-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hello Jie-rong-- > I have some questions about tensor calculation and C++/Fortran code. > > Is there any difference between "calcTensor" in varTensorTools.py and > "TENSOr" (Sass, 2001) in conventional Xplor-NIH? Are they the same thing? > They are not the same. > Does the potential energy terms in python modules e.g. NOEPot, RDCPot, still > call the Fortran code of old X-plor to calculate the structure and penalty > function? or these new potential functions are only coded in C++ (e.g. > NOEPot.cc) and nothing to do with the original Fortran? > They have nothing to do with the Fortran code. > If the calTensor is called at every cooling cycle to update the alignment > tensor during simulated annealing, would it be computationally expensive? > Since it calculates the tensor "outside" the core and put result back to the > core to do the structure calculation. Will it be more efficient if the > function of calcTensor is in the source code and recompiled, or it doesn't > matter? Probably too slow, and also not necessary. It suffices to periodically recalculate the tensor (not every timestep). > (I am writing a python helper, which is similar to calcTensor. Steric > alignment tensor can be obtained by sampling the structure over an > isotropically distributed space (The result is the same as PALES multiplied > by a constant). The method is computationally expensive, so I am wondering > if it can be improved.) Most likely, it can be improved. The first task is to locate bottlenecks- you might use a Python profiler for this. Once the bottlenecks are identified, I might be able to give further advice. best regards-- Charles -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Processed by Mailcrypt 3.5.8+ <http://mailcrypt.sourceforge.net/> iD8DBQFIVqlHPK2zrJwS/lYRArrOAJ0XTo++d7nTan3ZvYwEOiv1UzGwhACfZvO3 HyW3w6rhma6SGAYHZ5h7Gx0= =zO0j -----END PGP SIGNATURE-----
