Artificial intelligence based stable materials discovery process

Patent metadata

Field	Value
Publication number	US-2019171793-A1
Application number	US-201715832278-A
Country	US
Kind code	A1
Filing date	Dec 5, 2017
Priority date	Dec 5, 2017
Publication date	Jun 6, 2019
Grant date	—

Abstract

An existing materials database (EMDB) is a compilation of inorganic materials with composition and crystal structures known from prior experimental synthesis and characterization reports, or from ab initio or other computational studies, and includes a composition, structure, and stability value for each material. A hypothetical materials database (HMDB) is an extremely large compilation of materials of unknown stability and synthesizability, with no explicitly available or accessible prior experimental or computational report of their structure-composition combinations. An automated process for efficiently expanding the size of an EMDB includes a cyclical sub-process in which a rapid algorithm provides preliminary stability estimates for hypothetical materials selected from an HMDB, and those materials with a favorable stability prediction undergo a full ab initio analysis to obtain quantitative stability values and are then added to the EMDB. During each iteration of the cyclical sub-process, the rapid algorithm is trained on the EMDB, so that it becomes more effective at providing preliminary stability estimates with each iteration, as the EMDB expands.

First claim

what is claimed is:

1 . An automated process for expanding an existing materials database (EMDB), the process comprising:
providing a hypothetical materials database (HMDB), the HMDB comprising a compilation of hypothetical materials, each hypothetical material having:
a hypothetical composition; and
a crystal structure, including atomic coordinates in a unit cell and dimensions and shape of the unit cell; and
performing a cyclical sub-process at least two times, the cyclical sub-process comprising:
training a predictive model (PM) to generate a preliminary stability value for a material based on inputs of structure and composition, the training based on contents of the EMDB, the contents of the EMDB comprising a compilation of inorganic materials, and for each material in the compilation of inorganic materials:
a composition;
a crystal structure, including atomic coordinates in a unit cell and dimensions and shape of the unit cell; and
a stability value;

applying the PM to a set of hypothetical materials comprising at least a portion of the HMDB to produce a preliminary stability value for each hypothetical material in the set of hypothetical materials;
selecting a subset of the set of hypothetical materials, the subset consisting of all hypothetical materials in the set having a preliminary stability value meeting a predetermined threshold value requirement;
performing ab initio calculations on the subset, to determine a quantitative stability value for each hypothetical material in the subset, thereby converting each hypothetical material in the subset to a new material; and
adding each new material in the subset, with its composition and crystal structure, to the EMDB to generate a larger EMDB.

2 . The process as recited in claim 1 , further comprising:
generating the HMDB by decorating one or more prototype crystal structures with selected chemical elements.

3 . The process as recited in claim 2 , wherein generating the HMDB comprises restricting the selected chemical elements to elements having electronegativity within a specified number of Pauling units of an average electronegativity of elements occupying the same site in known materials having the same prototype structure.

4 . The process as recited in claim 2 , wherein generating the HMDB comprises restricting the selected chemical elements to elements having atomic radius within a specified percentage of a radius of an element occupying the same site in known materials having the same prototype structure.

5 . The process as recited in claim 1 , wherein the set consists of a portion of the HMDB.

6 . The process as recited in claim 1 , further comprising creating the set of hypothetical materials by randomly selecting a portion of the HMDB.

7 . The process as recited in claim 1 , further comprising creating the set of hypothetical materials by selecting only hypothetical materials in the HMDB having an identical prototype crystal structure.

8 . The process as recited in claim 1 , further comprising creating the set of hypothetical materials by selecting only hypothetical materials in the HMDB having one or more of the same chemical elements.

9 . The process as recited in claim 1 , wherein the set is provided by selecting only hypothetical materials in the HMDB having all chemical elements in common.

10 . The process as recited in claim 1 , wherein the cyclical sub-process is performed for at least ten iterations.

11 . The process as recited in claim 1 , wherein the cyclical sub-process is performed for at least a hundred iterations.

12 . The process as recited in claim 1 , wherein the cyclical sub-process is repeated until the hypothetical materials available in the HMDB are exhausted.

13 . The process as recited in claim 1 , wherein the cyclical sub-process is repeated until the PM reaches a predictive accuracy identical to the ab initio calculations, as measured with respect to a ground-truth in the EMDB.

14 . The process as recited in claim 1 , wherein the cyclical sub-process further comprises removing each new material from the HMDB.

15 . The process as recited in claim 1 , wherein selecting the subset comprises creating a discard set including all hypothetical materials in the set not having a preliminary stability value meeting the predetermined threshold value requirement.

16 . The process as recited in claim 15 , further comprising deleting all hypothetical materials in the discard set from the HMDB.

17 . An automated process for expanding an existing materials database (EMDB), the process comprising:
performing a cyclical sub-process at least two times, the cyclical sub-process comprising:
training a predictive model (PM) to generate a preliminary stability value for a material based on inputs of structure and composition, the training based on contents of the EMDB, the contents of the EMDB comprising a compilation of inorganic materials, and for each material in the compilation of inorganic materials:
a composition;
a crystal structure, including atomic coordinates in a unit cell, and dimensions and shape of the unit cell; and
a stability value;

applying the PM to a set of hypothetical materials comprising a portion of a hypothetical materials database (HMDB) to produce a preliminary stability value for each hypothetical material in the set of hypothetical materials, the HMDB a hypothetical materials database (HMDB), the HMDB comprising a compilation of hypothetical materials, each hypothetical material having:
a hypothetical composition; and
a crystal structure, including atomic coordinates in a unit cell; and

selecting a subset of the set of hypothetical materials, the subset consisting of all hypothetical materials in the set having a preliminary stability value meeting a predetermined threshold value requirement;
performing ab initio calculations on the subset, to determine a quantitative stability value for each hypothetical material in the subset, thereby converting each hypothetical material in the subset to a new material; and
adding each new material in the subset, with its composition and crystal structure, to the EMDB to generate a larger EMDB.

18 . The process as recited in claim 17 , further comprising generating the HMDB prior to performing the cyclical sub-process.

19 . The process as recited in claim 17 , further comprising:
creating the set of hypothetical materials; and removing from the HMDB all hypothetical materials in the set of hypothetical materials.

20 . The process as recited in claim 19 , wherein creating the set comprises randomly selecting a portion of the HMDB.

Assignees

• Toyota Res Inst Inc

Inventors

• Aykol Muratahan

Classifications

• G06F19/702

  Physics · mapped topic

• G06F19/707

  Physics · mapped topic

• G06F19/709 Primary

  Physics · mapped topic

• G16C60/00

  Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation · CPC title

• G16C20/30

  Prediction of properties of chemical compounds, compositions or mixtures · CPC title