MatIntel Scoring Categories — Full Implementation Guide
This document defines all 31 application scoring categories for MatIntel. Each entry covers the scientific rationale, the scoring criteria as Python logic, element/structural rules, viability modifiers, known caveats, and example compounds from the GNoME dataset.
How Scores Are Structured
Every category function receives:
- composition: pymatgen.Composition object
- structure: pymatgen.Structure object (or None if unavailable)
- band_gap: float in eV (from CHGNet or matminer)
- formation_energy: float in eV/atom
- e_hull: float in eV/atom (energy above convex hull from GNoME CSV)
- features: dict of matminer ElementProperty features
Every function returns a float 0.0–1.0.
Scores are then multiplied by the viability score (which incorporates element cost, abundance, radioactive filter, rare earth penalty, CLscore penalty, and e_hull penalty separately).
DOMAIN 1 — ENERGY STORAGE
1. Battery Cathode (Li-ion)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Positive electrode in lithium-ion batteries. Must intercalate Li⁺ reversibly at high voltage. Commercially dominant examples: LiCoO₂, LiFePO₄, NMC.
Key requirements: - Must contain Li in the structure (working ion present) - Transition metal redox center: Mn, Fe, Co, Ni, V, Cr (change oxidation state reversibly) - Band gap: 0 (metallic) to ~3 eV acceptable; fully insulating hurts rate performance - Polyanionic framework (phosphate, silicate, sulfate) or layered oxide preferred for stability - Avoid: purely metallic conductors with no intercalation sites
def score_battery_cathode_liion(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Hard requirement: must contain Li
if 'Li' not in elements:
return 0.0
# Redox-active transition metal present
redox_metals = {'Mn', 'Fe', 'Co', 'Ni', 'V', 'Cr', 'Cu', 'Mo', 'Ti'}
if elements & redox_metals:
score += 0.35
# Polyanionic framework bonus (P, Si, S, B as framework formers with O)
polyanion_formers = {'P', 'Si', 'S', 'B'}
if (elements & polyanion_formers) and ('O' in elements):
score += 0.20
# Layered oxide bonus (many cathodes are AMO2 type)
if 'O' in elements and not (elements & polyanion_formers):
score += 0.10
# Band gap: conductivity needed but not fully metallic
if band_gap is not None:
if 0.01 <= band_gap <= 3.0:
score += 0.25
elif band_gap < 0.01: # too metallic — might not have intercalation sites
score += 0.10
# Formation energy: should be strongly negative (stable)
if formation_energy is not None and formation_energy < -1.0:
score += 0.10
# Penalise fluorides (stability issues in electrolyte)
if 'F' in elements:
score *= 0.7
# Penalise sulfur without phosphorus (polysulfide dissolution)
if 'S' in elements and 'P' not in elements and 'O' not in elements:
score *= 0.6
return min(score, 1.0)
Viability notes: Co is a critical mineral (~$33/kg, high supply risk). Ni is moderate. Fe and Mn are excellent from a supply perspective — LFP (LiFePO₄) is the fastest-growing cathode chemistry for this reason.
Example targets from GNoME: NaCa(Cu9Si4)₂ after Na-ion adaptation; any Li-Mn-P-O compound.
2. Battery Anode
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Negative electrode in Li/Na-ion batteries. Stores ions at low voltage vs. Li/Li⁺. Commercial: graphite (372 mAh/g), silicon (~3500 mAh/g but volume expansion problem). Fe-Si alloys suppress Si expansion; iron silicides (FeSi₂) are studied as inactive matrix.
Key requirements: - No working ion required in structure (anode inserts Li/Na from electrolyte) - Metallic or near-metallic (band gap < 0.5 eV preferred for electronic conductivity) - Contains Si, Sn, Sb, P, Ge (alloying anodes) OR conversion-type (Fe, Co, Ni oxides/sulfides) - High theoretical capacity elements preferred - Must NOT be purely an oxide without a metal — pure oxides are conversion type only
def score_battery_anode(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# High-capacity alloying anode elements
alloying = {'Si', 'Sn', 'Sb', 'P', 'Ge', 'Bi', 'Al'}
if elements & alloying:
score += 0.40
# Conversion anode elements (transition metal + anion)
conversion_metals = {'Fe', 'Co', 'Ni', 'Cu', 'Mn', 'Mo', 'W'}
conversion_anions = {'O', 'S', 'Se', 'F', 'N'}
if (elements & conversion_metals) and (elements & conversion_anions):
score += 0.25
# Metallic character preferred
if band_gap is not None:
if band_gap < 0.1:
score += 0.20
elif band_gap < 0.5:
score += 0.12
elif band_gap < 1.5:
score += 0.05
# Iron silicide specific bonus (well-studied anode family)
if 'Fe' in elements and 'Si' in elements:
score += 0.15
# Penalise if Li or Na already in structure
# (anode materials don't contain the working ion in their discharged state)
if 'Li' in elements or 'Na' in elements:
score *= 0.5
return min(score, 1.0)
Viability notes: Si is abundant and cheap. Fe is excellent. Sn is moderate cost. Sb and Ge carry supply risk. Avoid Co-heavy anodes.
Example targets from GNoME: Fe5Si3H (if the H variant is stable), Fe8Si15P, Cu30Si7P.
3. Battery Cathode (Na-ion)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Positive electrode specifically for sodium-ion batteries. Na-ion is commercially emerging (CATL producing since 2023). Key difference from Li-ion: Na⁺ is larger (1.02 Å vs 0.76 Å for Li⁺), requiring a more open framework. Prussian blue analogues, layered oxides, and polyanionic compounds are the main families.
Key requirements: - Must contain Na in the structure - Transition metal redox center (same as Li-ion cathode) - More open framework preferred (larger interlayer spacing than Li-ion cathodes) - Mn and Fe strongly preferred (abundant, Na-ion's cost advantage relies on cheap elements) - Prussian blue analogue structure bonus (known excellent Na-ion cathode family)
def score_battery_cathode_naion(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Hard requirement: must contain Na
if 'Na' not in elements:
return 0.0
# Preferred redox metals for Na-ion (Fe and Mn especially)
preferred_redox = {'Fe', 'Mn'}
other_redox = {'Co', 'Ni', 'V', 'Cr', 'Cu', 'Ti'}
if elements & preferred_redox:
score += 0.40
elif elements & other_redox:
score += 0.25
# Polyanionic framework (very common in Na-ion cathodes)
polyanion_formers = {'P', 'Si', 'S', 'B'}
if (elements & polyanion_formers) and ('O' in elements):
score += 0.20
# Oxide layered structure (P2 and O3 type)
if 'O' in elements and not (elements & polyanion_formers):
score += 0.12
# Band gap
if band_gap is not None:
if 0.0 <= band_gap <= 3.0:
score += 0.20
# Bonus: no Li (pure Na-ion, not mixed)
if 'Li' not in elements:
score += 0.08
return min(score, 1.0)
Viability notes: Na-ion's commercial case rests on avoiding Li, Co, and expensive REEs. Mn and Fe are ideal. NaCa(Cu9Si4)₂ from your results is a candidate worth investigating.
4. Solid Electrolyte
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Solid-state ionic conductor replacing liquid electrolyte in all-solid-state batteries. Must conduct the working ion (Li⁺ or Na⁺) while being electronically insulating. Key families: NASICON (Na₃Zr₂Si₂PO₁₂ type), garnet (Li₇La₃Zr₂O₁₂), LLZO, argyrodite (Li₆PS₅Cl).
