Model documentation

Each registered model in the Atlas modelling framework with its source reference, assumptions, limitations, and implementation status.

Melt & thermal

74 models

Melt viscosity

High-temperature viscosity

AvailableNeeds review

A calculator for estimating high-temperature melt viscosity over temperature from selected simulant oxide chemistry.

Model family

High-temperature silicate melt model

Source

Implemented methods: Shaw, Hui-Zhang, GRD08, Duan, Sehlke, Langhammer (2021), Visc_Calc (Langhammer et al. 2022), and i-melt (Le Losq et al.)

Source note

Only composition-derived equations are enabled. Measured VFT curves remain excluded.

Assumptions

Input compositions are wt% oxide values using an FeO-basis chemistry table.
Iron input is shown as FeO and Fe2O3; Fe2O3 is converted to FeO-equivalent iron inside the viscosity equations.
Missing optional oxides such as H2O and F2O_1 are treated as 0 wt% and shown in warnings.

Limitations

Powder flow viscosity and high-temperature melt viscosity will be treated as different properties.
Measured VFT, Cassar, GlassNet (https://github.com/drcassar/glasspy, Cassar 2023), and gpvisc are not enabled in this calculator. GlassNet integration is saved for later.
Literature comparison, fitting workflows, and RMSE validation plots are reserved for a later version.

Parameters

Symbol	Label	Unit	Description
T	Temperature	K or deg C	The independent variable used for the viscosity curve. Atlas will show any unit conversion before running a model.
eta	Dynamic viscosity	Pa s or log10(Pa s)	The target output for high-temperature melt viscosity models.
x_i	Oxide fraction	wt%	Composition terms such as SiO2, Al2O3, CaO, MgO, FeO, Na2O, K2O, TiO2, and related model inputs.

Required inputs

Simulant composition — wt%
Temperature range — K or deg C

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Melt density

AvailableNeeds review

Temperature- and pressure-dependent silicate melt density from oxide composition using Lange & Carmichael (1990).

Model family

Composition-dependent silicate melt density

Source

Lange & Carmichael (1990) — DensityX model

Assumptions

Input compositions are wt% oxide values.
Total iron is treated as Fe2O3 for the density calculation.
Pressure defaults to 1 bar (ambient).

Limitations

Valid for silicate melts above the liquidus only.
Crystal-bearing magmas and partial melts are not represented.
Only the 9-oxide subset from the original paper is supported.

Parameters

Symbol	Label	Unit	Description
T	Temperature	K	Melt temperature.
P	Pressure	bar	Ambient pressure in bar (default 1).
rho	Density	g/cm3	Calculated silicate melt density.

Required inputs

Simulant composition — wt%
Temperature — K

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Liquidus temperature

AvailableNeeds review

Predicts liquidus temperature as the maximum crystallization temperature of 9 mineral phases (pseudoliquidus approach) from oxide composition using Nathan & Van Kirk (1978) regression coefficients.

Model family

Composition-dependent liquidus prediction

Source

Nathan & Van Kirk (1978) — Pseudoliquidus mineral regression

Source note

Regression coefficients for 9 mineral phases (magnetite, olivine, hypersthene, augite, quartz, plagioclase, orthoclase, leucite, nepheline). The liquidus is the highest of the 9 calculated phase temperatures.

Assumptions

Input compositions are wt% oxide values.
The liquidus is the maximum of 9 mineral-phase regression equations.
Applies to anhydrous silicate systems only.

Limitations

Does not account for volatile-bearing compositions.
Regression coefficients are calibrated on terrestrial basalt suites.
Crystal fractionation and undercooling effects are not modelled.

Parameters

Symbol	Label	Unit	Description
T_liq	Liquidus temperature	K	Calculated liquidus temperature.

Required inputs

Simulant composition — wt%

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Specific heat capacity

Heat capacity

AvailableNeeds review

Specific heat capacity of silicate melts and glasses from oxide composition using Stebbins (1984) and Bychkov & Koptev-Dvornikov (2019).

Model family

Silicate melt heat capacity

Source

Stebbins et al. (1984), Bychkov & Koptev-Dvornikov (2019)

Source note

Independently published Cp models for glass, liquid, and fully molten silicate phases. Berman (1988) crystalline Cp is available in the hybrid Cp model via mineralogy input.

Assumptions

Input compositions are wt% oxide values.
Stebbins glass Cp is valid below the glass transition.
Stebbins liquid Cp is valid above the liquidus.
Bychkov melt Cp applies to fully molten silicate mixtures.

Limitations

Model validity ranges depend on the original calibration data.
Phase transition enthalpies are only included in the hybrid Cp model.

Parameters

Symbol	Label	Unit	Description
T	Temperature	K	Temperature for the Cp calculation.
Cp	Specific heat capacity	J/kgK	Calculated specific heat capacity.

Required inputs

Simulant composition — wt%
Temperature — K

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Surface tension

AvailableNeeds review

Surface tension of silicate melts using Murase & McBirney (1973) linear regression with temperature. Two model variants: Slag (SLS) and Gob (GOB).

Model family

Silicate melt surface tension

Source

Murase & McBirney (1973); Kucuk et al. (1999) regression planned.

Source note

Murase model computes surface tension purely from temperature; Kucuk composition-dependent regression is planned.

Assumptions

Surface tension decreases linearly with temperature per Murase & McBirney (1973).

Limitations

Does not yet use composition-dependent Kucuk regression.
Atmosphere effects not included.

Parameters

Symbol	Label	Unit	Description
gamma	Surface tension	N/m	Surface tension of the simulant melt.

Required inputs

Simulant composition (optional) — wt%

Modelling outputs are research aids, not certified material specifications.

Glass transition temperature

AvailableNeeds review

Glass transition temperature Tg from empirical additive wt% model (Mazurin 2007, Fluegel 2007). Pure SiO₂ gives ≈1100°C.

Model family

Composition-dependent glass transition

Source

Mazurin (2007); Fluegel (2007); Volf (1988)

Source note

Empirical linear additive model on wt% basis.

Assumptions

Input compositions are wt% oxide values.
Tg is composition-dependent only.
Standard DSC heating rate assumed.

Limitations

±50 °C typical error.
Simplified linear model; does not capture non-linear modifier interactions.
Does not account for thermal history or heating rate.

Parameters

Symbol	Label	Unit	Description
Tg	Glass transition temperature	K	Glass transition temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Coefficient of thermal expansion

AvailableNeeds review

Linear CTE (20–100°C) calculated from oxide composition via Winkelmann & Schott (1893) additive model.

Model family

Composition-dependent thermal expansion

Source

Winkelmann & Schott (1893), compiled by Fluegel (2007)

Source note

Additive model valid for silicate glasses; FeO coefficient estimated.

Assumptions

Input compositions are wt% oxide values.
CTE is linear sum of oxide weight-fraction coefficients (Winkelmann & Schott).
Room-temperature CTE at 20–100°C.

Limitations

Does not predict temperature dependence of CTE.
Unsupported oxides (e.g., P2O5, Cr2O3 above standard coverage) are excluded with warning.
Simple additive model; Fluegel (2007) quadratic/interaction model is more accurate.

Parameters

Symbol	Label	Unit	Description
CTE	Thermal expansion coefficient	10^-6/K	Coefficient of thermal expansion of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal conductivity

AvailableNeeds review

First-order estimate of room-temperature thermal conductivity from an empirical correlation with SiO₂ weight fraction (Ratcliffe 1963; Fluegel 2006 review). Pure fused silica → 1.40 W/mK; modifier oxides reduce conductivity proportionally.

Model family

Composition-dependent thermal conductivity

Source

Ratcliffe (1963); Fluegel (2006) review

Source note

SiO₂-wt% linear correlation, ±0.15 W/mK typical error.

Assumptions

Input compositions are wt% oxide values.
Thermal conductivity depends primarily on SiO₂ content at room temperature.

Limitations

Room temperature (300 K) only.
±0.15 W/mK typical error.
Does not capture individual oxide effects.
Extrapolation unreliable below 40 wt% SiO₂.

Parameters

Symbol	Label	Unit	Description
k	Thermal conductivity	W/mK	Thermal conductivity at 300 K.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal shock resistance

AvailableReviewed

Thermal shock resistance parameter from oxide composition using GlassNet.

Model family

Composition-dependent thermal shock

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Empirical parameter, not a fundamental property.

Parameters

Symbol	Label	Unit	Description
TSR	Thermal shock resistance	K	Thermal shock resistance of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Annealing temperature

AvailableNeeds review

Annealing temperature (log η = 13 Pa·s) from Tg + 30°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent annealing point

Source

Mazurin (2007); standard glass technology

Source note

Derived from Tg with standard offset.

Assumptions

Input compositions are wt% oxide values.
T_anneal ≈ Tg + 30°C.

Limitations

±50 °C typical error.
Offset varies with composition; 30°C is a nominal value.

