3.7. Running DeepMD-Kit with Pharmaforge

The goal of this tutorial is to demonstrate the use of the pharmaforge package in conjunction with dftb+ and deepmd-kit to take a query from the database (see the previous example on database querying.

3.7.1. Learning Objectives

  • Learn hopw to use the pharmaforge package to run a deep potential on queried configurations from the database.

  • Learn how to use the dftb+ package to run a DFTB calculation on queried configurations from the database using pharmaforge.

  • Calculate the correction energy using the QDPi1 model.

3.7.2. Required Files

  • The tutorial python script is located in examples/CalculateMLP

  • The QDPi1 deepmd-kit model is located at https://gitlab.com/RutgersLBSR/qdpi, and should be downloaded into inputs.

  • DFTB+ should be installed (https://dftbplus.github.io/).

  • DFTB+ slater-Koster files must be downloaded somewhere. The path to these files should be pointed to within the run_qdpi1.py script.

3.7.3. Tutorial

First thing, make sure all the required files are in the correct location (See above) and that DFTB+ is installed.

Then, set the skf file locations as follows (or putting them into this location):

from pharmaforge.interfaces import DeepMDInterface
from pharmaforge.interfaces import DFTBInterface

skf_location="./inputs/3ob-3-1"

Now, we will first calculate the deep potential energy using the QDPi1 model, which if you recall is a deepmd-kit \(\Delta MLP\) model. The script to do this is as follows:

print("Test DeepMdKit")
test = DeepMDInterface(model="inputs/qdpi.pb")
test.ObtainDeepData("inputs/saved_model_qdpi1.hdf5")
test_energies, test_forces = test.calculate(limit=2, verbose=True)
print(test_energies)

As with the previous examples, you will see the energies and forces printed out.

Next, we will calculate the DFTB energy using the dftb+ package. The script to do this is as follows:

print("Test DFTB3")
test2 = DFTBInterface(skf_loc=skf_location, default=True)
test2.ObtainDeepData("inputs/saved_model_qdpi1.hdf5")
energies, forces = test2.calculate(limit=2, verbose=True)
print(energies)

Now, for a \(\Delta MLP\) the higher level potential energy is described by the following

(1)\[E_{higher level} = E_{MLP} + E_{DFTB}\]

So, we can calcualte the correction energy as follows:



for key in energies:
    for i, ener in enumerate(energies[key]):
        diff_e = test_energies[key][i] + ener
        diff_f = test_forces[key][i] + forces[key][i]
        print(f"Higher Level Energy for system {key} structure {i}: {diff_e}")

Note that these numbers are very similar to the numbers you found during your data_labeling tutorial!

3.7.4. Full Code

from pharmaforge.interfaces import DeepMDInterface
from pharmaforge.interfaces import DFTBInterface

skf_location="./inputs/3ob-3-1"

print("Test DeepMdKit")
test = DeepMDInterface(model="inputs/qdpi.pb")
test.ObtainDeepData("inputs/saved_model_qdpi1.hdf5")
test_energies, test_forces = test.calculate(limit=2, verbose=True)
print(test_energies)


print("Test DFTB3")
test2 = DFTBInterface(skf_loc=skf_location, default=True)
test2.ObtainDeepData("inputs/saved_model_qdpi1.hdf5")
energies, forces = test2.calculate(limit=2, verbose=True)
print(energies)


for key in energies:
    for i, ener in enumerate(energies[key]):
        diff_e = test_energies[key][i] + ener
        diff_f = test_forces[key][i] + forces[key][i]
        print(f"Higher Level Energy for system {key} structure {i}: {diff_e}")