Making Glass — Batching, Melting, and Cooling Explained

Back to: Mochizuki Research Group Education — Inorganic Materials Science > Virtual Experiments, Inorganic Materials Science resources

This page is the explanation that pairs with glassmelt-simulator-en.html. It collects the conversion formulas, reaction equations, and models used in the simulator’s ① Prepare the batch, ② Melt, and ③ Cool, together with their sources, assumptions, and tentative values. The principles of the glass transition and crystallization themselves (the VFT equation for viscosity, nucleation and growth, the JMAK equation, and the TTT diagram) are derived step by step in the explanation of binary phase diagrams, §7, so here we use only the results. How to read the screen of each tab is explained in Appendix A, and the items that we would like Prof. Mochizuki to confirm are in Items to Confirm.

The backbone of this page Glassmaking is easier to follow if you divide it into three stages: “what to weigh” (the oxide composition and the raw materials are different things), “how it melts” (decomposition → liquid → dissolution → bubble removal → homogenization), and “what it becomes when cooled” (if the structure freezes before crystals grow, it is glass). The answer to the last stage is decided by both the composition and the cooling rate, and not every composition turns into glass just because it is quenched.
Kinds of values (the same marks as in the simulator) Literature Values with a source (or values calculated from them). Schematic Educational rules or equations that show a trend (not quantitative predictions). Tentative Values that need to be checked against Prof. Mochizuki’s materials or the literature.

1. Introduction — The Steps of Making Glass

The basic steps of making glass in the laboratory are weighing and mixing → loading into the crucible → heating and holding (melting) → homogenizing the melt → quenching or slow cooling → observation. The three tabs of the simulator are arranged in this order and pass the state of one sample (the batch, how well it is mixed, and the state of the melt) on to the next tab.

The six components are specified in the instruction sheet. CaO, which is indispensable for soda-lime glass, and Li2O, which is used in the existing Binary Phase Diagrams ④, are not included. For this reason, the reference compositions were chosen from those for which literature values are available within these six components alone (§9).

2. Expressing Composition — mol% and mass%, and the 100 % Total

The composition of a glass is usually expressed as oxides (even if the actual raw material is a carbonate, we count the oxide that remains after melting). There are two bases, the fraction of the amount of substance (mol%) and the fraction of mass (mass%, also written wt%), and you must always state which one is used. If the mole fraction of component i is xi, its mass fraction is wi, and its molar mass is Mi, then the two are related as follows:

(1) wi=xiMi∑jxjMj , xi=wi/Mi∑jwj/Mj

The molar masses used are those calculated from the IUPAC standard atomic weights: SiO2 60.083, Al2O3 101.961, MgO 40.304, B2O3 69.617, Na2O 61.979, and K2O 94.195 g/mol Literature. For example, Na2O·2SiO2 (33.3/66.7 in mol%) is 34.0/66.0 in mass%. The light MgO has a larger value in mol%, while the heavy K2O and Al2O3 have larger values in mass%.

2.1 Handling the 100 % total

The six components always add up to 100 %. In the simulator, when the slider of component k is moved to the value v, the total is kept by increasing or decreasing the other components while keeping their current ratios:

(2) pk′=v, pj′=pj×100−v∑l≠kpl (j≠k)

Here p is the percentage on the chosen basis (mol% or mass%). When all the other components are 0 (when you move away from a pure substance), the remainder is divided equally among the other five. v is limited to 0–100, and an input that is not a number is treated as 0. Switching the basis does not change the composition itself (the same composition, linked by Eq. (1)).

3. Raw Materials and Oxides — Reaction Equations, Conversion Factors, and Mass Balance

Na2O and K2O absorb moisture and are hard to handle, so they are weighed as carbonates and decomposed by heating to give the oxides. B2O3 is often added as boric acid, H3BO3, and MgO as MgCO3 (or as MgO itself). Therefore “the composition expressed as oxides” and “the raw materials actually weighed out” are different things: the mass of the raw materials is larger than the mass of the oxides, and the difference escapes as gas.

(3) Na2CO3→Na2O+CO2↑ K2CO3→K2O+CO2↑ MgCO3→MgO+CO2↑ 2H3BO3→B2O3+3H2O↑

In reality, Na2CO3 reacts with the silica sand before it decomposes on its own (Na2CO3 + nSiO2 → Na2O·nSiO2 + CO2) and forms a silicate liquid (§4). Whichever route is taken, the amount of CO2 that leaves is the same.

Table 1. Conversion factors and mass loss of the raw materials Calculated from literature valuesFactor = grams of raw material needed to make 1 g of oxide. Mass loss = fraction of the raw material that escapes as gas.
Raw materialOxide formedFactorMass lossGas released
SiO2 (silica sand), Al2O3, MgO, anhydrous B2O3As is1.0000 %None
MgCO3 (84.313 g/mol)MgO2.09252.2 %CO2
H3BO3 (61.831 g/mol)B2O31.77643.7 %H2O
Na2CO3 (105.988 g/mol)Na2O1.71041.5 %CO2
K2CO3 (138.204 g/mol)K2O1.46731.8 %CO2

When a mass mglass of glass is made, the mass of raw material r and the mass of gas are obtained from the mass of oxide i, mglasswi, as

(4) mr=mglasswiMinrMr, mgas=mglasswiMingMg, ∑rmr−∑mgas=mglass

where nr and ng are the moles of raw material and of gas per mole of oxide (2 and 3 for H3BO3). The last equation is the mass balance, which the simulator shows in its readouts every time.

Example: per 100 g of glass (raw materials: MgCO3 and H3BO3) Na2O·2SiO2: silica sand 65.97 g + Na2CO3 58.19 g = 124.16 g. 24.16 g of CO2 escapes (19.5 % of the batch).
DURAN type: silica sand 81.00 g + H3BO3 23.09 g + Na2CO3 6.84 g + Al2O3 2.00 g = 112.93 g. Gas: 12.93 g.
Cordierite type: silica sand 51.36 g + Al2O3 34.86 g + MgCO3 28.83 g = 115.05 g. CO2: 15.05 g.
Pure B2O3 (from H3BO3): H3BO3 177.63 g. 77.63 g of H2O escapes, and the mass falls to 56.3 %.

4. What Happens During Heating — Decomposition, Liquid, Dissolution, Bubble Removal, and Homogenization

When the batch (the mixture of raw materials) is heated, it changes roughly in the following order. The temperature and speed of each change vary greatly with the particle size of the raw materials, the way they are mixed, and the heating rate, so ② of the simulator uses a schematic model that shows the order and the trends correctly Schematic.

  1. Decomposition and dehydration: H3BO3 dehydrates stepwise at 100–400 °C to become B2O3 (a multistep reaction passing through HBO2; abstract of Huber et al. 2020), and MgCO3 releases CO2 at a few hundred °C.
  2. First liquid: B2O3 melts at 450 °C. In a batch containing alkali carbonate, a eutectic liquid forms at around 700–800 °C and reacts with the silica sand to form a film of alkali silicate (abstract of Hong & Speyer 1993, for soda-lime batches). A large amount of CO2 is released at this stage.
  3. Dissolution of the sand: The grains of the remaining silica sand (and Al2O3) dissolve into the liquid. The lower the viscosity, the faster this goes.
  4. Fining (bubble removal): The gas bubbles trapped in the liquid rise and escape. If the viscosity is high, they do not escape.
  5. Homogenization: The compositional streaks (cords) left by uneven mixing fade through diffusion and convection.
  6. Volatilization: If the melt is kept at high temperature for a long time, B2O3 and the alkalis volatilize and the composition drifts (the certificate of NIST SRM 717a warns that borosilicate glass volatilizes especially readily above 1400 °C).

4.1 Speed of decomposition

The degree of decomposition ξk of raw material k is treated as a first-order reaction, with an Arrhenius-type rate constant:

(5) dξkdt=kk(T)(1−ξk), kk=k1/2exp⁡[−EkR(1T−1T1/2)], k1/2=ln2·EkβrefRT1/22

k1/2 is the value fixed, from the approximation −ln(1 − ξ) ≈ kRT2/(Eβ) for heating at a constant rate (βref = 10 K/min), so that exactly half of the material has decomposed at T1/2. T1/2 and E are 180 °C and 50 kJ/mol for H3BO3, 500 °C and 150 kJ/mol for MgCO3, 800 °C and 250 kJ/mol for Na2CO3, and 830 °C and 250 kJ/mol for K2CO3 Tentative (taken as a rough guide from the temperature ranges of Hong & Speyer and commonly quoted values; confirmation in primary sources is needed). Numerical calculation with Eq. (5) shows that the temperature at which Na2CO3 is half decomposed is 748 °C at 2 K/min, 803 °C at 10 K/min, and 828 °C at 20 K/min, which reproduces the trend familiar from thermal analysis that faster heating shifts decomposition to higher temperatures.

4.2 The liquid and the dissolution of the sand

The temperature TL1 of the first liquid is taken to be 450 °C if the batch contains B2O3, 780 °C if it contains alkali carbonate, and the liquidus Tl itself if it contains neither Schematic. The equilibrium fraction of liquid Leq is capped by the amount of flux Schematic. Let f be the mole fraction of flux (B2O3 + R2O) and c1 the fraction of flux contained in the first liquid (1 for a B2O3 liquid; about 0.26, close to the eutectic of the Na2O–SiO2 system, for a liquid of alkali carbonate and silica sand; if both are present, the average weighted by the mole ratio), and assume that the fraction of flux in the liquid changes linearly from c1 at TL1 to f at the liquidus; the lever rule then fixes the amount of liquid:

(6a) Leq=min⁡(s,fc1+(f−c1)s), s=T−TL1Tl−TL1(0≤s≤1)

For compositions rich in flux (f ≥ c1, for example Na2O·2SiO2), the straight line of the first term applies, and for compositions poor in flux, the second term takes effect. Examples: even if Al2O3 97 and B2O3 3 mol% is brought to 1600 °C, only about 10 % is liquid; if 3 % B2O3 is added to the cordierite type and the temperature is 1300 °C, about 18 % is liquid (the liquidus is 1438 °C with the upper limit for flux in §7.2); and if 43 % is added, the temperature exceeds the liquidus (1038 °C), so everything is liquid. This form is meant to avoid the error of predicting that a refractory oxide becomes almost entirely liquid after only a little flux is added. For compositions without a literature value of Tl (the DURAN type and the Bisbrouck series), the liquid is regarded as spread throughout at the temperature at which the melt viscosity reaches 103 Pa s (and at least TL1 + 150 K) Tentative. For the DURAN type this gives 1260 °C (the working point in the datasheet), and Leq at the softening point of 825 °C is about 0.3. This assumed temperature is shown in the readouts of ② with “(tentative)” attached.

The fraction u of the solid that remains undissolved approaches the equilibrium value ueq = 1 − Leq according to the following equation:

(6) dudt=−u−ueqτd, τd=20min×(η10Pa s)1/2

Assuming that diffusion through the boundary layer around a grain is rate-limiting, we made it proportional to η1/2. The reference value of 20 min is Tentative. If the melt is held at a temperature below the liquidus, u does not reach 0, and the result is “not fully melted.” If the cordierite type is held at 1400 °C it does not melt, but since its liquidus of 1467 °C is lower than the furnace limit (1600 °C), it does melt when held at 1550 °C. Pure SiO2 (1723 °C) and Al2O3 (2072 °C) have liquidus temperatures above the furnace limit, so they do not melt without flux. The message in ② distinguishes the case where the liquidus is below the furnace limit (the batch melts if the holding temperature is raised) from the case where it is above the limit (a composition that does not melt in this furnace). The liquidus of an arbitrary batch is determined by the interpolation of §7.2 (schematic). Flux-free SiO2–Al2O3 batches come out at 1723 °C or above and do not melt at the furnace limit. Most SiO2–MgO batches are also above 1600 °C, but near enstatite (SiO2 about 46–56 mol%) the liquidus is 1558–1600 °C, at or below the furnace limit, so these batches melt if the holding temperature is raised. The viscosity is given by the equation from the literature for the reference compositions and by the empirical equation of §7.1 (Fluegel 2007) for arbitrary batches.

4.3 Bubble removal — Stokes’ law

A bubble (radius r) in the liquid rises at the speed at which buoyancy and viscous drag balance (Stokes’ law):

(7) v=2ρgr29η, dbdt=Leq×(gas release rate)−vhb

b is the amount of gas trapped in the liquid (as a fraction of the total gas). Of the gas released by decomposition, the portion equal to the liquid fraction Leq at that moment is assumed to be trapped Schematic. With a bubble radius r = 0.2 mm, a melt depth h = 20 mm, and a density ρ = 2400 kg/m3 Tentative, η = 10 Pa s gives v = 2.1×10−5 m/s, and the bubble rises through the 20 mm depth in about 16 minutes. If the viscosity is 100 times higher, it takes 100 times longer, so bubbles are hard to remove from a viscous melt (the albite type has η ≈ 102.9 Pa s even at 1600 °C). In actual melting, fining agents (As2O3, Sb2O3, sulfates, and so on) are used to create large bubbles that are then removed, but the simulator does not treat this.

4.4 Homogenization — cords disappear by diffusion

Unevenness from mixing remains in the melt as compositional streaks (cords) with spacing λ. A sinusoidal concentration variation decays by diffusion as exp(−4π2Dt/λ2). The diffusion coefficient of the atoms that form the network of the melt is estimated from the Eyring relation, which links it to the viscosity:

(8) D≈kBTηd, τh=λ24π2D, λ=2mm×(1−0.75h)

Here d = 0.3 nm (the hopping distance) and h is the homogeneity in the mortar in ① (0–1). The equation for λ is Tentative. When Na2O·2SiO2 is melted at 1200 °C (η ≈ 12 Pa s), D ≈ 5.6×10−12 m2/s, and τh ≈ 19 min if it was mixed well (λ = 0.5 mm) and about 5 hours if it was put in without mixing (λ = 2 mm). Mixing well in the mortar makes the melt homogeneous sooner. In reality, convection and stirring stretch the streaks and make λ smaller, but the model does not include this.

4.5 Volatilization and mass

The volatilized fraction obeys dv/dt = kv(1 − v), with kv = 1.39×10−6 s−1·exp[−(Ev/R)(1/T − 1/1773 K)] and Ev = 250 kJ/mol (for B2O3; the alkalis volatilize at half that rate) Tentative. The magnitude is such that holding at 1500 °C for 2 hours loses about 1 % of the B2O3. The mass of the sample (as a fraction of the batch) is given by

(9) m(t)mbatch= mglass(1−wBvB−wRvR)+∑kmgas,k(1−ξk)mbatch

This is the brown dashed line in the graph of ② (corresponding to the thermogravimetry (TG) curve).

5. Structure of Glass — Network, Modification, and the Coordination of Al and B

5.1 Network formers, modifiers, and intermediates

SiO2 and B2O3 are network formers: SiO4 tetrahedra and BO3 triangles share their corner oxygens (bridging oxygens) and join into a network, and they can become glass by themselves (why they form glass easily is explained as Zachariasen’s rules in the phase diagram explanation, §7.9). Na2O, K2O, and MgO are modifiers: the oxygen they bring in breaks Si–O–Si bonds and creates non-bridging oxygens (O−), which are bonded to only one side. A broken network moves more easily, so the viscosity of the melt decreases (which is why Na2O acts as a flux), but if it is broken too much, crystallization becomes easier. Al2O3 is an intermediate: it enters the network when the conditions are met and cannot when they are not. How far the network is broken is expressed by NBO/T, the number of non-bridging oxygens per network-forming cation (Si, four-coordinate Al, and B; high-coordinate Al, which does not enter the network, is not counted). It is 1 for Na2O·2SiO2 and 2 for Na2O·SiO2. The atom diagrams of the simulator (Appendices A.3 and A.4) are drawn with the same color and size conventions as the figure “Molecular structure of glass (plan view)” in Lecture 01 (O is the largest red sphere among the atoms of the network (the modifier ions Na+, K+, and Mg2+ are in principle drawn even larger, but they are shrunk in compositions with many modifier ions so that they fit in the figure; see Appendix A.3), Si is a small blue sphere, Al is green, B is orange, and Na+ is a yellow sphere), and the arrangement of atoms is generated from the composition each time (it is not a copy of the lecture figure itself). A network whose rings differ in size and shape follows the idea of the two-dimensional random network that Zachariasen (1932) drew for the structure of glass, while a crystal of the same composition is drawn as the same network arranged regularly. The sphere sizes are ratios chosen so that the atoms can be told apart (Si 0.30, Al 0.41, B 0.33, Na 1.5, K 1.7, and Mg 1.2, with O taken as 1), and they differ from the actual ionic radii (Shannon 1976: O2− 1.40 Å, Na+ 1.02 Å, K+ 1.38 Å, and Mg2+ 0.72 Å (all six-coordinate); Si4+ 0.26 Å and Al3+ 0.39 Å (four-coordinate)). Ti (which enters the network) and Ca2+ (a modifier ion) in the lecture figure are not among the six components of this teaching material, so they are not drawn, and Mg2+ is drawn as an alkaline-earth modifier ion in the same way as Ca2+.

5.2 The coordination of Al and B changes with composition

An AlO4 tetrahedron has a charge of −1, so it enters the network only when a nearby cation such as Na+ compensates its charge. B also changes from BO3 (three-coordinate) to BO4− (four-coordinate) when it is compensated by a cation. As to which of the two takes the compensating ions first, Bruns et al. (2020) showed by NMR that Al takes Na first, and only the remaining Na forms BO4; that is, that the relation

(10) N4=[BO4][BO3]+[BO4]≈R′=R2O−Al2O3B2O3

holds well (Table 2). In the Na2O–B2O3–SiO2 system without Al, the following is known as the Dell–Bray rules: with R = Na2O/B2O3, N4 = R up to Rmax = 0.5 + K/16 (K = SiO2/B2O3), N4 stays flat beyond that up to Rd1 = 0.5 + K/4, and when there is still more modifier, non-bridging oxygens attach to three-coordinate B and N4 decreases (as summarized by Lu et al. 2021; the original sources have not been checked). In the simulator, for R > Rd1 we take N4 = Rmax(2 + K − R)/(1.5 + 0.75K) and lower it along a straight line until it reaches 0 at R = 2 + K (the straight-line form is Tentative). For example, in the Na2O–B2O3 system (K = 0), the value is 0.44 at 40 mol% Na2O and 0.33 at 50 mol% (the composition of NaBO2).

