jonpvandermause
commented
2 days ago Thanks for reporting. Could you please post the input script? I will try to reproduce.
Open rsdmse opened 2 days ago
jonpvandermause
commented
2 days ago Thanks for reporting. Could you please post the input script? I will try to reproduce.
mohan-s1
commented
1 day ago I apologize if this is a stupid way to upload our scripts; GitHub appears not to support .yaml, .slurm nor .py file types. All of the following files are placed in the same directory and submitted using the slurm script attached below.
fresh_online_10_03_24_agocl.yamlsupercell:
file: POSCAR_ag_2o_2cl
format: vasp
index: 0
replicate: [1, 1, 1]
jitter: 0.0
flare_calc:
gp: SGP_Wrapper
kernels:
- name: NormalizedDotProduct # select kernel for comparison of atomic environments
sigma: 2.0 # signal variance, this hyperparameter will be trained, and is typically between 1 and 10.
power: 2 # power of the kernel, influences body-order
descriptors:
- name: B2 # Atomic Cluster Expansion (ACE) descriptor from R. Drautz (2019). FLARE can only go from B1 up to B3 currently.
nmax: 12 # Radial fidelity of the descriptor (higher value = higher cost)
lmax: 4 # Angular fidelity of the descriptor (higher value = higher cost)
cutoff_function: quadratic # Cutoff behavior
radial_basis: chebyshev # Formalism for the radial basis functions
cutoff_matrix: [[7.0, 4.0, 4.0],[4.0, 4.0, 4.0],[4.0, 4.0, 4.0]] # In angstroms. NxN array for N_species in a system.
energy_noise: 0.096 # Energy noise hyperparameter, will be trained later. Typically set to 1 meV * N_atoms.
forces_noise: 0.1 # Force noise hyperparameter, will be trained later. System dependent, typically between 0.05 meV/A and 0.2 meV/A.
stress_noise: 0.001 # Stress noise hyperparameter, will be trained later. Typically set to 0.001 meV/A^3.
energy_training: True
force_training: True
stress_training: True
species:
- 47
- 8
- 17 # Atomic number of your species (here, 13 = Al).
single_atom_energies:
- 0
- 0
- 0 # Single atom Es to bias the E prediction of the model. Can help in systems with poor initial E estimations. Length must equal the number of species.
cutoff: 7.0 # Cutoff for the (ACE) descriptor. Typically informed by the RDF of the system. Should equal the maximum value in the cutoff_matrix.
variance_type: local # Calculate atomic uncertainties.
max_iterations: 20 # Maximum steps taken during each hyperparameter optimization call.
use_mapping: True # Print mapped model (ready for use in LAMMPS) during trajectory. Model is re-mapped and replaced if new DFT calls are made throughout the trajectory.
dft_calc:
name: Vasp
kwargs:
command: "srun vasp_gam"
xc: PBE
kpts: [1, 1, 1]
istart: 0
ediff: 1.0e-5
encut: 400
ismear: 0
sigma: 0.2
ispin: 2
lreal: Auto
prec: Accurate
algo: Very_Fast
ncore: 10
nelm: 500
nelmdl: -9
nelmin: 6
lcharg: False
lwave: False
lscalapack: False
params: {}
otf:
mode: fresh
md_engine: PyLAMMPS
md_kwargs:
command: "srun lmp"
specorder: [Ag, O, Cl]
dump_period: 5
pair_style: flare
fix:
- "1 all nvt temp 523 523 0.1"
keep_alive: False
initial_velocity: 500 # Initialize the velocities (units of Kelvin)
dt: 0.001 # Set the time step in picoseconds ( 0.001 = 1 fs here)
number_of_steps: 5000 # Total number of MD steps to be taken
output_name: 0_ase
init_atoms: [0] # Initial atoms to be added to the sparse set
std_tolerance_factor: -0.05 # The uncertainty threshold above which the DFT will be called
max_atoms_added: -1 # Allow for all atoms in a given frame to be added to the sparse set if uncertainties permit
train_hyps: [0,0] # Define range in which hyperparameters will be optimized. Here, hyps are optimized at every DFT call after the 5th call.
write_model: 3
store_dft_output: [[OUTCAR,OSZICAR], ./]
update_style: threshold # Sparse set update style. Atoms above a defined "threshold" will be added using this method
update_threshold: 0.01 #Threshold (Thr) if "update_style = threshold". Thr represents relative uncer-ty to mean atomic uncer-ty, where atoms above are added to sparse set
force_only: False
min_steps_with_model: 1POSCAR_ag_2o_2clAg O Cl
1.0000000000000000
25.1000000000000014 0.0000000000000000 0.0000000000000000
0.0000000000000000 25.1000000000000014 0.0000000000000000
0.0000000000000000 0.0000000000000000 25.1000000000000014
13 2 2
Cartesian
12.5500000000000007 12.5500000000000007 12.5500000000000007
14.5999999999999428 12.5500000000000007 10.5000000000000586
14.5999999999999428 10.5000000000000586 12.5500000000000007
14.5999999999999428 14.5999999999999428 12.5500000000000007
14.5999999999999428 12.5500000000000007 14.5999999999999428
16.6500000000000625 12.5500000000000007 12.5500000000000007
10.5000000000000586 12.5500000000000007 10.5000000000000586
12.5500000000000007 14.5999999999999428 10.5000000000000586
12.5500000000000007 10.5000000000000586 10.5000000000000586
12.5500000000000007 12.5500000000000007 8.4499999999999353
10.5000000000000586 10.5000000000000586 12.5500000000000007
12.5500000000000007 10.5000000000000586 14.5999999999999428
12.5500000000000007 8.4499999999999353 12.5500000000000007
16.8667008375160989 12.3696592239405732 14.7305931547166988
10.9347671194563478 14.1739099868444587 12.2650128878102169
14.9891129587497858 10.1595157419203890 10.2760135607241931
13.8123448220109104 11.1801335934864152 16.4371512753725746save_final_file_w_velocities_10_03_24.pyimport numpy as np
from numpy.random import random
import matplotlib.pyplot as plt
import matplotlib
from ase.build import molecule
from ase.io import write, read
from ase.io.trajectory import TrajectoryWriter
from ase.io.lammpsdata import write_lammps_data
a = read('0_ase_md.xyz','-1')
a.write('final_otf_structure.traj')
write_lammps_data('./initial.data',a,specorder=['Ag','O','Cl'],velocities=1)
print('save file done')fresh_gamma.slurm to submit the job to our cluster#!/bin/bash
#SBATCH --account paolucci
#SBATCH --nodes=4
#SBATCH --ntasks-per-node=96
#SBATCH --time=3-00:00:00
#SBATCH --partition=parallel
#SBATCH -x udc-ba04-30c1
##SBATCH --mail-type=END
#SBATCH --no-requeue
export MODULEPATH=/project/paolucci/apps/modulefiles:$MODULEPATH
module purge
module load vasp/6.3.0
module load gcc openmpi/4.1.4-nofabric lammps_flare/29Aug2024u1_1.4.2
unset VASP_SCRIPT
export VASP_COMMAND=/home/mjs7eek/setup_vasp6_gamma.sh
which python ase lmp flare-otf
set -x
flare-otf fresh_online_10_03_24_agocl.yaml && \
python save_final_file_w_velocities_10_03_24.py && \
srun lmp -in go_nvt.in && \
echo "finish!"
go_nvt.inunits metal
atom_style atomic
newton on
boundary p p p
read_data ./initial.data
#read_restart new.rest
mass 1 107.8682
mass 2 15.999
mass 3 35.453
pair_style flare
pair_coeff * * ./lmp.flare
########################### Geo Opt ######################################
compute unc all flare/std/atom L_inv_lmp.flare sparse_desc_lmp.flare
compute MaxUnc all reduce max c_unc
#reset_timestep 0
thermo 500
thermo_style custom step time temp ke pe etotal press c_MaxUnc
############################# MD #########################################
timestep 0.001 #ps
fix nvt all nvt temp 523 523 0.1
dump all all custom 1000 dump.nvt id type x y z fx fy fz vx vy vz c_unc
dump pos all custom 50 dump.pos id type x y z
dump xyz all xyz 500 dump-nvt.xyz
dump_modify xyz element Ag O Cl
run 100000000 upto every 100000 "if '$(c_MaxUnc) > 0.1' then quit"
write_data 0.data
write_restart 0.rest
########################################################################
#fix nvt all nvt temp 523 523 0.05
#run 20000000
#write_data 1.data
#write_restart 1.rest
########################################################################
########################################################################
#fix nvt all nvt temp 523 523 0.05
#run 80000000
#write_data 2.data
#write_restart 2.rest
########################################################################
Thank you for sending your files! GitHub makes it difficult to attach files, so I think copy/pasting is the only way to do it.
I believe I've root caused the issue. The run_dft method of the OTF class attempts to compute energies, forces, and stresses as follows:
# Calculate DFT energy, forces, and stress.
# Note that ASE and QE stresses differ by a minus sign.
if "forces" in self.dft_calc.implemented_properties:
if "forces" in self.atoms.calc.results:
forces = self.atoms.get_forces()
else:
forces = None
else:
forces = None
if "stress" in self.dft_calc.implemented_properties:
if "stress" in self.atoms.calc.results:
stress = self.atoms.get_stress()
else:
stress = None
else:
stress = None
if "energy" in self.dft_calc.implemented_properties:
if "energy" in self.atoms.calc.results:
energy = self.atoms.get_potential_energy()
else:
energy = None
else:
energy = NoneBut they're not actually get computed here, because self.atoms.calc.results is an empty dict. This is why the OUTCAR in your example isn't getting generated.
I have a fix for this in #423. If the build passes, I will merge into master, and you should be good to go.
jonpvandermause
commented
12 hours ago OK, the fix is now merged into master. Can you please give it a try and let me know how it goes?
A user is testing 1.4.2 (commit cb57c6b on Sep 30, 2024) using the same compilers, vasp, input files, and Slurm script as 1.3.3 (official release). The calculation runs fine in 1.3.3 but not for the newer version. Specifically, the first DFT call completes immediately without generating any output files:
which causes the subsequent command to fail:
Since we have not modified the VASP installation we don't think the DFT run failure is due to VASP. Do you know what could be causing this behavior (e.g. changes to the input file syntax)?