Parameters

Symbol	Label	Unit	Description
T_anneal	Annealing temperature	K	Annealing temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Strain temperature

AvailableNeeds review

Strain temperature (log η = 14.5 Pa·s) from Tg − 30°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent strain point

Source

Mazurin (2007); standard glass technology

Source note

Derived from Tg with standard offset.

Assumptions

Input compositions are wt% oxide values.
T_strain ≈ Tg − 30°C.

Limitations

±50 °C typical error.
Offset varies with composition; 30°C is a nominal value.

Parameters

Symbol	Label	Unit	Description
T_strain	Strain temperature	K	Strain temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Softening temperature

AvailableNeeds review

Dilatometric softening temperature from Tg + 10°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent softening point

Source

Mazurin (2007); standard glass technology

Source note

Derived from Tg with standard offset.

Assumptions

Input compositions are wt% oxide values.
T_soft ≈ Tg + 10°C (dilatometric).

Limitations

±50 °C typical error.
Offset varies with composition.

Parameters

Symbol	Label	Unit	Description
T_soft	Softening temperature	K	Softening temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Littleton softening point

AvailableNeeds review

Littleton softening point (log η = 7.6 Pa·s) from Tg + 220°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent Littleton point

Source

Mazurin (2007); ASTM C338

Source note

Derived from Tg with standard offset for fibre elongation method.

Assumptions

Input compositions are wt% oxide values.
T_Littleton ≈ Tg + 220°C.

Limitations

±50 °C typical error.
Offset varies with melt fragility.

Parameters

Symbol	Label	Unit	Description
T_Littleton	Littleton softening point	K	Littleton softening point of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Melting temperature

AvailableNeeds review

Melting temperature from empirical additive wt% model. Pure SiO₂ gives 1700°C; modifiers reduce Tmelt proportionally.

Model family

Composition-dependent melting point

Source

Schairer (1942); Levin (1964); Mazurin (2007)

Source note

Empirical additive model approximating liquidus of primary crystalline phase.

Assumptions

Input compositions are wt% oxide values.
Tmelt approximates the liquidus of the primary silicate phase.

Limitations

±75 °C typical error.
Simplified linear model; eutectic minima not captured exactly.

Parameters

Symbol	Label	Unit	Description
Tmelt	Melting temperature	K	Melting temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T0 (log η = 0) (GlassNet)

AvailableReviewed

T0 (log η = 0) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T0	T0 (log η = 0)	—	T0 (log η = 0) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T1 (log η = 1) (GlassNet)

AvailableReviewed

T1 (log η = 1) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T1	T1 (log η = 1)	—	T1 (log η = 1) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T2 (log η = 2) (GlassNet)

AvailableReviewed

T2 (log η = 2) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T2	T2 (log η = 2)	—	T2 (log η = 2) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T3 (log η = 3) (GlassNet)

AvailableReviewed

T3 (log η = 3) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T3	T3 (log η = 3)	—	T3 (log η = 3) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T4 (log η = 4) (GlassNet)

AvailableReviewed

T4 (log η = 4) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T4	T4 (log η = 4)	—	T4 (log η = 4) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T5 (log η = 5) (GlassNet)

AvailableReviewed

T5 (log η = 5) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T5	T5 (log η = 5)	—	T5 (log η = 5) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T6 (log η = 6) (GlassNet)

AvailableReviewed

T6 (log η = 6) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T6	T6 (log η = 6)	—	T6 (log η = 6) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T7 (log η = 7) (GlassNet)

AvailableReviewed

T7 (log η = 7) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T7	T7 (log η = 7)	—	T7 (log η = 7) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T8 (log η = 8) (GlassNet)

AvailableReviewed

T8 (log η = 8) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T8	T8 (log η = 8)	—	T8 (log η = 8) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T9 (log η = 9) (GlassNet)

AvailableReviewed

T9 (log η = 9) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T9	T9 (log η = 9)	—	T9 (log η = 9) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T10 (log η = 10) (GlassNet)

AvailableReviewed

T10 (log η = 10) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T10	T10 (log η = 10)	—	T10 (log η = 10) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T11 (log η = 11) (GlassNet)

AvailableReviewed

T11 (log η = 11) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T11	T11 (log η = 11)	—	T11 (log η = 11) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T12 (log η = 12) (GlassNet)

AvailableReviewed

T12 (log η = 12) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T12	T12 (log η = 12)	—	T12 (log η = 12) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 773K (GlassNet)

AvailableReviewed

log₁₀(η) at 773K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity773K	log₁₀(η) at 773K	—	log₁₀(η) at 773K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 873K (GlassNet)

AvailableReviewed

log₁₀(η) at 873K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity873K	log₁₀(η) at 873K	—	log₁₀(η) at 873K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 973K (GlassNet)

AvailableReviewed

log₁₀(η) at 973K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity973K	log₁₀(η) at 973K	—	log₁₀(η) at 973K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1073K (GlassNet)

AvailableReviewed

log₁₀(η) at 1073K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1073K	log₁₀(η) at 1073K	—	log₁₀(η) at 1073K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1173K (GlassNet)

AvailableReviewed

log₁₀(η) at 1173K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1173K	log₁₀(η) at 1173K	—	log₁₀(η) at 1173K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1273K (GlassNet)

AvailableReviewed

log₁₀(η) at 1273K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1273K	log₁₀(η) at 1273K	—	log₁₀(η) at 1273K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1373K (GlassNet)

AvailableReviewed

log₁₀(η) at 1373K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1373K	log₁₀(η) at 1373K	—	log₁₀(η) at 1373K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1473K (GlassNet)

AvailableReviewed

log₁₀(η) at 1473K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1473K	log₁₀(η) at 1473K	—	log₁₀(η) at 1473K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1573K (GlassNet)

AvailableReviewed

log₁₀(η) at 1573K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1573K	log₁₀(η) at 1573K	—	log₁₀(η) at 1573K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1673K (GlassNet)

AvailableReviewed

log₁₀(η) at 1673K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1673K	log₁₀(η) at 1673K	—	log₁₀(η) at 1673K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1773K (GlassNet)

AvailableReviewed

log₁₀(η) at 1773K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1773K	log₁₀(η) at 1773K	—	log₁₀(η) at 1773K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1873K (GlassNet)

AvailableReviewed

log₁₀(η) at 1873K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1873K	log₁₀(η) at 1873K	—	log₁₀(η) at 1873K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 2073K (GlassNet)

AvailableReviewed

log₁₀(η) at 2073K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity2073K	log₁₀(η) at 2073K	—	log₁₀(η) at 2073K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 2273K (GlassNet)

AvailableReviewed

log₁₀(η) at 2273K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity2273K	log₁₀(η) at 2273K	—	log₁₀(η) at 2273K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 2473K (GlassNet)

AvailableReviewed

log₁₀(η) at 2473K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity2473K	log₁₀(η) at 2473K	—	log₁₀(η) at 2473K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Tg (K) (GlassNet)

AvailableReviewed

Tg (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tg	Tg (K)	—	Tg (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Tmelt (K) (GlassNet)

AvailableReviewed

Tmelt (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tmelt	Tmelt (K)	—	Tmelt (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Tliquidus (K) (GlassNet)

AvailableReviewed

Tliquidus (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tliquidus	Tliquidus (K)	—	Tliquidus (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Littleton point (K) (GlassNet)

AvailableReviewed

Littleton point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TLittletons	Littleton point (K)	—	Littleton point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Annealing point (K) (GlassNet)

AvailableReviewed

Annealing point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TAnnealing	Annealing point (K)	—	Annealing point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Strain point (K) (GlassNet)

AvailableReviewed

Strain point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tstrain	Strain point (K)	—	Strain point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Softening point (K) (GlassNet)

AvailableReviewed

Softening point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tsoft	Softening point (K)	—	Softening point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Dilatometric softening (K) (GlassNet)

AvailableReviewed

Dilatometric softening (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TdilatometricSoftening	Dilatometric softening (K)	—	Dilatometric softening (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 293K (g/cm³) (GlassNet)

AvailableReviewed

Density at 293K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density293K	Density at 293K (g/cm³)	—	Density at 293K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1073K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1073K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1073K	Density at 1073K (g/cm³)	—	Density at 1073K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1273K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1273K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1273K	Density at 1273K (g/cm³)	—	Density at 1273K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1473K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1473K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1473K	Density at 1473K (g/cm³)	—	Density at 1473K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1673K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1673K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1673K	Density at 1673K (g/cm³)	—	Density at 1673K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Thermal conductivity κ (W/mK) (GlassNet)

AvailableReviewed

Thermal conductivity κ (W/mK) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
ThermalConductivity	Thermal conductivity κ (W/mK)	—	Thermal conductivity κ (W/mK) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE below Tg (K⁻¹) (GlassNet)