Key requirements: - Wide band gap (>3 eV) — must be electronically insulating - Contains Li or Na (working ion) - Contains Zr, La, Al, P, Si, S as framework elements (common in known solid electrolytes) - Sulfide-based: lower band gap acceptable (2–4 eV), generally higher conductivity - Oxide-based: higher band gap (>4 eV), better stability vs Li metal - Halide-based: Cl, Br, I with Li/Na, very active research area
def score_solid_electrolyte(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Hard requirement: Li or Na (or both)
if 'Li' not in elements and 'Na' not in elements:
return 0.0
# Must be electronically insulating
if band_gap is not None:
if band_gap >= 4.0:
score += 0.35
elif band_gap >= 3.0:
score += 0.25
elif band_gap >= 2.0:
score += 0.10
else:
return 0.0 # too conducting electronically
# Known solid electrolyte framework elements
oxide_framework = {'Zr', 'La', 'Al', 'Ta', 'Nb', 'Ti'}
sulfide_framework = {'P', 'Si', 'Ge', 'Sn', 'As'}
halide_elements = {'Cl', 'Br', 'I', 'F'}
if elements & oxide_framework:
score += 0.25
if elements & sulfide_framework and 'S' in elements:
score += 0.20
if elements & halide_elements:
score += 0.20
# Oxygen presence (most known solid electrolytes are oxides)
if 'O' in elements:
score += 0.10
# Formation energy: must be very stable
if formation_energy is not None and formation_energy < -2.0:
score += 0.10
return min(score, 1.0)
5. Hydrogen Storage
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Materials that absorb and release hydrogen reversibly for fuel cell vehicle storage. Target: >6.5 wt% gravimetric capacity, release at 60–120°C. Known families: complex hydrides (NaBH₄, LiAlH₄ type), metal hydrides (FeTiH₂, LaNi₅H₆), chemical hydrides.
Key requirements: - Contains H in the structure - Light metals preferred (Mg, Li, Na, Al, Ca) for gravimetric density - Transition metal hydrides (Fe, Ti, Ni, La) for kinetics - Moderate stability — too stable = can't release H₂ at practical temperatures - Formation energy: moderately negative (−0.3 to −1.5 eV/atom ideal)
def score_hydrogen_storage(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Hard requirement: must contain H
if 'H' not in elements:
return 0.0
# Light metals (good for gravimetric density)
light_metals = {'Li', 'Na', 'Mg', 'Al', 'Ca', 'K'}
if elements & light_metals:
score += 0.30
# Transition metal hydrides (good kinetics)
kinetic_metals = {'Fe', 'Ti', 'Ni', 'V', 'Zr', 'La', 'Ce', 'Mm'}
if elements & kinetic_metals:
score += 0.25
# Borohydrides / alanates (complex hydrides)
if 'B' in elements:
score += 0.15
if 'Al' in elements and 'H' in elements:
score += 0.10
# Formation energy: moderate stability preferred
if formation_energy is not None:
if -1.5 <= formation_energy <= -0.3:
score += 0.20
elif -0.3 < formation_energy <= 0.0:
score += 0.10 # too weak binding
return min(score, 1.0)
DOMAIN 2 — ENERGY CONVERSION
6. Solar Absorber (Single Junction)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Photovoltaic absorber layer for single-junction solar cells. Must absorb visible light efficiently and generate electron-hole pairs. Target band gap: 1.0–1.8 eV (Shockley-Queisser optimum ~1.34 eV). Known materials: Si (1.1 eV), GaAs (1.4 eV), CdTe (1.45 eV), CIGS (1.0–1.7 eV), perovskites (1.2–1.7 eV). Your Cu6SiMoS8 (1.03 eV) and Cu6SiWSe8 (0.96 eV) land here.
Key requirements: - Band gap: 1.0–1.8 eV (tight range) - Direct band gap strongly preferred over indirect - High optical absorption coefficient - No highly toxic elements (Cd, Pb penalised but not eliminated) - Earth-abundant elements preferred (commercial viability)
def score_solar_absorber_singlejunction(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Core band gap requirement — tight Shockley-Queisser window
if 1.0 <= band_gap <= 1.8:
# Peak score at ~1.34 eV
deviation = abs(band_gap - 1.34)
score += max(0.50 - deviation * 0.5, 0.25)
elif 0.8 <= band_gap < 1.0:
score += 0.10 # too low but could work in tandem
else:
return 0.0 # outside useful range
# Known good absorber element families
chalcopyrite = {'Cu', 'In', 'Ga', 'Se', 'S'} # CIGS family
kesterite = {'Cu', 'Zn', 'Sn', 'S', 'Se'} # CZTS family
chalcogenide_metals = {'Mo', 'W', 'Sb', 'Bi', 'Ge'}
if len(elements & chalcopyrite) >= 3:
score += 0.25
if len(elements & kesterite) >= 3:
score += 0.20
if elements & chalcogenide_metals:
score += 0.10
# Penalise very toxic elements (commercial barrier)
if 'Cd' in elements:
score *= 0.6
if 'As' in elements or 'Hg' in elements:
score *= 0.5
# Penalise Pb (regulatory pressure, though perovskites still studied)
if 'Pb' in elements:
score *= 0.7
return min(score, 1.0)
7. Solar Absorber (Tandem Top Cell)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Top cell in a silicon tandem solar cell stack. Must absorb high-energy photons while transmitting lower-energy light to the Si bottom cell. The perovskite-silicon tandem is the most commercially important variant, with target top cell band gap 1.6–2.0 eV. World record efficiency (2025): ~34%.
Key requirements: - Band gap: 1.6–2.0 eV specifically - Direct gap strongly preferred - Must be processable as thin film (not bulk crystal only) - Halide perovskite structure type is a bonus - Pb-free preferred but not required (Pb-perovskites still dominate research)
def score_solar_absorber_tandem(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Core band gap: tandem top cell window
if 1.6 <= band_gap <= 2.0:
deviation = abs(band_gap - 1.75)
score += max(0.50 - deviation * 0.8, 0.25)
elif 1.5 <= band_gap < 1.6:
score += 0.15
else:
return 0.0
# Halide perovskite indicators
halides = {'Cl', 'Br', 'I'}
a_site = {'Cs', 'Rb', 'K', 'Na'}
b_site = {'Pb', 'Sn', 'Ge', 'Bi', 'Sb', 'In'}
if (elements & halides) and (elements & a_site) and (elements & b_site):
score += 0.30
elif (elements & halides) and (elements & b_site):
score += 0.15
# Pb-free bonus (strong regulatory and environmental driver)
if 'Pb' not in elements and 'Cd' not in elements:
score += 0.15
return min(score, 1.0)
8. Thermoelectric
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Converts heat directly to electricity (or electricity to cooling) via Seebeck effect. Figure of merit ZT = S²σT/κ. Need: high Seebeck coefficient S, high electrical conductivity σ, low thermal conductivity κ. Best materials: PbTe, BiSb₂Te₃, SnSe, halide perovskites. Near-zero band gap with heavy atoms is ideal.
