Questions on Using Te-PIGS with MBX, i-PI, and LAMMPS

[Siyuan Nie]

Dear developers and users,

Thank you so much for sharing your excellent work! They really help a lot!

I’m currently trying to perform a Te-PIGS simulation (for liquid water) following your recent publications:

https://doi.org/10.1021/jacs.4c14610 and https://doi.org/10.1021/acsnano.4c16486.

I’ve reviewed the shared data in your repository (https://github.com/paesanilab/Data_Repository/tree/main/ice_SFG) and would like to confirm whether the .pb file provided there corresponds to the QC_MB-pol(2023) quantum correction potential used in your JACS paper?

If so, I’m trying to set up i-PI as the server, with MBX and LAMMPS as clients: MBX handles the MB-pol force field, and LAMMPS calls this .pb file.

I’d really appreciate your guidance on a few points:

1. MBX and MB-pol(2023) with i-PI

(1) To use the MB-pol(2023)_2c3b4a version, I installed the MBX v1.1.3 branch from https://github.com/PhilipZhu/MBX/tree/mbpol23. I followed the README and used the following setup in LAMMPS:

fix 1 all mbx 1 h2o_2c3b4a 1 2 3 1 2 2 json mbx.json

And in the .nrg file, I changed monomer h2o to h2o_2c3b4a. Is this correct to activate the MB-pol(2023)_2c3b4a version?

(2) How is the periodic boundary condition (PBC) handled in MBX when called by i-PI?

Is it automatically enabled if box_size and alpha_ewald_{elec/disp} (e.g. 0.6) are set in .json? Also, does the pbc='True' flag in the <ffsocket> section of the .xml file have any effect on PBC in MBX?

(3) If using <ffsocket mode='unix'>, can the port field in mbx.json be ignored?

(4) What unit system does MBX use (are they consistent with LAMMPS's real unit style)? Is manual unit conversion required when using i-PI?

2. Using .pb file in LAMMPS

(1) To call the .pb model in LAMMPS, is this syntax correct?

pair_style deepmd graph_qc-compress_mbpol23_type1.pb 

pair_coeff * * O H

(2) Is it correct to use units real in the LAMMPS input file when using this .pb file (same as MB-pol)?

3. Te-PIGS setup in i-PI

I’ve defined two force fields in the .xml following:

<ffsocket mode='unix' name='mbpol' pbc='True'> 

 <address> localhost_MBX </address> 

</ffsocket> 

<ffsocket mode='unix' name='lammps_qc_mbpol' pbc='True'>

 <address> localhost_LAMMPS </address> 

</ffsocket>

 <forces>

 <force forcefield='mbpol' weight='1'> </force>

 <force forcefield=' lammps_qc_mbpol' weight='1'> </force> 

</forces>

Does this correctly represent the sum of the Born–Oppenheimer PES and the quantum correction PES as required for Te-PIGS?

Thank you in advance for your assistance!

Best regards,

Siyuan

[Richa Rashmi]

Dear Siyuan,

Thank you for your interest in running Te-PIGS simulations with MB-pol(2023). If you are planning to simulate only water, you can do so directly using MBX + i-PI + DeePMD i.e. there is no need to involve LAMMPS.

i-PI communicates independently with both MBX and DeePMD through two separate JSON files. This setup is more straightforward and typically easier to debug. However, if you are planning hybrid classical force field simulations with MBX, then including LAMMPS in the workflow is recommended. I have attached the input files that we used for one of the Te-PIGS simulations from the ice SFG paper. Please take a look, and feel free to reach out with any questions.

A few additional clarifications that may not be evident from the input files:
1. The GitHub repository contains the correct .pb graph file for Te-PIGS simulations with MB-pol(2023).
2. Periodic boundary conditions are handled via mbx.json. The pbc tag in the i-PI input.xml file is only used for wrapping positions and does not control whether the simulation is periodic. For more details, see: https://docs.ipi-code.org/units.html.
3. The port number is not important, it is important to use consistent socket address in mbx.json and input.xml.
4. You do not need any unit conversions as long as you are specifying the units correctly in i-PI and LAMMPS (in case you are using LAMMPS).

Please feel free to reach out if you have any further questions.

Best regards,
The MBX Team

[Siyuan Nie]

Hi Richa,

Thank you so much for your explanation and for sharing all the input files, this is incredibly helpful! I’ve gone through them and things are much clearer now.  

Really appreciate your support!

Best regards,
Siyuan