Table 2. Comparison of the R′ rule with NMR LiteratureBruns et al. 2020. Compositions are in mol% (SiO2/B2O3/Na2O/Al2O3). The values from the simulator’s rule are the same.
CompositionN4 (NMR)R′R ignoring Al
74.00/20.70/4.30/1.000.12 ± 0.030.160.21
73.62/20.60/4.28/1.500.11 ± 0.030.130.21
73.25/20.49/4.26/2.000.07 ± 0.030.110.21
71.76/20.07/4.17/4.000.02 ± 0.030.010.21

5.3 Mg is poor at compensating B — an example showing that the coordination must not be fixed

The series of Bisbrouck et al. (2022) cited in the instruction sheet fixes SiO2 at 51, B2O3 at 15, and Al2O3 at 7.7 mol% and replaces Na2O with MgO (Table 3). Mg is also a divalent cation, so if it were used for compensation in the same way as Na, N4 should not change; however, the measured fraction of BO4 falls sharply from 57 % to 7 %, and Tg rises from 523 to 642 °C. Mg contributes little to the compensation of BO4 (abstract of a 2021 paper by the same authors). Furthermore, when there is more Mg, Na alone cannot fully compensate Al, and five- and six-coordinate Al increases. In other words, the coordination of Al and B changes with composition, and one cannot uniformly assume “Al is four-coordinate” and “B is three-coordinate”.

Table 3. Measurements for the Bisbrouck series and the simulator’s rule Literature SchematicThe measurements are from Table 1 of Bisbrouck et al. 2022. N4 (rule) is R′ counted from Na (and K) only, capped at the Dell–Bray Rmax and at 1.
LabelMgO/Na2O (mol%)Tg (°C)BO4 (measured)N4 (rule)NBO/T (rule)
N26M00/26.152357 %72 %0.16
N19M87.5/18.757337 %71 %0.16
N13M1313.1/13.159424 %36 %0.27
N8M1918.7/7.56427 %0 %0.39

The rule reproduces the trend (BO4 decreases as Mg increases), but it gives values larger than the measurements for N26M0 and N19M8 (Lu et al. 2021 also point out cases in which Dell–Bray-type rules fail for Al-containing systems). The simulator therefore shows the measured value alongside for the reference compositions, and shows the rule’s value, marked “Schematic,” for arbitrary batches.

Fig. 1. BO4 fraction N4 versus R′ (our own). The line is the simulator’s rule (for Rmax = 0.72; for R′ > Rd1 = 1.36 it falls along the straight line of §5.2), ● are the measurements of Bruns et al. 2020, and ■ those of Bisbrouck et al. 2022 (the horizontal axis is R′ counted from Na only).

5.4 Summary of the rules used in the simulator Schematic

  1. Al is first compensated by R2O (Na2O + K2O) and becomes four-coordinate AlO4−. When R2O is insufficient, MgO compensates a fraction φ = MgO/(MgO + R2O) of the remaining Al (the form of φ is schematic), and Al that is not compensated is taken to be five- or six-coordinate. In competition with Na, the compensation by Mg is weak (Bisbrouck et al. 2021/2022; the larger the field strength of the cation, the more five-coordinate Al: Thompson & Stebbins 2011), and in an alkali-free metaluminous composition (the cordierite type, MgO = Al2O3) all the Al is four-coordinate. Some five-coordinate Al has been reported in real MgO–Al2O3–SiO2 glasses (to be confirmed), so the “100 % four-coordinate” of the cordierite type is a value from the rule.
  2. The R2O that remains converts BO3 into BO4−: N4 = clamp(R′, 0, min(1, Rmax)). For R′ > Rd1 it is lowered along the straight line of §5.2 (tentative). Mg does not compensate BO4.
  3. The R2O that still remains, and the MgO not used for Al, create non-bridging oxygens (2 per mole). T in NBO/T includes three-coordinate B and four-coordinate Al but not high-coordinate Al (if there is no network former, NBO/T is undefined and is treated as being on the side where the network is broken).

6. Cooling — The Competition Between Crystallization and Vitrification

When a melt is cooled below the liquidus Tl, the crystal becomes the more stable phase. But for a crystal to form, it takes time for nuclei to form (nucleation) and grow (growth). If, before that, the temperature falls, the viscosity grows, and the rearrangement of atoms can no longer keep up with the cooling, the liquid freezes with its structure intact and becomes glass. Tab ③ of the simulator is a direct port of the model of Binary Phase Diagrams ④ (the functions at lines 1681–1809 of phasediagram-simulator.html), with only the material parameters replaced. For the derivation of the equations, see the phase diagram explanation, §7. Only the main points are given here.

(11) log10⁡η=A+BT−T0(VFT equation, Pa s) I=σkBTd05ηexp⁡(−16πσ33ΔGv2kBT), ΔGv=ΔHm(Tm−T)TmVm u=gkBTd02η[1−exp⁡(−ΔHm(Tm−T)RTTm)], X=1−exp⁡[−4π3∫0tI(t′)(∫t′tuds)3dt′]

In cooling at a constant rate q, the exponent of the crystal volume fraction is proportional to 1/q4, so this determines the critical cooling rate Rc at which X = 10−6 (Uhlmann’s criterion). If q > Rc, glass is obtained, and if q < Rc, crystals appear. The simulator’s verdict uses the same criterion: X < 10−6 is “Glass (amorphous),” 10−6 ≤ X < 1 % is “Nearly glass (with a trace of crystals; hard to see in X-ray diffraction),” 1–50 % is “Partly crystallized,” and 50 % or more is “Crystallized” (below 99.9 % it is written “Mostly crystallized,” which indicates that some glass phase remains). The freezing of the structure is solved with a single relaxation time (τ = η/G∞, G∞ = 1010 Pa), and the temperature at which the structural temperature (fictive temperature) Tf freezes is taken as Tg(q).

The solid red line in the graph of ③ connects the points at which X = 10−6 is reached while cooling at a constant rate (the continuous cooling transformation diagram, CCT), and it uses the same criterion as the verdict. In cooling at a constant rate q, Xext from the moment the temperature falls below the liquidus down to T is F(T)/q4 (F is the value calculated at q = 1 K/s), so the line is drawn by solving F(T)/q4 = 10−6 for q and then computing the time (Tl − T)/q (the same method as in Binary Phase Diagrams ④). The largest q that touches this line is exactly Rc. The cooling curve is also drawn, like this line, with the time at which the temperature falls below the liquidus as the origin. The dotted line is the time to reach X = 10−6 when the temperature is dropped suddenly to a given value and held there (the isothermal TTT curve), shown for reference. Even if the cooling curve crosses the isothermal TTT curve, the sample does not necessarily crystallize in continuous cooling (the same caution as in the phase diagram explanation, §7). While cooling, the time spent at the nose temperature is short, so the CCT line lies to the lower right of the TTT curve.

Table 4. Parameters used in the cooling kineticsOnly the reference compositions with complete literature values. σ and g of Na2O·2SiO2 and the cordierite type are the model values fitted to quantities in the literature (fits); σ of the albite type is the calculated value from the literature; “tentative” marks values for which there is nothing to fit to.
CompositionVFT (A, B, T0)Tm (K)ΔHm (kJ/mol)σ (J/m2)gRc (model)
Na2O·2SiO2−3.00, 4254, 429.3 (Jiusti et al. 2020)1148 (Tl)35.564 (NIST-JANAF)0.07910.23280.034 K/s
Cordierite type−3.97, 5316, 762.0 (Jiusti et al. 2020)1740 (Tl)225.9 (Navrotsky & Kleppa 1973)0.13500.03111.05 K/s
Albite type−5.44, 12700, 356 (Zanotto & Cassar 2017)139159.3 (same source)0.16 (same source, calculated)0.2373×10−22 K/s
Pure SiO2−2.94, 12997, 610 (Binary Phase Diagrams ④)19968.90.06518.6×10−5 K/s
Pure B2O3−0.97, 1580, 433 (Binary Phase Diagrams ④)72324.070.100 (tentative)17.8×10−8 K/s

Details of the fits: g for Na2O·2SiO2 and the cordierite type was chosen so that the maximum growth rate of the model matches Umax of Jiusti et al. 2020 (9.92×10−7 m/s and 9.05×10−6 m/s) (the temperatures of the maximum are 1105 K and 1687 K in the model and 1087 K and 1533 K in the literature, so the cordierite type is off by 150 K). σ was chosen so that the model’s Rc matches 0.034 K/s and 1.05 K/s, estimated with the surface-crystallization formula of the same paper, Rc = (Tl − TUmax)/tn Tentative (this estimate comes out several times larger than the measured value for Li2O·2SiO2, so it serves only as a guide for ranking). For the albite type, we use σ = 0.16 J/m2 from Table 2 of Zanotto & Cassar 2017 (a value calculated from the classical nucleation equation), Vm = 100.3 cm3/mol (density 2.614 g/cm3), and the liquid Cp = 369 J/(mol K), and g was chosen to match the maximum growth rate of the same paper, Umax ≈ 1 Å/s (1×10−10 m/s) (the model’s maximum is at 1359 K). Rc is smaller than 10−21 K/s, so the melt becomes glass with any cooling method in the laboratory (consistent with the fact that crystallization is not observed). Cp comes from the literature only for the liquids of Na2O·2SiO2 and the albite type (261.2 J/(mol K) from JANAF and 369 J/(mol K) from Zanotto & Cassar); for the others, the Dulong–Petit value of 3R × the number of atoms is used, together with a schematic value 1.25 times that. The densities of Na2O·2SiO2 and the cordierite type (2.50 and 2.60 g/cm3) are Tentative. The model’s Tg (10 K/min) is 715 K for Na2O·2SiO2 (literature: 713 K) and 1096 K for the cordierite type (literature: 1072 K).