AvailableReviewed

CTE below Tg (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTEbelowTg	CTE below Tg (K⁻¹)	—	CTE below Tg (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 328K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 328K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE328K	CTE at 328K (K⁻¹)	—	CTE at 328K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 373K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 373K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE373K	CTE at 373K (K⁻¹)	—	CTE at 373K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 433K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 433K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE433K	CTE at 433K (K⁻¹)	—	CTE at 433K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 483K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 483K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE483K	CTE at 483K (K⁻¹)	—	CTE at 483K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 623K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 623K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE623K	CTE at 623K (K⁻¹)	—	CTE at 623K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 293K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 293K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp293K	Cp at 293K (J/mol·K)	—	Cp at 293K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 473K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 473K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp473K	Cp at 473K (J/mol·K)	—	Cp at 473K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 673K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 673K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp673K	Cp at 673K (J/mol·K)	—	Cp at 673K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1073K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1073K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1073K	Cp at 1073K (J/mol·K)	—	Cp at 1073K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1273K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1273K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1273K	Cp at 1273K (J/mol·K)	—	Cp at 1273K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1473K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1473K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1473K	Cp at 1473K (J/mol·K)	—	Cp at 1473K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1673K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1673K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1673K	Cp at 1673K (J/mol·K)	—	Cp at 1673K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension above Tg (N/m) (GlassNet)

AvailableReviewed

Surface tension above Tg (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTensionAboveTg	Surface tension above Tg (N/m)	—	Surface tension above Tg (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1173K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1173K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1173K	Surface tension at 1173K (N/m)	—	Surface tension at 1173K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1473K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1473K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1473K	Surface tension at 1473K (N/m)	—	Surface tension at 1473K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1573K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1573K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1573K	Surface tension at 1573K (N/m)	—	Surface tension at 1573K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1673K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1673K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1673K	Surface tension at 1673K (N/m)	—	Surface tension at 1673K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

8 models

Elastic modulus

Young's modulus

AvailableNeeds review

Young's modulus from additive mole-fraction model (Winkelmann & Schott 1894). Coefficients for 18 oxides. ±8% typical accuracy for 55-80 mol% SiO₂.

Model family

Composition-dependent elastic modulus

Source

Winkelmann & Schott (1894, Ann. Phys. 51, 697); Rouxel (2007) review

Source note

Additive mole-fraction coefficients.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.

Limitations

±8% typical error.
Best for 55-80 mol% SiO₂.
Does not account for processing history or porosity.

Parameters

Symbol	Label	Unit	Description
E	Young's modulus	GPa	Young's modulus at 300 K.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Shear modulus

AvailableNeeds review

Shear modulus derived from Young's modulus (Winkelmann & Schott 1894) and Poisson's ratio (additive model) via G = E / 2(1+ν).

Model family

Composition-dependent shear modulus

Source

Winkelmann & Schott (1894); isotropic elasticity

Source note

Derived from Young's modulus and Poisson's ratio.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.
Isotropic elastic behaviour assumed.

Limitations

±5 GPa typical error.
Dependent on accuracy of both E and ν models.

Parameters

Symbol	Label	Unit	Description
G	Shear modulus	GPa	Shear modulus of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Poisson's ratio

AvailableNeeds review

Poisson's ratio from additive weight-fraction model, calibrated on literature data (Rouxel 2007, Makishima & Mackenzie 1975).

Model family

Composition-dependent Poisson's ratio

Source

Rouxel (2007); Makishima & Mackenzie (1975)

Source note

Additive weight-fraction model calibrated on binary and multicomponent silicate data.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.
Linear additivity of oxide-specific Poisson ratios.

Limitations

±0.02 typical error.
Simplified additive model; actual ν depends on network topology.

Parameters

Symbol	Label	Unit	Description
nu	Poisson's ratio	dimensionless	Poisson's ratio of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Microhardness

AvailableNeeds review

Vickers microhardness (Hv) from additive mole-fraction model calibrated on literature data (Volf 1988, Yamane & Mackenzie 1974).

Model family

Composition-dependent microhardness

Source

Volf (1988); Yamane & Mackenzie (1974)

Source note

Additive mole-fraction model for Vickers hardness.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.
Linear additivity of oxide-specific hardness contributions.

Limitations

±1 GPa typical error.
Does not account for indentation size effect or crack resistance.

Parameters

Symbol	Label	Unit	Description
HV	Vickers microhardness	GPa	Vickers microhardness of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Young's modulus E (GPa) (GlassNet)

AvailableReviewed

Young's modulus E (GPa) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
YoungModulus	Young's modulus E (GPa)	—	Young's modulus E (GPa) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Shear modulus G (GPa) (GlassNet)

AvailableReviewed

Shear modulus G (GPa) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
ShearModulus	Shear modulus G (GPa)	—	Shear modulus G (GPa) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Vickers hardness Hv (GPa) (GlassNet)

AvailableReviewed

Vickers hardness Hv (GPa) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Microhardness	Vickers hardness Hv (GPa)	—	Vickers hardness Hv (GPa) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Poisson's ratio ν (GlassNet)

AvailableReviewed

Poisson's ratio ν predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
PoissonRatio	Poisson's ratio ν	—	Poisson's ratio ν of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

9 models

Refractive index

AvailableNeeds review

Refractive index at 589 nm (nD) from oxide composition via Appen (1949) additive wt-fraction coefficients for 19 oxides. Pure SiO₂ gives 1.475; typical soda-lime ≈1.52.

Model family

Composition-dependent refractive index

Source

Appen (1949); Demkina (1958); Fluegel (2007) review

Source note

Linear additive model on weight-fraction basis.

Assumptions

Input compositions are wt% oxide values.
Coefficients apply to silicate glasses at room temperature.

Limitations

±0.005 typical error.
Single wavelength (589 nm, sodium D line) only.
Valid primarily for silicate glasses with >40 wt% SiO₂.

Parameters

Symbol	Label	Unit	Description
n_d	Refractive index	dimensionless	Refractive index of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Abbe number

AvailableNeeds review

Abbe number νD = (nD − 1)/(nF − nC) derived from Appen (1949) nD and dispersion coefficients (weight-fraction basis).

Model family

Composition-dependent Abbe number

Source

Appen (1949); Demkina (1958)

Source note

Derived from Appen nD and dispersion additive models.

Assumptions

Input compositions are wt% oxide values.
Standard Abbe number definition (nD − 1)/(nF − nC).

Limitations

±2 typical error.
Depends on accuracy of both Appen nD and dispersion coefficients.

Parameters

Symbol	Label	Unit	Description
V_d	Abbe number	dimensionless	Abbe number of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mean dispersion

AvailableNeeds review

Mean dispersion (nF − nC) of glasses from oxide composition using Appen (1949) additive dispersion coefficients (weight-fraction basis).

Model family

Composition-dependent dispersion

Source

Appen (1949); Demkina (1958)

Source note

Additive dispersion coefficients (×10⁴) on weight-fraction basis.

Assumptions

Input compositions are wt% oxide values.
Standard dispersion definition (nF − nC).

Limitations

±5% typical error.
Single dispersion range only; does not cover partial dispersion.

Parameters

Symbol	Label	Unit	Description
n_F - n_C	Mean dispersion	dimensionless	Mean dispersion (n_F - n_C) of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Reflectivity

AvailableNeeds review

Normal-incidence reflectivity R = ((nD − 1)/(nD + 1))² calculated from Appen (1949) refractive index estimate.

Model family

Composition-dependent reflectivity

Source

Fresnel (1823); Appen (1949)

Source note

Derived from Appen refractive index and Fresnel equation.

Assumptions

Input compositions are wt% oxide values.
Normal incidence, air–glass interface.
Single wavelength (589 nm).

Limitations

±0.2% absolute error.
Does not account for surface roughness or angle dependence.

Parameters

Symbol	Label	Unit	Description
R	Reflectivity	%	Reflectivity of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Abbe number νD (GlassNet)

AvailableReviewed

Abbe number νD predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
AbbeNum	Abbe number νD	—	Abbe number νD of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Refractive index nD (GlassNet)

AvailableReviewed

Refractive index nD predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
RefractiveIndex	Refractive index nD	—	Refractive index nD of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Refractive index (low λ) (GlassNet)

AvailableReviewed

Refractive index (low λ) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
RefractiveIndexLow	Refractive index (low λ)	—	Refractive index (low λ) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Refractive index (high λ) (GlassNet)

AvailableReviewed

Refractive index (high λ) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
RefractiveIndexHigh	Refractive index (high λ)	—	Refractive index (high λ) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Mean dispersion nF−nC (GlassNet)

AvailableReviewed

Mean dispersion nF−nC predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
MeanDispersion	Mean dispersion nF−nC	—	Mean dispersion nF−nC of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical

15 models

Electrical resistivity

AvailableNeeds review

Electrical resistivity at 200°C via additive mole-fraction model (Fluegel 2005). Includes Arrhenius-based estimate of critical temperature T(ρ=10⁸ Ω·cm).

