jlaw9 / crystal-gnn Goto Github PK
View Code? Open in Web Editor NEWThis project forked from pstjohn/crystal-gnn
Work in progress on graph neural networks for crystal energy prediction
This project forked from pstjohn/crystal-gnn
Work in progress on graph neural networks for crystal energy prediction
One thing I noticed after Jiaxing ran the DFT relaxations for the 128 structures from the MCTS run was that 7 of the structures already had DFT relaxations. I checked the difference of eV/atom between the two runs (see table below). 1 structure had a difference of 220 meV/atom, 2 had > 20 meV/atom, and 4 had a difference < 2 meV/atom.
Jiaxing, can you double check the DFT settings?
id | Q4_energyperatom | orig_energyperatom | diff_energyperatom |
---|---|---|---|
K5O1P1_sg15_icsd_156312_2 | -2.790 | -2.791 | 0.001 |
K5Zn2N3_sg159_icsd_087349_1 | -2.797 | -2.576 | -0.220 |
Zn5S2P2_sg194_icsd_049912_1 | -3.157 | -3.155 | -0.002 |
Zn5S2N2_sg194_icsd_049912_1 | -3.581 | -3.581 | 0.000 |
K5S1N1_sg15_icsd_156312_1 | -2.677 | -2.654 | -0.022 |
Zn3Sc1P3_sg198_icsd_000645_2 | -3.948 | -3.984 | 0.036 |
Zn4Ti1N4_sg62_icsd_095649_1 | -5.100 | -5.100 | 0.000 |
Prashun's reply: If the starting structures are identical, then they should converge to within 1-2 meV/atom (assuming same DFT parameters are used). Could you add this list and question to the stability Gist? I find it hard to keep track through emails. I've informed Jiaxing where she can find the original calculations.
Question: How do we know that MCTS labeling of the decoration i.e., _1, _2, ... is the same as the decoration labeling by Shubham's script?
@prashungorai You're correct that there could be cases where the decoration labeling originally used by Shubham's script corresponds to a different elemental substitution than the decoration labeling used by my script. I double checked these 7 structures and in this case, they all have identical starting structures.
@prashungorai, @pstjohn, While we're on the subject, I was thinking about a more informative decoration label to remove this potential confusion. Rather than a decoration index, we could directly list the elemental substitutions. For example for Zn4Ti1N4_sg62_icsd_095649_1
, Ti
replaces Ge
, N
replaces S
, and Zn
replaces Li
. We could use the label Zn4Ti1N4_sg62_icsd_095649_N-S_Ti-Ge_Zn-Li
(the substitution string could be sorted by battery element).
Once we agree on a nomenclature, I can go back and create the mappings for the battery dataset and update the MCTS action space to use the new labels.
There are three composition types without any DFT relaxations: _1_1_1
, _1_1_1_1
, and _1_1_2_2_2
.
@prashungorai Can you copy over the rest of the DFT relaxations to our shared /projects/rlmolecule
space? Or maybe Shubham's location here: /projects/rlmolecule/shubham/file_transfer/decorations/relaxed/battery_hypotheticals
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.