6.1 Cooling methods and cooling rates

Table 5. Rough cooling rates for each cooling method TentativeThey change by more than an order of magnitude with the size (thickness) of the sample. These are representative values to show how the speeds compare.
Cooling methodq (K/s)Na2O·2SiO2Cordierite typePure SiO2 (from the melt)
Pour into water (frit)300GlassGlassGlass
Pour onto a metal plate and press30GlassGlassGlass
Cool in air, crucible and all3GlassGlass (X ≈ 10−8)Glass
Switch off the furnace (furnace cooling)0.05Glass (X ≈ 2×10−7)Partly crystallized (X ≈ 17 %)Glass
Programmed slow cooling (0.5 K/min)0.0083Nearly glass (X ≈ 3×10−4, trace of crystals)CrystallizedGlass

Na2O·2SiO2 forms glass easily and is glass even with furnace cooling (X < 10−6). Programmed slow cooling (q = 0.0083 K/s) is slower than Rc ≈ 0.034 K/s, so a small amount of crystal appears (X ≈ 3×10−4), but the amount is small, so it is nearly glass both in appearance and in X-ray diffraction. The cordierite type crystallizes with furnace cooling, so it is a composition that “needs quenching.” Pure SiO2 has a very viscous melt, so it becomes glass even when cooled slowly (quenching is not needed). However, no melt forms unless the temperature is raised above 1723 °C.

7. Handling Arbitrary Batches — Viscosity, Liquidus, and Glass-Forming Tendency

For a batch of six freely chosen components, there is no method to construct the VFT equation, Tl, ΔHm, and σ all quantitatively from the composition. Also, following the policy of the instruction sheet, we avoid constructing Tg or the viscosity from a weighted average of the oxides (an additivity rule) and presenting it as a predicted value. Therefore, when the composition is changed from a reference composition, the viscosity is handled by an empirical equation from the literature (§7.1), the liquidus by interpolation from the reference points (§7.2, schematic), and crystallization by the glass-forming tendency (§7.3, schematic), and the screen shows which of these was used. All three are built so that the values do not jump when the composition is changed slightly (so that they are continuous).

7.1 Viscosity: the empirical equation of Fluegel 2007 Literature and interpolation outside its range Schematic

From more than 2200 composition–viscosity data sets in SciGlass, Fluegel 2007 expressed the temperatures at which η = 101.5, 106.6, and 1012 Pa s (isokom temperatures) as polynomials in the composition (standard errors of 9–17 °C). From its Tables 4–6, the simulator uses only the terms involving SiO2, Al2O3, MgO, B2O3, Na2O, and K2O (the terms for CaO, Li2O, and so on are set to 0):

(12) Tk(°C)=b0+∑ibiCi+∑i≤jbijCiCj+∑i≤j≤lbijlCiCjCl, k=1.5,6.6,12

Ci is the mol% of the components other than SiO2. Each time, a VFT equation passing through the three isokom temperatures (the first line of Eq. (11)) is constructed and used for the dissolution, bubbles, and cords in ② and for the freezing of the structure in ③. K2O and Na2O enter as separate components, so replacing Na2O with K2O still works as a flux (for SiO2 66.7 and K2O 33.3 mol%, log10(η/Pa s) ≈ 1.2 at 1400 °C, compared with 0.4 from the literature equation for Na2O·2SiO2). As a check of the transcription, Eq. (12) gives 1114/585/439 °C for Na2O·2SiO2 (1101/599/440 °C from the VFT of Jiusti et al. 2020) and 1539/815/585 °C for the DURAN type (in the datasheet, 106.6 Pa s is at 825 °C and 1012 Pa s at 560 °C).

The range of validity is taken from Table 3 of the same paper (the maximum concentrations in the source data), using for each component the narrowest of the values for the three viscosities (the smaller one): Al2O3 ≤ 10, B2O3 ≤ 16.97, MgO ≤ 16.61, Na2O ≤ 42, K2O ≤ 30, and SiO2 42.62–87.10 mol%. When the amount δ by which the composition falls outside the range (a distance in mol% points) exceeds 3, the isokom temperatures are shifted to those interpolated from the literature VFT equations of the reference compositions (Na2O·2SiO2, the DURAN type, the albite type, the cordierite type, SiO2, B2O3, and Na2O·2B2O3), with the weight w = clamp((δ − 3)/12, 0, 1) Schematic. Also, when extrapolation squeezes the spacing of the isokom temperatures, they are spread out until T12 ≤ T6.6 − 80 K and T1.5 ≥ T6.6 + 250 K (the spacing is wider than this in the literature VFT equation of every reference composition; for example, for N26M0 of the Bisbrouck series, T6.6 − T12 from Eq. (12) is only 16 K). The interpolation weights have the same form as in Eq. (13) (here Na2O and K2O are taken together and measured as R2O). In addition, around a reference composition that has a literature value, the discrepancy at that point is added with the weight g = a exp[−(d/6 mol%)2], so that the result agrees with the literature value exactly at the reference composition (for the Bisbrouck series, for which only Tg is known, the three temperatures are shifted by the same amount). For this reason, even if 1 % of Na2O is replaced with SiO2 in the albite type, the viscosity at 1600 °C is almost the same as from the literature equation (102.95 and 102.93 Pa s).

Because the interpolation is an average over the reference points, adding flux can increase the weight of distant refractory reference points (SiO2 and others) and the viscosity may actually rise (for example, when B2O3 is added to the cordierite type). Therefore, to make the viscosity decrease in the direction of adding flux, we impose the following two upper limits Schematic Tentative. Here the free flux is f = B2O3 + max(0, R2O − Al2O3) (the R2O used to compensate Al links the network, so it is not counted), and the base composition is the batch with the free flux removed. (i) Isokom temperature (K) ≤ the value of the base composition × (1 − 0.5f) (not used inside the range of the empirical equation). (ii) The upper limit is the minimum of the values for the base composition to which the same combination of flux has been added by s = 0, 0.02, 0.04, …, each multiplied by (1 − 0.1(f − s)) (this prevents reversals at the boundary between the range of the empirical equation and the range of interpolation). Both are phased out as f goes from 0.6 to 0.8, and (ii) is not used around the reference compositions that have been pulled toward the literature values. Example: when 0/10/30 % B2O3 is added to the cordierite type, the temperature at η = 103 Pa s falls from 1252 °C to 1175 °C and 1023 °C, and when it is added to SiO2, it falls from 2525 °C to 1303 °C and 1149 °C. For a batch for which only one reference composition can be used (for example, MgO–B2O3 or Al2O3–B2O3, which contain neither SiO2 nor Na2O; only pure B2O3 can be used), that viscosity is used as it is, and the screen says “differences due to composition cannot be shown.” For the Na2O–B2O3 system, the VFT equation of Na2O·2B2O3 (Jiusti et al. 2020: A = −3.03, B = 1279 K, T0 = 677.1 K) has been added, so log10(η/Pa s) at 1000 °C changes with composition as 0.59/0.12/−0.82 at 15/20/30 % Na2O. A batch in which the network formers (SiO2 + B2O3) make up less than 0.30 (Na2O alone, Al2O3–MgO, and so on) cannot have its viscosity estimated, so it is shifted to the VFT equation of a free-flowing melt (A = −4, B = 2000 K, T0 = 300 K; about 10−2.3 Pa s at 1200 °C), and at 0.15 or below only that equation is used (Tentative; the screen says “viscosity cannot be estimated (outside the scope of this glass teaching tool),” and Tf is not shown).

7.2 Liquidus: interpolation from reference points Schematic

The liquidus of an arbitrary batch is the average of the liquidus temperatures of the reference points in Table 6, with the following weights:

(13) Tl=∑pwpTl,p∑pwp, wp=apdp2+1, ap=∏c∈pclamp(xc0.5xc,p,0,1)

dp is the distance between the compositions of the batch and the reference point p (in mol% points), and ap is the product of “how much of each component c contained in the reference point is present in the batch.” A reference point that contains a component that the batch lacks has ap = 0 and is not used. For example, a flux-free SiO2–Al2O3 batch never borrows the liquidus of albite (1118 °C), which contains Na. When a component is increased gradually from 0, the weight also increases continuously from 0, so the value does not jump. This interpolation cannot represent the shape of the liquidus (the eutectic dip, which crystal appears first), so it is shown as schematic together with “which reference point is nearby.” Also, adding flux can increase the weight of distant refractory reference points and raise the liquidus, so we impose the following two upper limits (both Tentative). (i) For the flux f = B2O3 + R2O, Tl ≤ the interpolated value for the base composition (SiO2, Al2O3, MgO) − 1000 K × f (phased out as f goes from 0.6 to 0.8). (ii) For the amount b of B2O3, from the interpolated value Tl(s) of the composition in which B2O3 has been reduced to s (0 ≤ s ≤ b) without changing the ratios of the other components, Tl ≤ mins [Tl(s) − 300 K × (b − s)]. The 300 K is a value chosen to match the magnitude of the freezing-point depression of an ideal solution, RTl2/ΔHm (about 310 K for Na2O·2SiO2). It is phased in where SiO2 in the composition without B2O3 is 0.30–0.45, and phased out as b goes from 0.6 to 0.8. With this form, the liquidus falls when flux is added to a flux-free base composition and when B2O3 is added to a batch containing SiO2 (example: adding 10 % and 20 % (mol) of B2O3 to Na2O·2SiO2 gives 875 → 846 and 816 °C), and the reference points with literature values (Na2O·2SiO2, Na2O·SiO2, albite, Na2O·2B2O3, and so on) are not affected by the upper limits. On the other hand, when R2O is added to a batch that contains alkali, the liquidus may rise. In the Na2O–SiO2 system, on the side richer in Na2O than the eutectic (793 °C), the liquidus actually rises to Na2O·2SiO2 (875 °C) and Na2O·SiO2 (1089 °C), so the treatment (ii) is not applied to R2O (example: adding 20 % of Na2O or K2O to Na2O·2SiO2 gives about 1070 °C and 1000 °C, respectively).

Table 6. Reference points for interpolating the liquidusCompositions are in mol%. “Literature” is a value from the same source as the reference compositions, and “Tentative general value” is a value commonly quoted from binary and ternary phase diagrams (confirmation in primary sources is needed).
Reference pointCompositionTl (°C)Kind
Na2O·2SiO2SiO2 66.7, Na2O 33.3875Literature Jiusti et al. 2020
Eutectic of Na2O·2SiO2–SiO2SiO2 74.6, Na2O 25.4793Tentative general value
Liquidus of the Na2O–SiO2 system (quartz–tridymite)SiO2 76.1, Na2O 23.9 (75.5 mass% SiO2)870Literature Kracek 1930 (confirmed in the summary by Ryś 2007)
Liquidus of the Na2O–SiO2 system (tridymite–cristobalite)SiO2 89.0, Na2O 11.0 (88.7 mass% SiO2)1470Literature Kracek 1930 (same as above)
Liquidus of the Na2O–SiO2 system (midpoint of the two points above)SiO2 82.55, Na2O 17.451170Tentative value obtained by assuming a straight line between the two points
Na2O·SiO2SiO2 50, Na2O 501089Tentative general value
K2O·2SiO2, K2O·4SiO2K2O 33.3, 201045, 770Tentative general value
AlbiteSiO2 75, Al2O3 12.5, Na2O 12.51118Literature Zanotto & Cassar 2017
CordieriteSiO2 55.6, Al2O3 22.2, MgO 22.21467Literature Jiusti et al. 2020
Mullite 3Al2O3·2SiO2SiO2 40, Al2O3 601850Tentative general value
Forsterite 2MgO·SiO2SiO2 33.3, MgO 66.71890Tentative general value
Enstatite MgO·SiO2 (incongruent melting)SiO2 50, MgO 501557Tentative general value
Two-liquid region of MgO–SiO2SiO2 80, MgO 201695Tentative general value (above this temperature the melt separates into two liquids)
Spinel MgO·Al2O3Al2O3 50, MgO 502135Tentative general value
SiO2, B2O3, Na2O·2B2O3—1723, 450, 743Literature Binary Phase Diagrams ④, Jiusti et al. 2020
MgO, Al2O3—2852, 2072Tentative general value

Examples (compositions in mol%): a flux-free batch of SiO2 60 and Al2O3 40 has a liquidus of 1841 °C, a batch of SiO2 70 and MgO 30 has 1692 °C, and one of SiO2 40, Al2O3 30, and MgO 30 has 1650 °C; none of them melts at the upper limit of the furnace (1600 °C). In the SiO2–Na2O system the liquidus is 1458 °C at 10 % Na2O, 1222 °C at 15 %, 1061 °C at 20 %, and 820 °C at 25 %, and for SiO2 74, MgO 10, and Na2O 16 it is 1110 °C (the two Kracek points on the SiO2 side have been added; because this is interpolation, it undulates by up to about 20 K between the two points). Adding 0/10/30 % B2O3 to the cordierite type lowers the liquidus to 1468/1368/1168 °C. On the SiO2 side, where the liquidus is steep (12–15 % Na2O), a change of 0.5 mol% changes it by up to 45 K.

7.3 Glass-forming tendency Schematic

The speed of crystallization (ΔHm, σ, and the growth rate) cannot be fixed for an arbitrary batch, so from the structural rules (§5.4) we express the glass-forming tendency by a score s and show it in three levels:

(14) s=clamp(sN+sF+sM+sA+sMA+sB,−1.5,4.5)+s0, sN=clamp(1.5(NBO/T−1),−1.5,3), sF=clamp((0.55−F)/0.15,0,2) sM=clamp((xMgO−0.10)/0.10,0,2)×xMgOxMgO+xR2O, sA=clamp(xAl,high/0.08,0,3), sMA=clamp(xAl,Mg comp./0.15,0,1.5) sB=4clamp((R′−Rd1)/0.5,0,1)×xB2O3xSiO2+xB2O3, s0=4clamp((0.35−F)/0.10,0,1) A: s<1 (forms glass easily), B: 1≤s<2.5 (needs quenching), C: s≥2.5 (hard to form glass)

F is the mole fraction of the oxides that act as network formers (SiO2, B2O3, and four-coordinate Al2O3). Four-coordinate Al is counted only up to the amounts of SiO2 and B2O3 (Al alone does not form a network). The terms correspond, in order, to textbook factors that affect how easily a composition crystallizes: “the network is cut too much (the better connected it is, the more easily it forms glass),” “few formers,” “Mg, which has a large field strength, is the main modifier,” “uncompensated Al,” “Al compensated by Mg (the cordierite type),” “a borate rich in modifiers (non-bridging oxygen attached to three-coordinate B),” and “almost no network former.” The coefficients of sB and s0 are Tentative, and since only s0 is added outside the upper limit of 4.5, for a composition with no network former the rough guide to Rc becomes much larger than the upper limit of the slider (103 K/s). sN is continuous in NBO/T, so even within tendency A the rough guide to Rc differs with composition (1.6×10−4 for the DURAN type, 1.0×10−2 for the composition of Na2O·2SiO2, and 2.5×10−3 K/s for SiO2 75 and Na2O 25). In the cooling calculation, the rough guide to Rc is taken as 10−2 + 1.2s K/s, and Xext = 10−6(Rc/q)4 W(T). W is the fraction of the temperature range in which crystallization occurs (Tl–Tg) that has already been passed (a smooth step from 0 to 1). This is an approximation that borrows only the property Xext ∝ q−4 for cooling at a constant rate, and the numbers are meant to show the ranking. Examples: pure Al2O3, MgO, and Na2O have no network former, so s = 8.5 (rough guide to Rc 1.6×108 K/s), and MgAl2O4 has 8.0 (4.0×107 K/s); they crystallize even in water quenching (300 K/s) or at 103 K/s, and do not become glass even when quenched. Mg2SiO4 has s = 5.2 (C, 1.6×104 K/s; it crystallizes even in water quenching. In the literature it becomes glass only when a containerless droplet 1 mm across is cooled at 700 K/s (from the summary of Tangeman et al. 2001). In a crucible, crystals tend to form from the container wall, so the schematic value is on the large side), the composition of NaBO2 (Na2O 50 and B2O3 50) has 3.8 because sB takes effect (C, 4.0×102 K/s; it crystallizes in air cooling, whereas with 40 % Na2O the score is 0.2 and the tendency is A), a composition of Na2O 56, K2O 6.7, and SiO2 37 mol% has s ≈ 4.2 (C; a trace of crystals appears even in water quenching), one of SiO2 70 and MgO 30 has 1.8 (B), a slightly altered cordierite composition has 1.2 (B), and a flux-free batch of SiO2 60 and Al2O3 40 has 1.5 (B; the term sA for uncompensated Al takes effect).

8. Common Mistakes in Observation

9. Reference Compositions and Literature Values

Table 7. Reference compositions of the simulatorTg, Tx, Tc, and Tl are in K (DURAN and the Bisbrouck series are left in °C). KH = (Tx − Tg)/(Tl − Tx) (Hrubý’s index of glass stability). Treatment in ③: quantitative = the nucleation-and-growth model; schematic = §7.
NameCompositionCharacteristic temperatures, etc.③Source
Na2O·2SiO2mol% 33.3/66.7Tg 713, Tx 897, Tc 918, Tl 1148, KH 0.73QuantitativeJiusti et al. 2020
DURAN-type borosilicatemass% SiO2 81, B2O3 13, Na2O 4, Al2O3 2Tg 525 °C; 560/825/1260 °C at 1012/106.6/103 Pa sSchematic (the VFT equation is fitted to the three points of the datasheet: A −3.155, B 7254.8 K, T0 354.4 K)SCHOTT datasheet
Bisbrouck N26M0–N8M19Table 3Tg 523–642 °C, BO4 57–7 %. Homogeneous glass without nanocrystals according to EPMA and TEMSchematicBisbrouck et al. 2022
Albite typemol% Na2O 12.5, Al2O3 12.5, SiO2 75Tg 1086, Tm 1391, ΔHm 59.3 kJ/mol. No report of crystallization at ordinary pressure without seedsQuantitative (σ is the calculated value from the same paper)Zanotto & Cassar 2017
Cordierite typemol% MgO 22.2, Al2O3 22.2, SiO2 55.6Tg 1072, Tx 1203, Tc 1238, Tl 1740, KH 0.24QuantitativeJiusti et al. 2020
Na2O·SiO2 typemol% 50/50Melting point 1089 °C (general value). NBO/T = 2Schematic—
Pure SiO2, pure B2O3—Same values as in Binary Phase Diagrams ④QuantitativeThe phase diagram explanation, §7

10. Simplifications and Limits of the Model

Appendix A. What the Simulator (glassmelt-simulator-en.html) Shows

A.1 The whole page

There are three tabs: ① Prepare the batch (key batch), ② Melt (melt), and ③ Cool (cool). Adding #batch or the like to the end of the URL opens that tab. With ?only=cool only that tab is shown, and with ?embed=1 the headings and the explanatory text are hidden for an embedded display. The three tabs share one state (the batch, the raw materials, the arrangement of the grains in the mortar and the homogeneity, and the state of the melt), and what is changed in an earlier tab is carried over to the later tabs. When the batch, the raw materials, or the mass is changed, the results of ② and ③ are cleared and everything starts over. The animation stops when its tab is hidden, or when the screen (the browser tab) is hidden.