Model family

Composition-dependent resistivity

Source

Fluegel (2005); Seddon (1954); Bockris (1948)

Source note

Linear additive model for log10(ρ₂₀₀) on mole-fraction basis.

Assumptions

Input compositions are wt% oxide values.
DC resistivity, no frequency dependence.
Arrhenius temperature dependence with Ea = 100 kJ/mol.

Limitations

±0.5 in log10(ρ) typical error.
Single reference temperature (200°C) with linear Arrhenius extrapolation.

Parameters

Symbol	Label	Unit	Description
rho	Resistivity	Ohm·m	Electrical resistivity of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Permittivity

AvailableNeeds review

Optical permittivity ε∞ = nD² calculated from Appen (1949) refractive index. Represents electronic polarisation contribution at optical frequencies.

Model family

Composition-dependent permittivity

Source

Maxwell (1865); Appen (1949)

Source note

Maxwell relation ε∞ = n² with Appen additive nD.

Assumptions

Input compositions are wt% oxide values.
Optical-frequency permittivity from nD; no ionic contribution.

Limitations

±0.01 typical error.
Optical frequency only; static permittivity is typically higher.
Single wavelength (589 nm).

Parameters

Symbol	Label	Unit	Description
epsilon	Relative permittivity	dimensionless	Relative permittivity of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Dielectric loss tangent

AvailableNeeds review

Dielectric loss tangent (tan δ) at 1 MHz, 25°C from additive mole-fraction model. Alkali oxides increase loss; Al₂O₃ and ZrO₂ reduce it.

Model family

Composition-dependent loss tangent

Source

Shackelford (2005); Volf (1988)

Source note

Additive log10(tan δ) model on mole-fraction basis.

Assumptions

Input compositions are wt% oxide values.
Room temperature, 1 MHz measurement frequency.

Limitations

Factor-of-2–3× typical error.
Single frequency; frequency dispersion of tan δ not captured.

Parameters

Symbol	Label	Unit	Description
tan_delta	Loss tangent	dimensionless	Dielectric loss tangent of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Critical resistivity temperature

Temperature for 1 MOhm·m resistivity

AvailableNeeds review

Temperature at which DC resistivity reaches 1 MΩ·m (10⁸ Ω·cm), derived from the Fluegel (2005) log10(ρ) additive model with Arrhenius extrapolation (Ea = 100 kJ/mol).

Model family

Composition-dependent T-rho

Source

Fluegel (2005); Arrhenius (1889)

Source note

Derived from electrical resistivity model with Arrhenius temperature dependence.

Assumptions

Input compositions are wt% oxide values.
Arrhenius temperature dependence with Ea = 100 kJ/mol.

Limitations

±30°C typical error.
Extrapolation assumes Arrhenius behaviour below 200°C.

Parameters

Symbol	Label	Unit	Description
T_rho1	Temperature at 1 MOhm·m	K	Temperature at which resistivity reaches 1 MOhm·m.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Dielectric

Permittivity ε (GlassNet)

AvailableReviewed

Permittivity ε predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Permittivity	Permittivity ε	—	Permittivity ε of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Dielectric

Dielectric loss tan δ (GlassNet)

AvailableReviewed

Dielectric loss tan δ predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TangentOfLossAngle	Dielectric loss tan δ	—	Dielectric loss tan δ of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

T(ρ = 1 MΩ·m) (K) (GlassNet)

AvailableReviewed

T(ρ = 1 MΩ·m) (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TresistivityIs1MOhm.m	T(ρ = 1 MΩ·m) (K)	—	T(ρ = 1 MΩ·m) (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 293K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 293K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity293K	log₁₀(ρ) at 293K	—	log₁₀(ρ) at 293K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 373K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 373K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity373K	log₁₀(ρ) at 373K	—	log₁₀(ρ) at 373K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 423K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 423K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity423K	log₁₀(ρ) at 423K	—	log₁₀(ρ) at 423K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 573K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 573K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity573K	log₁₀(ρ) at 573K	—	log₁₀(ρ) at 573K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1073K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1073K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1073K	log₁₀(ρ) at 1073K	—	log₁₀(ρ) at 1073K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1273K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1273K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1273K	log₁₀(ρ) at 1273K	—	log₁₀(ρ) at 1273K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1473K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1473K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1473K	log₁₀(ρ) at 1473K	—	log₁₀(ρ) at 1473K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1673K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1673K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1673K	log₁₀(ρ) at 1673K	—	log₁₀(ρ) at 1673K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization & durability

11 models

Crystallization onset

Crystallization onset temperature

AvailableReviewed

Crystallization onset temperature of glasses from oxide composition using GlassNet.

Model family

Composition-dependent crystallization onset

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value — model trained on SciGlass data including DSC/DTA measurements.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Heating rate dependence not captured.

Parameters

Symbol	Label	Unit	Description
T_x	Crystallization onset	K	Crystallization onset temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization peak

Crystallization peak temperature

AvailableReviewed

Crystallization peak temperature of glasses from oxide composition using GlassNet.

Model family

Composition-dependent crystallization peak

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value — model trained on SciGlass data including DSC/DTA measurements.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Heating rate dependence not captured.

Parameters

Symbol	Label	Unit	Description
T_p	Crystallization peak	K	Crystallization peak temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Max growth velocity

Maximum crystal growth velocity

AvailableReviewed

Maximum crystal growth velocity from oxide composition using GlassNet.

Model family

Composition-dependent growth velocity

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Temperature of max growth velocity available separately.

Parameters

Symbol	Label	Unit	Description
U_max	Max growth velocity	m/s	Maximum crystal growth velocity of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Max growth velocity temperature

Temperature of max growth velocity

AvailableReviewed

Temperature at which crystal growth velocity is maximum, from oxide composition using GlassNet.

Model family

Composition-dependent T(Umax)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.

Parameters

Symbol	Label	Unit	Description
T_Umax	Temperature of max growth velocity	K	Temperature of maximum crystal growth velocity.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Chemical durability

Hydrolytic class (ISO 719)

AvailableNeeds review

Hydrolytic class (ISO 719 classes 1–5) from additive mole-fraction model. Alkali oxides increase leachate; Al₂O₃ and B₂O₃ improve resistance.

Model family

Composition-dependent hydrolytic class

Source

Fluegel (2002); ISO 719; DIN 12111

Source note

Additive model for log10(extracted Na₂O) on mole-fraction basis, mapped to ISO 719 classes.

Assumptions

Input compositions are wt% oxide values.
Standard ISO 719 test conditions (98°C water, 60 min).

Limitations

±1 class typical error.
Calibrated on industrial silicate glass compositions.

Parameters

Symbol	Label	Unit	Description
HC	Hydrolytic class	ISO 719 class	Hydrolytic class (ISO 719) of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Water solubility

Water solubility in melts

AvailableNeeds review

Water solubility (H₂O ppm) in silicate melts at 1200°C under 1 atm H₂O, from additive mole-fraction model (Fluegel 2002, Scholze 1959).

Model family

Composition-dependent water solubility

Source

Fluegel (2002); Scholze (1959)

Source note

Additive model for log10(H₂O/ppm) on mole-fraction basis at 1200°C.

Assumptions

Input compositions are wt% oxide values.
Atmospheric pressure, 1200°C, 1 atm H₂O (steam).

Limitations

±30% typical error.
Single temperature / pressure; pressure dependence not included.

Parameters

Symbol	Label	Unit	Description
H2O	Water solubility	wt%	Water solubility in the simulant melt.

Required inputs

Simulant composition — wt%
Temperature — K

Modelling outputs are research aids, not certified material specifications.

Crystallization

Thermal shock resistance R (GlassNet)

AvailableReviewed

Thermal shock resistance R predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
ThermalShockRes	Thermal shock resistance R	—	Thermal shock resistance R of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

T(Umax) (K) (GlassNet)

AvailableReviewed

T(Umax) (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TMaxGrowthVelocity	T(Umax) (K)	—	T(Umax) (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

Umax (m/s) (GlassNet)

AvailableReviewed

Umax (m/s) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
MaxGrowthVelocity	Umax (m/s)	—	Umax (m/s) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

Tp (K) (GlassNet)

AvailableReviewed

Tp (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CrystallizationPeak	Tp (K)	—	Tp (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

Tx (K) (GlassNet)

AvailableReviewed

Tx (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CrystallizationOnset	Tx (K)	—	Tx (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Wiki

Model documentation

Each registered model in the Atlas modelling framework with its source reference, assumptions, limitations, and implementation status.

Melt & thermal

74 models

Melt viscosity

High-temperature viscosity

AvailableNeeds review

A calculator for estimating high-temperature melt viscosity over temperature from selected simulant oxide chemistry.

Model family

High-temperature silicate melt model

Source

Implemented methods: Shaw, Hui-Zhang, GRD08, Duan, Sehlke, Langhammer (2021), Visc_Calc (Langhammer et al. 2022), and i-melt (Le Losq et al.)

