Data for Deep Learning of ab initio Hessians for Transition State Optimization

<pre>This repo contains data generated in a reasearch regarding machine learning transition state search.<br><br><br>Data (outputs.zip) are in form of json files corresponding to inititial transition state guess structures, optimized transition states, and corresponding reactants and products. The naming of the files contains six pieces of information. <br><br>1. Reaction ID (total 265, listed in Reaction.pdf)<br>2. Noise level added onto KinBot-generated initial transition state guess structures (0, 1, 2, 5, 10, 20, 50 pm)<br>3. Molecule type (TS: transition state, R: reactant, P: product, no label: initial guess structure)<br>4. Transition state optimization method (dft1: Quasi-Newton DFT optimization, nn1: Quasi-Newton NewtonNet optimization, nn0: Full-Hessian NewtonNet optimization)<br>5. (optional) IRC optimization method (dft1: Quasi-Newton DFT optimization, nn1: Quasi-Newton NewtonNet optimization)<br>6. (optional) Frequency calculation method (dft1: single point DFT vibrational analysis)<br><br>For example, file 210noise05_TS_nn0.json is the data of transition state for reaction 210 with 5 pm noise onto the initial guess structures optimized by full-Hessian NewtonNet method.<br><br>All data are stored in json format. They can be loaded through python by using:<br><br>import json<br>json.load(open(f'{rxn:03}noise{noise}<i>_TS_</i>{ts_method}.json', 'r'))<br><br>Details of each file are described in README file.<br>Scripts for using this data can be found at:https://github.com/THGLab/MLHessian-TSopt</pre>