Key requirements: - Near-metallic band gap (< 0.5 eV optimal, up to ~1.0 eV acceptable) - Heavy atoms present (atomic mass > 100) — key for low thermal conductivity - Anharmonic bonding (chalcogenides, halides especially) - Complex unit cell or multiple inequivalent sites (phonon scattering)
def score_thermoelectric(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Band gap: narrow semiconductor or semimetal
if band_gap is not None:
if band_gap < 0.1:
score += 0.30 # semimetallic — good electrical conductivity
elif band_gap < 0.5:
score += 0.35 # ideal narrow gap
elif band_gap < 1.0:
score += 0.20
else:
score += 0.05 # too insulating
# Heavy elements (critical for phonon scattering / low κ)
heavy_elements = {'Pb', 'Bi', 'Sb', 'Te', 'Se', 'Tl', 'In', 'Sn', 'Ge',
'Ba', 'Cs', 'I', 'Br', 'Ag', 'Hg'}
heavy_count = len(elements & heavy_elements)
score += min(heavy_count * 0.12, 0.35)
# Chalcogenide / halide framework (highly anharmonic — key for low κ)
if 'Te' in elements:
score += 0.10
elif 'Se' in elements or 'S' in elements:
score += 0.07
if elements & {'I', 'Br', 'Cl'}:
score += 0.08
# Penalise very light element-dominated structures (high κ)
light_only = elements - {'H', 'Li', 'Be', 'B', 'C', 'N', 'O', 'F', 'Na', 'Mg', 'Al', 'Si', 'P', 'S', 'Cl'}
if not light_only:
score *= 0.4 # all light elements = high thermal conductivity
return min(score, 1.0)
Example targets from GNoME: Ba12SiSn3I8 (#1 in your thermoelectric results), Cs3CeI6.
9. OER Electrocatalyst (Oxygen Evolution)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Catalyst for the oxygen evolution half-reaction in water electrolysis (2H₂O → O₂ + 4H⁺ + 4e⁻). The rate-limiting step in green hydrogen production. Best known: IrO₂ and RuO₂ (expensive), Fe/Co/Ni oxides (cheap but less active), Fe5Si3 (your result). Operates in acidic (PEM electrolyzer) or alkaline (AEL) conditions.
Key requirements: - Metallic or near-metallic (needs to conduct electrons to electrode) - OER-active transition metals: Ir, Ru, Fe, Co, Ni, Mn - Oxide, hydroxide, or oxyhydroxide structure preferred - For acidic OER: Ir and Ru almost essential (stability requirement) - For alkaline OER: Fe, Co, Ni oxides well-established
def score_oer_electrocatalyst(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# OER-active transition metals
acidic_active = {'Ir', 'Ru'} # stable in acid
alkaline_active = {'Fe', 'Co', 'Ni', 'Mn'} # good in alkaline
if elements & acidic_active:
score += 0.40
elif elements & alkaline_active:
score += 0.30
# Must be electronically conductive
if band_gap is not None:
if band_gap < 0.5:
score += 0.25
elif band_gap < 2.0:
score += 0.10
else:
score *= 0.3 # insulating = bad catalyst support
# Oxide / mixed oxide (most OER catalysts activate to oxide in situ)
if 'O' in elements:
score += 0.15
# Silicide electrocatalyst bonus (your Fe5Si3 finding)
if 'Si' in elements and (elements & alkaline_active):
score += 0.10
# Bonus for known catalyst structural families
perovskite_b_sites = {'Fe', 'Co', 'Ni', 'Mn', 'Ir', 'Ru'}
if 'O' in elements and len(elements & perovskite_b_sites) >= 1:
score += 0.10
return min(score, 1.0)
10. HER Electrocatalyst (Hydrogen Evolution)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Catalyst for the hydrogen evolution half-reaction (2H⁺ + 2e⁻ → H₂). The other half of water splitting. Best: Pt (expensive), MoS₂ edge sites, Ni, Co-P. Operates at cathode; different element preferences from OER.
def score_her_electrocatalyst(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# HER-active elements
noble = {'Pt', 'Pd', 'Rh', 'Ir'}
earth_abundant = {'Mo', 'W', 'Ni', 'Co', 'Fe', 'Cu'}
if elements & noble:
score += 0.40
elif elements & earth_abundant:
score += 0.30
# Sulfides and phosphides are excellent HER catalysts (MoS₂, CoP, Ni₂P)
if 'S' in elements and (elements & earth_abundant):
score += 0.20
if 'P' in elements and (elements & earth_abundant):
score += 0.20
if 'Se' in elements and (elements & earth_abundant):
score += 0.15
if 'N' in elements and (elements & earth_abundant):
score += 0.10
# Metallic or semimetallic
if band_gap is not None:
if band_gap < 0.5:
score += 0.20
elif band_gap < 1.5:
score += 0.08
return min(score, 1.0)
11. CO₂ Reduction Catalyst
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Electrochemical or photochemical reduction of CO₂ to fuels/chemicals (CO, formate, methanol, ethanol, ethylene). Cu is uniquely selective for multi-carbon products. Active area of research for carbon utilisation.
def score_co2_reduction(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Cu is uniquely active for C-C coupling in CO2RR
if 'Cu' in elements:
score += 0.35
# Other CO2RR-active metals
co2rr_metals = {'Ag', 'Au', 'Zn', 'Sn', 'Bi', 'In', 'Pb', 'Pd'}
if elements & co2rr_metals:
score += 0.25
# Metallic character (needs to conduct electrons)
if band_gap is not None:
if band_gap < 0.5:
score += 0.20
elif band_gap < 2.0:
score += 0.08
# Oxide surface (many CO2RR catalysts are oxides or oxide-derived)
if 'O' in elements and (elements & {'Cu', 'Ag', 'Zn', 'Sn', 'Bi'}):
score += 0.15
# Penalise if noble-metal free (earth-abundant bonus)
noble = {'Pt', 'Pd', 'Rh', 'Ir', 'Au', 'Ru'}
if not (elements & noble):
score += 0.05
return min(score, 1.0)
12. Photocatalyst (Water Splitting)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Light-driven water splitting using semiconductor photocatalyst suspended in water or as thin film. Both OER and HER occur on the same particle. Band edges must straddle water redox potentials: conduction band < −0.41 V vs NHE (for H₂), valence band > +0.82 V vs NHE (for O₂). Band gap must be > 1.23 eV (thermodynamic minimum) but ideally < 3 eV to absorb visible light.
def score_photocatalyst_water_splitting(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Band gap must straddle water redox: 1.8–3.2 eV optimal for visible light
if 1.8 <= band_gap <= 3.2:
# Peak at ~2.2 eV (best balance of absorption and driving force)
deviation = abs(band_gap - 2.2)
score += max(0.40 - deviation * 0.15, 0.20)
elif 1.23 <= band_gap < 1.8:
score += 0.10 # thermodynamically possible but low driving force
else:
return 0.0
# Known photocatalyst families
oxide_photocatalysts = {'Ti', 'Zn', 'Ga', 'In', 'Nb', 'Ta', 'W', 'Mo', 'Fe', 'Bi'}
if 'O' in elements and (elements & oxide_photocatalysts):
score += 0.25
# Nitride photocatalysts (GaN, Ta₃N₅ family)
if 'N' in elements and (elements & {'Ga', 'Ta', 'Ge', 'C'}):
score += 0.20
# Sulphide photocatalysts (CdS, ZnIn₂S₄ family — but penalise Cd)
if 'S' in elements:
score += 0.10
if 'Cd' in elements:
score *= 0.6
return min(score, 1.0)
DOMAIN 3 — ELECTRONICS AND OPTOELECTRONICS
13. Semiconductor (General)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Your existing semiconductor category, tightened. General-purpose scoring for any semiconductor application — catch-all for compounds that don't fit more specific categories.
def score_semiconductor_general(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
if band_gap is None:
return 0.0
# Semiconductor range
if 0.1 <= band_gap <= 4.0:
# Penalise extremes
if 0.5 <= band_gap <= 3.0:
score += 0.50
else:
score += 0.25
else:
return 0.0
elements = {str(e) for e in composition.elements}
# Known semiconductor families
iv = {'Si', 'Ge', 'C'}
iii_v = {'Ga', 'In', 'Al', 'N', 'P', 'As', 'Sb'}
ii_vi = {'Zn', 'Cd', 'Hg', 'O', 'S', 'Se', 'Te'}
if elements & iv:
score += 0.20
if len(elements & iii_v) >= 2:
score += 0.20
if len(elements & ii_vi) >= 2:
score += 0.15
# Transition metal oxides and chalcogenides
tmo = {'Ti', 'Mo', 'W', 'V', 'Cr', 'Fe', 'Cu', 'Ni', 'Co'}
if 'O' in elements and (elements & tmo):
score += 0.10
return min(score, 1.0)
14. LED / Light Emitter
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Light-emitting diode material. Requires direct band gap in visible range. Different from solar absorber even with same band gap: LED needs high quantum yield, while solar needs high absorption coefficient. GaN (blue LED), AlGaInP (red/green), halide perovskites (tunable colour).
def score_led(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Visible range: 1.77–3.1 eV (700–400 nm)
if 1.77 <= band_gap <= 3.1:
# Preference for middle visible (green ~2.3 eV)
score += 0.45
elif 3.1 < band_gap <= 4.0:
score += 0.20 # near-UV LED applications
else:
return 0.0
# III-V semiconductors (commercial LED family)
iii_v_metals = {'Ga', 'In', 'Al'}
iii_v_anions = {'N', 'P', 'As'}
if (elements & iii_v_metals) and (elements & iii_v_anions):
score += 0.30
# II-VI compounds (ZnSe, ZnS, CdS etc.)
if elements & {'Zn', 'Cd'} and elements & {'S', 'Se', 'Te'}:
score += 0.15
# Halide perovskites for LEDs (rapidly growing)
halides = {'Cl', 'Br', 'I'}
if elements & halides and elements & {'Cs', 'Rb'} and elements & {'Pb', 'Sn', 'In'}:
score += 0.20
# Penalise indirect gap indicators (common in Si-type structures)
# (pymatgen space group analysis could improve this)
if 'Si' in elements and len(elements) == 1:
score *= 0.3 # pure Si is indirect gap
return min(score, 1.0)
15. Photodetector
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Semiconductor device that converts light to electrical signal. Wider band gap range is acceptable than for solar or LED — applications span UV, visible, and infrared. Key families: Si (visible/NIR), InGaAs (telecom NIR), HgCdTe (IR), perovskites (fast response).
def score_photodetector(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Broad range: UV to SWIR
if 0.3 <= band_gap <= 4.5:
score += 0.40
# Sub-ranges have specific applications
if 0.3 <= band_gap < 1.0:
score += 0.10 # IR photodetector
elif 1.0 <= band_gap < 2.0:
score += 0.15 # visible
elif 2.0 <= band_gap <= 3.5:
score += 0.10 # UV
else:
return 0.0
# High-mobility semiconductor families
iii_v = {'Ga', 'In', 'Al'}
anions = {'N', 'P', 'As', 'Sb'}
if (elements & iii_v) and (elements & anions):
score += 0.25
# Perovskites (fast response, high sensitivity)
if elements & {'Pb', 'Sn'} and elements & {'I', 'Br', 'Cl'} and elements & {'Cs', 'Rb'}:
score += 0.20
# HgCdTe family (IR detectors)
if 'Hg' in elements and 'Cd' in elements and 'Te' in elements:
score += 0.15 # technically excellent but toxic
return min(score, 1.0)
16. Transparent Conductor
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Wide band gap material that is also electrically conductive. Seemingly contradictory — achieved by degenerate doping. ITO (In₂O₃:Sn) dominates commercially but In is expensive and scarce. Alternatives: AZO (Al-doped ZnO), FTO (F-doped SnO₂), Ga₂O₃, amorphous oxides.
Key requirements: - Band gap > 3.0 eV (transparent to visible light) - Low effective electron mass (for high conductivity when doped) - Oxides strongly preferred (most known transparent conductors are oxides) - Avoid highly toxic elements
def score_transparent_conductor(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Wide band gap requirement
if band_gap >= 3.5:
score += 0.45
elif band_gap >= 3.0:
score += 0.30
else:
return 0.0 # would absorb visible light
# Known transparent conductor families
tco_metals = {'In', 'Sn', 'Zn', 'Ga', 'Cd', 'Al', 'Ti'}
if 'O' in elements and (elements & tco_metals):
score += 0.35
# Bonus for earth-abundant ITO alternatives (In is scarce)
if 'Zn' in elements and 'O' in elements and 'In' not in elements:
score += 0.15 # AZO/GZO family
if 'Sn' in elements and 'O' in elements:
score += 0.10 # FTO family
# Penalise non-oxides (transparent conductors are almost all oxides)
if 'O' not in elements:
score *= 0.4
return min(score, 1.0)
17. Ferroelectric
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Material with spontaneous electric polarisation that can be reversed by applied field. Used in capacitors, memory devices (FeRAM), sensors, actuators. Key requirement: non-centrosymmetric crystal structure. BaTiO₃ is the archetypal example. Perovskite ABO₃ structure is strongly associated.
def score_ferroelectric(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Non-centrosymmetric space groups if structure available
if structure is not None:
try:
from pymatgen.symmetry.analyzer import SpacegroupAnalyzer
sga = SpacegroupAnalyzer(structure)
sg_num = sga.get_space_group_number()
# Non-centrosymmetric space groups (rough filter)
# Full list would be 68 of 230 space groups
centrosymmetric_sg = set(range(1, 3)) | {10,11,12,13,14,15} | \
set(range(47, 75)) | set(range(83, 89)) | \
set(range(123, 143)) | set(range(147, 149)) | \
set(range(162, 168)) | set(range(175, 177)) | \
set(range(191, 195)) | set(range(200, 207)) | \
set(range(221, 231))
if sg_num not in centrosymmetric_sg:
score += 0.35
except:
pass
# Perovskite-type elements (ABO₃ structure)
a_site = {'Ba', 'Sr', 'Ca', 'Pb', 'Na', 'K', 'Bi'}
b_site = {'Ti', 'Zr', 'Nb', 'Ta', 'Fe', 'Mn', 'W'}
if 'O' in elements and (elements & a_site) and (elements & b_site):
score += 0.35
# Band gap: ferroelectrics are insulators
if band_gap is not None:
if band_gap >= 2.5:
score += 0.20
elif band_gap >= 1.5:
score += 0.10
# Pb-free ferroelectrics (regulatory bonus)
if 'Pb' not in elements and 'O' in elements:
score += 0.10
return min(score, 1.0)
18. Piezoelectric
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Converts mechanical stress to electrical charge (and vice versa). Used in sensors, actuators, ultrasound transducers, energy harvesters. Same non-centrosymmetric requirement as ferroelectric. PZT (Pb-Zr-Ti oxide) dominates industrially but is lead-heavy. KNbO₃, BaTiO₃, AlN are lead-free alternatives.
def score_piezoelectric(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Non-centrosymmetric check (same as ferroelectric)
if structure is not None:
try:
from pymatgen.symmetry.analyzer import SpacegroupAnalyzer
sga = SpacegroupAnalyzer(structure)
sg_num = sga.get_space_group_number()
# Polar space groups are a subset of non-centrosymmetric
polar_sg = {1, 2, 3, 4, 6, 7, 8, 9, 10, 11, 13, 14, 17, 18, 19, 26,
28, 29, 30, 31, 32, 33, 34, 36, 38, 39, 40, 41, 42, 44,
45, 46, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108,
109, 110, 111, 113, 115, 117, 119, 121, 156, 157, 158,
159, 160, 161, 166, 168, 169, 170, 171, 172, 173, 174,
177, 182, 183, 184, 185, 186, 187, 188}
if sg_num in polar_sg:
score += 0.40
except:
pass
# Known piezoelectric element combinations
if 'O' in elements:
piezo_metals = {'Ti', 'Zr', 'Nb', 'Ta', 'Al', 'Ga', 'Zn', 'Li', 'Ba', 'Pb', 'K'}
if elements & piezo_metals:
score += 0.30
# AlN — excellent for high-frequency MEMS
if 'Al' in elements and 'N' in elements:
score += 0.20
# Wide band gap (insulators preferred)
if band_gap is not None and band_gap >= 3.0:
score += 0.15
# Lead-free bonus
if 'Pb' not in elements:
score += 0.15
return min(score, 1.0)
19. Topological Insulator
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Material that is insulating in the bulk but has conducting surface/edge states protected by time-reversal symmetry. Applications in quantum computing, spintronics, low-dissipation electronics. Key families: Bi₂Te₃, Bi₂Se₃, Bi₂Se₂Te, SnTe, Pb₁₋ₓSnₓSe. Heavy elements with strong spin-orbit coupling are essential.
def score_topological_insulator(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Must be an insulator but small gap
if 0.05 <= band_gap <= 0.5:
score += 0.35
elif 0.5 < band_gap <= 1.5:
score += 0.15
else:
return 0.0
# Heavy elements with strong spin-orbit coupling (essential for topology)
soc_heavy = {'Bi', 'Sb', 'Pb', 'Sn', 'Te', 'Se', 'Tl', 'Hg', 'In'}
heavy_count = len(elements & soc_heavy)
score += min(heavy_count * 0.15, 0.40)
# Quintuple-layer structure indicator (Bi₂X₃ family)
if 'Bi' in elements and elements & {'Te', 'Se', 'S'}:
score += 0.20
# IV-VI topological crystalline insulators (SnTe, PbTe type)
if elements & {'Sn', 'Pb', 'Ge'} and elements & {'Te', 'Se', 'S'}:
score += 0.15
return min(score, 1.0)
DOMAIN 4 — MAGNETICS
20. Permanent Magnet
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Hard magnet with high coercivity and remanence — stays magnetised without applied field. Applications: EV motors, wind turbines, hard drives. Nd₂Fe₁₄B (neodymium magnet) dominates. Sm₂Co₁₇ is the high-temperature alternative. Both contain critical REEs.
Key requirements: - Must contain Fe, Co, or Ni (magnetic moment carriers) - REE for anisotropy (Nd, Sm, Dy, Pr) — but score accordingly - No band gap (must be metallic) - High density of magnetic atoms
def score_permanent_magnet(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Core magnetic elements
magnetic_metals = {'Fe', 'Co', 'Ni'}
if not (elements & magnetic_metals):
return 0.0
score += 0.30
# REE for magnetocrystalline anisotropy (what makes a hard magnet)
anisotropy_ree = {'Nd', 'Sm', 'Dy', 'Pr', 'Tb', 'Ho'}
if elements & anisotropy_ree:
score += 0.35 # essential for permanent magnet — despite REE cost
# Boron is almost universal in top permanent magnets (Nd₂Fe₁₄B type)
if 'B' in elements:
score += 0.15
# Must be metallic
if band_gap is not None:
if band_gap < 0.1:
score += 0.15
else:
score *= 0.5 # semiconducting permanent magnets are rare
# Nitrogen interstitials (Sm₂Fe₁₇N₃ type — emerging class)
if 'N' in elements and 'Fe' in elements and 'Sm' in elements:
score += 0.05
return min(score, 1.0)
Viability notes: REE penalty from your existing filter applies here and is correct. Permanent magnets legitimately require REEs — this is the domain where the REE cost is most commercially accepted. Consider reducing the REE penalty multiplier specifically for this category in your viability.py.
21. Soft Magnet
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Low coercivity magnetic material — easy to magnetise and demagnetise. Used in transformer cores, inductors, magnetic shielding, sensors. Fe-Si alloys are the commercial workhorse (electrical steel). Ferrites used in high-frequency applications. No REE required.
def score_soft_magnet(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
magnetic_metals = {'Fe', 'Co', 'Ni'}
if not (elements & magnetic_metals):
return 0.0
score += 0.25
# Fe-Si alloys (electrical steel — dominant soft magnet)
if 'Fe' in elements and 'Si' in elements:
score += 0.30
# Ferrites (Mn-Zn, Ni-Zn, spinel ferrites)
ferrite_metals = {'Mn', 'Zn', 'Ni', 'Cu'}
if 'Fe' in elements and 'O' in elements and (elements & ferrite_metals):
score += 0.25
# Amorphous / nanocrystalline indicators (Fe + B, or Fe + P + C)
if 'Fe' in elements and 'B' in elements:
score += 0.10
if 'Fe' in elements and 'P' in elements:
score += 0.10
# Metallic
if band_gap is not None and band_gap < 0.5:
score += 0.15
# Penalise REE (soft magnets don't need them and they add cost)
ree = {'Nd', 'Sm', 'Dy', 'Pr', 'Tb', 'Gd', 'La', 'Ce', 'Er', 'Ho', 'Tm', 'Yb', 'Lu'}
if elements & ree:
score *= 0.7
return min(score, 1.0)
22. Magnetic Semiconductor / Spintronics
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Material combining semiconductor band gap with magnetic ordering. Used in spin-polarised LEDs, magnetic memory, quantum computing. Dilute magnetic semiconductors (DMS) like (Ga,Mn)As are the main family. Rare earth hydrides from your results fall here.