Source note

Only composition-derived equations are enabled. Measured VFT curves remain excluded.

Assumptions

Input compositions are wt% oxide values using an FeO-basis chemistry table.
Iron input is shown as FeO and Fe2O3; Fe2O3 is converted to FeO-equivalent iron inside the viscosity equations.
Missing optional oxides such as H2O and F2O_1 are treated as 0 wt% and shown in warnings.

Limitations

Powder flow viscosity and high-temperature melt viscosity will be treated as different properties.
Measured VFT, Cassar, GlassNet (https://github.com/drcassar/glasspy, Cassar 2023), and gpvisc are not enabled in this calculator. GlassNet integration is saved for later.
Literature comparison, fitting workflows, and RMSE validation plots are reserved for a later version.

Parameters

Symbol	Label	Unit	Description
T	Temperature	K or deg C	The independent variable used for the viscosity curve. Atlas will show any unit conversion before running a model.
eta	Dynamic viscosity	Pa s or log10(Pa s)	The target output for high-temperature melt viscosity models.
x_i	Oxide fraction	wt%	Composition terms such as SiO2, Al2O3, CaO, MgO, FeO, Na2O, K2O, TiO2, and related model inputs.

Required inputs

Simulant composition — wt%
Temperature range — K or deg C

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Melt density

AvailableNeeds review

Temperature- and pressure-dependent silicate melt density from oxide composition using Lange & Carmichael (1990).

Model family

Composition-dependent silicate melt density

Source

Lange & Carmichael (1990) — DensityX model

Assumptions

Input compositions are wt% oxide values.
Total iron is treated as Fe2O3 for the density calculation.
Pressure defaults to 1 bar (ambient).

Limitations

Valid for silicate melts above the liquidus only.
Crystal-bearing magmas and partial melts are not represented.
Only the 9-oxide subset from the original paper is supported.

Parameters

Symbol	Label	Unit	Description
T	Temperature	K	Melt temperature.
P	Pressure	bar	Ambient pressure in bar (default 1).
rho	Density	g/cm3	Calculated silicate melt density.

Required inputs

Simulant composition — wt%
Temperature — K

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Liquidus temperature

AvailableNeeds review

Predicts liquidus temperature as the maximum crystallization temperature of 9 mineral phases (pseudoliquidus approach) from oxide composition using Nathan & Van Kirk (1978) regression coefficients.

Model family

Composition-dependent liquidus prediction

Source

Nathan & Van Kirk (1978) — Pseudoliquidus mineral regression

Source note

Assumptions

Input compositions are wt% oxide values.
The liquidus is the maximum of 9 mineral-phase regression equations.
Applies to anhydrous silicate systems only.

Limitations

Does not account for volatile-bearing compositions.
Regression coefficients are calibrated on terrestrial basalt suites.
Crystal fractionation and undercooling effects are not modelled.

Parameters

Symbol	Label	Unit	Description
T_liq	Liquidus temperature	K	Calculated liquidus temperature.

Required inputs

Simulant composition — wt%

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Specific heat capacity

Heat capacity

AvailableNeeds review

Specific heat capacity of silicate melts and glasses from oxide composition using Stebbins (1984) and Bychkov & Koptev-Dvornikov (2019).

Model family

Silicate melt heat capacity

Source

Stebbins et al. (1984), Bychkov & Koptev-Dvornikov (2019)

Source note

Independently published Cp models for glass, liquid, and fully molten silicate phases. Berman (1988) crystalline Cp is available in the hybrid Cp model via mineralogy input.

Assumptions

Input compositions are wt% oxide values.
Stebbins glass Cp is valid below the glass transition.
Stebbins liquid Cp is valid above the liquidus.
Bychkov melt Cp applies to fully molten silicate mixtures.

Limitations

Model validity ranges depend on the original calibration data.
Phase transition enthalpies are only included in the hybrid Cp model.

Parameters

Symbol	Label	Unit	Description
T	Temperature	K	Temperature for the Cp calculation.
Cp	Specific heat capacity	J/kgK	Calculated specific heat capacity.

Required inputs

Simulant composition — wt%
Temperature — K

Default simulant

TUBS-M

Modelling outputs are research aids, not certified material specifications. Always check the cited model source and valid range.

Surface tension

AvailableNeeds review

Surface tension of silicate melts using Murase & McBirney (1973) linear regression with temperature. Two model variants: Slag (SLS) and Gob (GOB).

Model family

Silicate melt surface tension

Source

Murase & McBirney (1973); Kucuk et al. (1999) regression planned.

Source note

Murase model computes surface tension purely from temperature; Kucuk composition-dependent regression is planned.

Assumptions

Surface tension decreases linearly with temperature per Murase & McBirney (1973).

Limitations

Does not yet use composition-dependent Kucuk regression.
Atmosphere effects not included.

Parameters

Symbol	Label	Unit	Description
gamma	Surface tension	N/m	Surface tension of the simulant melt.

Required inputs

Simulant composition (optional) — wt%

Modelling outputs are research aids, not certified material specifications.

Glass transition temperature

AvailableNeeds review

Glass transition temperature Tg from empirical additive wt% model (Mazurin 2007, Fluegel 2007). Pure SiO₂ gives ≈1100°C.

Model family

Composition-dependent glass transition

Source

Mazurin (2007); Fluegel (2007); Volf (1988)

Source note

Empirical linear additive model on wt% basis.

Assumptions

Input compositions are wt% oxide values.
Tg is composition-dependent only.
Standard DSC heating rate assumed.

Limitations

±50 °C typical error.
Simplified linear model; does not capture non-linear modifier interactions.
Does not account for thermal history or heating rate.

Parameters

Symbol	Label	Unit	Description
Tg	Glass transition temperature	K	Glass transition temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Coefficient of thermal expansion

AvailableNeeds review

Linear CTE (20–100°C) calculated from oxide composition via Winkelmann & Schott (1893) additive model.

Model family

Composition-dependent thermal expansion

Source

Winkelmann & Schott (1893), compiled by Fluegel (2007)

Source note

Additive model valid for silicate glasses; FeO coefficient estimated.

Assumptions

Input compositions are wt% oxide values.
CTE is linear sum of oxide weight-fraction coefficients (Winkelmann & Schott).
Room-temperature CTE at 20–100°C.

Limitations

Does not predict temperature dependence of CTE.
Unsupported oxides (e.g., P2O5, Cr2O3 above standard coverage) are excluded with warning.
Simple additive model; Fluegel (2007) quadratic/interaction model is more accurate.

Parameters

Symbol	Label	Unit	Description
CTE	Thermal expansion coefficient	10^-6/K	Coefficient of thermal expansion of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal conductivity

AvailableNeeds review

Model family

Composition-dependent thermal conductivity

Source

Ratcliffe (1963); Fluegel (2006) review

Source note

SiO₂-wt% linear correlation, ±0.15 W/mK typical error.

Assumptions

Input compositions are wt% oxide values.
Thermal conductivity depends primarily on SiO₂ content at room temperature.

Limitations

Room temperature (300 K) only.
±0.15 W/mK typical error.
Does not capture individual oxide effects.
Extrapolation unreliable below 40 wt% SiO₂.

Parameters

Symbol	Label	Unit	Description
k	Thermal conductivity	W/mK	Thermal conductivity at 300 K.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal shock resistance

AvailableReviewed

Thermal shock resistance parameter from oxide composition using GlassNet.

Model family

Composition-dependent thermal shock

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Empirical parameter, not a fundamental property.

Parameters

Symbol	Label	Unit	Description
TSR	Thermal shock resistance	K	Thermal shock resistance of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Annealing temperature

AvailableNeeds review

Annealing temperature (log η = 13 Pa·s) from Tg + 30°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent annealing point

Source

Mazurin (2007); standard glass technology

Source note

Derived from Tg with standard offset.

Assumptions

Input compositions are wt% oxide values.
T_anneal ≈ Tg + 30°C.

Limitations

±50 °C typical error.
Offset varies with composition; 30°C is a nominal value.

Parameters

Symbol	Label	Unit	Description
T_anneal	Annealing temperature	K	Annealing temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Strain temperature

AvailableNeeds review

Strain temperature (log η = 14.5 Pa·s) from Tg − 30°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent strain point

Source

Mazurin (2007); standard glass technology

Source note

Derived from Tg with standard offset.

Assumptions

Input compositions are wt% oxide values.
T_strain ≈ Tg − 30°C.

Limitations

±50 °C typical error.
Offset varies with composition; 30°C is a nominal value.

Parameters

Symbol	Label	Unit	Description
T_strain	Strain temperature	K	Strain temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Softening temperature

AvailableNeeds review

Dilatometric softening temperature from Tg + 10°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent softening point

Source

Mazurin (2007); standard glass technology

Source note

Derived from Tg with standard offset.

Assumptions

Input compositions are wt% oxide values.
T_soft ≈ Tg + 10°C (dilatometric).

Limitations

±50 °C typical error.
Offset varies with composition.

Parameters

Symbol	Label	Unit	Description
T_soft	Softening temperature	K	Softening temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Littleton softening point

AvailableNeeds review

Littleton softening point (log η = 7.6 Pa·s) from Tg + 220°C offset, where Tg is from empirical additive model.

Model family

Composition-dependent Littleton point

Source

Mazurin (2007); ASTM C338

Source note

Derived from Tg with standard offset for fibre elongation method.

Assumptions

Input compositions are wt% oxide values.
T_Littleton ≈ Tg + 220°C.

Limitations

±50 °C typical error.
Offset varies with melt fragility.

Parameters

Symbol	Label	Unit	Description
T_Littleton	Littleton softening point	K	Littleton softening point of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Melting temperature

AvailableNeeds review

Melting temperature from empirical additive wt% model. Pure SiO₂ gives 1700°C; modifiers reduce Tmelt proportionally.

Model family

Composition-dependent melting point

Source

Schairer (1942); Levin (1964); Mazurin (2007)

Source note

Empirical additive model approximating liquidus of primary crystalline phase.

Assumptions

Input compositions are wt% oxide values.
Tmelt approximates the liquidus of the primary silicate phase.

Limitations

±75 °C typical error.
Simplified linear model; eutectic minima not captured exactly.

Parameters

Symbol	Label	Unit	Description
Tmelt	Melting temperature	K	Melting temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T0 (log η = 0) (GlassNet)

AvailableReviewed

T0 (log η = 0) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T0	T0 (log η = 0)	—	T0 (log η = 0) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T1 (log η = 1) (GlassNet)

AvailableReviewed

T1 (log η = 1) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T1	T1 (log η = 1)	—	T1 (log η = 1) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T2 (log η = 2) (GlassNet)

AvailableReviewed

T2 (log η = 2) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T2	T2 (log η = 2)	—	T2 (log η = 2) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T3 (log η = 3) (GlassNet)

AvailableReviewed

T3 (log η = 3) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T3	T3 (log η = 3)	—	T3 (log η = 3) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T4 (log η = 4) (GlassNet)

AvailableReviewed

T4 (log η = 4) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T4	T4 (log η = 4)	—	T4 (log η = 4) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T5 (log η = 5) (GlassNet)

AvailableReviewed

T5 (log η = 5) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T5	T5 (log η = 5)	—	T5 (log η = 5) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T6 (log η = 6) (GlassNet)

AvailableReviewed

T6 (log η = 6) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T6	T6 (log η = 6)	—	T6 (log η = 6) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T7 (log η = 7) (GlassNet)

AvailableReviewed

T7 (log η = 7) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T7	T7 (log η = 7)	—	T7 (log η = 7) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T8 (log η = 8) (GlassNet)

AvailableReviewed

T8 (log η = 8) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T8	T8 (log η = 8)	—	T8 (log η = 8) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T9 (log η = 9) (GlassNet)

AvailableReviewed

T9 (log η = 9) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T9	T9 (log η = 9)	—	T9 (log η = 9) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T10 (log η = 10) (GlassNet)

AvailableReviewed

T10 (log η = 10) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T10	T10 (log η = 10)	—	T10 (log η = 10) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T11 (log η = 11) (GlassNet)

AvailableReviewed

T11 (log η = 11) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T11	T11 (log η = 11)	—	T11 (log η = 11) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

T12 (log η = 12) (GlassNet)

AvailableReviewed

T12 (log η = 12) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
T12	T12 (log η = 12)	—	T12 (log η = 12) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 773K (GlassNet)

AvailableReviewed

log₁₀(η) at 773K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity773K	log₁₀(η) at 773K	—	log₁₀(η) at 773K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 873K (GlassNet)

AvailableReviewed

log₁₀(η) at 873K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity873K	log₁₀(η) at 873K	—	log₁₀(η) at 873K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 973K (GlassNet)

AvailableReviewed

log₁₀(η) at 973K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity973K	log₁₀(η) at 973K	—	log₁₀(η) at 973K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1073K (GlassNet)

AvailableReviewed

log₁₀(η) at 1073K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1073K	log₁₀(η) at 1073K	—	log₁₀(η) at 1073K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1173K (GlassNet)

AvailableReviewed

log₁₀(η) at 1173K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1173K	log₁₀(η) at 1173K	—	log₁₀(η) at 1173K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1273K (GlassNet)

AvailableReviewed

log₁₀(η) at 1273K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1273K	log₁₀(η) at 1273K	—	log₁₀(η) at 1273K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1373K (GlassNet)

AvailableReviewed

log₁₀(η) at 1373K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1373K	log₁₀(η) at 1373K	—	log₁₀(η) at 1373K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1473K (GlassNet)

AvailableReviewed

log₁₀(η) at 1473K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1473K	log₁₀(η) at 1473K	—	log₁₀(η) at 1473K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1573K (GlassNet)

AvailableReviewed

log₁₀(η) at 1573K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1573K	log₁₀(η) at 1573K	—	log₁₀(η) at 1573K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1673K (GlassNet)

AvailableReviewed

log₁₀(η) at 1673K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1673K	log₁₀(η) at 1673K	—	log₁₀(η) at 1673K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1773K (GlassNet)

AvailableReviewed

log₁₀(η) at 1773K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1773K	log₁₀(η) at 1773K	—	log₁₀(η) at 1773K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 1873K (GlassNet)

AvailableReviewed

log₁₀(η) at 1873K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity1873K	log₁₀(η) at 1873K	—	log₁₀(η) at 1873K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 2073K (GlassNet)

AvailableReviewed

log₁₀(η) at 2073K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity2073K	log₁₀(η) at 2073K	—	log₁₀(η) at 2073K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 2273K (GlassNet)

AvailableReviewed

log₁₀(η) at 2273K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity2273K	log₁₀(η) at 2273K	—	log₁₀(η) at 2273K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Viscosity

log₁₀(η) at 2473K (GlassNet)

AvailableReviewed

log₁₀(η) at 2473K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Viscosity2473K	log₁₀(η) at 2473K	—	log₁₀(η) at 2473K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Tg (K) (GlassNet)

AvailableReviewed

Tg (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tg	Tg (K)	—	Tg (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Tmelt (K) (GlassNet)

AvailableReviewed

Tmelt (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tmelt	Tmelt (K)	—	Tmelt (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Tliquidus (K) (GlassNet)

AvailableReviewed

Tliquidus (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tliquidus	Tliquidus (K)	—	Tliquidus (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Littleton point (K) (GlassNet)

AvailableReviewed

Littleton point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TLittletons	Littleton point (K)	—	Littleton point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Annealing point (K) (GlassNet)

AvailableReviewed

Annealing point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TAnnealing	Annealing point (K)	—	Annealing point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Strain point (K) (GlassNet)

AvailableReviewed

Strain point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tstrain	Strain point (K)	—	Strain point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Softening point (K) (GlassNet)

AvailableReviewed

Softening point (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Tsoft	Softening point (K)	—	Softening point (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Characteristic temperatures

Dilatometric softening (K) (GlassNet)

AvailableReviewed

Dilatometric softening (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TdilatometricSoftening	Dilatometric softening (K)	—	Dilatometric softening (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 293K (g/cm³) (GlassNet)

AvailableReviewed

Density at 293K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density293K	Density at 293K (g/cm³)	—	Density at 293K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1073K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1073K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1073K	Density at 1073K (g/cm³)	—	Density at 1073K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1273K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1273K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1273K	Density at 1273K (g/cm³)	—	Density at 1273K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1473K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1473K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1473K	Density at 1473K (g/cm³)	—	Density at 1473K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Density

Density at 1673K (g/cm³) (GlassNet)

AvailableReviewed

Density at 1673K (g/cm³) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Density1673K	Density at 1673K (g/cm³)	—	Density at 1673K (g/cm³) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Thermal conductivity κ (W/mK) (GlassNet)

AvailableReviewed

Thermal conductivity κ (W/mK) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
ThermalConductivity	Thermal conductivity κ (W/mK)	—	Thermal conductivity κ (W/mK) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE below Tg (K⁻¹) (GlassNet)

AvailableReviewed

CTE below Tg (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTEbelowTg	CTE below Tg (K⁻¹)	—	CTE below Tg (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 328K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 328K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE328K	CTE at 328K (K⁻¹)	—	CTE at 328K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 373K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 373K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE373K	CTE at 373K (K⁻¹)	—	CTE at 373K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 433K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 433K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE433K	CTE at 433K (K⁻¹)	—	CTE at 433K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 483K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 483K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE483K	CTE at 483K (K⁻¹)	—	CTE at 483K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

CTE at 623K (K⁻¹) (GlassNet)

