Hands-on to APEX (Alloy Properties EXplorer using simulations)

Background

After obtaining a DP model for an alloy, it is important to not only calibrate the model's energies, forces, and virials using dp test, but also to test its accuracy in describing other properties. This requires a convenient software package (workflow) to compute and compare a series of properties computed with DFT and existing interatomic potential functions (such as EAM and MEAM). The agreement between DP and DFT in various properties reflects the accuracy of DP and guides training strategies and parameter settings.

The previous version of alloy property test package autotest is integrated inside the DP-GEN that has following drawbacks:

1. Inefficiency when testing multiple models simultaneously. For each model, the testing process involves make, run, and post, with each test waiting for the previous step to finish before the user's intervention to kick off the following step, which significantly reduces the efficiency of model testing.

2. Difficulty in locating specific tasks when errors occur during the running process, indicating weak error handling capabilities.

With the help of recently developed tools such as dflow and fpop, these issues can be effectively addressed. Thus, based on the previous codes, the new Python package: APEX (Alloy Properties Explorer using simulations) is introduced to provide an integrated alloy property test tool-kits based on cloud-native workflow with high efficiency.

Install APEX

Download APEX packedge via git clone

APEX can be easily installed via pip, as well as other dependencies such as pydflow.

check if apex is installed

LAMMPS example

Here we take a case study example for demonstration. This test is to evaluate preformance of a recently published pre-trained Deep Potential model: DPA-1 on description of interatomic interation of HCP Titanium metal through MD simulation using LAMMPS.

Figure 1. DPA model

Firstly, go to the prepared case directory:

tree to check current local working directory:

global_bohrium.json contains necessary config for Bohrium platform (e.g. account infomation; machine type; image address et al.), and should be prepared under current work directory in the first place.

Please replace with your own Borium username, password and program_id via following code:

A complete APEX framework to test alloy properties contains two individual workflows (relaxation and property) joint together.

Three types of test parameters json files are prepared here for demonstration:

1. param_relax.json contains structures path, interaction infomation and relaxation parameters:

2. param_props.json contains structures path, interaction infomation and property test parameters:

As can be seen, the eos and elastic have been turned on by indicating value of "skip" to be false.

3. param_joint.json contains all the information including parameters regarding relaxation and property test:

Submit relaxation workflow

This can be done via any of following commands, for example:

• apex param_relax.json
• apex param_joint.json --relax
• apex param_relax.json param_props.json --relax

For example: (it may takes a few minutes to submit to Bohrium depending on one's network connection)

Once Workflow has been submitted (ID: ... appears, you can monitor the workflow on website.

While waiting...

APEX Introduction

APEX inherits the functionality of the second version of alloy properties calculations and is developed based on the dflow. By incorporating the advantages of cloud-native workflow features, APEX simplifies the complex process to automatically test multiple configurations and properties within one workflow. It offers users an more intuitive and easy-to-use interaction, making the overall user experience more straightforward.

Overall framework of APEX:

Figure 1｜A illustrtion of APEX workflow

Three types of workflow achieved on the Argo UI:

• relaxation mode:

• property mode:

• joint mode:

• The architechture of relaxation and property-test workflows follows the similar three-steps style of previous codes:

Figure 2｜Three steps in Relaxation flow

When the workflow finished, the work directory will be downloaded automatically, containing finished relaxation tasks:

Submit property-test workflow

Then, we can continue submiting a property-test workflow. This could be realized via one of following commands (note that you should have the relaxation workflow finished in the first place before continue this step):

• apex param_props.json
• apex param_joint.json --props
• apex param_relax.json param_props.json --props

for example: (it may takes a few minutes to submit to Bohrium depending on one's network connection)

you can monitor this workflow on website once submitted.

Basics for property tests

• The equation of state (EOS):

Figure 3｜Three steps in EOS calculation

Figure 4｜Example of EOS plot (from Computer Physics Communications 253 (2020) 107206)

• The elastic modulus:

Figure 5｜Three steps in elastic constants calculation

• The surface energy:

  

• The point defect energy:

  

• The stacking fault energy (Gamma line):

  

When the workflow completed, the work directory, together with property results, will be downloaded automatically to current directory:

Submit a joint workflow (relaxation + property-test)

You can also submit an overall joint workflow to complete equivalent jobs of above two workflow.

Let's start over by reset the work direction:

Submit a joint workflow by:

• apex param_joint.json
• apex param_relax.json param_props.json

for example: (it may takes a few minutes to submit to Bohrium depending on one's network connection)

you can monitor this workflow on website once submitted.

When the workflow completed, the work directory, together with property results, will be downloaded automatically to current directory:

We can now extract the results of EOS and elastic tensor corresponding to each configuration: