Table 2 Subset of Potential Stable Compounds Predicted using ElemNet. Out of the 450 M predictions, we determined the number of systems where ElemNet identifies at least one new potential stable compound.

From: ElemNet: Deep Learning the Chemistry of Materials From Only Elemental Composition

Category

Binary

Ternary

Quaternary

Count

Examples

Count

Examples

Count

Examples

[Li,K,Na]-Containing

4

KF6 NaF8

707

NaY 2 F 7 KY 2 F 7

18446

CsNa2CdF4 Na2CrPbF5

Chalco-/oxyhalides

5

OF9 SeF9

522

Y2OF6 Sc2OF7

17184

Sr3Cu2IO4 Zr6RhIO2

Metal Oxides

1

Cu 2 O

81

KTi4O5 ReAu2O5

501

YAlV2O6 Y4FeBi2O3

3d Metal Oxides

1

Cu 2 O

3

Zn2(CuO)3 Ti5CuO2

1

TiZnCrO5

Intermetallics

11

Nb5Sn3 Al 5Ir3

123

HfAl5Ir3 YAl4Ir3

425

Sc5NiSn3Mo ZrAl5OsRh

Intermetallics HHIp < 2500

0

 

0

 

1

NaMn2AlAu6

  1. We list the number of binary, ternary, and quaternary systems for several categories of compounds along with the two most stable predictions. We validated some of the these compounds- NaY2F7 and KY2F7 using DFT computations by leveraging crystal structures of existing materials with similar stoi-chemistry; we found them to be stable using DFT, further literature search revealed that they have already been synthesized recently. Our model predicts Cu2O as the only new binary oxide which is a known compound but was not in our training set.
Back to article page