Inconsistent client state in dispatch thread! (III)

[htana...@gmail.com]

Dear All,

I plan to integrate the neural network potential (NNP) compatible version of LAMMPS provided by Advance Soft Co. with i-PI.

The LAMMPS code is publicly available at the following link:

https://github.com/advancesoftcorp/lammps/tree/based-on-lammps_2Jun2022

Regarding the example 'h2o-piglet.4,' I confirmed that the calculation runs when  integrating the mentioned LAMMPS version with i-PI.

However, when applying NNP and performing calculation with a different structural model, the following error occurred:

!W!  @SOCKET:   Inconsistent client state in dispatch thread! (III)
!W!  @SOCKET:   Client died or got unresponsive(C). Removing from the list.

I have attached the input files for calculation without the NNP file (ffield.sannp). If you notice any points of concern, could you please provide guidance?

Thank you very much in advance.

Best regards,

Hideki

[htana...@gmail.com]

Dear All,

The cause of the error "Inconsistent client state in dispatch thread! (III)" is not known, but it has been observed that when specifying the lattice vectors in the input.xml file as <cell mode='manual'> [10.71568112, 0, 0, 0, 13.95756416, 0, -2.940193993, 0, 15.11886684] </cell>, the resulting *.chk file shows the following:

<cell shape='(3, 3)'>
   [1.07156811e+01, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 1.39575642e+01,
   0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 1.51188668e+01]
</cell>

"bx = -2.940193993" is replaced with "bx = 0".

Is it defined in i-pi that the c-vector is along the z-axis, and the b-vector lies in the yz plane? In Lammps, it is required that the a-vector is along the x-axis, and the b-vector lies in the xy plane. This discrepancy might be causing the issue.

Any comments or insights on this matter would be greatly appreciated.

Best regards,

Hideki

2023年12月27日水曜日 13:29:50 UTC+9 htana...@gmail.com:

[HT]

I noticed that the conversion from lattice constants in angstroms to Bohr was incorrect, but upon correction, the following errors no longer occur:

!W!  @SOCKET:   Inconsistent client state in dispatch thread! (III)
!W!  @SOCKET:   Client died or got unresponsive(C). Removing from the list.

However, after configuring as follows in the *.xml:
<initialize nbeads='6'>
  <file mode='xyz'>init.xyz</file>
  <velocities mode='thermal' units='kelvin'> 298.15 </velocities>
  <cell mode='manual'> [38.26639194, 0, 0, 0, 49.84336643, 0, -10.49962335, 0, 53.99045358] </cell>
</initialize>

In the *.chk file, the output was as follows:
<cell shape='(3, 3)'>
   [   3.82663919e+01,   0.00000000e+00,   0.00000000e+00,   0.00000000e+00,   4.98433664e+01,
       0.00000000e+00,   0.00000000e+00,   0.00000000e+00,   5.39904536e+01 ]
</cell>

The lattice vectors were not correctly recognized in ipi.

Also, in the coordinate output file, it was output as follows:
# CELL(abcABC):   38.26639    49.84337    53.99045    90.00000    90.00000    90.00000  Step:           0  Bead:       0 kinetic_od{atomic_unit}  cell{atomic_unit}

However, it should be 101.0 instead of 90 for beta. This is an error.

On the other hand, when entering lattice constants in the *.xyz coordinate file as follows:
# positions{angstrom} CELL{abcABC}: 20.24970256 26.375973630000004 29.10576245 90.0 101.0050538 90.0 cell{angstrom}

In the *.chk file, the lattice vectors were output as follows, and the lattice vectors were correctly converted:
<cell shape='(3, 3)'>
   [   3.82663914e+01,   3.05202592e-15,  -1.04996232e+01,   0.00000000e+00,   4.98433658e+01,
       4.01081270e-15,   0.00000000e+00,   0.00000000e+00,   5.39904529e+01 ]
</cell>

In conclusion, when lattice constants are entered in *.xyz coordinate file, the lattice vectors are correctly recognized. However, when entering the lower triangular lattice vectors in the *.xml file, the lattice vectors are not correctly recognized.

I would like to ask everyone for guidance on the following: When reading a coordinate file in Lammps, the atomic coordinates is necessary to correspond to the lower triangular lattice vectors. However, when loading a coordinate file *.xyz in ipi, should the atomic coordinates be specified with respect to the upper triangular or lower triangular lattice vectors?

Best regards,

Hideki

2024年1月2日火曜日 16:48:14 UTC+9 htana...@gmail.com:

[Michele Ceriotti]

i-PI stores the cell with the vectors in the columns (a convention that is often used in theoretical work). this was not very well documented. 

rather than giving you full details here, I am making a pull request with improved documentation 

https://github.com/i-pi/i-pi/actions/runs/7472035123/job/20333410697?pr=301

PLS have a look and let me know if the documentation in there would have been enough to clarify your doubts.

Cheers

M

[htana...@gmail.com]

Dear Michele,

Thank you for your reply.

As for "LAMMPS and i-PI have the same convention," in LAMMPS, it is assumed to be a lower triangular lattice vector (not upper triangular lattice vector), as stated in the following manual, though it might be my misunderstanding....
https://docs.lammps.org/Howto_triclinic.html

Then, I tried to view the link https://github.com/i-pi/i-pi/actions/runs/7472035123/job/20333410697?pr=301, but it appears as an attached file, and I could not see the document. I am not familiar with GitHub, so could you please show me how to view the document?

Thank you!

Best regards,

Hideki

2024年1月10日水曜日 17:16:33 UTC+9 michele....@gmail.com:

[Michele Ceriotti]

Sorry, wrong link https://github.com/i-pi/i-pi/pull/301 

This has since been merged. 

LAMMPS and i-PI use the same convention, but LAMMPS stores vectors in the rows, i-PI in the columns. 

Michele

[htana...@gmail.com]

Dear Michele,

I appreciate your guidance. I now have a good understanding of the format of lattice vectors.

Thank you!

Best regards,

Hideki

2024年1月15日月曜日 21:10:20 UTC+9 michele....@gmail.com: