom of the tank, which heat up the glass at the bottom. Blowing in air ("bubbling")
or a mechanical forced upward flow through e.g. a step in the bottom of the tub are further
possibilities.
 Page 141

The gradient from the "hot spot" to the (cold) side walls of the tank is one of the main driving forces
for the formation of convection rolls. In addition, there is a strong downward flow induced by the
support of the cold batch. A comparison with the values of the radiant thermal conductivity for green
and white glass shows that, at least in the first case, convection as a heat transport mechanism cannot
be dispensed with. The absorption properties of the respective melts must therefore also be
considered when designing appropriate melting furnaces.

9.11. Alternative approaches to CO2 avoidance

In progress, see lectures

10. Energy consideration of the melting process

The energy requirement for a glass tank is based on two different aspects: The actual energy required
to melt a batch and on the other hand the heat losses. The picture below shows an energy breakdown
for a (moderately efficient) container glass furnace:

Roughly, it can be deduced from this breakdown that about 30% of the energy leaves the furnace with
the exhaust gas and 20% is structural losses in the furnace, top furnace and regenerator. 50% of the
energy goes into the glass as latent heat and reaction enthalpy and a small amount (about 2%) is
needed for water evaporation. The energy required to melt a batch is the sum of the latent heat of the
glass (44 %), the net reaction energy, i.e. the heat required for the melting reactions (endothermic
 Page 142

minus exothermic) (3 %) plus the latent heat of the batch gases (here considered under the item "flue
gas out").
Below variations of the quantities mentioned above are shown for different glass types, related to a
glass transition temperature of 1400 °C and an exhaust gas temperature before the regenerator of
1500 °C (here using 0 °C as the reference point!).

Sodium boro-
Soda lime Lead glass E-glass
silicate glass
glass (19% PbO)
(8% B2O3)
[kJ/kg] [kJ/kg] [kJ/kg]
[kJ/kg]
Tangible heat of melt (0-1400 °C) 1775 1603 1609 1588

Reaction heat 487 403 412 roughly 400

Tangible heat volatile reaction

293 166 140 577
products (0-1500 °C)

TOTAL 2555 2172 2161 2565

The picture below shows the mass flow through the reactor volume of an air-gas fired glass furnace.
The batch in this example contains 60 % (600 kg) cullet per 1000 kg of glass produced. In addition,
there are 480 kg of primary raw materials; 80 kg of the batch escape as batch gases (mostly CO2 from
the limestone CaCO3, the soda Na2CO3, etc.). The intrinsically converted heat or useful heat Hex is
composed of a proportion Hchem (fictitiously related to 25 °C) of the endothermic, chemical reactions
and the heat content of the processable melt extracted at Tex. Via the efficiency of the heat conversion
ηex of the plant, here assumed to be 42 %, an actual energy demand Hin is thus determined. In the
example, it is 1100 kWh per t of glass. Via the so-called lower calorific value 𝐻8ƒ (kWh per kg fuel), a
required fuel quantity mF can be calculated. In the example, natural gas is chosen with the
corresponding equation for combustion being (neglecting minor contributions by alkanes of higher
order):
CH4 + 2 O2 ⇌ 2 H2O + CO2
In the case of near-stoichiometric combustion with air, this also determines the air and flue gas
quantities (Air: approximately 23 wt.-% O2). The mass flows through the tank basin and the combustion
chamber are thus coupled via the useful heat Hex and the efficiency ηex. The calculation shows that
approx. 1.55 t of gases pass through the combustion chamber per t of glass and 300 kg of CO2 are
produced.
This should rather be a “l”
 Page 143

fuel specific data:

𝐻IJ = 15 kWh/kg
N
𝑉KLM = 14.7 𝑚O /𝑘𝑔
𝑉PQ = 3.1 𝑚O /𝑘𝑔
N
N
𝑉RPQ = 1.6 𝑚O /𝑘𝑔
N
𝑉SQP = 3.1 𝑚O /𝑘𝑔

The first step in balancing a tank is to obtain data that is relatively easily accessible:

l = Length [m]
w = Width [m]
h = depth [m]

A = Area of melt [m2] = l·w

V = Volume of melt bath [m3] = l·w·h
M = Capacity of melt bath [t] = ρ·V

p = Throughput [t·d–1]
F
r = Rate of throughput [t·m–2·d–1] = 6
m „ …·„
τg = nominal retention time [h] = ρ F = ρ f ⇒ 𝑟 = ‡(
 Page 144

The following quantities serve as a basis for description:

§ vk volume fraction (0.01 vol%) of gas species k, 0 ≤ vk ≤ 1;
§ yk mass fraction (0.01 mass-%) of species k, 0 ≤ yk ≤ 1;
§ xk mole fraction (0.01 mol-%) of species k; gases are treated as ideal; therefore vk ≈ xk;
.∙0&
§ VM molar volume of ideal gases; VM = ; at normal conditions (NTP, T0 = 298.15 K (25 °C) and
F&
p0 = 1 bar → VM = 24.788 ·10–3 m3·mol–1;

§ M molar mass in g·mol–1;

l
§ d0 density at NTP; for ideal gases d0 = m ;
T

§ VF volume related to 1 kg fuel in m3·kg–1; (m3 figures related to NTP);

§ V volume in m3·t–1 related to 1 t glass;
§ HF heat quantity related to 1 kg fuel;
§ H heat quantity related to 1 t glass;

§ 𝐻8ƒ lower calorific value (formerly: net calorific value HNCV) in kWh·kg–1 fuel;
§ 𝐻ˆƒ upper calorific value (formerly: gross calorific value HGCV) in kWh·kg–1 fuel;
§ mF specific fuel demand in kg fuel per t glass;
§ r throughput rate in t·m–2·h–1 or t·m–2·d–1;
§ A melting area in m2;
§ p throughput rate in t·h–1 or t·d–1; given by p = r·A;
§ q formal heat flux normalized to the melting surface A in kW·m–2; given by q = H·r;
§ cP molar heat capacity in Wh·kg–1·K–1; with 1 J·mol–1·K–1 = M·3.6·Wh·kg–1·K–1, M = numerical
value of the molar mass in g·mol–1.

The total balance is then composed of the theoretical heat demand of the glass (Hex) and the heat loss
Hloss. The energy Hin, which must be supplied, is therefore Hin = Hex + Hloss or the following applies

𝐻_rrIrI
𝐻Y_ = , with 𝜂r= as efficiency of the system.
𝜂r=

The determination of the individual variables is in the next chapters.

10.1. Determination of Hneeded

The energy required to melt a batch Hneeded is the sum of the latent heat of the glass ΔHT,liq, the net
enthalpy of reaction ΔHchem, i.e. the heat required for the melting reactions (endothermic minus
exothermic) and the latent heat of the (hot) batch gases ΔHT,gas:

𝐻9hhHhH = Δ𝐻A‰h[ + Δ𝐻0,8DP + Δ𝐻0,:K] = Δ𝐻A‰h[ + 𝑐F (𝑇h\ − 25 °C) + 𝑐𝑝 8𝑇gas − 25 °C9

 Page 145

The enthalpy of reaction Δ𝐻A‰h[ can be determined according to Hess' heat theorem from the
&
standard enthalpy of formation of the glass 𝐻:8K]] plus the standard enthalpy of formation of the
& &
mixture gases 𝐻:K] minus the standard enthalpy of formation of the batch 𝐻ŠKEA‰ :
& & & &
∆𝐻A‰h[ = 𝐻:8K]] + 𝐻:K] − 𝐻ŠKEA‰

The enthalpies of formation of the mixture gases and the reactants are obtained simply by summing
up the individual enthalpies of formation of the respective component as a function of their respective
amount of substance nk:
&
𝐻:K] = r 𝑛( 𝐻&(
b

&
𝐻ŠKEA‰ = r 𝑛b 𝐻&b
b

For the raw materials in the batch, this may be a great simplification in the case of mixed crystals and
mixed phases. For the mixture gases, the assumption of an ideal gas approximation is sufficient.
The determination of H0glass is based on the approach of constitutional phases according to R. Conradt,
which is preferred here (see earlier chapter). There are other approaches (for example according to
Kröger and Illner in Salmang "Die Glasfabrikation"), but these are much more limited and consider
many glass-relevant aspects only conditionally.
The standard enthalpy of formation of the glass H0glass is thus the sum of the standard enthalpy of
formation of the corresponding constitutional component H0k plus the vitrification enthalpy of the
respective component 𝐻‹CDE , taking into account the respective amount of each component nk:
&
𝐻:8K]] = r 𝑛X (𝐻X& + 𝐻XCDE )
X

The latent heat of the glass ΔH𝑇,glass at 1400 °C (often tabulated value) is given below as the sum of the
constitutional components with H1673,liq. For any given temperature, one can then simply derive the
energy content of a melt HT,liq, taking into account the heat capacities of the individual components.

