PWscf to yambo interface (p2y): tips and troubleshooting
File formats in general are a nasty business, so you can imagine how converting between different formats can easily lead to problems. PWscf itself underwent many changes from 2.1.5 to 4.0 in the output data format, although the 4.0 file format should be quite stable. It's likely that issues remain with the p2y convertor. Please consider the following points when using the routine.
- The direct method is usually the fastest. However, in some specific cases it might be better to use the indirect method, e.g. if there are thousands of k-points.
- Kleinman-Bylander form factors are not created with p2y: if these are important to you, you might be better using the Abinit convertor, a2y. Otherwise, you might think about writing the code yourself...!
- Yambo doesn't like non-symmorphic symmetries (those including translations).
There are a few ways to remove them:
- Change the cutoff, and hence, the FFT mesh slightly, until you see some "symmetry not allowed" in the pw.x output;
- Shift your cell so that all symmetries are real (if possible);
- Break the symmetry by giving equivalent atoms different names, e.g. Si1, Si2;
- Fix it at source: in PW/sgam_at.f90, initialize fractional_translations to .false. (thanks to Paolo Giannozzi for that one).
- Make sure that you have used the correct configuration option (--with-p2y = xxx) appropriate for use with the version of PWscf you have.
- Make sure you have used BOTH --with-p2y= AND --with-iotk= options during configure.
- Make sure you have actually asked for the wavefunctions to be written! Check for the presence of evc.dat files in the .save/K00001/ directories.
- Be consistent when using "-assume byterecl" with ifort.
- Since libiotk.a is compiled with the Quantum-ESPRESSOpackage, it is often compiled as a parallel library: if so, the p2y compilation must also be done using a parallel compiler, or else there will be complaints about missing MPI libraries, etc.
- The direct interface is quite slow for versions 3.1 and 3.1.1 due to re-reading of the data-file.xml file.
- Although pw_export.x is supposed to work in parallel with many processors, it sometimes does not run smoothly. For instance, the command structure pw_export.x -input cheese.export might not work, which may cause some redirection problems. Direct your woes to pw_forum.
Still having problems?
Conversion from PWscf files can be a bit tricky, and it is possible that bugs remain in the code. If you are sure that you have tried everything, send us a mail, attaching the data-file.xml file, the first few lines of ncdump SAVE/ns.db1 and the error messages and standard output, as well as anything else that might be useful to know.