I'm working on a Machine Learning project applied to the materials science field, but without having consolidated domain knowledge of the latter. What I'm trying to figure out is the following: let's say we are given a (for example) quaternary chemical system made by $A_xB_yC_zD_w$ where capital letters represent chemical elements and lower cases represent the number of atoms. Is there a way I could classify a priori what kind of material I have? (for example metal, semiconductor or insulator..?) Could we potentially rely on the set of elements (phase field) that form the composition? Even just giving an estimate with a certain probability would be interesting.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
No, there is no way to do this, at least not in the way you have in mind. As a simple example, consider the unary system of elemental carbon:
If the carbon is in the diamond structure, you will get a wide band-gap insulator;
If the carbon is in the graphite (or graphene) structure, you will have 2D planar conductivity;
If the carbon forms nanotubes, the conductivity depends on how the hexagons wrap around the tube (multiples of three lead to metallic nanotubes; others are insulating);
If the carbon forms nanoribbons, then the conductivity depends on the direction and edges of the ribbon.
Note that all of these forms are stable at ambient pressure and temperature. If you consider more extreme conditions, even richer behaviour can arise!
-
1$\begingroup$ Nice example! Just one additional comment: if the machine learning model is limited to predicting the property of the most stable compound and phase of the given chemical composition, then the question becomes well-defined, but still extremely difficult. $\endgroup$– wzkchem5Sep 23, 2022 at 18:41
-
1$\begingroup$ @wzkchem5 true. If you can input the atomic coordinates that also gives the ML a chance, but species-along is extraordinarily difficult. $\endgroup$ Sep 23, 2022 at 21:05