𝐻%Œ•N,8DP = r 𝑛X ∙ 𝐻%Œ•N,8DP,X
X

𝑐F,8DP = r 𝑛X ∙ 𝑐F,8DP,X
X

𝐻0,8DP = 𝐻%Œ•N,8DP + 𝑐Ž,8DP ∙ (𝑇 − 1673)

Taking into account the energy that the glass already possesses at 25°C, the energy content of a glass
melt at any temperature is related to the glass at 25 °C:
∆𝐻0,8DP = 𝐻0,8DP − 𝐻&:8K]]

This is therefore the amount of energy extracted from a furnace by the glass throughput. Note: The
respective entropies are described analogously. A detailed description is therefore not given here. The
corresponding values can be looked up in tables, e.g. SGTE Handbook 2008 (excerpt below).
 Page 146

vit vit
K -H° S° H S -H1673,melt S1673,melt cP,melt

P2O5·3CaO 4117.1 236.0 135.1 51.5 3417.1 898.7 324.3

P2O5 1492.0 114.4 18.2 9.5 1138.5 586.6 181.6
Fe2O3 823.4 87.4 45.2 17.2 550.2 370.3 142.3
FeO·Fe2O3 1108.8 151.0 82.8 31.4 677.8 579.9 213.4
FeO·SiO2 1196.2 92.8 36.7 13.8 962.3 342.7 139.7
2FeO·SiO2 1471.1 145.2 55.2 20.5 1118.8 512.1 240.6
MnO·SiO2 1320.9 102.5 40.2 15.1 1085.3 345.2 151.5
2ZnO·SiO2 1643.1 131.4 82.4 31.4 1261.1 494.5 174.5
ZrO2·SiO2 2034.7 84.5 86.6 32.6 1686.2 381.2 149.4
CaO·TiO2 1660.6 93.7 67.4 25.5 1365.7 360.2 124.7
TiO2 903.7 185.4 40.2 19.7 741.0 335.6 87.9
BaO·Al2O3·2SiO2 4222.1 236.8 130.5 95.4 3454.3 1198.3 473.2
BaO·2SiO2 2553.1 154.0 81.6 26.8 2171.1 533.5 241.4
BaO·SiO2 1618.0 104.6 56.5 41.0 1349.8 361.1 146.4
Li2O·Al2O3·4SiO2 6036.7 308.8 184.1 12.1 5235.4 1173.2 498.7
Li2O·SiO2 1648.5 79.9 16.7 6.3 1416.7 339.7 167.4
K2O·Al2O3·6SiO2 7914.0 439.3 106.3 29.3 6924.9 1559.4 765.7
K2O·Al2O3·2SiO2 4217.1 266.1 80.4 22.1 3903.7 666.5 517.6
K2O·4SiO2 4315.8 265.7 26.4 21.3 3697.8 983.7 410.0
K2O·2SiO2 2508.7 190.6 12.6 23.9 2153.1 595.4 275.3
Na2O·Al2O3·6SiO2 7841.2 420.1 125.0 28.4 6870.1 1512.5 648.1
Na2O·Al2O3·2SiO2 4163.5 248.5 92.0 27.9 3614.1 856.9 423.8
B2O3 1273.5 54.0 18.2 11.3 1088.7 271.1 129.7
Na2O·B2O3·4SiO2 5710.9 270.0 42.7 21.1 4988.0 1090.2 637.6
Na2O·4B2O3 5902.8 276.1 58.3 40.1 4986.7 1275.5 704.2
Na2O·2B2O3 3284.9 189.5 48.8 26.6 2735.9 780.3 444.8
Na2O·B2O3 1958.1 147.1 43.6 19.5 1585.7 538.7 292.9
2MgO·2Al2O3·5SiO2 9113.2 407.1 135.8 41.4 7994.8 1606.2 1031.8
MgO·SiO2 1548.5 67.8 46.6 13.6 1318.0 296.2 146.4
2MgO·SiO2 2176.9 95.4 61.4 11.0 1876.1 402.9 205.0
CaO·MgO·2SiO2 3202.4 143.1 92.3 25.7 2733.4 621.7 355.6
2CaO·MgO·2SiO2 3876.9 209.2 106.7 32.0 3319.2 775.3 426.8
CaO·Al2O3·2SiO2 4223.7 202.5 103.0 37.7 3628.8 791.2 380.7
2CaO·Al2O3·SiO2 3989.4 198.3 129.9 49.4 3374.0 787.8 299.2
3Al2O3·2SiO2 6820.8 274.9 188.3 71.5 5816.2 1231.8 523.4
CaO·SiO2 1635.1 83.1 49.8 18.8 1382.0 329.7 146.4
2CaO·SiO2 2328.4 120.5 101.3 38.5 1868.2 509.2 174.5
Na2O·2SiO2 2473.6 164.4 29.3 13.2 2102.5 588.7 261.1
Na2O·SiO2 1563.1 113.8 37.7 9.8 1288.3 415.1 179.1
Na2O·3CaO·6SiO2 8363.8 461.9 77.3 20.5 7372.6 1555.6 786.6
Na2O·2CaO·3SiO2 4883.6 277.8 57.7 13.4 4240.9 990.4 470.3
2Na2O·CaO·3SiO2 4763.0 309.6 87.0 22.6 4029.6 1107.9 501.2
SiO2 908.3 43.5 6.9 4.0 809.6 157.3 86.2

In the table above, the thermodynamic data of the corresponding components are given in relation to
the crystalline reference state of the glass. This means that the H0 value refers to cristobalite as a low-
density polymorph and not to quartz. (Regarding the above table: In the systems that do not form a
single-component glass, Hvit = 0.6 Hm is applied).

Ex.: H0 (Quarz) = – 910,1 KJ/mol

H0 (Cristo.) = – 908,3 KJ/mol
Hvit (Glass) = – 901,4 KJ/mol, given by –908,3 KJ/mol + 6,9 KJ/mol (see table above)

Since, as mentioned above, H0 (the same applies to G and S) results from the weighted sums of the
molar fractions of the reference phases, the following applies:
 Page 147

&
𝐻:8K]] = r 𝑛X (𝐻X& + 𝐻XCDE )
X
In this system of reference phases in the glass, the mixing quantities are negligibly small. There is
calorimetric evidence for this. Of course, this does not apply in the mixture of pure oxides!

The latent heat of the mixture gases ΔH(𝑇,gas) can then be determined just as easily according to (with
cp,gas for the individual substance quantities i):
𝑐F,:K] = r 𝑛Y 𝑐F,:K],Y
Y
&
∆𝐻0,:K] = 𝐻%Œ•N,:K] + 𝑐F,:K] (𝑇 − 1673) − 𝐻:K]

The theoretical heat demand Hneeded, as seen, is composed of the chemical heat demand ΔH0chem of the
fictitious conversion of batch to glass and batch gases at 298 K. As is known, this assumption is
permissible, since Hex is a state function. The corresponding latent heats are then considered
independently. If a fraction of cullet is added, the overall equation changes. Taking into account the
mass fraction yC of the cullet, the following is valid (only for the mixture shown below):
𝐻9hhHhH = (1 − 𝑦A ) ∙ ∆𝐻A‰h[ + ∆𝐻0,8DP + (1 − 𝑦A ) ∙ ∆𝐻0,:K]

This is because there is no heat of transformation ΔHchem and no formation of batch gases for the
fraction of cullet.
If one is not in a position to look up or determine Hvit (or Svit), the approximation for "normal" cooling
rates between 10 K·s–1 < q < 100 K·s–1 is:
∆a 0-1 ∆R 0-1
∆a /
≈ 1/3 und ∆R /
≈ 1/2

10.2. Determination of Hin

In the previous chapter we got to know a very exact method to determine Hneeded. This provides a basis
for a reliable balancing of a glass tank. In the next step, Hin is determined, i.e. the energy required to
actually melt a certain amount of glass, taking into account the total efficiency ηex of the system.
Reminder:
R4UUVUV
𝐻D9 = wUW
, with ηex as the efficiency of the system

So, the question now is how much energy must be supplied by the combustion inside the furnace. For
the following consideration, let us assume an overall efficiency ηex of 50 %, so the total wall and flue
gas losses Hw + Hstack are as large as Hneeded. For the necessary amount of fuel mF that must be used, the
following then applies:
R-4 [‹: •ˆh8] ’9h“:” J• E‰h :8K]] [‹•‰/E]
mF =
R,X [E :8K]]] ‰hKED9: CK8ˆh [‹•‰/‹:]

with 𝐻8ƒ as the lower calorific value of the fuel, i.e. without taking into account the heat of
condensation when the combustion gases are cooled to 25 °C. mF is given as the amount of fuel
required in [kg] per mass of molten glass in [t]. The amount of fuel supplied is very well known (every
industry has its gas flow meters) and the lower heating value of the fuel can be calculated from its
composition. In general, no industry relies on the stated calorific values of the suppliers any more, as
significantly large ranges are always stated by the suppliers and the actual calorific value can fluctuate
 Page 148

within these ranges at short notice. Such fluctuations can lead, for example, to CO or NOx emission
limits being exceeded, to a change in flame length and shape, an increase in wall temperature and an
increase in velocity above the dusty mixture and have a corresponding influence on air demand and
flue gas temperature.
Therefore, the actual gas composition is measured directly by means of gas chromatographs integrated
in the gas flow. 𝐻8ƒ can then be calculated with corresponding accuracy.
For the overall analysis, it is important to know the energy input, the required air volume, the
maximum temperature level (availability of energy) and the radiation properties of the flame. These
values are provided by the combustion calculation.
In the first step, the reaction equation of the combustion is set up. A very general form for the
combustion of hydrocarbons is:
𝑦 𝑧 𝑦
𝐶= 𝐻# 𝑂– + !𝑥 + − " ∙ 𝑂G = 𝑥 ∙ 𝐶𝑂G + ∙ 𝐻G 𝑂
4 2 2

The more specific reaction equation for alkanes ( ) is:

2𝑛 + 2 2𝑛 + 2
𝐶_ 𝐻G_'G + 0𝑛 + 1 𝑂G → 𝑛 𝐶𝑂G + 𝐻G 𝑂
4 2

The more specific reaction equation for alcohols ( ) is:

2𝑛 + 2 1 2𝑛 + 2
𝐶_ 𝐻G_'G 𝑂 + 0𝑛 + − 1 𝑂G → 𝑛 𝐶𝑂G + 𝐻G 𝑂
4 2 2
The heat radiation from the flame then results from the oscillation and rotation of the molecules with
dipoles, e.g. H2O, CO2. N2 therefore behaves athermal and does not contribute to heat radiation. From
the reaction equations, one obtains the quantity ratios between fuel, oxidant and exhaust gas.
The lower heating value is calculated for a gas mixture according to (for values see lecture slides):

𝐻8ƒ = r 𝑥X ∙ ℎX ∙ 𝑑X

xk= Proportion mol-% of component k

hk= part. Lower calorific value of component k
dk= Density of the component k
The stoichiometric air demand and the exhaust gas volumes (dry/wet) per kg fuel for mixed fuels then
follow with the partial fuel proportions (xk and dk) and the partial volume proportions air/exhaust gas
per fuel k (VFK):

𝑉 ƒ (air) = r 𝑥X ∙ 𝑑X ∙ 𝑉Xƒ (air),

𝑉 ƒ (off) = r 𝑥X ∙ 𝑑X ∙ 𝑉Xƒ (off),

𝑉 ƒ (dry) = r 𝑥X ∙ 𝑑X ∙ 𝑉Xƒ (dry).

 Page 149

Then, for the exact calculation of the heat capacities, the mixed gases must be added to the waste
gases. For these, the calculation is based on volume and the quantities of raw materials in kg per t of
glass:
Vb = 112·(1-yC) or more precise Vb = 0.269·dolomite + 0.248·lime + 0.234·Soda

If non-stoichiometric combustion is then added, the air number must be taken into account. This
corrects the respective volumes of air and exhaust gas:
𝑉
𝑉 ƒ (dry) + 𝑚Š 𝑝—6
ƒ
𝑉(air) = 𝑚ƒ ∙ 𝜆 ∙ 𝑉 ƒ (air), with the air faktor λ = ∙
𝑉 ƒ (air) 0.21 − 𝑝—6

𝑉(off) = 𝑚ƒ ∙ !𝑉 ƒ (off) + (λ − 1) 𝑉 & (air)" + 𝑉ŠKEA‰ :K]h]

So, in summary: mF can be measured exactly, 𝐻8ƒ can be calculated exactly and thus Hin can be
determined exactly. And by this, the total quantity of all heat losses, Hw + Hstack, is also known:
Hw + Hstack = Hin – Hex
The ternary diagram of C, O and H may serve as a simple estimate of the volume ratios of the
combustion and exhaust gases:

This allows the volume ratio of CO2 to H2O in an exhaust gas to be determined graphically by
intersecting the fuel-oxygen connection line with the CO2–H2O exhaust gas line as a function of the air
number (assuming ideal gases, vk = xk, i.e. volume fractions equal to the molar fractions).

10.2.1. Heat capacities of air and exhaust gas

The heat capacities of air and exhaust gas can be calculated according to:
CP in Wh·m–3NTP·K–1 = S Xkvk + (S Ykvk) T + (S Zkvk)/T2
 Page 150

Using the factors mentioned below to calculate the heat capacities of air and exhaust gas according to
the above equation, volume fractions of the gas species vk (m3 per m3 air or exhaust gas) and T in units
of 1000 K (e.g. at 1200 °C, T = 0.001·(1200 + 273) = 1.473). The factors with the index k = air, gas-off,
oil-EL-off or oil-S-off serve to quickly arrive at valid estimates for the heat capacity of the air, cP(air), as
well as the exhaust gases cP(off) from the combustion of natural gas or oil (type EL “extra light” or S
“heavy”), even without a combustion calculation.

k Xk Yk Zk k Xk Yk Zk

O2 0.3357 0.0468 -0.00187 air 0.3398 0.0323 -0.00249

N2 0.3409 0.0285 -0.00266 gas-off 0.3546 0.0528 -0.00274

CO2 0.4946 0.1013 -0.00957 oil-EL-off 0.3606 0.0499 -0.00318

H2O 0.3362 0.1199 0.00037 oil-S-off 0.3609 0.0497 -0.00320

10.3. Determination of the exchanged heat Hexch

The efficiency of the heat exchanger ηre must of course also be taken into account. This is reflected in
the exchanged heat Hexch = Hoff – Hstack. It is the difference between the energy content of the exhaust
gas before entering the heat exchanger Hoff and the energy content of the exhaust gas after leaving
the heat exchanger Hstack. The preheated air reduces the energy demand and thus affects mF as reduces
the fuel demand related to the actual energy consumption:

𝐻™ + 𝐻h\
𝑚˜ =
𝐻8ƒ − (1 − ηfr ) ∙ (𝑉 ƒ (off) ∙ 𝑐Ž,J•• + (λ − 1) ∙ 𝑉 ƒ (air) ∙ 𝑐Ž,KD“ ) ∙ ∆𝑇J••

§ ΔT = T – 298K
§ Hex: theoretical heat demand
§ Hw: wall loss
§ λ: air factor
Where ηre is approximate without taking into account excess air, false air and ballast volumes carried
by mixture gases:
𝑉 ƒ (air) 𝑐F,KD“ ∆𝑇“h
η“h = D ƒ G∙D G∙
𝑉 (off) 𝑐F,J•• ∆𝑇J••

In the ideal case that Toff = Tre, i.e. the flue gas temperature is equal to the temperature of the
preheated combustion air, ηre is still determined by the smaller air volume compared to the flue gas
(approx. factor 0.85) and by the smaller heat capacity of the air compared to the flue gas (approx.
 Page 151

factor 0.9). This results in a maximum efficiency of the heat exchanger of approx. 76.5 %. In reality, the
efficiency of a regenerator is 55 – 68 %.
By preheating the combustion air, the adiabatic flame temperature, i.e. the temperature that would
be reached if the fuel burned isobaric but thermally isolated, can be increased considerably:

𝐻8ƒ + 𝑉 ƒ (air) ∙ 𝑐F,KD“ ∙ ∆𝑇“h

∆𝑇KH = ƒ
𝑉 (off) ∙ 𝑐F,J•• + (λ − 1) ∙ 𝑉 ƒ (air) ∙ 𝑐F,KD“
In the case of ideal preheating, adiabatic flame temperature increases to 2500 – 3000 °C compared to
1860 °C for non-preheated combustion air:

10.4. Determination of the heat loss Hw + Hstack

The heat losses consist of the structural losses (Hw) and the flue gas losses (Hstack). The Sankey diagram
below illustrates all the dependencies within this system of melter and regenerator (or general heat
exchanger). If you have understood these at least qualitatively, you are already a long way ahead. The
same description also applies to the heat flux densities q, according to H·r (r = spec. pull).
 Page 152

Hstack can still be determined relatively well. Determine the quantity and composition of the flue gas
and calculate the energy content of the flue gas according to Hstack = cP(off)·V(off)·(Tstack – 25 °C), with
Hexch = Hoff – Hstack = Hre + Hwx. This is the amount of energy that escapes from the stack.
It is more difficult with Hw (and this includes Hwx), not because the theory is more complicated, but
because it is much more difficult to obtain reliable production data. In simple terms, heat transport
through a solid wall at high temperatures is by conduction and radiation. Convection only plays a role
at the outer boundaries. At the inner boundaries of a multi-layered wall, the transfer occurs by
conduction and radiation, and inside a compact layer by conduction alone. In principle, the layers have
different thicknesses L and thermal conductivities λ° and the heat conduction (the heat flux density q)
across one layer occurs according to:
λ&
𝑞= ∙ ∆𝑇
𝐿
Taking into account the respective heat transfer coefficient α, the effective total heat flux density q
perpendicular through a wall composed of radiation and conduction of the individual layers and the
convective heat transfer at the two boundary surfaces (hot and cold side of the lining, h and c) results
according to
𝑇‰ − 𝑇A
𝑞=
1 1 𝐿% 𝐿G 𝐿_
𝑎‰ + 𝑎A + λ% + λG + ∙∙∙ + λ_
with Th / Tc as the temperature of the hot and cold sides. And where here λ represents the sum of the
effective thermal conductivities from radiation and conduction. Relevant for a design (material
selection) are the temperatures at the respective interfaces. For the temperature Ti of the i-th layer in
a system of n layers, the following applies (without derivation):
1 𝐿% 𝐿G 𝐿
+ + + ⋯ + Y$%
𝑎‰ λ% λG λY$%
𝑇Y = 𝑇‰ − (𝑇‰ − 𝑇< ) ∙
1 1 𝐿% 𝐿G 𝐿_
𝑎‰ + 𝑎A + λ% + λG + ⋯ + λ_
Thus, a simple and manageable description of Hw would be possible. However, all of the above idealizes
the following points: There is rarely an ideal thermal contact. Likewise, no losses due to "gaps" and
false air are taken into account in the above and a steady state or at least constant convection is
assumed. Finally, the thermal conductivities λ and thicknesses L are well known, the heat transfer
coefficients α at the hot (and cold) side of the lining are difficult to determine (αh >> αc). Since α =
ε·Cs·T3 applies to α, a misdetermination of Th is clearly more critical than a misdetermination of Tc. And
with that, have fun determining Th in a hot tub while it is running. However, it must be stated that the
total (limiting) influence of αh on the temperature distribution is less than that of αk.
Metrologically, it is difficult to reliably determine the important parameters (e.g. via pyrometer
measurements). The deviations from reality become even more serious when using material
characteristics provided by the refractory supplier (α and λ). These are generally far off from the actual
values.
Further deviations from the theoretical values can arise from joints between the individual refractory
blocks, which are not taken into account. For example, a furnace has about 10% expansion joints on
the sides and in the top furnace relative to the total area. These should either be avoided or insulated.
Non-insulated joints radiate with approx. T = 700 – 750 °C and have a heat flow of 50 kW/m2. Insulated
 Page 153

joints radiate at approx. T = 225 °C and have a heat flux of 7 kW/m2. For every 100 kW of structural
heat loss, energy consumption increases by approx. 170 kW. This means an additional energy
consumption of 40–50 MJ/tglass for a 350 t/d end-port tub.
In the figure below, the heat losses in the ceiling are exemplarily shown as a function of the thermal
conductivity λ of an insulation layer. If, for example, the thermal conductivity of the insulation layer
changes (e.g. locally due to degradation) from 0.5 W·m–1·K–1 to 1,1 W·m–1·K–1, this only leads to what
at first sight appears to be a small increase from approx. 85 °C to 110 °C when the temperature on the
outside of the wall is checked. However, the heat flow increases significantly from about 1.5 kW·m–2
to almost 2.2 kW·m–2. The pure outside temperature therefore only provides limited information about
the actual heat losses.

