Phonons does NOT work in parallel, you cannot use mpirun on it. Phonon calculations simply do require huge amounts of memory, this is a current concern in the development of Inelastica, some initiatives have been taken to minimize memory usage.
Please update to the latest Inelastica version if you already haven't. If this does not help you, then try and run the calculation in single precision (check the flags for phonons). 2015-08-27 14:41 GMT+02:00 Jingxian Yu <[email protected]>: > Dear All, > > I have successfully generated all required *.TSHS files by dividing the > large system into small fragments, plus required only-S files. When I > attempt to calculate phonon-electron coupling using the following command > line > ____________________________________________________________ > Phonons -c --DeviceFirst=65 --DeviceLast=226 --FCfirst=81 --FClast=210 > --FCwildcard="./FCrun*" PHrun > ____________________________________________________________ > > our computer starts the calculation without any problem, but always > terminates the job after 4-7 minutes no matter whether I run the > calculation in a parallel or sequential mode. But the computer echoes > different error messages. It shows "=>> PBS: job killed: mem 34582064kb > exceeded limit 31457280kb" in the sequential mode, while it is "mpirun > noticed that process rank 17 with PID 25741 on node r2949 exited on signal > 9 (Killed)" in the parallel mode. > > It is clear that a sequential calculation requires a greater memory size. > I am not sure whether the same memory size issue results in the termination > of a parallel job because the message echoed from a parallel is not > informative to me. > > Could I have your comments and suggestions on this calculation? Your help > will be sincerely appreciated. Cheers! > > Jin > > > > On 14/08/2015 6:11 PM, Nick Papior wrote: > > You shouldn't need to rename the files, nor delete files, try first and do > the straightforward thing. Do them separately in their own directories: > FCrun_11-11 > FCrun_12-12 > Then do > Phonons --FCwildcard="./FCrun_*" ... > > This should work fine. > > 2015-08-14 8:26 GMT+00:00 Jingxian Yu <[email protected]>: > >> Dear All, >> >> I need to calculate force constants for a large system (containing more >> than 300 atoms). It would take more than 1000 hours on a 16-core node. So I >> attempt to divide the large job into small pieces. My plan is to divide >> device atoms into 10 consecutive sections, and calculate the Hamiltonian >> (TSHS files) for each section. Then I will rename them, so they can link >> together. >> >> I had a test with the AuH2 example. Unfortunately, my idea was not >> successful. >> >> In brief, I set the following lines to calculate the Hamiltonian for the >> atom 11, >> MD.FCfirst 11 >> MD.FClast 11 >> >> then I set the following lines to calculate the Hamiltonian for the atom >> 12, >> MD.FCfirst 12 >> MD.FClast 12 >> >> After two jobs completed, I renamed "Chain 12 1.TSHS" to "Chain 11 >> 7.TSHS", and so on. Also, I deleted the file "Chain 12 0.TSHS". >> >> Regarding overlap calculations, I followed the setting suggested on the >> tutorial. >> >> However I got the following message when I ran "Phonons -c >> --DeviceFirst=10 --DeviceLast=13 --FCfirst=11 --FClast=12 PHrun", >> ... >> Phonons.GetFileLists: WARNING - Inconsistent number of *.TSHS files in >> directory ./FCrun >> ... >> ValueError: could not broadcast input array from shape (0,3) into shape >> (22,3) >> ... >> >> Is my idea too naive? Your comments will be appreciated! Cheers! >> >> Jin >> >> >> >> >> >> >> >> >> >> > > > -- > Kind regards Nick > > > -- Kind regards Nick