A.2 ① Prepare the batch

Controls. Choose a reference composition or set the batch with the six sliders (the 100 % total is kept by Eq. (2)). Switching between mol% and mass% displays the same composition on the other basis. Choose the MgO source (MgCO3 or MgO), the B2O3 source (H3BO3 or anhydrous B2O3), and the mass of glass to make (10–300 g). “Weigh” puts the raw materials on the balance one by one and then into the mortar, “Mix in the mortar” swaps the grains around, and “Put in the crucible” makes everything ready to go on to ②. “Keep this batch for comparison” leaves the previous batch in the graph as dotted lines.

Apparatus figure. The raw-material bottles (the height of the contents is proportional to the mass of the raw material), the electronic balance (the grams of the raw material being weighed), the mortar (the mixing time), and the platinum crucible. Internal figure. A schematic of the inside of the mortar, about 5 mm wide. The number of grains is proportional to the volume of the raw material (mass/density), and the raw materials are distinguished by color and shape. Immediately after weighing, the grains lie in layers in the order in which they were added. The homogeneity h is calculated by dividing a 24 × 10 grid into 6 × 5 sections and using the variation M of the proportions of the raw materials among the sections, as h = (M0 − M)/(M0 − Mr) (M0: for the layered state, Mr: for a random arrangement), and it is smoothed slightly for display. One frame is 2 s of mixing, and it comes to about 0.85 after about 10 minutes. Graph. The left graph shows the oxide composition (dark bars are mol%, light bars are mass%; the colors of the bars are the same as in the atomic figure of ②: SiO2 blue, Al2O3 green, MgO pink, B2O3 orange, Na2O yellow, K2O light purple), and the right graph shows the masses of the raw materials (the colored part is the oxide that remains, and the gray part is the gas that escapes). Readouts. The grams of each raw material, the total, the grams of CO2 and H2O, the mass balance (Eq. (4)), the estimate of the network (NBO/T, the fraction of four-coordinate Al, and N4; for the reference compositions the measured value is shown alongside), and the glass-forming tendency (§7).

A.3 ② Melt

Controls. Set the heating rate (1–20 K/min), the holding temperature (800–1600 °C; 1600 °C is assumed to be the upper limit of the furnace), and the holding time (0–10 h), and press “▶ Heat” to start heating. “Advance 30 min” and “Advance to the end” move time forward. If a condition is changed, the heating starts over from the beginning. The time of one frame is chosen from 0.25 to 30 minutes to suit the length of the schedule and is shown on the screen. When a reference composition is chosen, the default holding temperature and time are filled in (Na2O·2SiO2 1200 °C for 2 h, the DURAN type 1600 °C for 4 h, the cordierite type 1550 °C for 2 h, the albite type 1600 °C for 10 h, Bisbrouck N8M19 1500 °C (the other Bisbrouck series 1400 °C), and so on; all are tentative). For the albite type, about 4 % of the sand remains undissolved even after 10 h at the upper limit of the furnace (the melt is very viscous), so the simulator guides you to compare how hard it is to crystallize with the “Idealized” option in ③. When you come to ② after only weighing (and mixing) in ①, the powder is treated as already put into the crucible, and the step display of ① is advanced as well.

Apparatus figure. A cross-section of the electric furnace (heating elements, insulating bricks, thermocouple, exhaust port), the contents of the crucible (powder → melt; bubbles), the gas coming out (H2O or CO2), and the temperature controller. The color inside the furnace is a schematic of the temperature. Internal figure. “Grains, bubbles, cords” (about 5 mm wide; the spacing of the stripes is drawn to the same scale as λ, and the sizes of grains and bubbles are schematic) uses the arrangement of the grains of ① as it is, and shows the carbonates decomposing and shrinking, the liquid spreading, the sand grains dissolving and shrinking, the bubbles rising, and the stripes fading (the speed at which a bubble rises depends on the viscosity). The grains other than carbonates (silica sand, Al2O3, MgO, B2O3, and dehydrated H3BO3) shrink, once the liquid has formed, by the same rule as the undissolved solid u (size ∝ u1/3). As in ①, the legend is made for each raw material present in the batch, and the gas of the bubbles (CO2, H2O) is decided from the raw materials that release it. For a batch that contains no SiO2, the sentences do not say “sand” but use the names of the solids present in the batch (grains of Al2O3 and MgO, lumps of B2O3). The volatilization notice also lists only the components present in the batch (B2O3, Na2O, K2O). “Network of the melt” (a two-dimensional schematic; the width is 3–5 nm, taking the T–T distance as 0.31 nm) draws, with the color and size conventions of §5.1, an irregular network obtained from a hexagonal network in which each T (Si, Al, B) has three bonds in the plane, smoothed by Stone–Wales-type bond rotations (which create five- and seven-membered rings) and springs. The number of T atoms (50–60) is chosen for legibility, and the proportions of the kinds (Si, AlO4−, high-coordinate Al, BO3, BO4−) are matched to the rules of §5.4. For the non-bridging oxygen, the fraction NBO/T ÷ 4 of the four bonds is applied to the three bonds in the plane, and the number of T–O–T linkages to cut is round(0.375 N NBO/T) (N: the number of T atoms). Cutting one linkage produces two O− (R2O + ≡Si–O–Si≡ → 2 ≡Si–O− + 2R+), so the number of O− in the figure is 0.75 NBO/T per T atom, which is 3/4 of the NBO/T of the readout (the actual value). O− is also drawn as a red sphere like the bridging oxygen, and is connected by a black line to the T atom on one side only. The cations are placed to match the negative charges in the figure: one R+ for each O− (connected by a thin red line); for a linkage cut by MgO, the two O− are bent to the same side with one Mg2+ between them (≡Si–O− Mg2+ −O–Si≡); one R+ for each AlO4−, or one Mg2+ for two of them; and one R+ for each BO4− (these last two are charge compensation that does not create non-bridging oxygen, so they are connected to the T atom by a red dashed line). The ratio of R to Mg is the ratio of R2O to MgO that create non-bridging oxygen in §5.4 (for AlO4−, the ratio of what R2O and MgO have compensated), and the ratio of K to Na is K2O/(Na2O + K2O). In a composition without R2O, all the AlO4− is compensated in sets of two AlO4− and one Mg2+, so the number of AlO4− in the figure is made even (one is exchanged with another kind), and Na+ and K+ are not drawn. When there are more modifier ions than rings, all the cation spheres are reduced by the same factor, down to a minimum of 0.6 times (the diameters relative to O being at most 0.9 times for Na+, 1.02 times for K+, and 0.72 times for Mg2+). For this reason, in a composition with many modifier ions such as Na2O·2SiO2, Na+ looks almost as large as O. Once the batch is molten, the atoms vibrate, and bonds are rearranged at a frequency proportional to the mobility of the atoms (calculated from the viscosity): an O− cuts the neighboring T–O–T, and the original break is reconnected; the number of O− does not change, and a cation moves along with its partner O−. Rearrangements that would increase the overlap of the new O− with other atoms or cations, and rearrangements that would increase the number of nearby breaks, are not carried out (the same criterion as when the network is made; for a composition that has overlaps from the time it was made, rearrangements are made within the range that does not increase the overlaps). Since only the breaks at O− are drawn as rearrangements, for a composition without O− (pure SiO2, pure B2O3, the DURAN type, the albite type, the cordierite type, and so on) the network merely vibrates and no rearrangement is drawn, and for a composition in which O− occurs only in pairs bound to Mg2+, there is almost no non-overlapping place in the figure, so almost no rearrangement can be drawn (the heading says so; in an actual melt the bonds around Mg2+ are rearranged as well). In an actual melt, flow also occurs through rearrangement of the bonds of the bridging oxygen. The network figure is animated when the melt fraction is one half or more (the verdict of ② is neither “Still a powder” nor “Not fully melted”). On Reset or when the heating is restarted, the network is returned to its initial arrangement (the network just after it was made is kept, and it is restored to that). The frame of the network is the same size as in ③. Graph. The upper graph shows the temperature (solid line), the schedule (dashed line), and the mass of the sample (brown dashed line, Eq. (9)); the lower graph shows the undecomposed raw material, the undissolved solid u, the bubbles b (on a scale in which the larger of the maximum values of the current and the kept results is 1; the maximum is written in the legend), and the strength of the cords. “Keep this result for comparison” overlays the previous result as pale lines (when it was kept partway through the heating, the legend says “stopped partway at … h”). Verdict. When the holding is over, the verdict is “Not fully melted” if the melt fraction is below 50 %; stones are reported if u > 0.02, bubbles if b > 0.01, inhomogeneity if the cords > 0.1, and a composition shift due to volatilization > 0.3 mol% is shown as a caution. When the batch is not fully melted, the reason is shown separately: the temperature of the first liquid is not reached; there is no flux and the holding temperature is below the liquidus (the guidance differs according to whether the liquidus is above or below the furnace limit of 1600 °C); there is little flux and the liquid is capped by Eq. (6a); or the melt is viscous and dissolution is slow (when the holding temperature is below the liquidus, it is also shown that part remains solid even at equilibrium; for a congruently melting composition such as Na2Si2O5, almost no liquid actually forms below the liquidus). The reason why cords remain is likewise split into two cases: “how it was mixed” when the homogeneity h is small, and “slow diffusion in a viscous melt” when it was mixed well. Readouts. The first liquid (the mol% of flux), the liquidus (with a distinction between literature value, general value, interpolation, and assumption; for interpolation, the names of the nearby reference points), Leq at the holding temperature (whether it is capped by the amount of flux), and the origin of the viscosity (a literature equation, a VFT equation fitted to the three points of the datasheet, the empirical equation of Fluegel 2007, or interpolation from the reference compositions). Redoing “Weigh,” “Mix in the mortar,” or “Put in the crucible” clears the results of ② and ③.

