On Thu, Jun 22, 2006 at 02:23:22PM +0200, Francesco Pietra wrote: > Hi Lennart: > This is a most welcome invitation. > > Below is pasted a message that I addressed this morning (8.38 UTC) to the > mpqc community. As such, it could only trigger response from experts in mpqc. > However, to let you investigating the matter, I could send you files of > successful and unsuccessful computations. The version of mpqc is just the one > currently on debian etch amd64. > > I don't know if it is a problem of software. It may well be that I am putting > mpqc to work the wrong way. In particular, I do not undestand whether the > flexibility of the molecule poses problems to the software, stopping it under > certain conditions. I do not know how exactly mpcq works, the manual is > detailed as to commands but I have not found a description of how it works. > Perhaps, to understand fully, one should look at the source. But, as an > object oriented program, it must be a hard way. I must also add that what I > am trying to do is putting quantum calculation on the very hard way; I wonder > whether other dare so. Not evident from the publications. The problems I am > investigating are rather for molecular mechanics, and I do so. But molecular > mechanics tells nothing about most properties of the molecule, except the > shape and relative energies of conformers. > > What will never finish to surprise me is that tout le monde - or about so - > does such computations with a commercial software, released (at a very high > price) for commercial Linux under the understanding that installation of > Linux must be that of the CD, no variations. Sic. It surprises me because I > can not figure how users can understand what they are doing with such > inmperscrutable packages. At any event, these computations end on the most > respectable chemical journals. > > Tell me what you want. > > cheers > francesco > ______________ > Although this may well be a premature ask for help, it might result in > suggesting me a general approach. > > mpqc 2.3.0 runs correctly (amd64 debian etch, parallel mode with four nodes, > dual opteron) for simple molecules, B3LYP simple input, as given in the > manual with water.in. For example, the energy difference between the two ax > and eq conformers of methylcyclohexane (input from molecular-mechanics > minimized structures) is evaluated 1.8 kcal/mol, in accordance with the > experimental value. In both cases the calculation reached convergence > extremely rapidly and the energy was furnished as "Molecular Energy". > > With molecules of the size C22H25NO6 involving benzene ring, unsaturated > conjugated ring and helicity problems, so far I was unable to reach > convergence. Either one of the two most stable conformers used for input with > the water template (basis set 3-21G* or higher) with coordinates furnished as > described above for the methylcyclohexanes, was not brought to conclusion. In > most instances the calculation hangs after a few cycles of iterations and the > prompt is not furnished; I can examine from the *.out file "each ?molecule" > for each iteration set and found it correct, with all atoms - including > hydrogens - where they should be. Also, the general conformation was not much > changed, except for minor details. > > In another instance the above calculations went farther (about four hours), > for some twenty cycles of iteration and it did not hang: the prompt was > furnished. However, no "Molecular Energy" was furnished, only scf energy and > it was clearly stated that convergence was not reached. Comparing scf energy, > from input as bove for the two most stable conformers, resulted in a far too > large energy difference between them (5 kcal/mol; absurd because I can see > both them by NMR at r.t.). > > I was unable to tell the story in fewer words. I do not expect anything easy > for complex systems. Before putting the hands further on files I am trying to > figure out a plan how to move. > > Thanks for your kind attention.
Well there is a version 2.3.1 which is supposed to contain some bug fixes. I wonder if there is any chance that one of the fixed bugs is what is causing your problem. The mpqc changelog has this: 2006-03-22: MPQC-2.3.1 * MCSearch added for cubic interpolation during quasi-Newton line searches. * Added KMLYP method. * Updated libtool to version 1.5.22. * SumDenFunctional returns correct HF exchange coefficient when using nested ACM functionals. This could change the results from certain, uncommon inputs. * Other minor bug fixes, enhancements, and * documentation improvements. Debian doesn't have 2.3.1 built yet, but it wouldn't be hard to do. I can try and build a debian package of 2.3.1 if you want to try that. Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]