The influence of false air on the furnace efficiency is also not to be neglected. There will always be
different local pressures in the furnace. Too high a furnace pressure is generally to be avoided, because
corrosive waste gases are forced into the joints of the refractory material and can thus cause faster
degradation. Negative pressure, in turn, results in an increase in cold-air leakage in the furnace and
thus in energy consumption. 500 Nm3/h of cold-air leakage, which is quite a low value, increases energy
consumption by approx. 0.05 – 0.06 GJ/ton ≅ 1.5 %, see graph below. As you can see, insulation and
furnace pressure are very important.

Usually, the pressure is set at one point in the oven (between 0.03 and 0.07 mbar at the doghouse, i.e.
a slight overpressure) and this point is taken as a reference. The different pressures are then set in the
 Page 154

rest of the oven. The oven pressure is checked, for example, by throwing paper into the doghouse and
observing whether the fire is drawn into the tank or blown out.
 Page 155

11. Plant engineering

There are melting tanks (note: tank and furnace are often used synonymously) for continuous and
discontinuous melting. Their capacities range from 100 kg/d up to 1 000 000 kg/d with working
temperatures of 1450 °C – 1700 °C. Melting furnaces are always built specifically for the type of glass
(SLS, fibre [E, S, C, M,...] optical glasses, boro-silicates, alkali-free, alkali-borates,...) and their heating is
done with gas, oil, electric, petroleum coke, bio-gas or hydrogen. Air or pure oxygen can be used as
oxidising agents. The type and structure of the furnace is further determined by the raw materials,
cullet, batch and product requirements (quality, price, homogeneity, material properties, ...). In the
following, we will only consider the continuously operated melting furnaces.
A furnace consists of refractory material. It is continuously loaded with premixed batch (generally a
mix of raw materials and cullet). The batch is loaded via the dog house (or via two dog houses in the
case of large furnaces), which can be designed differently.

Heat is transferred from the combustion chamber (upper furnace) by means of fossil combustion with
pure oxygen or preheated air (regenerator / recuperator) or by electrodes. It is characteristic that all
basic process steps take place at the same time in different zones, whereby the individual zones can
be more or less clearly separated from each other. This results in a multitude of flow lines from batch
feed to exit from the furnace and a wide dwell time distribution with its respective quality
dependencies. For economic reasons, the aim is of course to achieve a long service life for the tank
(tank travel / furnace travel). In continuous operation, this normally is 5 – 15 years for typical mass
glass furnaces. Continuously operated furnaces are used for container glass, hollow glass, flat glass
(float & rolled), "tableware" production (e.g. drinking glasses), fiber & glass wool and for many special
glasses (tubes, display glass, glass ceramics, lamps,...). They are generally not used for hand-formed
glasses, quartz glass (Tm !) and optical glass fibers.
 Page 156

Exhaust gas is always produced in fossil-fuel heated tanks. Each tank therefore has an exhaust gas
cleaning system (shown below as an example), which is intended to ensure that the statutory emission
limits are complied with.

For example, so-called "scrubbing" is carried out to precipitate SO2. For this purpose, Ca(OH)2 or CaCO3
(wet scrubbing) is injected into the flue gas stream, which after
Ca(OH)2 + SO2 ⇌ CaSO3 + H2O
CaCO3 + SO2 ⇌ CaSO3 + CO2
reacts to form calcium sulfite and is reacted with oxygen and water to form gypsum:
CaSO3 + 2H2O + ½O2 ⇌ CaSO4 · 2H2O
If batch preheating is used (see later chapter), in which the batch is preheated in direct contact by the
hot exhaust gas, the acidic components of the exhaust gas react with the basic components of the
batch. The exhaust gas is then kind of “purified”.
Refractory linings in tanks are not discussed in detail here. This topic is too large to be dealt with in
sufficient depth here. Reference is made to other lectures or the secondary literature (e.g. TRIER). Only
this much about it:
A glass tank consists of a multi-layer structure in the basin area, i.e. where the glass is melted. Logically,
the refractory quality "downstream" should always be better, as the glass has then reached a higher
quality by being free of bubbles and no more bubbles should occur due to refractory corrosion.
For direct glass contact corrosion-resistant refractory material (often fused cast aluminum zirconium
silicate bricks (AZS) or chromite bricks with no porosity, high density, high chemical resistance to the
respective glass chemistry) is used. On the outside, the insulation layer is made of refractory material
with low thermal conductivity coefficients and high porosity.
 Page 157

The vault is built up of silica stones. These consist of mostly not completely avoidable quartz < 3 %,
tridymite (approx. 40 %), cristobalite (approx. 50 %), silica glass < 5 % (+ dissolved impurities) and often
wollastonite < 5 % (as CaCO3 is generally used as a sintering aid). Here, the porosity and the residual
quartz content are decisive because of many phase transformations and corresponding expansion
behavior. Al2O3 must be avoided at all costs, as it significantly lowers the application temperature of
the stones due to a eutectic even at low contents (see below).

The roof must be heated up and cooled down very slowly, the application temperature of silica goes
up to 1650 °C.

11.1. Air-gas firing / Oxy-gas firing

Before going into detail about the individual tank types, two fundamental concepts are discussed.
Air or oxygen can be used as the oxidant. The differences are significant. With air-gas firing, the usually
preheated air (regenerative or recuperative, see later, the regenerative concept is the most
widespread (1856 (SIEMENS)) is mixed with the fuel gas and burnt. There is a high thermal ballast (79
vol-% N2 content in the air, this is athermal, i.e. non-radiative) and also high NOX emissions. The
combustion equation for the simple reaction with methane is as follows:
CH4 + 2O2 + 2·(80/20)·N2 ⇌ CO2 + 2H2O + 8 N2
 Page 158

For every 1 m3 of fuel gas, there are 10 m3 of air and approx. 11 m3 of waste gas. The combustion air
is generally preheated as mentioned. This can be done with a regenerator (regenerative, indirect heat
exchanger) or a recuperator (recuperative, direct heat exchanger). The concepts will be explained in
detail later.
With oxy-gas firing, pure O2 is mixed with the fuel gas and burnt. This results in lower NOX emissions,
theoretically slightly higher flame temperatures and more aggressive flue gas, as no N2 dilutes the flue
gas. The combustion equation for the simple reaction with methane is as follows:
CH4 + 2O2 + = CO2 + 2H2O
For every 1 m3 of fuel gas, there are 2 m3 of air and 3 m3 of waste gas. The quantities of waste gas
produced are therefore significantly smaller. The necessary oxygen is supplied by on-site O2
production, pipelines or trucks (10 –100 t/d). Of course, this has an impact on the total energy costs.
The costs/energy for oxygen production are currently around 20 %, which is largely responsible for
why oxy-gas concepts have not yet become widely established.
A rough energy comparison of different types of tanks shows very different efficiencies (efficiency in
%):
§ Recuperative furnaces: 20 – 35 %.
§ Regenerative furnaces: 35 – 50 %.
§ Oxy-fuel furnaces (without O2-generation): 40 – 55 %.
§ Regenerative + batch preheater: 50 – 65 %
§ Oxy-fuel without O2-generation + preheater: 50 – 65 %
§ LoNox® melter: 48 – 53 % (this is a special recuperative melter type with cullet preheating to
minimize NOX emissions)
§ All-electric melting: up to 90 %
At first glance, it is astonishing that oxy-fuel tanks are no more inefficient than regeneratively heated
tanks, even when they are operated without heat recovery, i.e. when the exhaust gas is discharged
from the tank at a high temperature. This is simply due to the more compact design of these tanks, but
above all due to the significantly lower exhaust gas volume due to the absence of nitrogen:

[MJ/ton of Glass] at [MJ/ton of Glass] at

Air-Fuel Furnace Oxy-fuel Furnace
50% cullet 50% cullet

Total 4.300 Total 3.436

Glass inherent Glass inherent
2.107 2.100
(Water, Hchem, Hex) (Water, Hchem, Hex)
Structural losses,
Structural losses,
817 upper and lower 636
upper and lower tank
tank
Structural losses, Structural losses,
129 0
Regenerator Regenerator
Flue gas losses 1.247 Flue gas losses 700
 Page 159

Unfortunately, the good performance of the oxy-fuel tanks is clouded by the costs and energy that
have to be expended to produce the oxygen via VSA or VPSA (vapor [pressure] swing adsorption)
systems. These plants are based on the process of gas separation by adsorption. Here, air flows through
a molecular sieve in which nitrogen and carbon dioxide are bound due to their physical properties.
Oxygen and argon, on the other hand, flow through the molecular sieve. The costs and energy for this
add up to about 20 %, which almost wipes out the advantage. Oxygen separation via ceramic
membranes has the potential to save 40 - 60 % energy compared to conventional VSA/VPSA air
separation systems. If such a technology becomes commercially available on a large scale, it would
mean a major breakthrough for oxy-fuel firing, which would be preferable anyway from an
environmental point of view due to the greatly reduced NOx values.

