Katsuhiko Nishimra
ktns.****@gmail*****
2014年 1月 25日 (土) 22:04:07 JST
Mr. Fujii, First, your criteria for ending optimization seems not to be inappropriate. And this error seems to be related to the ticket #30862. Ultimately, the problem is how the RFO scaling factor should be determined. At present, I implemented the scaling code under precondition that the RFO step scales linearly to the scaling factor. But, in fact, the RFO step scales apparently quadratically to the scaling factor. So, if the Hessian approximation goes too poor, the RFO step goes too small unexpectedly, and perhaps due to numeric unstablity, Hessian update routine fails to generate well defined Hessian matrix, I guess. I presently don't have any idea why step goes too small. #30862 is only a workaround for this problem, but should I implement it? 2014/1/25 Mikiya Fujii <mikiy****@gmail*****>: > And, is the conditions of optimization is too strict? > > M > *************************************** > Mikiya Fujii, Ph.D. > > Yamashita & Ushiyama Laboratory, > Department of Chemical System Engineering, > Faculty of Engineering, > The University of Tokyo. > > tel&fax: +81-3-5841-7286 > email: fujii****@tcl***** > HP: http://www.tcl.t.u-tokyo.ac.jp/ > *************************************** > > > 2014/1/25 Mikiya Fujii <mikiy****@gmail*****>: >> Dear ktns, >> >> Is this related to ticket #30862? >> https://sourceforge.jp/ticket/browse.php?group_id=6231&tid=30862 >> >> M >> *************************************** >> Mikiya Fujii, Ph.D. >> >> Yamashita & Ushiyama Laboratory, >> Department of Chemical System Engineering, >> Faculty of Engineering, >> The University of Tokyo. >> >> tel&fax: +81-3-5841-7286 >> email: fujii****@tcl***** >> HP: http://www.tcl.t.u-tokyo.ac.jp/ >> *************************************** >> >> >> 2014/1/25 Mikiya Fujii <mikiy****@gmail*****>: >>> Dear ktns, >>> >>> Hi, Please check an error written in d0_am1_opt.dat. >>> The corresponding input is also attached. >>> The used revision is r1659. >>> >>> I'm not hurrying up. >>> >>> My best, >>> Mikiya >>> >>> >>> >>> >>> *************************************** >>> Mikiya Fujii, Ph.D. >>> >>> Yamashita & Ushiyama Laboratory, >>> Department of Chemical System Engineering, >>> Faculty of Engineering, >>> The University of Tokyo. >>> >>> tel&fax: +81-3-5841-7286 >>> email: fujii****@tcl***** >>> HP: http://www.tcl.t.u-tokyo.ac.jp/ >>> *************************************** > > _______________________________________________ > MolDS-dev mailing list > MolDS****@lists***** > http://lists.sourceforge.jp/mailman/listinfo/molds-dev -- Katsuhiko Nishimra ktns.****@gmail*****
