Use this short script to drive CP2K
#!/bin/bash # for ii in 2 4 6 8 10 12 do sed -e "s/MY_SUPERCELL/${ii}/g" template.inp > input_${ii}.inp cp2k.popt input_${ii}.inp > NaCl_supercell_${ii}.out done
where the template input is this geometry optimization using the classical forcefield FIST module of CP2K.
@SET NREP MY_SUPERCELL @SET OPTIMIZER LBFGS # BFGS &FORCE_EVAL METHOD Fist &MM &FORCEFIELD &CHARGE ATOM Na CHARGE +1.000 &END CHARGE &CHARGE ATOM Cl CHARGE -1.000 &END CHARGE &NONBONDED &BMHFT map_atoms NA NA atoms NA NA RCUT 10.0 &END BMHFT &BMHFT map_atoms NA CL atoms NA CL RCUT 10.0 &END BMHFT &BMHFT map_atoms CL CL atoms CL CL RCUT 10.0 &END BMHFT &END NONBONDED &END FORCEFIELD &POISSON &EWALD EWALD_TYPE spme ALPHA .35 GMAX 12*${NREP} O_SPLINE 6 &END EWALD &END POISSON &END MM &SUBSYS &CELL #ABC 5.620 5.620 5.620 ABC 2*5.620 2*5.620 2*5.620 MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP} &END CELL &TOPOLOGY COORD_FILE_NAME NaCl.pdb COORDINATE PDB CONN_FILE_FORMAT OFF MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP} &END TOPOLOGY &END SUBSYS &END FORCE_EVAL &GLOBAL PROJECT NaCl RUN_TYPE GEO_OPT &END GLOBAL &MOTION &GEO_OPT OPTIMIZER ${OPTIMIZER} &END &END MOTION
The script runs very quickly when the LBFGS optimizer is used. But see what happens if we switch to the BFGS optimizer instead (change the OPTIMIZER variable in the cp2k input file - you might want to reduce the size of the supercells in the driver script - NREP varying from 1 to 6 perhaps). Look at the timings that cp2k prints at the end of a run and see if you can see the culprit. Also look for warnings in your outputs (this is a good habit to get into).
How does the conjugate gradients optimizer compare to LBFGS in efficiency for this system?
For studying many properties of solid materials it is important that the lattice parameters used in a simulation are close to equilibrium for the model chemistry (Hamiltonian) used. Otherwise, large stresses can be present that complicate comparison to experiment. Successful cell optimization requires that the energy changes smoothly with cell volume - and for this the energy cutoff is the most important parameter. The input file template below can be used with driver script to examine how the energy volume curve of NaCl changes with the PW cutoff.
@SET SCALE_FACTOR MY_SCALING @SET NREP 1 @SET OPTIMIZER BFGS # LBFGS @SET CUTOFF MY_CUTOFF @SET SAFTEY_CUTOFF 1.1 &FORCE_EVAL METHOD QS &DFT BASIS_SET_FILE_NAME BASIS_MOLOPT POTENTIAL_FILE_NAME GTH_POTENTIALS &MGRID CUTOFF ${CUTOFF} REL_CUTOFF 60 &END MGRID &QS EPS_DEFAULT 1.0E-12 &END QS &SCF SCF_GUESS RESTART &OT ON MINIMIZER DIIS &END OT &END SCF &XC &XC_FUNCTIONAL Pade &END XC_FUNCTIONAL &END XC &END DFT &SUBSYS &CELL ABC 5.620*${SCALE_FACTOR} 5.620*${SCALE_FACTOR} 5.620*${SCALE_FACTOR} MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP} #&CELL_REF # ABC 5.620*${SAFETY_FACTOR} 5.620*${SAFETY_FACTOR} 5.620*${SAFETY_FACTOR} # MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP} #&END &END CELL &COORD scaled Na 0.000 0.000 0.000 Cl 0.500 0.500 0.500 Na 0.000 0.500 0.500 Cl 0.500 0.000 0.000 Na 0.500 0.000 0.500 Cl 0.000 0.500 0.000 Na 0.500 0.500 0.000 Cl 0.000 0.000 0.500 &END &TOPOLOGY MULTIPLE_UNIT_CELL ${NREP} ${NREP} ${NREP} &END TOPOLOGY &KIND Na BASIS_SET DZVP-MOLOPT-SR-GTH POTENTIAL GTH-PADE-q9 &END KIND &KIND Cl BASIS_SET DZVP-MOLOPT-SR-GTH POTENTIAL GTH-PADE-q7 &END KIND &END SUBSYS &END FORCE_EVAL &GLOBAL PROJECT NaCl RUN_TYPE ENERGY &END GLOBAL &MOTION &GEO_OPT OPTIMIZER ${OPTIMIZER} &END &CELL_OPT KEEP_SYMMETRY &END &END MOTION
The driver script could be something like
#!/bin/bash # CUTOFF="280" for ii in 0.800 0.825 0.850 0.875 0.900 0.925 0.950 0.975 1.000 1.025 1.050 1.075 1.100 do #this assumes that the input template is in NaCl_QS.inp sed -e "s/MY_SCALING/${ii}/g" NaCl_QS.inp > temp.inp sed -e "s/MY_CUTOFF/${CUTOFF}/g" temp.inp > input_${ii}.inp #this line should changed to point to your cp2k executable mpirun -np 2 cp2k.popt input_${ii}.inp > NaCl_${CUTOFF}_${ii}.out done
You can extract energies from the outputs with a command like
grep 'ENERGY|' *out | awk '{print $10}' > NaCl_energy_volume.dat
and you can plot the results in your favourite graphing software.
What is happening here? Try changing the PW cutoff (defined in the driver script) and using the CELL_REF variable.
Copy the input template to a new file and change the RUN_TYPE to CELL_OPT. You'll also need to ask the code to calculate the stress tensor (in FORCE_EVAL section) ANALYTICALLY! Also define the CUTOFF and SCALING_FACTOR. Start the cell optimization from small cells (SCALING_FACTOR 0.85) or large cells (SCALING_FACTOR 1.10) - do you get the same results?
If you have access a machine with several cores (16 or so ideally) check whether increasing the supercell size (NREP variable) affects the results.