euphonic-optimise-dipole-parameter

This program is useful for users wanting to efficiently calculate phonon frequencies on many q-points for polar materials.

Polar materials have a long range force constants matrix, so an analytic correction must be applied, which is computationally expensive. The correction is calculated using an Ewald sum, and the balance between the real and reciprocal sums can be tuned to reduce the computational cost without affecting the result. This is done with the dipole_parameter argument to ForceConstants.calculate_qpoint_phonon_modes

The program runs a calculation for a small test number of q-points (100 by default) repeatedly for different values of dipole_parameter, and times both the initialisation time, and time per q-point. Euphonic precalculates as much of the correction as it can to minimise the calculation time per q-point, so although some dipole_parameter values will have higher initialisation time, the time per q-point may be lower, which is what’s important when calculating for many q-points.

Usage

Simply run the program on a .castep_bin file to get an output suggesting the optimum dipole_parameter. For example, for quartz.castep_bin:

euphonic-optimise-dipole-parameter quartz.castep_bin

Results for dipole_parameter  0.25
Ewald init time     :  0.39 s
Ewald time/qpt      :  27.69 ms

Results for dipole_parameter  0.50
Ewald init time     :  0.08 s
Ewald time/qpt      :  5.57 ms

Results for dipole_parameter  0.75
Ewald init time     :  0.03 s
Ewald time/qpt      :  4.17 ms

Results for dipole_parameter  1.00
Ewald init time     :  0.03 s
Ewald time/qpt      :  6.79 ms

Results for dipole_parameter  1.25
Ewald init time     :  0.06 s
Ewald time/qpt      :  10.30 ms

Results for dipole_parameter  1.50
Ewald init time     :  0.06 s
Ewald time/qpt      :  15.04 ms

******************************
Suggested optimum dipole_parameter is  0.75
init time           :  0.03 s
time/qpt            :  4.17 ms

As you can see above, the time per q-point has been reduced from 6.79ms per q-point to 4.17ms per q-point, just by correctly tuning the real and reciprocal sums.

You can change the number of q-points to test, the minimum dipole_parameter, maximum dipole_parameter, and step size between dipole_parameter values. To see the command line options, run:

euphonic-optimise-dipole-parameter -h

Or see the command line options in more detail below

Command Line Options

Run and time an interpolation calculation for a small number of q-points for different values of dipole_parameter to determine dipole_parameter’s optimum value for this material

usage: euphonic-optimise-dipole-parameter [-h]
                                          [--asr [{reciprocal,realspace}]]
                                          [--use-c | --disable-c]
                                          [--n-threads N_THREADS] [-n N]
                                          [--dipole-parameter-min DIPOLE_PARAMETER_MIN]
                                          [--dipole-parameter-max DIPOLE_PARAMETER_MAX]
                                          [--dipole-parameter-step DIPOLE_PARAMETER_STEP]
                                          filename

Named Arguments

-n

The number of times to loop over q-points. A higher value will get a more reliable timing, but will take longer

Default: 500

--dipole-parameter-min, --min

The minimum value of dipole_parameter to test

Default: 0.25

--dipole-parameter-max, --max

The maximum value of dipole_parameter to test

Default: 1.5

--dipole-parameter-step, --step

The difference between each dipole_parameter to test

Default: 0.25

File I/O arguments

filename: Phonon data file. This should contain force constants data. Accepted formats: .yaml, force_constants.hdf5 (Phonopy); .castep_bin, .check (Castep); .json (Euphonic).

Force constants interpolation arguments

--asr

Possible choices: reciprocal, realspace

Apply an acoustic-sum-rule (ASR) correction to the data: “realspace” applies the correction to the force constant matrix in real space. “reciprocal” applies the correction to the dynamical matrix at each q-point.