optimise_eta.py Script

This script 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 eta_scale argument to ForceConstants.calculate_qpoint_phonon_modes

This script runs a calculation for a small test number of q-points (100 by default) repeatedly for different values of eta_scale, 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 eta 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 script on a .castep_bin file to get an output suggesting the optimum eta_scale. For example, for quartz.castep_bin:

optimise_eta.py quartz

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

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

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

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

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

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

******************************
Suggested optimum eta 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 eta, maximum eta, and step size between eta values. To see the command line options, run:

optimise_eta.py -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 eta to determine eta’s optimum value for this material

usage: optimise_eta.py [-h] [-n N] [--eta_min ETA_MIN] [--eta_max ETA_MAX]
                       [--eta_step ETA_STEP]
                       filename

Positional Arguments

filename

The .castep_bin file to extract the data from

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: 100
--eta_min	The minimum value of eta to test Default: 0.25
--eta_max	The maximum value of eta to test Default: 1.5
--eta_step	The difference between each eta to test Default: 0.25