11.2. Regenerative and recuperative air preheating

If the furnace is operated with air, regenerative or recuperative air preheating can be considered. With
regenerative air preheating, two alternating burner systems must always be installed. While one
burner system is operating, the hot exhaust gas is passed through refractory grate in the regenerator
of the other system and heats it up. After a certain time, this refractory grate is "charged" (heated up).
Then the burner system is swapped and the system with the charged regenerator is used. Cold air
flows into the regenerator, warms up by passing through the grate, is mixed with the fuel gas and
burnt. This “changing process” is carried out every 15 to 30 min, depending on the system or the
company's philosophy (a cross-heated furnace is shown schematically on the left, the firing pictures
on the right show an end-port furnace):

The refractory material in a regenerator has to meet three requirements: mechanical stability (a
regenerator can be more than 20 m high, which means that the lower grids are subjected to a heavy
mechanical load), chemical stability due to the corrosive attacks of the exhaust gas (particles and
condensates) and thermal load due to the thermal cycling.
 Page 160

The refractory material used in the regenerator is crucial for efficiency. Different materials and
geometries [e.g. cross bricks (cruciform) or pot bricks (chimney blocks)] are used, depending on the
temperatures to which the zone is exposed and the condensates that can form from the flue gases
(keyword: refractory corrosion). In general, the refractory quality decreases from top to bottom, as
the loading temperature does in the same way.

The decisive factor is how well and quickly the refractory material can absorb the heat of the exhaust
gas and release it back into the air. This is dependent on its material properties (primarily cp, λ, α). Of
equal importance is a uniform flow of the flue gas and the air through the regenerator in order to
ensure a uniform heating of the grids. If, for example, the exhaust gas coming out of the tank enters
the regenerator head (chamber head) too quickly, it bounces against the back wall and "falls" down
there without using the entire volume of the latticework. This reduces the efficiency and locally
increases the load on the latticework. Overall, however, a regenerator is characterized by a high
efficiency. However, a flame / flue gas exchange is necessary and high investment costs are incurred
 Page 161

due to the large amount of refractory material. For a consideration of the efficiency of a regenerator,
see the chapter "Determination of the exchanged heat Hexch". Preheating temperatures of 1200 °C can
be achieved at flue gas temperatures of 1400 °C.
A recuperator is a direct heat exchanger using the counterflow principle. It is more stable in operation
because it is operated without flame change, requires lower investment costs, but also only allows a
lower preheating temperature compared to the regenerator. At exhaust gas temperatures of 1400 °C,
an air preheating of approx. 700 °C is achieved.

Compared to the regenerator, however, the required base area for a recuperator is also significantly
smaller.

11.3. U-flame furnace, regenerative

The following terms are synonymous: U-flame furnace, end-port furnace, horseshoe-flame furnace.
These furnaces are the most energy-efficient furnaces, with tonnages up to approx. 450 t/d (normally
approx. 300 t/d) and a melting capacity of 2– 3.5 t/m2. They are often used for hollow glass. Their
energy efficiency is 3 – 4 GJ/tglass at a melting and refining temperature < 1600 °C and a glass
temperature in the working tank of 1300 – 1350 °C. The lifetime of such a furnace can be up to 15
years.
There are 2 – 4 burners per burner port; however, not all of them are always fired. Air preheating is
done by regenerators. Due to the design and firing from the front, the flame expands along the length
of the furnace and "turns" at / in front of the baffle wall ("end of the furnace"). This results in the name
U-flame furnace.
 Page 162

The combustion gases thus have a long residence time in the top furnace and there is intensive heat
transfer to the batch and the glass. These tanks have less structural losses compared to cross-heated
tanks, but also less melting capacity than those.
As can be seen below, the areas of the melting tank, refining tank and working tank are quite clearly
separated from each other by the throat. Between the melting area and the fining area, a wall (fining
ridge) is usually installed. This forces the glass to flow upwards and then over an area with a low glass
melt depth, so that bubbles only have to overcome a small height to escape from the melt. The glass
is then passed through the passage that delimits the working end. This is where the secondary fining
takes place and later, in the forehearths, the conditioning and temperature homogenization is
achieved. This is followed by the feeder, which transfers the molten glass to the respective shaping
section.
 Page 163

11.4. Cross-heated furnace, regenerative

The terms cross-fired furnace and x-fired furnace are synonymous. These furnaces allow a more
precise temperature setting over the length of the furnace compared to U-flame furnaces, as here the
flame length only has to overcome the width of the furnace and not the entire length. This means that
these furnaces can also be designed for larger tonnages (up to 1 000 t/d at a melting rate of 2 – 3 t/m2).
However, they are not as energy efficient, their energy efficiency is 5 – 6 GJ/tglass. This type of furnace
is often used for float glass. In this case, the furnaces have an open design between the melting and
working area to avoid turbulence and streaks and to ensure a layered glass quality. Melting and refining
temperatures are up to 1650 °C and a tank can last up to 15 years.

Float tanks are the largest tanks in the glass industry. The total length of the tank (without float bath)
can be 80 m or more. The total length of the production plant with all downstream processes then
quickly exceeds 500 m. If a cross-heated furnace is used for other types of glass, the passage is basically
the same as for a U-flame furnace.
There is another special feature with float glass due to the chemistry of the glass. In principle, float
glass and container glass do not differ very much in their chemistry. However, due to the application
conditions, float glass allows a higher proportion of Fe(II) in the glass. From a purely chemical point of
view, float container glass is therefore the higher-quality glass. You can see this if you look at a float
glass pane in the direction of the edges. You can see a clear greenish tint, caused by the Fe(II). This
coloration is not present in flint container glass. However, the higher Fe(II) content is desirable because
it improves the absorption of thermal radiation from the flame (see chapter "Optical properties of
glass").

11.5. Special glass furnaces

Special glass furnaces can be designed very differently and their construction is determined by the
glass chemistry. Thus, there are very different requirements, e.g. for aggressive alkali-borate melts,
corrosion in the regenerator, etc... Generally, they have a tonnage of less than 80 t/d and a rather poor
energy efficiency of up to 10 GJ/tglass and more. Such furnaces are often subject to the highest demands
on glass quality, therefore high refractory qualities are necessary. They are often oxy-fueled and cross-
heated to ensure an exact temperature profile for melting and refining temperatures even above
1650 °C. The lifetime of such tanks is correspondingly short at 2 to 8 years. They can also be
 Page 164

recuperatively or regeneratively heated. An example of an oxy-fuel heated fiberglass furnace is shown

below. The exhaust gases are simply drawn off at the beginning of the tank ("flue gas channel").

In some cases, especially in fiber-glass production, so-called "top burners" are also used, which
generate a very high heat flux onto the melt (800 kW·m–2) and support the melting process (see below).

11.6. Electrically heated tanks

It has already been mentioned that electrodes are often used to support fossil firing in a tank. However,
there are also fully electrically operated tanks. The principle is the same: A strong current is passed
through the molten glass by means of electrodes. The electrical resistance of the melt (Rmelt)
determines the heat input according to P = I2 – Rmelt .
 Page 165

These tanks have high efficiencies of approx. 90 %. The electrodes are made of Mo or, for strongly
oxidised As/Sb refined glass made of SnO2. All-electric furnaces have a high specific melting power and
the potential for CO2 reduction, or for a "zero emission" concept. However, problems often exist in
such furnaces due to hot-spots, poor degassing, high electricity costs, difficulties in adjusting the redox
for brown glass and electrode configuration for furnaces larger than 150 t/d.
These furnaces are usually loaded with batch from the top, hence the name "cold-top furnace".

11.7. General information on the design of furnaces

The design of glass furnaces is a complex matter that can only be discussed here on short and in
extremely general terms. In general, nowadays a tank design is always preceded by several simulations,
e.g. on temperature distribution, flow distribution, glass quality, residence time distribution, energy
input, regenerator efficiency etc. Today's simulation models have reached a very high level and are
constantly being refined as computer performance increases.
Various general aspects have to be taken into account. For example, the length-to-width ratio should
preferably be as large as possible (up to 1.8), because the length of the tank is included in the residence
time of the melt and thus determines the glass quality. However, if the tank is too narrow, there is a
risk that the flame burnout will take place very close to the side wall and exceed the maximum thermal
load of the refractory material (which must also be considered in other areas).

The flame should not touch the baffle wall. It must be remembered, however, that what is visually
observable as a flame in the furnace does not represent the actual heat-transferring "flame". This is
 Page 166

much larger, since the essential, effective energy is transferred by the radiating exhaust gases (CO2 and
H2O) in the non-visible IR range.

The picture above shows a burner port. The angle of inclination of the burner and the burner neck are
also very important for the melting process. The latter cannot be changed after the furnace has been
constructed and annealed, so that one then has to "live" with a potential design flaw. The angle
determines how quickly fuel gas and fuel air (or oxygen) mix. The faster the combustion gas and air
mix, the faster the burnout and the higher the local temperatures. As a result, the NOX values also rise.
Likewise, the flame is not distributed far enough across the furnace and the energy transfer is not
uniform and inefficient.
There are various geometric solutions for the chamber connection to the burner throat via the burner
neck. An attempt is made to achieve the widest possible incident flow to the chamber.
The dwell time of bubbles and the flow directions of the glass are very important, as they determine
the glass quality. Short cuts should be avoided as well as unnecessary prolongation of the dwell time.

Similar considerations apply to the regenerator. An optimum flow velocity in the regenerator must be
achieved. If the exhaust gas velocity is too low, the exhaust gas flow is distributed to the front chamber
area, the rear part is not effective. If the exhaust gas velocity is too high, the exhaust gas flow is
 Page 167

distributed to the rear chamber area. This leads to a heavy load on the baffle wall and increased danger
of blockages in the lower area (example in the picture below, flow distribution in a regenerator).

