TiO2-Water DP Trainig Data

This repo contains the data used to train a DPMD potential for the TiO2-water interface, as described in

Calegari Andrade, M. F., Ko, H.-Y., Zhang, L., Car, R. & Selloni, A. Free energy of proton transfer at the water–TiO2 interface from ab initio deep potential molecular dynamics. Chem. Sci. 11, 2335–2341 (2020).

Before you start, please make sure you have DeepMD-Kit and Lammps installed. You can find good tutorials on how to compile DeepMD-kit and Lammps at the DeepMD-Kit Documentation Page

Downloading the dataset

Please make sure you have git-lfs running on your local machine. The raw data used to train the DNN can only be downloaded with git-lfs.

git clone https://github.com/CSIprinceton/TiO2-Water.git
cd TiO2-Water
git lfs fetch --all
git lfs pull

Training your own DP models

I have provided trained DNN graphs that are ready to use. These graphs are located at "train/?/". In case you want to train your own DNN model, please do the following:

cd raw_data
./raw_to_set.sh
cd ../train/1
dp train tio2-water.json

Running a simulation with Lammps

There is a simple lammps input example at lammps/lammps.in and a initial condition at lammps/pos.in. This example runs a NVT simulation of the TiO2-water interface at 330 K. 3 different DNN models are used to evaluate the interatomic interactions, and the deviation between the energies and forces predicted by these potentials will be outputed to the file model_devi.out.

Running Lammps is simple

lmp < lammps.in > lammps.out

Note: there is no guarantee that the DNN potential will predict reasoable energy and forces if you start your dynamics from a unphysical configuration (or too far from the configurations sampled in the training data).

Detailed description of the dataset

This data contains atomic coordinates, energy and forces of TiO2 anatase, liquid water and the TiO2-water interface used to train a Deep Neural Network Potential. Data set was collected via a iterative training scheme described in [2]. In this repo you will find 3 main folders:

1. lammps: example of an input script (lammps.in) used to run Deep Potential Molecular Dynamics (DPMD) using the Lammps package. The file "pos.in" contain a initial configuration of the TiO2-water interface;

2. raw_data: in this root folder you will find 5 directories containing the raw data of different systems. These systems are:

   • "tio2/": bulk anatase TiO2 (162 atoms);
   • "tio2_1water_vac/": anatase (101) surface with one adsorbed water molecule in vacuum (219 atoms);
   • "tio2_2water_vac/": anatase (101) surface with two adsorbed water molecules in vaccum (222 atoms);
   • "tio2_water/": TiO2-water interface (426 atoms);
   • "water/": bulk liquid water (192 atoms)

3. train: DeepMD-kit [3] input files used to train a Deep Neural Network potential (smooth edition) [4]. There are directories inside this folder, named from 1 to 4. These folders differ only in the random initialization of the initial NN parameters. We used the prediction deviation from models 1-4 as a coarse error estimatior of forces (or energy). NOTE: This input script was used with DeepMD-Kit version 1.2. Different versions of the code have different input flags. Please check DeepMD-kit documentation for more details.

4. PW: example of PWscf [5] input file used to compute energy and forces of TiO2-water. Please note: if you want to expand this training data you HAVE to use the same pseudo-potential, wave-function cutoff and DFT functional (SCAN) described in the PW input file. It is also higly recommended to use PWscf as your force and energy evaluator.

Note on the format of raw files

Please check the DeepMD-kit documentation for more details about the raw files format. Briefly, each raw file contains the full information of one snapshot in one line. So, if a "prefix.raw" contains 100 lines it means that this file has 100 different snapshots. Below we give a brief description of each raw file in our data:

1. coord.raw: atomic coordinates in angstrom units. Format: C(1,x) C(1,y) C(1,z) ... C(N,x) C(N,y) C(N,z). C(i,j) is the cartesion component j of atom with index i.
2. force.raw: atomic forces in eV/angstrom units. Format: F(1,x) F(1,y) F(1,z) ... F(N,x) F(N,y) F(N,z). F(i,j) is the cartesion component j of atom with index i.
3. energy.raw: potential energy in eV units.
4. box.raw: unit cell tensor in angstrom units.
5. type.raw: index assigned for each atomic species. Format: I(1) I(2) ... I(N). I(i) is the label of atomic species of atom with index i. In this data set we use the following convention: Ti=0, H=1 and O=2.

References

[1] Calegari Andrade, M. F., Ko, H.-Y., Zhang, L., Car, R. & Selloni, A. Free energy of proton transfer at the water–TiO2 interface from ab initio deep potential molecular dynamics. Chem. Sci. 11, 2335–2341 (2020).

[2] Zhang, L., Lin, D.-Y., Wang, H., Car, R. & E, W. Active learning of uniformly accurate interatomic potentials for materials simulation. Phys. Rev. Mater. 3, 023804 (2019).

[3] Wang, H., Zhang, L., Han, J. & E, W. DeePMD-kit: A deep learning package for many-body potential energy representation and molecular dynamics. Comput. Phys. Commun. 228, 178–184 (2018).

[4] Zhang, L. et al. End-to-end Symmetry Preserving Inter-atomic Potential Energy Model for Finite and Extended Systems. in Advances in Neural Information Processing Systems 31 (eds. Bengio, S. et al.) 4436–4446 (Curran Associates, Inc., 2018).

[5] Giannozzi, P. et al. Quantum ESPRESSO: a Modular and Open-Source Software Project for Quantum Simulations of Materials. J. Phys. Condens. matter 21, 395502 (2009).