A.4 ③ Cool

Controls. Choose a cooling-method button (Table 5) or the cooling-rate slider (10−4–103 K/s), and press “▶ Cool” to start cooling. If there is a result from ②, its holding temperature, undissolved material, bubbles, and cords are carried over; if there is none (or if the “Idealized” box is checked), the cooling starts from a homogeneous melt above the liquidus. “Keep this result for comparison” keeps up to four results in the table and the graph (if the cooling is not yet finished, the calculation is carried through to the end at once under those conditions and then kept). The display shows the whole of the cooling in about 8 seconds, and the elapsed time (in real seconds) appears in the readouts.

Apparatus figure. The cooling method (in water, on a metal plate, on a brick, or in the furnace) and the temperature of the sample; at the upper right, the appearance of the resulting sample (glass is transparent, crystals are white grains, and bubbles and cords are also drawn); and at the lower right, a schematic of the X-ray diffraction. Internal figure. It is the same as the first network of ② (the same arrangement for the same composition and the same screen width; what was rearranged in ② returns to the initial arrangement in ③). As the temperature falls, the mobility of the atoms (the bar at the lower right; calculated from the viscosity) decreases, and it stops once Tg is passed (the positional disorder remains). When crystals form, inside the light green region (whose area fraction is X) the network becomes a regular arrangement of regular hexagons, and the kinds of T, O−, and cations are arranged periodically with their numbers unchanged. O− is not removed (a crystal that contains modifiers, such as Na2Si2O5, also has one non-bridging oxygen per Si; the difference between a crystal and a glass is not the presence of non-bridging oxygen but the long-range order). When only part of the sample has crystallized, the pairs of T atoms whose bonding partner changed through a Stone–Wales rotation are brought into the same region, and only the glass nodes within 2.5 d of the crystalline region are smoothed again with the same springs as when the network is made (so that the bonds at the boundary are not stretched long and the O atoms do not overlap). The cations on the glass side are placed away from the atoms of the crystalline region. When a set of two AlO4− and one Mg2+ straddles the boundary so that only one side is visible, the sets are re-formed among the visible AlO4− that have no partner (in order of proximity; when a set would become long, it is also exchanged with nearby sets that are being drawn), and the charge in the figure is balanced (an AlO4− that does not enter a set is compensated by R+, and in a composition without R2O one T atom at the boundary is put into the other region so that the number is made even). For a sample whose melt did not reach one half in ② (the verdict of ② is “Still a powder” or “Not fully melted”), the glass network is not drawn, and the melted fraction and a note to that effect are shown. Graph. In the “CCT/TTT with cooling curve” graph, the horizontal axis is the logarithm of the time since the temperature fell below the liquidus, and the vertical axis is the temperature. The red solid line is the line along which X = 10−6 is reached during cooling (CCT, §6), the red dashed line is the line along which X = 50 % is reached, the red dotted line is the TTT curve for isothermal holding (for reference), and the blue line is the cooling curve of the current run. For an arbitrary batch, the red lines are drawn by the same method from the schematic equations of §7.3 (the length of the dashes is changed to distinguish them as schematic). If the cooling curve touches the red solid line, the crystallized fraction reaches 10−6 or more (the same as the verdict). Crossing the isothermal dotted line does not necessarily mean that crystallization occurs. The “Enthalpy vs. T (reference compositions only)” graph is for the reference compositions only, and it overlays the path H = (1 − X)Hglass + X Hcrystal on the lines of the equilibrium liquid and the crystal. Verdict. With the volume fraction of crystals (including the undissolved material of ②) X < 10−6 the verdict is glass (the same criterion as Rc and the red solid line of the CCT); below 1 % it is nearly glass (with a trace of crystals); below 50 % it is partly crystallized; above that it is crystallized (below 99.9 % it is “mostly crystallized”); when crystallization during cooling is below 1 % and undissolved material is the main part, it is stones; and if the batch was not melted in ②, it is not melted. When the batch is not melted, the cooling is not calculated, and neither the values of Tg and Tf nor the CCT/TTT graph is shown (because the glass transition is the freezing of a melt). For a composition whose liquidus is above the furnace limit, the simulator explains that it will not melt even if the conditions of ② are changed, and that the “Idealized” check lets you see the case cooled from a melt. Tg and Tf are also not shown when most of the sample (50 % or more) crystallizes during cooling. “There is no glass transition” is written only when the whole sample (99.9 % or more) has crystallized; below 99.9 % it says “mostly crystalline (the remaining glass phase has a Tg)” (for partial crystallization, the note says that the value is that of “the remaining glass phase”). In this case the Tg(q) line and its legend are also not drawn in the CCT/TTT graph. A numerical value of Tg is given in the text only for the reference compositions (Tg(q) of the model) and for compositions that have a literature Tg (the DURAN type and the Bisbrouck series; the rough guide from the viscosity equation and the literature value are shown side by side); for all other compositions it says “There is no literature Tg. The rough guide from the viscosity equation is Tf ≈ … °C (schematic)” (for a batch for which only one reference composition can be used, “differences due to composition cannot be shown” is added, and for a batch whose viscosity cannot be estimated (§7.1), Tf is not shown). When the rough guide to Rc of an arbitrary batch exceeds 104 K/s, it says “far above the upper limit of the cooling rate: it will not become glass even with laboratory quenching.” For a sample whose melt did not reach one half in ②, no cooling is calculated, so the heading of the apparatus figure also becomes “Not fully melted (does not become glass)” (and the explanation for pouring into water becomes “sintered lump of powder (not glass)”). On a narrow screen (narrower than 560 px), the comparison table is shown as one card per result. The q of a cooling-method button is exactly the value in Table 5 (the slider knob is rounded to steps of 0.05, but the heading and the calculation use the value in the table).

A.5 Checking the calculations

The calculation part of the simulator (the functions that do not use the DOM) was extracted and run with node, and the following was confirmed (2026-09-24): the conversion factors of Table 1; the round trip between mol% and mass% for the reference compositions, and their totals; the mass balance of the batch (raw materials − gas = glass); the proportional distribution and the upper and lower limits of Eq. (2) (clamping of non-numeric input and out-of-range values); the reproduction of R′ in Table 2; Rc, Umax, and Tg in Table 4; the verdicts in Table 5; the monotonic decrease of X as q is raised; the temperature of half decomposition in Eq. (5) (803 °C at 10 K/min), the Stokes and Eyring values, and the consistency of the mass in Eq. (9). The repairs made after the first independent verification (2026-09-24) added the following to the checks: the capping of Eq. (6a) (Leq increases monotonically as flux is increased; Al2O3 and MgO do not melt with 3 % flux; the assumed liquidus of the DURAN type is 1260 °C), all the Al in the cordierite type being four-coordinate, the agreement of the verdict with Rc (the verdict switches between glass and not glass at 10±0.05 times Rc), σ and Umax of the albite type, and the mass balance of pure B2O3 made from H3BO3, which is 56.3 % of the batch mass (all 150 items passed). The repairs made after the second independent verification added the following: the transcription of Eq. (12) (the isokom temperatures of Na2O·2SiO2 and the DURAN type), that five flux-free refractory batches do not melt at 1600 °C, that K2O·2SiO2 melts at 1400 °C, that the viscosity and the liquidus do not jump at each 0.5 mol% step in SiO2–Na2O and SiO2–K2O (including the neighborhood of 22 % Na2O), that the VFT equation is normal and continuous for 20000 random batches, that the values are drawn toward the literature values around the reference compositions, that the rough guide to Rc differs with composition within tendency A, and that the largest q that touches the CCT line agrees with Rc (all 182 items passed). The repairs made after the third independent verification added the following: that pure Al2O3, MgO, MgAl2O4, Na2O, and K2O crystallize when cooled at 103 K/s, that Tl and T(η = 103 Pa s) do not rise when B2O3 is added to the six base compositions from 0 to 30 % (and also Na2O from 0 to 20 % to the cordierite type and enstatite) in steps of 0.5 mol%, that the cordierite type with 3–20 % B2O3 added melts homogeneously at 1500 °C for 2 h, that Tf changes with composition in the Na2O–B2O3 system, and N4 of the NaBO2 composition and its crystallization on slow cooling, that a batch of Na2O alone becomes a freely flowing melt, the reproduction of the two Kracek points, and that the values are sensible for 20000 random batches (together with the previous 182 items, all 233 items passed).

Items to Confirm (with Prof. Mochizuki)

  1. Compositions handled in classes and student experiments: is the range of the six oxides acceptable, or should CaO and Li2O (soda-lime glass, and the existing Li2O·2SiO2) be added? How the reference compositions are chosen (Table 7).
  2. Raw materials: the grain size of the silica sand; MgO or MgCO3; H3BO3 or anhydrous B2O3; whether Na2CO3 is anhydrous. The amount of glass to make (default 100 g).
  3. The crucible (platinum, alumina, and so on) and the upper limit temperature of the electric furnace (assumed to be 1600 °C).
  4. The actual conditions for the heating rate, holding temperature, and holding time (all default values are tentative).
  5. The cooling methods (water, metal plate, crucible and all, furnace cooling, slow cooling) and the size of the sample. The rough cooling rates in Table 5. Whether to anneal (holding near the annealing point).
  6. Observation and evaluation: which of appearance, X-ray diffraction, and differential scanning calorimetry to carry out.
  7. The rough decomposition temperatures (table: H3BO3 180 °C, MgCO3 500 °C, Na2CO3 800 °C, K2CO3 830 °C), and the reference values for sand dissolution, bubbles, cords, and volatilization (the tentative values of §4).
  8. The scope of the structural explanation: whether the R′ rule and the Dell–Bray rules, and the treatment of Mg (the compensation ratio φ = MgO/(MgO + R2O)), may be taught in this form.
  9. The schematic of the amount of liquid in ② (Eq. (6a)): the flux fraction of the first liquid, 0.26 (alkali carbonate) and 1 (B2O3), and the assumption of using the temperature at η = 103 Pa s in place of the liquidus for compositions without a literature liquidus.
  10. The form and the boundaries (A/B/C) of the score of the glass-forming tendency (§7.3), and the rough guide to Rc, 10−2 + 1.2s K/s.
  11. Confirm the general liquidus values in Table 6 (Na2O·SiO2 1089 °C, eutectic 793 °C, K2O·2SiO2 1045 °C, K2O·4SiO2 770 °C, mullite 1850 °C, forsterite 1890 °C, enstatite 1557 °C, the two-liquid region of MgO–SiO2 1695 °C, spinel 2135 °C) against a collection of phase diagrams. Whether the interpolation (Eq. (13)) may be shown in this form.
  12. Whether it is acceptable to use the empirical equation of Fluegel 2007 for the viscosity of an arbitrary batch and to switch to interpolation from the reference compositions outside its range (§7.1).
  13. The default of ② for the albite type (a little remains undissolved even at 1600 °C for 10 h) and the criterion for sand dissolution (20 minutes, tentative).
  14. The third region of Dell–Bray (the form in which N4 is lowered along a straight line for R > Rd1) and the coefficients of the tendency terms sB (borates rich in modifiers) and s0 (almost no network former) (all tentative). Whether alkali-rich borates (such as the composition of NaBO2) may be treated in this form.
  15. The upper limits on the effect of flux (0.5 and 0.1 in §7.1, 1000 K and 300 K in §7.2), and the VFT equation that regards a batch with few network formers as a freely flowing melt (all tentative). Whether it is acceptable to display a batch consisting only of alkali or B2O3 with MgO and Al2O3 as “outside the scope of this glass teaching tool.”
  16. The liquidus on the SiO2 side of the Na2O–SiO2 system: the two points of Kracek 1930 (870 °C at 75.5 mass%, 1470 °C at 88.7 mass%) are values confirmed in the summary in a doctoral thesis (Ryś 2007), and the original has not been checked. The midpoint of the two points (1170 °C) is a tentative value obtained by treating the line as straight.

References

Among the numbers on this page, the “literature values” are the values from the sources listed here (or values calculated from them). Among the model equations, those of §4, §7.2, and §7.3 are educational schematics and not reproductions of a specific experiment (the empirical equation for viscosity in §7.1 is that of Fluegel 2007). Fig. 1 is our own drawing.

Companion simulator: Making Glass — Mix Oxides, Melt, and Cool. Principles of the glass transition and crystallization: the explanation of binary phase diagrams, §7.

  1. A. Fluegel, “Glass viscosity calculation based on a global statistical modelling approach”, Glass Technol.: Eur. J. Glass Sci. Technol. A 48, 13–30 (2007). Table 3 (the maximum concentrations in the source data) and Tables 4–6 (the coefficients) were used. glassproperties.com/viscosity (the author’s public version)
  2. J. Jiusti, D. R. Cassar, E. D. Zanotto, “Which glass stability parameters can assess the glass-forming ability of oxide systems?”, Int. J. Appl. Glass Sci. 11, 612 (2020). doi:10.1111/ijag.15416 (the arXiv:2001.01113 version was consulted; Table 2 for Tg, Tx, Tc, Tl, Umax, and VFT)
  3. E. D. Zanotto, D. R. Cassar, “The microscopic origin of the extreme glass-forming ability of Albite and B2O3”, Sci. Rep. 7, 43022 (2017). doi:10.1038/srep43022
  4. N. Bisbrouck, M. Micoulaut, J. M. Delaye, S. Gin, F. Angeli, “Structure–property relationship and chemical durability of magnesium-containing borosilicate glasses with insight from topological constraints”, npj Mater. Degrad. 6, 58 (2022). doi:10.1038/s41529-022-00268-8
  5. N. Bisbrouck et al., J. Am. Ceram. Soc. 104, 4518 (2021). doi:10.1111/jace.17876 (consulted only to the extent of the abstract)
  6. S. Bruns et al., “Impact of Al2O3 addition on the structure of sodium borosilicate glasses”, Front. Mater. 7, 189 (2020). doi:10.3389/fmats.2020.00189
  7. X. Lu, L. Deng, J. Du, J. D. Vienna, “Predicting boron coordination in multicomponent borate and borosilicate glasses using analytical models and machine learning”, J. Non-Cryst. Solids 553, 120490 (2021). doi:10.1016/j.jnoncrysol.2020.120490 (a summary of the Dell–Bray and Du–Stebbins rules)
  8. A. Navrotsky, O. J. Kleppa, “Estimate of enthalpies of formation and fusion of cordierite”, J. Am. Ceram. Soc. 56, 198 (1973). doi:10.1111/j.1151-2916.1973.tb12455.x (abstract: heat of fusion 54.0 ± 2.0 kcal/mol)
  9. M. W. Chase, Jr., NIST-JANAF Thermochemical Tables, 4th ed., J. Phys. Chem. Ref. Data Monograph 9 (1998). The table for Na2Si2O5 (cr,l) (35.564 kJ/mol for melting at 1147 K, and Cp of the liquid 261.2 J/(mol K)). janaf.nist.gov/tables/Na-030.html
  10. SCHOTT, DURAN (borosilicate glass 3.3) Technical Data (Tg, characteristic temperatures of the viscosity, density).
  11. K. S. Hong, R. F. Speyer, “Thermal analysis of reactions in soda-lime silicate glass batches containing melting accelerants: I, one- and two-component systems”, J. Am. Ceram. Soc. 76, 598 (1993). K. S. Hong, S. W. Lee, R. F. Speyer, ibid. 76, 605 (1993) (consulted only to the extent of the abstract).
  12. C. Huber et al., Energy Sci. Eng. 8, 1650 (2020). doi:10.1002/ese3.622 (multistep dehydration of boric acid; consulted only to the extent of the abstract)
  13. NIST, Standard Reference Material 717a (Borosilicate Glass) certificate (caution on volatilization at high temperature).
  14. L. M. Thompson, J. F. Stebbins, Am. Mineral. 96, 841 (2011). doi:10.2138/am.2011.3680 (compensating ions and five-coordinate Al; consulted only to the extent of the abstract)
  15. C. J. Tangeman et al., Geophys. Res. Lett. 28, 2517 (2001). doi:10.1029/2000GL012222 (Mg2SiO4 glass; consulted only to the extent of the summary)
  16. A. Hrubý, Czech. J. Phys. B 22, 1187 (1972). D. Turnbull, Contemp. Phys. 10, 473 (1969) (confirmed from the citation in Jiusti et al. 2020).
  17. F. C. Kracek, “The system sodium oxide–silica”, J. Phys. Chem. 34, 1583–1598 (1930). M. Ryś, Investigation of Thermodynamic Properties of Alkali Metals in Oxide Systems Relevant to Coal Slags, doctoral thesis, RWTH Aachen (2007); confirmed in its literature survey (a summary of the liquidus of the Na2O–SiO2 system). d-nb.info/989424367
  18. W. H. Zachariasen, “The atomic arrangement in glass”, J. Am. Chem. Soc. 54, 3841–3851 (1932). doi:10.1021/ja01349a006 (the schematic of the two-dimensional irregular network)
  19. R. D. Shannon, “Revised effective ionic radii and systematic studies of interatomic distances in halides and chalcogenides”, Acta Cryst. A 32, 751–767 (1976). doi:10.1107/S0567739476001551 (ionic radii)
  20. Mochizuki Research Group, Explanation of Binary Phase Diagrams, §7 Glass Transition and Crystallization (derivation of the cooling model; sources and assumptions for the parameters of SiO2 and B2O3).

Mochizuki Research Group, Tokyo University of Science / Companion simulator: glassmelt-simulator-en.html. Until the items to confirm are resolved, tentative values are displayed as tentative.