A Deep Learning-Augmented Density Functional Framework for Reaction Modeling with Chemical Accuracy

this manual will tell you how to train and test with you datasets.

Install Requirements

Install package version

python==3.9.0
pyscf==2.2.1
torch==2.0.0
ruamel.yaml==0.17.21
numpy==1.24.2
scipy==1.10.1
paramiko==3.1.0

Then we will install package DeePKS-kit. DeePKS-kit is a pure python library so it can be installed following the standard git clone then pip install procedure. Note that the two main requirements pytorch and pyscf will not be installed automatically so you will need to install them manually in advance. Below is a more detailed instruction that includes installing the required libraries in the environment.

We use mamba here as an example. So first you may need to install Miniforge. and install the requirement package.

mamba create -n deepks python=3.9.0 paramiko=3.1.0 numpy scipy=1.10.1 h5py ruamel.yaml=0.17.21 paramiko=3.1.0
mamba activate deepks
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1
pip install pyscf==2.2.1

Then you can install deepks

git clone https://github.com/deepmodeling/deepks-kit
cd deepks/
python setup.py install

or you can install

pip install git+https://github.com/deepmodeling/deepks-kit/

Train our datasets

you can download our examples datasets from zenodo.14876882 The datasets is Below, QM document contain the each type DFT training parameterstrain_input.yaml and datasets path such as train.raw, and get_energy.py will give you output energy.

test_sets also have contain the "eigenvaleu of local density matrix" of test sets and the end reults

validate_sets only contain the results

projects
├── QM
│   ├── B3LYP
│   │   ├── GRAM
│   │   ├── GRAMandT1X
│   │   └── T1X
└── validate_sets
   └── WHG_BHRE
       └── result

Train

we use the DeePHF@B3LYP as a example.

cd QM/B3LYP/GRAMandT1X
deepks train train_input.yaml  -d train.raw -t valid.raw -o model.out/model.pth > model.out/log.iter 2> model.out/err.iter

Test

## you can check the test.raw error
deepks test -m model.out/model.pth -d test.raw -o test_corr/test >L1L2.out

## or you can use `get_energy.py` get the output `energy`, for example
python get_energy.py --model models/wb97M-V/GRAMandT1X/model.out/model.pth --raw RGD1_opt_wb97mv.raw

generate the dm_eig.npy from xyz file.

deepks scf scf_input.yaml -m None -s xyzfile/test.xyz -X M06-2X -d result/

Results

Model performance was evaluated on independent test sets RGD1 database(83 reactions)

Model performance was evaluated on independent test sets BH9 database(97 reactions)