11.8. Corrosion of melting tanks

A brief word on the corrosion of tanks. Corrosion occurs on every refractory material depending on its
load. Therefore, corrosion phenomena can be very diverse. It is always desirable to have consistent
corrosion throughout the whole tank so that there is no premature failure at any point of it. Frequent
corrosion is caused by bladder drilling, where rising bubbles stress the refractory material. There is
also "metal drilling" in the bottom of the tub. Pure metals (liquid, e.g. lead from impurities) are formed
by reduction and sink to the bottom. There they literally "drill" themselves into the bottom of the tank
through corrosion, so-called "downward-drilling".
A reaction zone always forms in the boundary layer between the molten glass and the refractory. In
this reaction zone, convections can arise that hollow out the purging edge (so-called Marangoni
convection). Marangoni convection is a flow at interfaces caused by local differences in the interfacial
tension σ. Since the interfacial tension of most substances decreases with increasing temperature T, a
flow occurs from warm to cold areas of the interface. This is therefore surface tension-driven
convection at the three-phase refractory-melt-atmosphere interface (so-called flux-line corrosion).
 Page 168

Finally, the regenerator is subject to severe corrosion from condensates (mentioned earlier) and carry-
over. Particles ("carry-over", esp. from the decripitation of dolomite) or condensates in the
corresponding temperature zones from evaporating species can cause serious problems if they clog
the regenerator. At first, "only" the efficiency of the regenerator is lowered, but eventually, not enough
combustion air can be supplied to the tank, which is fatal.
Corrosion of all parts lower the efficiency of the furnace, mostly through regenerator corrosion and
(to a lesser extent) through furnace corrosion (higher heat losses, false air). A furnace loses about 1.5
% of its initial efficiency per year due to the mentioned factors.
 Page 169

12. Glass Shaping

The chapter on shaping glass draws heavily on the chapter on viscosity, as the viscosity of a glass is the
key to shaping it in a way that is typical of glass. Shaping cannot be understood without a thorough
understanding of the viscosity of glass.
There are various processes to form a glass:
§ Blowing (container glass, tableware)
§ Pressing (glass for glasses, CRT-screens)
§ Rolling (cooktop, wire glass, ornamental glass)
§ Drawing (float, tubes, fiber)
§ Floating (float)
§ Casting (optical glass)
§ Centrifugation (CRT’s, tubes with very large diameter, glass wool)
§ others
In the course of this lecture, only blowing & pressing, floating, rolling and drawing will be dealt with.

12.1. Glass fiber drawing

Glass fibers have a wide range of properties, which are determined by their composition. Shaping is
done by drawing the glass, which is provided by the feeder, through a so-called bushing, a plate with
many small holes.

The bushing is made of platinum with the addition of 5 – 25 % rhodium for high-temperature stability.
It is usually equipped with a resistance heater. A homogeneous temperature distribution is crucial for
an even pull. There are 500 to 7000 tips (holes) in a bushing through which the glass is drawn. The
bushing is embedded in refractory material. Its thickness, dimensions, geometry and material are
decisive for the drawing process. The tip diameter is usually 1 – 2.5 mm, a fast beading is desired for a
quick release (in case of a necessary restart).
 Page 170

Technically, this process is based on the principle of nozzle flow. Accordingly, Hagen-Poiseuille's law
applies, which describes the laminar flow of a liquid through an opening:

Depending on the viscosity of the liquid and the length and width of the nozzle, corresponding
considerations can be made that determine the outflow of the glass (here from the bushing). However,
these considerations also apply in general to all feeder openings. For a large nozzle and a forced flow
by a pressure gradient or for a flow by gravity (with h as the height of the liquid level above the nozzle),
the mass flow rate dm/dt is defined by:

še o še o6 ∙(∙„
šZ
= 𝑓(r› ∙ w ∙ ∆𝑃 respectively šZ
= 𝑓(r› ∙ w

For the mass flow with a small nozzle applies:

𝜕𝑚 2 ∙ 𝑓 (r› 𝜌 ∙ 𝜎
= ∙
𝜕𝑡 𝑟 𝜂
If, however, the opening becomes very small, i.e. r2 << s/(r ·g) and thus falls below the critical length L
= (s/(r ·g))1/2, then it is no longer the pressure difference Δp that determines the mass flow, but the
surface tension. The particular geometry factor is determined according to:
𝜋 𝑟k 𝜋 𝑟k 𝜋 N
𝑓(r› = ∙ ; ∙ ; ∙𝑟
ÀÁÂ
8 𝑑 8 𝑑 + 𝑟 ÀÁÂ
ÀÄÁÄÂ 8
.›„f: I≫f I`f ŽUVZZr eYZ
n›<„: I≪f

The pure gravitationally induced dropping speed is approx. 12 m/h. However, the drawing speed per
tip can easily exceed 500 km/h.
Various aspects have to be considered when drawing glass fibers. The viscosity of the glass melt during
drawing is a very decisive parameter. It is easy to imagine that at such high drawing speeds, the
slightest instability or fluctuation can cause the fiber to break off (this is why it is rare to pass directly
by the bushing when being guided through a fiber drawing machine, as even a breeze can cause a
 Page 171

break-off). If the viscosity is too high, high stresses and tearing can occur (see picture on the left). If
the viscosity is too low, the fiber becomes unstable (see picture on the right). At ideal viscosity, the
drawing cone forms ideally (see picture in the middle).

A key point is also the relationship between Tliq and T(3). Generally, we try to put a large distance
between Tliq and T(3). This is to prevent the fiber from crystallizing as it is drawn. The smaller the delta,
the higher the risk that the fiber will crystallize ("devitrification") and break during drawing:

Unfortunately, the compositions (e.g. for E-glass) that show a large distance between Tliq and T(3)—
and are thus easier to process—require expensive raw materials. Low-cost or high-strength
compositions (S-glass, basalt glass fibers) are in the suboptimal range, i.e. show a high risk of
crystallization. Hence high cooling rates must be realized in the drawing process:
 Page 172

After the drawing process, the fibers (either continued as individual fibers or brought together to form
a fiber bundle) are rolled up, woven or they are separated into small pieces and processed further as
insulation fibers or reinforcing fibers. In general, the fiber is first coated with a sizing, an organic film,
which protects it. The individual possible processes are shown schematically below:
 Page 173

Fiber drawing has a very high potential to solidify rapidly crystallizing melts into glass, as very high
cooling rates can be achieved due to the low glass mass.

12.2. Hollow glass production

The hollow glass production (mainly for container glass) is divided into three processes:
1. BB – Blow Blow
2. NNPB – Narrow Neck Press Blow
3. PB – Press Blow
All three processes are two-stage processes. They are run on so-called IS machines (IS = individual
section). The name comes from the fact that each section, which consists of 1 – 4 moulds and thus
produces 1 – 4 bottles per step, can be stopped, maintained and restarted individually. In the past
rotary machines were used. In the event of a problem, the whole machine had to be stopped. A
schematic cross-section of an IS machine with one section is shown below. In a single machine, a row
of 8 – 12 sections can be installed.

The "cavity rate" C is a measure of the performance of an IS machine. It depends on how fast the
process is run (i.e. how well the line is controlled), but of course also on the glass weight and size of
the bottles. Large and heavy bottles are produced at a lower cavity rate than small, light bottles: C =
bottles per minute / number of cavities in the machine. Highly efficient IS machines with servo-electric
control achieve C = 25 for small bottles, i.e. more than 700 bottles per minute in the entire machine.
This means a daily production of more than 1 million bottles.
The three processes are presented in detail below:
Th blow-blow process is mainly used for narrow-neck bottles and heavy bottles, such as champagne
and wine bottles. The glass is made available in the feeder. What has just been said about the nozzle
flow also applies here. After the gob has been cut, it is loaded and settle-blown into the so-called
preform. This step already forms the bottle mouth in the preform. Then air is blown into the mould
from below ("counter-blowing") to form the "parison" (the preliminary stage of the actual bottle). This
two-step process, especially the settle-blowing, creates a so-called settle wave, an area of uneven glass
 Page 174

density distribution, which can be visible in the final product. Then, in a second stage, after the transfer
of the parison, the blowing into the final shape follows. This previous transfer, the "invert", is a very
decisive process, as the parison warms up again at the surface due to the thermal energy from the
interior and can thus be further formed without damaging the surface. After the transfer, the parison
remains in the open mould for a short time, heats up further at the surface and elongates ("reheat").
Then the mould closes and the blowing into the final mould takes place. The mould opens and the
bottle (still at 400 – 500 °C) can move on. The process is shown schematically below.

The press-blow process is mainly used for wide-mouth containers, such as baby food or canning jars.
The loading is identical to the one described before. The drop is pressed into the mould with a plunger.
This generally gives a better glass distribution compared to the blow-blow process. In the preform, the
neck finish is formed last. Then, in a second step, after the transfer of the parison, the reheating and
the blowing into the final mould follows, analogous to the blow-blow process.
 Page 175

The Narrow Neck Press and Blow is the newest shaping process (1974) and the fastest and most
efficient. It is used mainly for smaller bottles (up to about 0.5L). The gob is also pressed into the mould
with a punch. This gives a very good glass distribution compared to the blow-blow process. In the
preform the mouth is formed last. Then, in a second step, after transferring the parison, it is blown out
into the final mould, again analogous to the blowing-blowing process.
 Page 176

The wall thickness δ and the filling volume Vfill of the finished bottle are determined by the gob weight.
Correct temperature control is very important when moulding hollow glass. The glass starts to stick to
the mould material at a certain temperature. This is a function of the viscosity and thus only depends
on the glass chemistry and the interface temperature between glass and mould. The sticking
temperature is at η = 109.8 dPa·s. For a normal soda-lime-silica glass this means that from 645 °C on the
glass shows this tendency to stick. The moulding process is therefore controlled in such a way that this
temperature is not reached in the mould. However, degradation phenomena, namely the build-up of
oxide layers on the mould, can lead to a reduction in the thermal conductivity of the mould and thus
to an increase in the interfacial temperature. Undesired sticking can then occur. The interface
temperature TC can be determined with T1,2 as the temperature of the glass/mould according to:

(
𝑇% − 𝑇B v p∙<8 ∙o)@Y=Z
=
𝑇B − 𝑇G
v(p∙<8 ∙o)[=\]]

The moulds are regularly lubricated with an oil-graphite paste to facilitate loading. However, if the
oxide formation is too strong, this procedure is no longer sufficient and the mould has to be replaced
and reconditioned. The development of special low-friction coatings provides a remedy.
After moulding, the glass is coated with a hot end coating (usually a few nm SnO2 or TiO2) via a CVD
process and then stress-relieved in a lehr. Then a cold end coating of PE (also a few nm) is applied.
This is followed by a mostly highly automated 100 % (!) inspection.

12.3. Drawing tubular glass

Tube drawing is applicable for glasses with Tg of approx. 430 °C – 790 °C and thus for a variety of
different glass compositions. There are various processes, the best known being the Danner process.
The glass flows onto a so-called pipe. The feeder (C) controls the glass flow. The glass is not loaded
centrally onto the rotating (8 – 11 rpm) pipe (A), differences in thickness are evened out after a few
rotations. The pipe is always pulled back a little over time to reduce corrosion. Then the glass is pulled
off the pipe and a virtually infinitely long glass pipe is formed.
The diameter of a pipe can be up to 70 cm, the length up to 2 m. The bulb (B) at the tip of the pipe is
controlled by the blowing pressure from the blowing nozzle. The glass tube is deflected into the
horizontal by the Danner pipe in a tight pull and prevented from collapsing by the gas pressure applied
from behind by the Danner pipe. The drawn glass then enters the drawing channel (D) and is cooled
and stress relived.
 Page 177

The rotational movement ω of the pipe itself has no influence on the film thickness d. It only serves to
equalize (homogenize) the film. The thickness d of the film is given solely by the glass properties, the
mass flow 𝜕m/∂t and the diameter R and angle of incidence φ of the pipe.:

𝜕𝑚 2𝜋 ∙ 𝜌G ∙ 𝑔 ∙ sin 𝜑 ∙ 𝑅k 𝑑 N
= ∙0 1 ⟹
𝜕𝑡 3∙𝜂 𝑅

7 𝜕𝑚
3∙𝜂∙! "
𝑑=Ç 𝜕𝑡
2𝜋 ∙ 𝜌G ∙ 𝑔 ∙ sin 𝜑 ∙ 𝑅

The Danner process allows great variability in terms of tube diameters. It achieves mass throughputs
of up to 40 t/d.
Another process worth mentioning for drawing tubes is the Vello process. Here the glass is drawn
vertically, a higher throughput and better temperature homogeneity is achieved compared to the
Danner process.
 Page 178

The glass is drawn over a Pt-coated needle or cone (A) and ring (B) to form the bulb (C). The glass tube
hangs freely and is directed horizontally if the viscosity is appropriate (D). Rods can also be drawn
without a needle.
In another process, the down-draw process, the glass is not deflected, drawing speeds are then up to
20 m/min. Due to the non-deflection, the down-draw process allows greater flexibility than the Danner
and Vello processes.

12.4. Pressing glass

Glass is pressed to shape lenses, for example, and (in the past) to produce cathode ray screens and
funnels.
The thickness reduction during pressing effectively only occurs during the first fractions of a second of
the pressing process. If the viscosity is set incorrectly, even increasing the pressing force has little
effect. It all depends on the correct setting of the viscosity, i.e. the temperature during pressing. The
picture below shows the importance of a viscosity value of 7.6 as the limit of macroscopic
deformability.
 Page 179

12.5. Drawing / rolling of flat glass

There are also different processes for drawing or rolling flat glass. In rolling, the glass is rolled out
between two rollers to the desired thickness.

During drawing, the glass is drawn upwards or downwards from the melt at the appropriate viscosity.
The glass melt is incompressible as a good approximation. Therefore, a uniaxial state of stress is
formed. The mechanical stresses relax again spontaneously. As a result, the original cross-sectional
profile (with the exception of the rounding of sharp corners) is precisely reproduced with affinity.

With this process (up-draw or down-draw), extremely thin glasses can be produced. For displays, it is
possible to produce a glass thickness of 50 µm. These glasses are very strong in relation to their mass
and show extraordinary mechanical properties. If the drawing process is well controlled, i.e. no defects
are introduced into the glass, the finished glass can be rolled around a coin. This property is important
for flexible, foldable displays, for example.
Another drawing process (although the term drawing process is not quite correct here) is the overflow
fusion process, which is also used for display glasses. In this process, glass flows into a kind of trough
and is pulled over the sides of the trough; it spills over, so to speak. The two side streams are then
brought together under the trough.
 Page 180

The great advantage of this process is that the two outer sides have no contact with any refractory
material. This means that the highest surface qualities can be produced for display glass, currently in
a size of 294 cm x 337 cm (Gen 10.5).

12.6. Float process

The float process was developed by Sir A. Pilkington in the 1950s and finally widely used industrially in
1966. The story of this development is well worth reading and shows the staying power needed to
establish a new technology. Pilkington was on the verge of bankruptcy several times during the
development of the float process. Today, the float process is the process for efficient and cost-effective
production of high-quality flat glass.
The basic idea of this process is that, in order to avoid damage, the glass is not formed between
refractories by rolling or drawing into a 2-D sheet, but that the forming is carried out on a liquid. This
would mean that one side would only be in contact with a very smooth liquid surface and the other
would only be in contact with a gas atmosphere. The requirements for such a liquid would thus be:
§ Density higher than that of the glass
§ Minimal chemical reactivity with the glass
§ Non-wetting between glass and liquid
§ Boiling temperature > 1400 °C (max. flow temperature of the glass)
§ Melting point < 500 °C (lower than the Tg of the glass)
§ Low vapor pressure
§ Handleable chemistry (no highly toxic substances)
§ Economical
In the end, tin turned out to be the most suitable with MP = 231 °C, BP = 2620 °C, vapor pressure at
1100 °C = 0.002 mm Hg and a density of 6.5 g/cm3. However, 65 % gold / 35 % antimony mixtures (MP
 Page 181

< 400 °C), for example, were also considered, but then, surprisingly, rejected due to economic
considerations. It is true that such a bath filling (the glass flows out on a tub) lasts quite a long time.
But at some point, it will have to be replaced and that is an expensive business even with pure tin.
So, the glass flows onto a liquid tin bath during the float process. Since tin is easily oxidized, the bath
is operated under a forming gas atmosphere of nitrogen (N2) and hydrogen (H2). The set-up is as shown
below:

The glass flows into the bath via the lipstone, regulated by the tweel. So called Top rollers, which are
gear wheels that grip into the glass and thus regulate its speed and width, initially guide the glass
ribbon. The ribbon then cools down (ideally homogenously) on its way through the float bath and is
then pulled out of the bath at about Tg. The temperature at the entrance to the bath is about 1250 °C,
and 550 °C at the exit. The glass should pass homogeneously through this temperature range.
Adjusting the exact thickness of the glass must be thought of a bit like pulling on a rubber band: The
cooling furnace after the float bath pulls the glass out to a thickness smaller than the so-called
equilibrium thickness. The internal tension in the glass strives back to the equilibrium thickness. The
outward force of the top rollers counteracts this "collapse" if necessary and enables the desired glass
ribbon thickness.
The equilibrium thickness (d) of the glass ribbon is determined by the surface tension at the three-
phase interface Sn-atmosphere-glass.
 Page 182

It relates to the following equation:

With the above, the horizontal spreading force Fs can be calculated. Fs < 0 means a movement inwards,
Fs > 0 outwards (to the side).
𝐹W = 𝛾W_ − 8𝛾( + 𝛾(W_ 9 = 𝛾( ∙ (cos 𝛼 − 1)

For a soda-lime-silica glass with 𝛾g = 0.3 N/m and thus 𝛼 = approx. 110°, this means Fs < 0 and thus an
inward movement until F is in equilibrium with the hydrostatic pressure of the glass and tin. The
equilibrium thickness is thus approx. 6.3 mm (Fs > 0: spreading; Fs = 0: d = deq; Fs < 0: contraction). The
thickness of window panes is normally 4 mm. So here the top rollers have to slow down the glass and
pull it outwards so that it does not flow together and is pulled out too quickly.
Overall, thinner glass must also be pulled faster because the pull of the oven generally remains
constant (keyword: process stability in the oven). Glass in the middle of the ribbon usually flows faster
than glass at the edge.
The glass flows can be followed, for example, by scattering a tracer. After the glass ribbon has been
extracted, it is usually gassed with SO2 to create a protective alkali-sulphate layer on the underside,
transferred to the lehr, stress-relieved, inspected and separated.
Despite the high surface quality that can be achieved, defects can also occur on the glass during the
float process. Some tin always diffuses into the glass. Therefore, the glass bath side and the
atmosphere side (so-called "fire side") can always be distinguished.
Sn is an extremely easy metal to oxidize. The entire float chamber is therefore kept under an N2/H2-
atmosphere (5 to 10 % H2), as mentioned above. Nevertheless, traces of O2 always enter the chamber
via leaks. Oxygen entering the bath leads to the formation of SnO2, so-called "dross". This in turn can
deposit on the glass ribbon and create defects. Reactions with hydrogen in the atmosphere are also
possible. This then produces elemental Sn again, which can drip onto the glass. A major source of O2
input is, of course, the oxidic glass itself. The solubility of O2 in metallic Sn is very low and strongly
temperature-dependent. If the solubility is exceeded, crystalline SnO2 is formed, which appears as a
defect on the underside of the glass ribbon. Defects can also occur on the upper side of the ribbon via
the evaporation of SnO.
A similar cycle as with oxygen also exists with sulfur from the glass. And for defects on the upper side
of the ribbon, the evaporation of SnS is more critical (the source of entry of the sulfur into the tin bath
is again the glass itself). The sulfur reacts with the tin and can thus, e.g. via evaporation and
condensation after SnS(g) + H2 ⇌ Sn + H2S, lead to elemental Sn, which can drip onto the glass.
 Page 183