def score_magnetic_semiconductor(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
if band_gap is None:
return 0.0
# Must be a semiconductor
if not (0.1 <= band_gap <= 3.5):
return 0.0
score += 0.20
# Magnetic 3d transition metals (spin source)
magnetic_3d = {'Mn', 'Fe', 'Co', 'Ni', 'Cr', 'V'}
if elements & magnetic_3d:
score += 0.30
# Rare earth f-electron magnets (4f magnetic moments)
magnetic_ree = {'Gd', 'Eu', 'Dy', 'Nd', 'Sm', 'Tb', 'Ho', 'Er'}
if elements & magnetic_ree:
score += 0.25
# Semiconductor host matrix
semiconductor_host = {'Ga', 'In', 'Ge', 'Si', 'Zn', 'Cd'}
if elements & semiconductor_host:
score += 0.15
# Chalcogenide magnetic semiconductors (EuS, EuO type)
if elements & {'Eu', 'Gd'} and elements & {'S', 'Se', 'O', 'Te'}:
score += 0.10
return min(score, 1.0)
DOMAIN 5 — THERMAL AND STRUCTURAL COATINGS
23. Thermal Barrier Coating (TBC)
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Ceramic coating on turbine blades to insulate metal from hot combustion gases. Enables higher operating temperatures → better engine efficiency. YSZ (yttria-stabilised zirconia) is the current standard but degrades above 1200°C. Hexaaluminates (LaMgAl₁₁O₁₉ type) and pyrochlores (La₂Zr₂O₇) are next-generation candidates. Ba12PrY3Al8O30 from your results fits here.
def score_tbc(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Must be an oxide
if 'O' not in elements:
return 0.0
# Wide band gap (optical transparency, no electronic conduction at high T)
if band_gap is not None:
if band_gap >= 4.0:
score += 0.35
elif band_gap >= 3.0:
score += 0.20
else:
score += 0.05
# Zirconia family (current standard)
if 'Zr' in elements:
score += 0.25
if 'Y' in elements:
score += 0.10 # YSZ — the benchmark
# Hexaaluminate family (next generation)
if 'Al' in elements:
a_site_hex = {'Ba', 'Sr', 'La', 'Ce', 'Pr', 'Nd', 'Sm'}
if elements & a_site_hex:
score += 0.30 # magnetoplumbite/β-alumina structure
# Pyrochlore family (A₂B₂O₇)
pyro_a = {'La', 'Nd', 'Sm', 'Gd', 'Er', 'Yb', 'Y'}
pyro_b = {'Zr', 'Ti', 'Hf', 'Ce', 'Sn'}
if (elements & pyro_a) and (elements & pyro_b):
score += 0.25
# High melting point indicator: refractory metal oxides
refractory_ox = {'Hf', 'Ta', 'W', 'Nb', 'Mo'}
if elements & refractory_ox:
score += 0.10
# Very stable formation energy (must survive high temperature)
if formation_energy is not None and formation_energy < -3.0:
score += 0.10
return min(score, 1.0)
24. Thermal Interface Material
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: High thermal conductivity, electrically insulating material placed between heat source and heatsink. Used in power electronics packaging. Key: high κ (>10 W/m·K), wide band gap. AlN (κ ~180 W/m·K), BN (~300 W/m·K), diamond (2000 W/m·K), BeO (270 W/m·K).
def score_thermal_interface(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Must be electrically insulating
if band_gap is not None:
if band_gap >= 5.0:
score += 0.35
elif band_gap >= 3.5:
score += 0.25
elif band_gap >= 2.5:
score += 0.10
else:
return 0.0
# High thermal conductivity material families
# Diamond and cBN: highest thermal conductors
if elements == {'C'}: # diamond
score += 0.50
if 'B' in elements and 'N' in elements:
score += 0.40 # BN family
# Aluminium nitride (very common in power electronics)
if 'Al' in elements and 'N' in elements:
score += 0.35
# Silicon carbide (also high κ)
if 'Si' in elements and 'C' in elements:
score += 0.25
# Beryllium oxide (extremely high κ but toxic)
if 'Be' in elements and 'O' in elements:
score += 0.20
score *= 0.5 # BeO is highly toxic — flag this
# Simple binary/ternary structure (complex structures have lower κ)
n_elements = len(composition.elements)
if n_elements <= 2:
score += 0.10
elif n_elements >= 5:
score *= 0.7
return min(score, 1.0)
25. Hard Coating / Wear Resistant
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Hard ceramic coating on cutting tools, dies, engine components. TiN, TiAlN, CrN, DLC (diamond-like carbon), Al₂O₃ are commercial standards. Key property: high hardness (>20 GPa Vickers), which correlates with high bulk modulus and short strong bonds. Transition metal nitrides, carbides, and borides dominate.
def score_hard_coating(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Hard material element families (nitrides, carbides, borides)
transition_metals = {'Ti', 'Cr', 'W', 'Mo', 'V', 'Nb', 'Ta', 'Zr', 'Hf', 'Al'}
hard_anions = {'N', 'C', 'B'}
if (elements & transition_metals) and (elements & hard_anions):
score += 0.45
# Pure carbon (diamond, DLC)
if elements == {'C'}:
score += 0.40
# Al₂O₃ (alumina coatings — CVD on cutting tools)
if elements == {'Al', 'O'} or ('Al' in elements and 'O' in elements and len(elements) <= 3):
score += 0.25
# Ternary nitrides/carbides (better performance than binary)
if len(elements & transition_metals) >= 2 and (elements & hard_anions):
score += 0.15 # e.g. TiAlN better than TiN
# Wide band gap (hardness correlates with ionic/covalent bonding)
if band_gap is not None:
if band_gap >= 3.0:
score += 0.15
elif band_gap >= 1.5:
score += 0.08
# Very negative formation energy (stable at high temperatures)
if formation_energy is not None and formation_energy < -2.5:
score += 0.10
return min(score, 1.0)
26. Corrosion Resistant Coating
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Protective coating to prevent oxidation, acid attack, or electrochemical corrosion. Passive oxide layers (Cr₂O₃ on stainless steel), fluorides, and noble metal oxides. Chemically inert, wide band gap, thermodynamically stable.
def score_corrosion_resistant(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Chromia-forming / passivating oxide formers
passive_metals = {'Cr', 'Al', 'Ti', 'Zr', 'Ta', 'Nb', 'Hf', 'Si', 'W'}
if 'O' in elements and (elements & passive_metals):
score += 0.35
# Fluoride coatings (extremely chemically inert)
if 'F' in elements:
fluoride_metals = {'Ca', 'Ba', 'Sr', 'Mg', 'La', 'Ce', 'Al'}
if elements & fluoride_metals:
score += 0.30
# Wide band gap and chemically stable
if band_gap is not None and band_gap >= 3.5:
score += 0.20
# Very negative formation energy (thermodynamic stability = corrosion resistance)
if formation_energy is not None and formation_energy < -4.0:
score += 0.20
elif formation_energy is not None and formation_energy < -2.5:
score += 0.10
# Noble metals (excellent corrosion resistance)
noble = {'Pt', 'Pd', 'Au', 'Ir', 'Rh', 'Ru'}
if elements & noble:
score += 0.15
return min(score, 1.0)
27. Refractory / Ultra-High Temperature Structural
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Structural materials for use above 1500°C. Turbine hot sections, hypersonic vehicles, nuclear reactor components. Key families: HfC, ZrB₂, TaC, HfB₂ (UHTCs — ultra-high temperature ceramics), W and Re alloys, transition metal carbides.
