2. Iterations¶
Here we continue from the previous tutorial performing the initial setup and test run for a static volume for the FCC Si example in the VASP tutorials. Here we will perform the calculations of the total energies at different volumes. Please make sure you have completed the previous tutorial before continuing.
Let us first fetch the AiiDA-VASP run file for the multiple volumes:
$ wget https://github.com/aiida-vasp/aiida-vasp/raw/master/tutorials/run_fcc_si_multiple_volumes.py
Inspect the file, which has the following changes compared to the file used in the previous tutorial:
--- /home/docs/checkouts/readthedocs.org/user_builds/aiida-vasp/checkouts/latest/tutorials/run_fcc_si_one_volume.py +++ /home/docs/checkouts/readthedocs.org/user_builds/aiida-vasp/checkouts/latest/tutorials/run_fcc_si_multiple_volumes.py @@ -1,8 +1,8 @@ """ -Call script to calculate the total energies for one volume of standard silicon. +Call script to calculate the total energies for different volumes of the silicon structure. -This particular call script set up a standard calculation that execute a calculation for -the fcc silicon structure. +This particular call script set up a standard calculation for each structure and submits +each of them. """ # pylint: disable=too-many-arguments import numpy as np @@ -16,12 +16,12 @@ load_profile() -def get_structure(): +def get_structure(alat): """ Set up Si primitive cell fcc Si: - 3.9 + alat 0.5000000000000000 0.5000000000000000 0.0000000000000000 0.0000000000000000 0.5000000000000000 0.5000000000000000 0.5000000000000000 0.0000000000000000 0.5000000000000000 @@ -33,7 +33,6 @@ """ structure_data = DataFactory('structure') - alat = 3.9 lattice = np.array([[.5, .5, 0], [0, .5, .5], [.5, 0, .5]]) * alat structure = structure_data(cell=lattice) for pos_direct in [[0.0, 0.0, 0.0]]: @@ -110,13 +109,15 @@ OPTIONS = AttributeDict() OPTIONS.account = '' OPTIONS.qos = '' - OPTIONS.resources = {'num_machines': 1, 'num_mpiprocs_per_machine': 1} + OPTIONS.resources = {'num_machines': 1, 'num_mpiprocs_per_machine': 8} OPTIONS.queue_name = '' OPTIONS.max_wallclock_seconds = 3600 OPTIONS.max_memory_kb = 2000000 - # POSCAR equivalent - # Set the silicon structure - STRUCTURE = get_structure() + # Iterate over each lattice constant and pass it explicitly + for lattice_constant in [3.5, 3.6, 3.7, 3.8, 3.9, 4.0, 4.1, 4.2, 4.3]: + # POSCAR equivalent + # Set the silicon structure + STRUCTURE = get_structure(lattice_constant) - main(CODE_STRING, INCAR, KMESH, STRUCTURE, POTENTIAL_FAMILY, POTENTIAL_MAPPING, OPTIONS) + main(CODE_STRING, INCAR, KMESH, STRUCTURE, POTENTIAL_FAMILY, POTENTIAL_MAPPING, OPTIONS)
As you can see, we did only minimal changes, all which should be self explanatory. In essence we perform the same workchain as before, only that we now iterate over multiple workchains, each getting supplied a different volume of the unit cell.
Change the
optionsandcode_stringas you did in the previous tutorial to suit your setup.Save and execute the resulting run script by issuing:
$ python run_fcc_si_multiple_volumes.py
Let us check the progress:
$ verdi process list -a PK Created Process label Process State Process status ---- --------- --------------- ---------------- ----------------------------------- 1043 37s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1067 1054 37s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1102 1065 36s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1132 1067 36s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1078 36s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1139 1089 36s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1141 1100 35s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1143 1102 35s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1113 35s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1145 1124 35s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1147 1137 34s ago VaspWorkChain ⏵ Waiting Waiting for child processes: 1149 1132 34s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1139 33s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1141 32s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1143 31s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1145 30s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1147 29s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload 1149 27s ago VaspCalculation ⏵ Waiting Waiting for transport task: upload Total results: 18 Report: last time an entry changed state: 26s ago (at 09:24:10 on 2022-12-20) Report: Using 9% of the available daemon worker slots.
