READ ME File For 'Computational data for: "Machine Learned Potentials by Active Learning from Organic Crystal Structure Prediction Landscapes"'

Dataset DOI: 10.5258/SOTON/D2840

Date that the file was created: October, 2023

-------------------
GENERAL INFORMATION
-------------------

ReadMe Author: Patrick Butler, University of Southampton


This dataset supports the publication:
AUTHORS:Butler, Patrick W. V., Hafizi, Roohollah and Day, Graeme M.
TITLE:Machine Learned Potentials by Active Learning from Organic Crystal Structure Prediction Landscapes
JOURNAL:
PAPER DOI IF KNOWN:


--------------------------
SHARING/ACCESS INFORMATION
-------------------------- 

Licenses/restrictions placed on the data, or limitations of reuse: CC-BY

--------------------
DATA & FILE OVERVIEW
--------------------

Computational data for "Reducing overprediction of molecular crystals via threshold clustering"

This repository contains the landscapes described in the article. For each system the structures 
before and after threshold clustering are given. Additionally csv files are provided that contain 
the calculated energy and density for all structures found and for the basin minima identified 
from threshold clustering.


The dataset contains:

    Oxalic_Acid_landscape.zip
        committee_NNP:
            contains weights for each of the 5 folds in the 5 cross validation using 
            parameters listed in entry 6 of Table 1 in the manuscript. Also includes 
            the training sets and NNP settings.
        csp:
            contains the structures and data predicted for oxalic acid lattice 
            energy minimized with the FIT+DMA model

    Oxalic_Acid_on_the_fly.zip
        1_cNNP_before_OTF:
            contains weights for the 6 member committee neural network potential 
            trained on the predicted oxalic acid landscape before on-the-fly 
            training. Includes the training set and NNP settings.
        2_cNNP_after_OTF:
            contains weights for the 6 member committee neural network potential 
            trained on the predicted oxalic acid landscape after on-the-fly 
            training. Includes the training set and NNP settings.
        fit+dma_trajectories:
            contains the structures sampled from the FIT+DMA MC trajectories initiated 
            from the alpha and beta oxalic acid polymorphs and the energies estimated 
            by the FIT+DMA, the cNNP before on-the-fly training, and the cNNP after 
            on-the-fly training.
        Oxalic_acid_top300:
            contains the 300 lowest energy structures from the FIT+DMA CSP of oxalic 
            acid and the 10 structures sample from this set by farthest point sampling.
    
    Resorcinol_landscape.zip
        committee_NNP:
            contains weights for the 18 member committee neural network potential 
            used to correct the initial CSP landscape. Includes the training set 
            and NNP settings.
        csp:
            contains the structures and data predicted for resorcinol energy minimized 
            at with DFTB-D3
        predictions:
            The predicted corrections for the low level CSP landscape to PBE-D3 level

    TTBI_landscape.zip
        committee_NNP:
            contains weights for the 18 member committee neural network potential 
            used to correct the initial CSP landscape. Includes the training set 
            and NNP settings.
        csp:
            contains the structures and data predicted for TTBI lattice 
            energy minimized with the FIT+DMA model
        predictions:
            The predicted corrections for the low level CSP landscape to PBE-D3 level

    figures.zip
	contains jupyter notebooks for generating the figures in the main text