def score_refractory(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Refractory transition metals (melting point > 2000°C)
refractory_metals = {'W', 'Re', 'Os', 'Ta', 'Mo', 'Hf', 'Nb', 'Zr', 'V', 'Cr', 'Ti'}
if elements & refractory_metals:
score += 0.30
# UHTC carbides and borides (HfC, ZrB₂ family)
uhtc = {'Hf', 'Zr', 'Ta', 'Ti', 'Nb'}
if (elements & uhtc) and ('C' in elements or 'B' in elements or 'N' in elements):
score += 0.35
# Very stable thermodynamics
if formation_energy is not None and formation_energy < -3.0:
score += 0.20
elif formation_energy is not None and formation_energy < -1.5:
score += 0.10
# No volatile elements (must survive high temperature)
volatile = {'Na', 'K', 'Li', 'Rb', 'Cs', 'Zn', 'Cd', 'Hg', 'S', 'Se', 'Te'}
if elements & volatile:
score *= 0.4
# Wide band gap or metallic (both types exist in refractory materials)
if band_gap is not None:
if band_gap < 0.5 or band_gap > 4.0:
score += 0.15
return min(score, 1.0)
DOMAIN 6 — EMERGING AND SPECIALISED
28. Superconductor
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Zero electrical resistance below critical temperature Tc. Applications: MRI magnets, particle accelerators, quantum computing, lossless power transmission. Conventional BCS superconductors (Nb, NbTi, Nb₃Sn) require liquid He cooling. High-Tc cuprates (YBCO) work at 77K (liquid N₂). Your rare earth hydrides under pressure were showing up in superconductor prediction papers.
def score_superconductor(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Must be metallic
if band_gap is not None and band_gap > 0.5:
return 0.0
score += 0.20
# BCS/conventional superconductor elements
conventional = {'Nb', 'V', 'Pb', 'Sn', 'In', 'Al', 'Mo', 'Re', 'W'}
if elements & conventional:
score += 0.25
# Cuprate high-Tc indicators (Cu + O + REE or Ba/Sr)
if 'Cu' in elements and 'O' in elements:
cuprate_a_site = {'Ba', 'Sr', 'La', 'Y', 'Bi', 'Tl', 'Hg'}
if elements & cuprate_a_site:
score += 0.35
# Iron-based superconductor indicators (FeAs, FeSe planes)
if 'Fe' in elements and elements & {'As', 'Se', 'P'}:
score += 0.25
# Hydride superconductors (high-Tc under pressure)
if 'H' in elements:
hydride_hosts = {'La', 'Y', 'Ce', 'Th', 'Ca', 'Ba', 'Lu'}
if elements & hydride_hosts:
score += 0.20
return min(score, 1.0)
29. Radiation Detector / Scintillator
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Converts high-energy radiation (X-ray, gamma, neutron) into detectable signal. Medical imaging (CT, PET), security screening, nuclear monitoring. BaI₂:Eu, CsI:Tl, LYSO (Lu₂SiO₅:Ce), NaI:Tl are commercial scintillators. Your BaI₂ family compounds from the thermoelectric results are actually relevant here.
Key requirements: - High density (heavy elements for gamma absorption) - Wide enough band gap to be a scintillator (>3 eV) OR narrow gap for semiconductor detector - Contains luminescence-activatable sites (rare earth dopant sites) - High Z elements (Ba, I, Lu, Bi, Hg, Pb) for radiation stopping power
def score_radiation_detector(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# High-Z elements (high radiation stopping power)
high_z = {'Ba', 'I', 'Cs', 'Bi', 'Pb', 'Tl', 'Hg', 'W', 'Lu', 'Gd', 'Xe'}
high_z_count = len(elements & high_z)
score += min(high_z_count * 0.15, 0.40)
# Scintillator band gap (needs to emit visible photons)
if band_gap is not None:
if 3.0 <= band_gap <= 6.0:
score += 0.30 # scintillator window
elif band_gap < 3.0:
score += 0.10 # semiconductor detector (Ge, CdTe type)
# Known scintillator families
# Alkali halides (NaI, CsI, BaI₂ type)
alkali = {'Na', 'Cs', 'K', 'Ba', 'Sr'}
halides = {'I', 'Br', 'Cl', 'F'}
if (elements & alkali) and (elements & halides):
score += 0.25
# Oxide scintillators (LYSO, BGO, LSO type)
scint_oxides = {'Lu', 'Gd', 'Bi', 'Y', 'Ce'}
if 'O' in elements and (elements & scint_oxides):
score += 0.20
# Gd-containing for neutron detection
if 'Gd' in elements:
score += 0.10
return min(score, 1.0)
30. Solid Oxide Fuel Cell (SOFC) Component
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: High-temperature electrochemical device converting fuel to electricity. Operates at 600–1000°C. Three key components with different material requirements: electrolyte (oxide ion conductor, wide gap), cathode (mixed ionic-electronic conductor), anode (electronic conductor + fuel oxidation catalyst). YSZ is the standard electrolyte.
def score_sofc_electrolyte(composition, structure, band_gap, formation_energy, e_hull, features):
"""Scores for SOFC electrolyte component (oxide ion conductor)."""
score = 0.0
elements = {str(e) for e in composition.elements}
# Must be an oxide
if 'O' not in elements:
return 0.0
# Wide band gap (electronic insulator)
if band_gap is not None:
if band_gap >= 4.0:
score += 0.35
elif band_gap >= 3.0:
score += 0.20
else:
return 0.0
# Known oxide ion conductor families
# Fluorite structure (ZrO₂, CeO₂, Bi₂O₃)
fluorite = {'Zr', 'Ce', 'Hf', 'Th'}
if 'O' in elements and (elements & fluorite):
score += 0.30
if 'Y' in elements:
score += 0.15 # YSZ benchmark
# Perovskite oxide conductors (LaGaO₃ family)
perovskite_sofc = {'La', 'Sr', 'Ga', 'Mg'}
if 'O' in elements and len(elements & perovskite_sofc) >= 2:
score += 0.25
# BIMEVOX family (Bi₄V₂O₁₁ derivatives)
if 'Bi' in elements and 'V' in elements and 'O' in elements:
score += 0.20
# High temperature stability (very negative formation energy)
if formation_energy is not None and formation_energy < -3.5:
score += 0.10
return min(score, 1.0)
31. Multiferroic
Scoring Systems Order - Gaussian (continuous) scoring: listed first and treated as the primary model. - Discrete (legacy) scoring: listed second for backward-compatible comparison.
What it is: Rare class of material simultaneously showing ferroelectric AND magnetic ordering. Allows electric field control of magnetic properties — huge for low-power memory. BiFeO₃ is the main known room-temperature multiferroic. Very few compounds qualify.
