Accuracy¶
EPW accuracy on various quantities¶
Some physical quantities are more sensitive to numerical instabilities than other. Below is a list of quantities tested by the EPW Test Farm and their related accuracies.
Note that this is an upper bounds and show the maximum variation on a quantity due to different compiler, hardware, libraries, level of optimization and parallelism.
Physical quantity |
Absolute error |
Relative error [in %] |
Benchmark value |
Slave |
Note |
|---|---|---|---|---|---|
DOS |
0.00001 |
0.001 |
stable |
||
Eigenenergy (eV) |
0.001 |
0.5 |
-0.0219 (epw_soc) |
EPW-farmer_gcc485 |
stable |
Re[Sigma] (meV) |
0.02 |
2.0 |
-982.600493 (epw_soc) |
EPW-farmer_gcc485 |
stable |
Im[Sigma] (meV) |
0.04 |
0.04 |
1858.014377 (epw_soc) |
EPW-farmer_gcc485 |
stable |
Z |
0.08 |
0.08 |
-11.665563 (epw_soc) |
EPW-farmer_gcc485 |
stable |
PH lamda (lambda) |
0.00001 |
0.001 |
stable |
||
PH lamda (lambda_tr) |
0.00001 |
0.01 |
stable |
||
El lamda (lam) |
0.00001 |
3.0 |
-0.00015 (epw_base) |
EPW-farm2_pgi15 |
stable |
Integrated lamda |
0.0001 |
0.02 |
0.6690848 (epw_soc) |
EPW-farm2_pgi15 |
stable |
Integrated lamda TR |
3 |
100.0 |
2.7783115 (epw_soc) |
EPW-farm2_intel13 |
bug |
Ph gamma (meV) |
0.008 |
900 |
-0.001178 (epw_soc) |
EPW-farm2_intel13 |
stable |
Ph freq (meV) |
0.005 |
0.05 |
104.4255 (epw_trev) |
desktop |
stable |
logavg for Tc |
0.00001 |
0.2 |
0.005807 (epw_trev) |
EPW-farmer4_gcc485_openblas |
stable |
l_a2F for Tc |
0.001 |
0.2 |
0.0838588 (epw_trev) |
EPW-farm2_intel13 |
stable |
Fermi level (eV) |
0.00001 |
0.001 |
stable |
||
lambda_max |
0.0005 |
0.01 |
5.164161 (epw_super) |
desktop |
stable |
lambda_k_max |
0.0001 |
0.01 |
1.510491 (epw_super) |
desktop |
stable |
Electron-phonon coupling strength |
0.0005 |
0.05 |
0.939445 (epw_super) |
desktop |
stable |
Allen-Dynes Tc (K) |
0.02 |
0.05 |
31.8408371 (epw_super) |
EPW-farm2_intel13 |
stable |
BCS superconducting gap |
0.0001 |
0.05 |
0.0048291 (epw_super) |
EPW-farm2_intel13 |
stable |
The Benchmark value is the value from the tests that lead to the largest error. In parenthesis is the name of the test from the test suite.
If you want to run the test yourself, you can do so by going to
cd espresso/test-suite/
Then type
make prolog
to generate the userconfig file.
You can then run an epw test in sequential by doing
make run-custom-test testdir=epw_base
and in parallel by doing
make run-custom-test-parallel testdir=epw_base