[Wien] Can NMATMAX ba a universal number for MPI-version? / Thank you
Oh, now I understand. Thank you for the reply. Oleg Rubel On Fri, 25 Apr 2008, Peter Blaha wrote: > This is already done! > > nmatmax1=nmatmax1*sqrt(NPE*1.d0) > > But of course you still need a "meaningful" NMATMAX at the beginning. > > Oleg Rubel schrieb: >> Dear Wien2k Users and Developers, >> >> I calculated some case in which my RKmax was reduced due to 'Matrix size' >>> NMATMAX. That was a serial version. I moved to MPI and met the same >> problem. In the case of MPI, I would expect that NMATMAX applies to the >> piece of the global matrix stored on a particular node, and thus larger >> 'Matrix size' is possible. But it seems to be not a case. Of course I can >> recompile lapw1 with a larger NMATMAX for lapw1(c)_mpi, but again it >> depends on the number of nodes involved and cannot be a universal number. >> >> Should the rules of applying NMATMAX to MPI version be changed? >> >> >> Any suggestions are welcome. >> >> Thank you in advance >> >> Oleg Rubel >> >> === >> Faculty of Physics >> Philipps University Marburg >> ___ >> Wien mailing list >> Wien at zeus.theochem.tuwien.ac.at >> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien > > -- > > P.Blaha > -- > Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna > Phone: +43-1-58801-15671 FAX: +43-1-58801-15698 > Email: blaha at theochem.tuwien.ac.atWWW: > http://info.tuwien.ac.at/theochem/ > -- > ___ > Wien mailing list > Wien at zeus.theochem.tuwien.ac.at > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien > >
[Wien] Can NMATMAX ba a universal number for MPI-version?
This is already done! nmatmax1=nmatmax1*sqrt(NPE*1.d0) But of course you still need a "meaningful" NMATMAX at the beginning. Oleg Rubel schrieb: > Dear Wien2k Users and Developers, > > I calculated some case in which my RKmax was reduced due to 'Matrix size' >> NMATMAX. That was a serial version. I moved to MPI and met the same > problem. In the case of MPI, I would expect that NMATMAX applies to the > piece of the global matrix stored on a particular node, and thus larger > 'Matrix size' is possible. But it seems to be not a case. Of course I can > recompile lapw1 with a larger NMATMAX for lapw1(c)_mpi, but again it > depends on the number of nodes involved and cannot be a universal number. > > Should the rules of applying NMATMAX to MPI version be changed? > > > Any suggestions are welcome. > > Thank you in advance > > Oleg Rubel > > === > Faculty of Physics > Philipps University Marburg > ___ > Wien mailing list > Wien at zeus.theochem.tuwien.ac.at > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien -- P.Blaha -- Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna Phone: +43-1-58801-15671 FAX: +43-1-58801-15698 Email: blaha at theochem.tuwien.ac.atWWW: http://info.tuwien.ac.at/theochem/ --
[Wien] Can NMATMAX ba a universal number for MPI-version?
Dear Wien2k Users and Developers, I calculated some case in which my RKmax was reduced due to 'Matrix size' > NMATMAX. That was a serial version. I moved to MPI and met the same problem. In the case of MPI, I would expect that NMATMAX applies to the piece of the global matrix stored on a particular node, and thus larger 'Matrix size' is possible. But it seems to be not a case. Of course I can recompile lapw1 with a larger NMATMAX for lapw1(c)_mpi, but again it depends on the number of nodes involved and cannot be a universal number. Should the rules of applying NMATMAX to MPI version be changed? Any suggestions are welcome. Thank you in advance Oleg Rubel === Faculty of Physics Philipps University Marburg
[Wien] Re Re: integer elements in rotation operation
> However, I probably think that you misunderstood you question. > > My questions can be expressed as foloowings: > > 1) whether the basis vectors of the coordinate system on which the rotation > operation act >are unit cell or primary cell basis, which may be not orthogonal to each > other. As I mentioned, this depends on the specific lattice. Eg. for an F-centered structure, the "conventional" (cubic) cell is used as coordinate system, while for eg. a hex. lattice, the hexagonal basis will be used. > 2) for two vectors whose coordinates are expressed in non-orthogonal basis, > the dot-product operation > >between them will not be directly implemented as the dot-product operation > of two vectors whose coordintes > > are expressed in orthogonal system. I think that at least one should first > convert the coordinats of > >non-orthogonal systm to the corresponding coordinates in orthogonal > system, then dot-product > >operation will be carried out as usually.i.e > (x1i+y1j+z1k).dot.(x2i+y2j+z2k)=x1*x2+y1*y2+z1*z2 > > but thoroughly this conversion step is not found in stern.f > > so, what is goning on here about the basis vectors choosing techniques ? A dot product K.R is independent on the chosen coordinate system (as long as both K and R space are corresponding to the same system).