def score_multiferroic(composition, structure, band_gap, formation_energy, e_hull, features):
score = 0.0
elements = {str(e) for e in composition.elements}
# Must have magnetic transition metal
magnetic_metals = {'Fe', 'Mn', 'Co', 'Ni', 'Cr', 'V', 'Cu'}
if not (elements & magnetic_metals):
return 0.0
score += 0.20
# Non-centrosymmetric structure check
if structure is not None:
try:
from pymatgen.symmetry.analyzer import SpacegroupAnalyzer
sga = SpacegroupAnalyzer(structure)
if not sga.is_laue():
score += 0.25
except:
pass
# Perovskite-type structure with both magnetic and polar B-site
a_site = {'Bi', 'Pb', 'Ba', 'La', 'Ca', 'Sr'}
polar_b = {'Ti', 'V', 'Nb', 'Ta', 'W', 'Mo'}
magnetic_b = {'Fe', 'Mn', 'Co', 'Ni', 'Cr'}
if 'O' in elements and (elements & a_site):
if (elements & polar_b) and (elements & magnetic_b):
score += 0.35 # classic multiferroic setup
elif elements & magnetic_b:
score += 0.15
# BiFeO₃ family (the most studied multiferroic)
if 'Bi' in elements and 'Fe' in elements and 'O' in elements:
score += 0.20
# Semiconductor gap (multiferroics tend to be insulators/semiconductors)
if band_gap is not None and 1.0 <= band_gap <= 4.0:
score += 0.10
return min(score, 1.0)
Integration Into scoring.py
Replace your current scoring block with the full registry:
# scoring.py — Application Score Registry
SCORING_FUNCTIONS = {
# Energy Storage
'battery_cathode_liion': score_battery_cathode_liion,
'battery_anode': score_battery_anode,
'battery_cathode_naion': score_battery_cathode_naion,
'solid_electrolyte': score_solid_electrolyte,
'hydrogen_storage': score_hydrogen_storage,
# Energy Conversion
'solar_singlejunction': score_solar_absorber_singlejunction,
'solar_tandem': score_solar_absorber_tandem,
'thermoelectric': score_thermoelectric,
'oer_electrocatalyst': score_oer_electrocatalyst,
'her_electrocatalyst': score_her_electrocatalyst,
'co2_reduction': score_co2_reduction,
'photocatalyst_h2o': score_photocatalyst_water_splitting,
# Electronics and Optoelectronics
'semiconductor': score_semiconductor_general,
'led': score_led,
'photodetector': score_photodetector,
'transparent_conductor': score_transparent_conductor,
'ferroelectric': score_ferroelectric,
'piezoelectric': score_piezoelectric,
'topological_insulator': score_topological_insulator,
# Magnetics
'permanent_magnet': score_permanent_magnet,
'soft_magnet': score_soft_magnet,
'magnetic_semiconductor': score_magnetic_semiconductor,
# Thermal and Structural Coatings
'thermal_barrier': score_tbc,
'thermal_interface': score_thermal_interface,
'hard_coating': score_hard_coating,
'corrosion_resistant': score_corrosion_resistant,
'refractory': score_refractory,
# Emerging and Specialised
'superconductor': score_superconductor,
'radiation_detector': score_radiation_detector,
'sofc_electrolyte': score_sofc_electrolyte,
'multiferroic': score_multiferroic,
}
def score_all_applications(composition, structure, band_gap,
formation_energy, e_hull, features):
"""Run all 31 scorers and return a dict of application → score."""
return {
name: func(composition, structure, band_gap,
formation_energy, e_hull, features)
for name, func in SCORING_FUNCTIONS.items()
}
Streamlit Dropdown Labels
Add these human-readable labels to your app.py dropdown:
APPLICATION_LABELS = {
'battery_cathode_liion': 'Battery Cathode (Li-ion)',
'battery_anode': 'Battery Anode',
'battery_cathode_naion': 'Battery Cathode (Na-ion)',
'solid_electrolyte': 'Solid Electrolyte',
'hydrogen_storage': 'Hydrogen Storage',
'solar_singlejunction': 'Solar Absorber — Single Junction',
'solar_tandem': 'Solar Absorber — Tandem Top Cell',
'thermoelectric': 'Thermoelectric',
'oer_electrocatalyst': 'OER Electrocatalyst (Water Splitting)',
'her_electrocatalyst': 'HER Electrocatalyst (Green Hydrogen)',
'co2_reduction': 'CO₂ Reduction Catalyst',
'photocatalyst_h2o': 'Photocatalyst (Water Splitting)',
'semiconductor': 'Semiconductor (General)',
'led': 'LED / Light Emitter',
'photodetector': 'Photodetector',
'transparent_conductor': 'Transparent Conductor',
'ferroelectric': 'Ferroelectric',
'piezoelectric': 'Piezoelectric',
'topological_insulator': 'Topological Insulator',
'permanent_magnet': 'Permanent Magnet',
'soft_magnet': 'Soft Magnet',
'magnetic_semiconductor': 'Magnetic Semiconductor / Spintronics',
'thermal_barrier': 'Thermal Barrier Coating',
'thermal_interface': 'Thermal Interface Material',
'hard_coating': 'Hard Coating / Wear Resistant',
'corrosion_resistant': 'Corrosion Resistant Coating',
'refractory': 'Refractory / UHTC',
'superconductor': 'Superconductor',
'radiation_detector': 'Radiation Detector / Scintillator',
'sofc_electrolyte': 'Solid Oxide Fuel Cell Electrolyte',
'multiferroic': 'Multiferroic',
}
Viability Override: Permanent Magnet REE Exception
In viability.py, add a category-specific override so the rare earth penalty is softened for permanent magnets (REEs are legitimately required and commercially accepted there):
def apply_ree_penalty(composition, base_viability, application_category):
"""
Apply rare earth penalty with category-specific override.
Permanent magnets legitimately need REEs — reduce penalty for them.
"""
elements = {str(e) for e in composition.elements}
# REE penalty tiers (from the main viability scorer)
hard_ree = {'Dy', 'Tb', 'Eu', 'Ho', 'Er', 'Tm', 'Lu', 'Yb'}
moderate_ree = {'Nd', 'Pr', 'Sm', 'Gd', 'Sc'}
mild_ree = {'La', 'Ce', 'Y'}
if application_category == 'permanent_magnet':
# REE is expected here — halve the penalty
hard_mult = 0.65 # vs 0.3 normally
moderate_mult = 0.80 # vs 0.6 normally
mild_mult = 0.95 # vs 0.85 normally
else:
hard_mult = 0.30
moderate_mult = 0.60
mild_mult = 0.85
multiplier = 1.0
for e in elements:
if e in hard_ree:
multiplier *= hard_mult
elif e in moderate_ree:
multiplier *= moderate_mult
elif e in mild_ree:
multiplier *= mild_mult
return base_viability * multiplier
Known Caveats and GNoME-Specific Issues
Band gap underestimation: DFT-predicted band gaps (including CHGNet estimates) typically underestimate true values by 30–50%. A compound predicted at 0.8 eV may actually be 1.1–1.3 eV. Consider applying a 1.3× correction factor to band gaps when comparing against target ranges, or widen all target windows by ~0.3 eV on the lower bound.
Disordered structures: Some GNoME CIFs have partial occupancies. The space group symmetry
checks in ferroelectric/piezoelectric/multiferroic scorers will fail on disordered structures.
Wrap all SpacegroupAnalyzer calls in try/except and return a neutral 0.5 score if they fail.
Structure=None fallback: For categories that rely heavily on space group (ferroelectric, piezoelectric, multiferroic), the score will be substantially lower when structure analysis fails. This is intentional — these are structure-sensitive properties and a composition-only score is genuinely uncertain.
E_hull context: GNoME guarantees all 240K compounds are below the convex hull (E_hull ≤ 0). However, GNoME used its own DFT settings which may differ slightly from Materials Project. For very borderline compounds (E_hull between -5 and 0 meV/atom), treat stability as uncertain.
Radioactive elements: The radioactive element hard filter (atomic number > 83, plus Tc) should be applied BEFORE any application scoring — if a compound contains Np, Ac, Th, etc., return 0.0 viability regardless of how good the application score is. This is already implemented in your viability.py.
Document version: 1.0 — covers all 31 MatIntel application categories Last updated for GNoME dataset compatibility