All calculations are submitted to the cluster in parallel, which is pretty convenient and certainly an improvement compared to the regular tutorial, where you do the calculations sequentially. Not a problem for a few calculations, but maybe not ideal if you want to execute a few thousand calculations.
After a while, we again execute
verdi process list -a, hoping our runs are finished and get:$ verdi process list -a PK Created Process label Process State Process status ---- --------- --------------- ---------------- ----------------------------------- 1043 4m ago VaspWorkChain ⏹ Finished [0] 1054 4m ago VaspWorkChain ⏹ Finished [0] 1065 4m ago VaspWorkChain ⏹ Finished [0] 1067 4m ago VaspCalculation ⏹ Finished [0] 1078 4m ago VaspWorkChain ⏹ Finished [0] 1089 4m ago VaspWorkChain ⏹ Finished [0] 1100 4m ago VaspWorkChain ⏹ Finished [0] 1102 4m ago VaspCalculation ⏹ Finished [0] 1113 4m ago VaspWorkChain ⏹ Finished [0] 1124 4m ago VaspWorkChain ⏹ Finished [0] 1137 4m ago VaspWorkChain ⏹ Finished [0] 1132 4m ago VaspCalculation ⏹ Finished [0] 1139 4m ago VaspCalculation ⏹ Finished [0] 1141 3m ago VaspCalculation ⏹ Finished [0] 1143 3m ago VaspCalculation ⏹ Finished [0] 1145 3m ago VaspCalculation ⏹ Finished [0] 1147 3m ago VaspCalculation ⏹ Finished [0] 1149 3m ago VaspCalculation ⏹ Finished [0] Total results: 18 Report: last time an entry changed state: 30s ago (at 09:27:33 on 2022-12-20) Report: Using 0% of the available daemon worker slots.
All processes are in a finished state and we can extract the total energies for each step. However, it should be obvious that extracting the total energies from
miscfrom each step manually seems a bit inefficient. One could query (AiiDA has shortcuts for this), but one would still need to look up the values in some way. Surely the computer should be able to eject what we need, right? Maybe it is easier to be able to access them directly when all the calculations are complete? Let us do another modification to the run script above, namely:--- /home/docs/checkouts/readthedocs.org/user_builds/aiida-vasp/checkouts/latest/tutorials/run_fcc_si_multiple_volumes.py +++ /home/docs/checkouts/readthedocs.org/user_builds/aiida-vasp/checkouts/latest/tutorials/run_fcc_si_multiple_volumes_eos.py @@ -2,14 +2,15 @@ Call script to calculate the total energies for different volumes of the silicon structure. This particular call script set up a standard calculation for each structure and submits -each of them. +each of them. We, in addition collect the total energies and eject them to a file called +eos. """ # pylint: disable=too-many-arguments import numpy as np from aiida import load_profile from aiida.common.extendeddicts import AttributeDict -from aiida.engine import submit +from aiida.engine import run from aiida.orm import Bool, Code, Str from aiida.plugins import DataFactory, WorkflowFactory @@ -77,12 +78,13 @@ # Workchain related inputs, in this case, give more explicit output to report inputs.verbose = Bool(True) # Submit the workchain with the set inputs - submit(workchain, **inputs) + results = run(workchain, **inputs) + return results if __name__ == '__main__': # Code_string is chosen among the list given by 'verdi code list' - CODE_STRING = 'vasp@mycluster' + CODE_STRING = 'VASP/6.3.2-gompi-2021b-std-wannier90-libxc-hdf5-beef-d7238be44ec2ed23315a16cc1549a1e3@betzy' # INCAR equivalent # Set input parameters @@ -95,7 +97,7 @@ # POTCAR equivalent # Potential_family is chosen among the list given by # 'verdi data vasp-potcar listfamilies' - POTENTIAL_FAMILY = 'PBE.54' + POTENTIAL_FAMILY = 'PBE' # The potential mapping selects which potential to use, here we use the standard # for silicon, this could for instance be {'Si': 'Si_GW'} to use the GW ready # potential instead @@ -107,17 +109,27 @@ # AttributeDict is just a special dictionary with the extra benefit that # you can set and get the key contents with mydict.mykey, instead of mydict['mykey'] OPTIONS = AttributeDict() - OPTIONS.account = '' - OPTIONS.qos = '' + OPTIONS.account = 'nn9997k' + OPTIONS.qos = 'devel' OPTIONS.resources = {'num_machines': 1, 'num_mpiprocs_per_machine': 8} OPTIONS.queue_name = '' OPTIONS.max_wallclock_seconds = 3600 OPTIONS.max_memory_kb = 2000000 + EOS = [] # Iterate over each lattice constant and pass it explicitly for lattice_constant in [3.5, 3.6, 3.7, 3.8, 3.9, 4.0, 4.1, 4.2, 4.3]: # POSCAR equivalent # Set the silicon structure STRUCTURE = get_structure(lattice_constant) - main(CODE_STRING, INCAR, KMESH, STRUCTURE, POTENTIAL_FAMILY, POTENTIAL_MAPPING, OPTIONS) + results = main(CODE_STRING, INCAR, KMESH, STRUCTURE, POTENTIAL_FAMILY, POTENTIAL_MAPPING, OPTIONS) + + # The output are stored as AiiDA datatypes, which is Dict in this case. To obtain a regular + # dictionary, we use get_dict + misc = results['misc'].get_dict() + EOS.append([lattice_constant, misc['total_energies']['energy_extrapolated']]) + # Write volume and total energies to file + with open('eos', 'w', encoding='utf8') as file_object: + for item in EOS: + file_object.write(f'{item[0]} {item[1]}\n')
Notice that except for the iteration over the call to the
mainwhich eventually runs the workchain, we have replacedsubmitwithrun. The reason for this is that we have to make sure the VASP run completes before we can extract the total energies. This also has the consequence that the VASP calculations are now executed sequentially, instead of in parallel. We will later suggest how we can fix this, but for the time being, let us go with this approach. Save the new file asrun_fcc_si_multiple_volumes_eos.pyor fetch it with:wget https://github.com/aiida-vasp/aiida-vasp/raw/master/tutorials/run_fcc_si_multiple_volumes_eos.py
This file should enable automatic extraction of total energies versus the different volumes.
Execute
run_fcc_si_multiple_volumes_eos.py:$ python run_fcc_si_multiple_volumes_eos.py
Give the script time to complete. Contrary to before you will now see the output in the terminal (the similar output we previously got with
verdi process reportfor a given process). This is because we now useruninstead ofsubmit. More on this later. A fileeosshould be generated in the folder you executed the Python command.Inspect the
eosfile that was generated:$ more eos 3.5 -4.42341939 3.6 -4.66006381 3.7 -4.79595554 3.8 -4.86303429 3.9 -4.87588353 4.0 -4.8481407 4.1 -4.78451926 4.2 -4.69228837 4.3 -4.58122058
Plot the data with your favorite plotting tool, for instance Gnuplot:
$ gnuplot gnuplot> plot "eos" with lp
which should give you something similar that is shown in the FCC Si tutorial.
We have now completed the FCC Si tutorial in AiiDA-VASP. However, maybe not in the
best way. As you have already realized, we had to change submit to run in order
to make sure we got the output of each executed workchain. If we would have used submit
we would only have access to a link to some output that we did not know was occupied or not at the
point of inspection (unless we waited until we know the execution of the workchain
was complete). In doing so, all the volume steps now run sequentially and thus it takes
far longer for all the volume calculations to be finished (assuming your cluster started
each volume calculations at the time of submission).
In the next tutorial we will instead develop a workchain to perform the different volume calculations (in fact, more general, namely different structure calculations) and call that from the run script. In doing so, you will realize that we can keep the benefits of automatic extraction of the data and parallel execution. In addition, we get the benefit of storing the result in the database.