AvailableReviewed

CTE at 623K (K⁻¹) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CTE623K	CTE at 623K (K⁻¹)	—	CTE at 623K (K⁻¹) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 293K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 293K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp293K	Cp at 293K (J/mol·K)	—	Cp at 293K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 473K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 473K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp473K	Cp at 473K (J/mol·K)	—	Cp at 473K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 673K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 673K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp673K	Cp at 673K (J/mol·K)	—	Cp at 673K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1073K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1073K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1073K	Cp at 1073K (J/mol·K)	—	Cp at 1073K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1273K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1273K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1273K	Cp at 1273K (J/mol·K)	—	Cp at 1273K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1473K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1473K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1473K	Cp at 1473K (J/mol·K)	—	Cp at 1473K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Thermal

Cp at 1673K (J/mol·K) (GlassNet)

AvailableReviewed

Cp at 1673K (J/mol·K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Cp1673K	Cp at 1673K (J/mol·K)	—	Cp at 1673K (J/mol·K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension above Tg (N/m) (GlassNet)

AvailableReviewed

Surface tension above Tg (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTensionAboveTg	Surface tension above Tg (N/m)	—	Surface tension above Tg (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1173K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1173K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1173K	Surface tension at 1173K (N/m)	—	Surface tension at 1173K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1473K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1473K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1473K	Surface tension at 1473K (N/m)	—	Surface tension at 1473K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1573K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1573K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1573K	Surface tension at 1573K (N/m)	—	Surface tension at 1573K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Surface tension

Surface tension at 1673K (N/m) (GlassNet)

AvailableReviewed

Surface tension at 1673K (N/m) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
SurfaceTension1673K	Surface tension at 1673K (N/m)	—	Surface tension at 1673K (N/m) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

8 models

Elastic modulus

Young's modulus

AvailableNeeds review

Young's modulus from additive mole-fraction model (Winkelmann & Schott 1894). Coefficients for 18 oxides. ±8% typical accuracy for 55-80 mol% SiO₂.

Model family

Composition-dependent elastic modulus

Source

Winkelmann & Schott (1894, Ann. Phys. 51, 697); Rouxel (2007) review

Source note

Additive mole-fraction coefficients.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.

Limitations

±8% typical error.
Best for 55-80 mol% SiO₂.
Does not account for processing history or porosity.

Parameters

Symbol	Label	Unit	Description
E	Young's modulus	GPa	Young's modulus at 300 K.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Shear modulus

AvailableNeeds review

Shear modulus derived from Young's modulus (Winkelmann & Schott 1894) and Poisson's ratio (additive model) via G = E / 2(1+ν).

Model family

Composition-dependent shear modulus

Source

Winkelmann & Schott (1894); isotropic elasticity

Source note

Derived from Young's modulus and Poisson's ratio.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.
Isotropic elastic behaviour assumed.

Limitations

±5 GPa typical error.
Dependent on accuracy of both E and ν models.

Parameters

Symbol	Label	Unit	Description
G	Shear modulus	GPa	Shear modulus of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Poisson's ratio

AvailableNeeds review

Poisson's ratio from additive weight-fraction model, calibrated on literature data (Rouxel 2007, Makishima & Mackenzie 1975).

Model family

Composition-dependent Poisson's ratio

Source

Rouxel (2007); Makishima & Mackenzie (1975)

Source note

Additive weight-fraction model calibrated on binary and multicomponent silicate data.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.
Linear additivity of oxide-specific Poisson ratios.

Limitations

±0.02 typical error.
Simplified additive model; actual ν depends on network topology.

Parameters

Symbol	Label	Unit	Description
nu	Poisson's ratio	dimensionless	Poisson's ratio of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Microhardness

AvailableNeeds review

Vickers microhardness (Hv) from additive mole-fraction model calibrated on literature data (Volf 1988, Yamane & Mackenzie 1974).

Model family

Composition-dependent microhardness

Source

Volf (1988); Yamane & Mackenzie (1974)

Source note

Additive mole-fraction model for Vickers hardness.

Assumptions

Input compositions are wt% oxide values.
Room temperature, bulk glass.
Linear additivity of oxide-specific hardness contributions.

Limitations

±1 GPa typical error.
Does not account for indentation size effect or crack resistance.

Parameters

Symbol	Label	Unit	Description
HV	Vickers microhardness	GPa	Vickers microhardness of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Young's modulus E (GPa) (GlassNet)

AvailableReviewed

Young's modulus E (GPa) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
YoungModulus	Young's modulus E (GPa)	—	Young's modulus E (GPa) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Shear modulus G (GPa) (GlassNet)

AvailableReviewed

Shear modulus G (GPa) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
ShearModulus	Shear modulus G (GPa)	—	Shear modulus G (GPa) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Vickers hardness Hv (GPa) (GlassNet)

AvailableReviewed

Vickers hardness Hv (GPa) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Microhardness	Vickers hardness Hv (GPa)	—	Vickers hardness Hv (GPa) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mechanical

Poisson's ratio ν (GlassNet)

AvailableReviewed

Poisson's ratio ν predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
PoissonRatio	Poisson's ratio ν	—	Poisson's ratio ν of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

9 models

Refractive index

AvailableNeeds review

Refractive index at 589 nm (nD) from oxide composition via Appen (1949) additive wt-fraction coefficients for 19 oxides. Pure SiO₂ gives 1.475; typical soda-lime ≈1.52.

Model family

Composition-dependent refractive index

Source

Appen (1949); Demkina (1958); Fluegel (2007) review

Source note

Linear additive model on weight-fraction basis.

Assumptions

Input compositions are wt% oxide values.
Coefficients apply to silicate glasses at room temperature.

Limitations

±0.005 typical error.
Single wavelength (589 nm, sodium D line) only.
Valid primarily for silicate glasses with >40 wt% SiO₂.

Parameters

Symbol	Label	Unit	Description
n_d	Refractive index	dimensionless	Refractive index of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Abbe number

AvailableNeeds review

Abbe number νD = (nD − 1)/(nF − nC) derived from Appen (1949) nD and dispersion coefficients (weight-fraction basis).

Model family

Composition-dependent Abbe number

Source

Appen (1949); Demkina (1958)

Source note

Derived from Appen nD and dispersion additive models.

Assumptions

Input compositions are wt% oxide values.
Standard Abbe number definition (nD − 1)/(nF − nC).

Limitations

±2 typical error.
Depends on accuracy of both Appen nD and dispersion coefficients.

Parameters

Symbol	Label	Unit	Description
V_d	Abbe number	dimensionless	Abbe number of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Mean dispersion

AvailableNeeds review

Mean dispersion (nF − nC) of glasses from oxide composition using Appen (1949) additive dispersion coefficients (weight-fraction basis).

Model family

Composition-dependent dispersion

Source

Appen (1949); Demkina (1958)

Source note

Additive dispersion coefficients (×10⁴) on weight-fraction basis.

Assumptions

Input compositions are wt% oxide values.
Standard dispersion definition (nF − nC).

Limitations

±5% typical error.
Single dispersion range only; does not cover partial dispersion.

Parameters

Symbol	Label	Unit	Description
n_F - n_C	Mean dispersion	dimensionless	Mean dispersion (n_F - n_C) of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Reflectivity

AvailableNeeds review

Normal-incidence reflectivity R = ((nD − 1)/(nD + 1))² calculated from Appen (1949) refractive index estimate.

Model family

Composition-dependent reflectivity

Source

Fresnel (1823); Appen (1949)

Source note

Derived from Appen refractive index and Fresnel equation.

Assumptions

Input compositions are wt% oxide values.
Normal incidence, air–glass interface.
Single wavelength (589 nm).

Limitations

±0.2% absolute error.
Does not account for surface roughness or angle dependence.

Parameters

Symbol	Label	Unit	Description
R	Reflectivity	%	Reflectivity of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Abbe number νD (GlassNet)

AvailableReviewed

Abbe number νD predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
AbbeNum	Abbe number νD	—	Abbe number νD of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Refractive index nD (GlassNet)

AvailableReviewed

Refractive index nD predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
RefractiveIndex	Refractive index nD	—	Refractive index nD of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Refractive index (low λ) (GlassNet)

AvailableReviewed

Refractive index (low λ) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
RefractiveIndexLow	Refractive index (low λ)	—	Refractive index (low λ) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Refractive index (high λ) (GlassNet)

AvailableReviewed

Refractive index (high λ) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
RefractiveIndexHigh	Refractive index (high λ)	—	Refractive index (high λ) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Optical

Mean dispersion nF−nC (GlassNet)

AvailableReviewed

Mean dispersion nF−nC predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
MeanDispersion	Mean dispersion nF−nC	—	Mean dispersion nF−nC of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical

15 models

Electrical resistivity

AvailableNeeds review

Electrical resistivity at 200°C via additive mole-fraction model (Fluegel 2005). Includes Arrhenius-based estimate of critical temperature T(ρ=10⁸ Ω·cm).

Model family

Composition-dependent resistivity

Source

Fluegel (2005); Seddon (1954); Bockris (1948)

Source note

Linear additive model for log10(ρ₂₀₀) on mole-fraction basis.

Assumptions

Input compositions are wt% oxide values.
DC resistivity, no frequency dependence.
Arrhenius temperature dependence with Ea = 100 kJ/mol.

Limitations

±0.5 in log10(ρ) typical error.
Single reference temperature (200°C) with linear Arrhenius extrapolation.

Parameters

Symbol	Label	Unit	Description
rho	Resistivity	Ohm·m	Electrical resistivity of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Permittivity

AvailableNeeds review

Optical permittivity ε∞ = nD² calculated from Appen (1949) refractive index. Represents electronic polarisation contribution at optical frequencies.

Model family

Composition-dependent permittivity

Source

Maxwell (1865); Appen (1949)

Source note

Maxwell relation ε∞ = n² with Appen additive nD.

Assumptions

Input compositions are wt% oxide values.
Optical-frequency permittivity from nD; no ionic contribution.

Limitations

±0.01 typical error.
Optical frequency only; static permittivity is typically higher.
Single wavelength (589 nm).

Parameters

Symbol	Label	Unit	Description
epsilon	Relative permittivity	dimensionless	Relative permittivity of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Dielectric loss tangent

AvailableNeeds review

Dielectric loss tangent (tan δ) at 1 MHz, 25°C from additive mole-fraction model. Alkali oxides increase loss; Al₂O₃ and ZrO₂ reduce it.

Model family

Composition-dependent loss tangent

Source

Shackelford (2005); Volf (1988)

Source note

Additive log10(tan δ) model on mole-fraction basis.

Assumptions

Input compositions are wt% oxide values.
Room temperature, 1 MHz measurement frequency.

Limitations

Factor-of-2–3× typical error.
Single frequency; frequency dispersion of tan δ not captured.

Parameters

Symbol	Label	Unit	Description
tan_delta	Loss tangent	dimensionless	Dielectric loss tangent of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Critical resistivity temperature

Temperature for 1 MOhm·m resistivity

AvailableNeeds review

Temperature at which DC resistivity reaches 1 MΩ·m (10⁸ Ω·cm), derived from the Fluegel (2005) log10(ρ) additive model with Arrhenius extrapolation (Ea = 100 kJ/mol).

Model family

Composition-dependent T-rho

Source

Fluegel (2005); Arrhenius (1889)

Source note

Derived from electrical resistivity model with Arrhenius temperature dependence.

Assumptions

Input compositions are wt% oxide values.
Arrhenius temperature dependence with Ea = 100 kJ/mol.

Limitations

±30°C typical error.
Extrapolation assumes Arrhenius behaviour below 200°C.

Parameters

Symbol	Label	Unit	Description
T_rho1	Temperature at 1 MOhm·m	K	Temperature at which resistivity reaches 1 MOhm·m.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Dielectric

Permittivity ε (GlassNet)

AvailableReviewed

Permittivity ε predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Permittivity	Permittivity ε	—	Permittivity ε of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Dielectric

Dielectric loss tan δ (GlassNet)

AvailableReviewed

Dielectric loss tan δ predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TangentOfLossAngle	Dielectric loss tan δ	—	Dielectric loss tan δ of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

T(ρ = 1 MΩ·m) (K) (GlassNet)

AvailableReviewed

T(ρ = 1 MΩ·m) (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TresistivityIs1MOhm.m	T(ρ = 1 MΩ·m) (K)	—	T(ρ = 1 MΩ·m) (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 293K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 293K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity293K	log₁₀(ρ) at 293K	—	log₁₀(ρ) at 293K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 373K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 373K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity373K	log₁₀(ρ) at 373K	—	log₁₀(ρ) at 373K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 423K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 423K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity423K	log₁₀(ρ) at 423K	—	log₁₀(ρ) at 423K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 573K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 573K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity573K	log₁₀(ρ) at 573K	—	log₁₀(ρ) at 573K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1073K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1073K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1073K	log₁₀(ρ) at 1073K	—	log₁₀(ρ) at 1073K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1273K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1273K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1273K	log₁₀(ρ) at 1273K	—	log₁₀(ρ) at 1273K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1473K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1473K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1473K	log₁₀(ρ) at 1473K	—	log₁₀(ρ) at 1473K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Electrical resistivity

log₁₀(ρ) at 1673K (GlassNet)

AvailableReviewed

log₁₀(ρ) at 1673K predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
Resistivity1673K	log₁₀(ρ) at 1673K	—	log₁₀(ρ) at 1673K of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization & durability

11 models

Crystallization onset

Crystallization onset temperature

AvailableReviewed

Crystallization onset temperature of glasses from oxide composition using GlassNet.

Model family

Composition-dependent crystallization onset

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value — model trained on SciGlass data including DSC/DTA measurements.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Heating rate dependence not captured.

Parameters

Symbol	Label	Unit	Description
T_x	Crystallization onset	K	Crystallization onset temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization peak

Crystallization peak temperature

AvailableReviewed

Crystallization peak temperature of glasses from oxide composition using GlassNet.

Model family

Composition-dependent crystallization peak

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value — model trained on SciGlass data including DSC/DTA measurements.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Heating rate dependence not captured.

Parameters

Symbol	Label	Unit	Description
T_p	Crystallization peak	K	Crystallization peak temperature of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Max growth velocity

Maximum crystal growth velocity

AvailableReviewed

Maximum crystal growth velocity from oxide composition using GlassNet.

Model family

Composition-dependent growth velocity

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Temperature of max growth velocity available separately.

Parameters

Symbol	Label	Unit	Description
U_max	Max growth velocity	m/s	Maximum crystal growth velocity of the simulant.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Max growth velocity temperature

Temperature of max growth velocity

AvailableReviewed

Temperature at which crystal growth velocity is maximum, from oxide composition using GlassNet.

Model family

Composition-dependent T(Umax)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.

Parameters

Symbol	Label	Unit	Description
T_Umax	Temperature of max growth velocity	K	Temperature of maximum crystal growth velocity.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Chemical durability

Hydrolytic class (ISO 719)

AvailableNeeds review

Hydrolytic class (ISO 719 classes 1–5) from additive mole-fraction model. Alkali oxides increase leachate; Al₂O₃ and B₂O₃ improve resistance.

Model family

Composition-dependent hydrolytic class

Source

Fluegel (2002); ISO 719; DIN 12111

Source note

Additive model for log10(extracted Na₂O) on mole-fraction basis, mapped to ISO 719 classes.

Assumptions

Input compositions are wt% oxide values.
Standard ISO 719 test conditions (98°C water, 60 min).

Limitations

±1 class typical error.
Calibrated on industrial silicate glass compositions.

Parameters

Symbol	Label	Unit	Description
HC	Hydrolytic class	ISO 719 class	Hydrolytic class (ISO 719) of the simulant glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Water solubility

Water solubility in melts

AvailableNeeds review

Water solubility (H₂O ppm) in silicate melts at 1200°C under 1 atm H₂O, from additive mole-fraction model (Fluegel 2002, Scholze 1959).

Model family

Composition-dependent water solubility

Source

Fluegel (2002); Scholze (1959)

Source note

Additive model for log10(H₂O/ppm) on mole-fraction basis at 1200°C.

Assumptions

Input compositions are wt% oxide values.
Atmospheric pressure, 1200°C, 1 atm H₂O (steam).

Limitations

±30% typical error.
Single temperature / pressure; pressure dependence not included.

Parameters

Symbol	Label	Unit	Description
H2O	Water solubility	wt%	Water solubility in the simulant melt.

Required inputs

Simulant composition — wt%
Temperature — K

Modelling outputs are research aids, not certified material specifications.

Crystallization

Thermal shock resistance R (GlassNet)

AvailableReviewed

Thermal shock resistance R predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
ThermalShockRes	Thermal shock resistance R	—	Thermal shock resistance R of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

T(Umax) (K) (GlassNet)

AvailableReviewed

T(Umax) (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
TMaxGrowthVelocity	T(Umax) (K)	—	T(Umax) (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

Umax (m/s) (GlassNet)

AvailableReviewed

Umax (m/s) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
MaxGrowthVelocity	Umax (m/s)	—	Umax (m/s) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

Tp (K) (GlassNet)

AvailableReviewed

Tp (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CrystallizationPeak	Tp (K)	—	Tp (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.

Crystallization

Tx (K) (GlassNet)

AvailableReviewed

Tx (K) predicted from oxide composition using GlassNet multitask deep neural network (Cassar 2023).

Model family

GlassNet AI (multitask DNN)

Source

GlassNet — Cassar (2023)

Source note

AI-predicted value.

Assumptions

Input compositions are wt% oxide values.

Limitations

AI model accuracy varies by composition family.
Training data from SciGlass (218k+ compositions).

Parameters

Symbol	Label	Unit	Description
CrystallizationOnset	Tx (K)	—	Tx (K) of the glass.

Required inputs

Simulant composition — wt%

Modelling outputs are research aids, not certified material specifications.