13. Incremental systems

Incremental systems are a frequently used approach in glass technology to determine the properties
of a glass solely on the basis of its chemical composition. The first approaches to this go back to Otto
Schott, one of the most important glass technologists and pioneer of modern glass research. As early
as 1894, O. Schott found out that basic properties of glasses (density, modulus of elasticity, coefficient
of thermal expansion, viscosity, refractive index, strengths, ...) can be predicted empirically by means
of a (linear) mixing model.
The principle of incremental systems determines a property p of a material via its composition given
as mole fractions xj or mass fractions yj and via an increment aj assigned for the corresponding
composition component. Then a relationship applies approximately for many properties according to:

𝑝(𝑥^ ) ≈ r 𝑎 ^ ∙ 𝑥 ^
^

Properties that are linked to the molar volumes can be represented particularly well by incremental
systems. For more complex properties, one chooses non-linear approaches:

𝑝8𝑥^ 9 = r 𝑎^ ∙ 𝑥^ + r 𝑏 ^,X ∙ 𝑥 ^ ∙ 𝑥X + r 𝑐^,X,U ∙ 𝑥 ^ ∙ 𝑥X ∙ 𝑥U + ⋯

^ ^,X ^,X,U

This is used to describe curved hypersurfaces in the composition space. Disadvantage: The number of
increments (the so-called "adjustable parameters") increases rapidly. Thermochemical equilibria can
be described particularly badly by increment systems. A typical example is the liquidus temperature of
a multi-component system.
A good increment system always gives the range of validity and error. When extrapolating increment
systems beyond their range of validity, great care must be taken. Only in rare cases is this possible
without too much error. Some of the increment systems use mole fractions, some mass fractions.
Quick conversion should be mastered. The following rules are helpful:

𝑚^ mass [g]

𝑛^ quantity of material [mol] Material conversion (chemical

reactions) take place in
stoichiometric proportions.

𝑀^ molar Mass [g/mol] it is recommended to make your own

list of Mj, accurate to 0.001 mg/mol.

𝜌^ density [g/cm3] Whenever a V occurs in a

thermodynamic equation, one can
substitute ρ for condensed phases.

𝑉^Ÿ molar volume [mol/cm3] Molar volumes of mixed phases are

approximately additive. Therefore,
material properties can often be
 Page 184

represented by linear "incrementing

systems

𝑚^ Two simple but fundamental

= 𝑛^
𝑀^ relationships that greatly facilitate
𝑀^ calculation in material systems
𝑉Ÿ =
𝜌^

𝑛^ ∙ 𝑀^ = 𝑚^ = Relationship between
amount of substance n, mass
m and molar mass M

𝑚^ = mass fraction = 0.01 · mass %

𝑦^ =
∑^ 𝑚^

𝑛^ = mole fraction = 0.01 · mol %

𝑥^ =
∑^ 𝑛^

1 = mean molar mass

〈𝑀〉 = @ 𝑥/ 𝑀/ = 𝑦
/ ∑/ /
𝑀/

=^ · l ^ = Conversion rule
𝑦^ = ¡l¢
#
𝑥^ = l^ ∙ 〈𝑀〉
^
 Page 185

13.1. Modelling of viscosity

There are different approaches for modelling viscosity. A simple approach was proposed by Lakatos
(T. Lakatos - Glast. Tidskr. 31:2., 1976, 51–54):

𝑇(𝑁) = 𝑎(SiOG ) + 100 ∙ r 𝑎(𝑗) ∙ 𝑦(𝑗)/𝑦(SiOG )

A more complex approach has been suggested by Flügel (A. Flügel – Glass Sci. Techn., 2007, 48 (1),
13–30):
_ _ _

Isokom temperature = 𝑏› + r Ë𝑏Y 𝐶Y + r(𝑏YX 𝐶Y 𝐶X + r 𝑏YXe 𝐶Y 𝐶X 𝐶e )Ì

Y`U X`Y e`X

This method can therefore be used to model viscosity dependencies. The viscosity-temperature
function of a glass is set via the chemical composition. The picture below shows the effect of a series
of oxides on the different viscosity levels of an initial glass of the composition 74 SiO2, 10 CaO, 16 Na2O
(left: mass fraction contents, right: molar fraction contents). Now 1 % SiO2 is replaced by another oxide.
The resulting shift ΔT of the viscosity levels is shown by the curves. Example: The replacement of
1 mass-% SiO2 by Na2O lowers the level T (2.0) by 24 K, and the Tg = T (13.0) by 5 K. If the replacement
is carried out with Al2O3, T (2.0) and Tg increase by 14 and 5 K, respectively. Li2O is the strongest liquidus
flux; even small amounts strongly decrease the viscosity in the entire range. B2O3 has a similarly strong
effect in the melting and processing range, but not in the cooling range; B2O3 additions therefore make
a glass easily processable, but "short“.

(74-x) SiO2 + 10 CaO + 16 Na2O + x NN, x = 1

by weight by molar amounts
0
K2O 0 PbO
-10 PbO K2O
-10 B2O3
-20
DTnn in K

Na2O
-30 -20 Li2O
Li2O Na2O
-40 -30
B2O3
-50 -40

-60 -50
20 ZrO2 20
ZrO2
10 Al2O3 10
CaO CaO Al2O3
DTnn in K

0 MgO 0 MgO
BaO ZnO BaO ZnO
-10 -10

-20 -20

-30 -30
0 2 4 6 8 10121416 0 2 4 6 8 10121416 0 2 4 6 8 10121416 0 2 4 6 8 10121416
log h, h in dPa·s log h, h in dPa·s log h, h in dPa·s log h, h in dPa·s
 Page 186

13.2. Modelling of the mechanical properties

The elastic constants are linked to each other. If one knows two, the other two can be calculated from
them:
2 ∙ (1 + 𝜇) 𝐸 𝐸∙𝐺
𝐾= ∙𝐺 = =
3 ∙ (1 − 2𝜇) 3 ∙ (1 − 2𝜇) 9𝐺 − 3𝐸
3 ∙ (1 − 2𝜇) 𝐸 3𝐾 ∙ 𝐸
𝐺= ∙𝐾 = =
2 ∙ (1 + 𝜇) 2 ∙ (1 + 𝜇 ) 9𝑘 − 𝐸
9𝐾 ∙ 𝐺
𝐸 = 3 ∙ (1 − 2𝜇) ∙ 𝐾 = 2 ∙ (1 + 𝜇) ∙ 𝐺 =
3𝐾 + 𝐺
1 3𝐾 − 2𝐺 1 𝐸 𝐸
𝜇= ∙ = − = −1
2 3𝐾 + 𝐺 2 6𝐾 2𝐺

E and µ of a glass can be calculated from chemistry taking into account further empirical data V°(j) and
U°(j) according to an incremental model of Makishima & Mackenzie (1973/75):
Oxid 𝑗 = 𝑀V 𝑂b
o∙R °^
"Intrinsic" module 𝑈 ^ / 𝑉^Ÿ = ∑ in MPa
=^ ∙ l ^

k <e7
"Intrinsic" space requirements 𝑉 ^ = N 𝜋 ∙ 𝑁M ∙ 8𝑎(𝑗) ∙ 𝑟ŸN + 𝑏(𝑗) ∙ 𝑟&N 9 in e›U
∑ m^
∑ m^T
is a measure of lateral contraction

For the detailed procedure of the calculation, please refer to the lecture slides. Here it is again
interesting to remember that the modulus as a unit corresponds to MPa = N/m2 converted to J/m3, i.e.
an energy density; the energy that is stored in the structure through compression of the bonds in the
case of the compression modulus, for example.

13.3. Modelling of the thermal expansion

There are also incremental approaches to modelling thermal expansion. The best known and, despite
its simplicity, very successful approach is that of von Appen according to:

𝛼 = r 𝑎Y ∙ 𝑝Y

𝑎Y = oxide-specific coefficient
𝑝Y = mol-% of the oxidic component in the glass
Other models by Winkelmann/Schott, Gilard/Dubral or Naray-Szabo, etc. give improvements for
special cases, but there is no generally valid solution, as the determination in each case is based on
different glasses. Therefore, please always consider the chemical limits of the models.
The coefficient of thermal expansion also depends on the cooling history of the glass (Tf, influence of
structural relaxation). The models therefore only have a limited accuracy.
 Page 187

13.4. Modelling of the optical properties

In order to determine the refractive index nD and the Abbe number νD as a function of the composition,
one uses experimentally determined, oxide-specific factors starting from the molar refraction r* (for
details on values see lecture slides):
∑^ f^ ∙#^
𝑛c = 1 + 𝜌 ∙ ∑^ 𝑟^ ∙ 𝑦^ , respectively 𝑣c = ∑^ t^ ∙#^

To determine the course of the refractive index n as a function of the wavelength, the measured values
n(λ) are approximated by formulae fitted to the different data for different wavelengths λ. For
example, for the range between 365 – 2300 nm, the Sellmeier formula is suitable:
^`N 𝐴 𝜆G 𝜆G
G (𝜆) ^ ^
𝑛 −1= r G G
^`% 𝜆 − 𝜆^

Aj = experimentally determined Sellmeier coefficients, also "amplitude of resonant wavelength"

λj = experimentally determined Sellmeier coefficients, also "resonance wavelength"
In general, two of the resonance wavelengths are in the UV range and one in the IR range, since the
curve is steeper in the UV range, i.e. n varies more strongly with λ. With the formula mentioned above,
the refractive indices can be interpolated and generally also extrapolated.