From chemistry-request;at;ccl.net Fri Apr 17 18:01:20 1992 Date: Fri, 17 Apr 1992 15:24 CST From: Andy Holder Subject: Semiempirical errors. To: CHEMISTRY-!at!-ccl.net Status: R Hello netters. Jack Houser asked anout some error messages that were generated when running an AM1 calculation. While many of these types of errors are relics of older codes, this one sounds pretty common, so I'll address it to the net in general. Large molecules sometimes have gradient problems, and this sounds a great deal like on of those. The FIRST thing that should be done is to check the actaul values of the individual gradients and the gnorm itself. (This is done in AMPAC by using the keyword GRAD.) If one of these gradients is enormous, there is the problem. The molecule can be redefined in order that the offending parameter is removed. In the case of such a large system, I am leery of using an automatic z-matrix generation routine. (Topical question: Why use it at all if you don't use it for big guys?) Defining symmetry to a large gradient parameter only worsens the situation in that the bad parameter is even more important now than when it caused generation of the first error! There are several ways to approach one of these. I'll list but a few below in no particular order. (Others please participate as well. There needs to be more discussion of "real" chemistry (i.e. quantum (haha)) on here anyhow.) 1. Redefine the molecule to get rid of the bad geomtric variable. This is not always possible and may lead to other problems. 2. Optimize a biggy in pieces. Divide the system into easily handled chunks and only set optimization flags on those pieces. After this is accomplished, put the optimized chunks back in and let the whole thing go again. This is a bit tedious, but I've had good success with it. 3. Use dummy atoms to redefine the parameters with big gradients. 4. The wavefunction for the molecule may not be stable. It may be that the system is a biradical and you are trying to define it via RHF. Polyenes also suffer from nonconvergence along this line as well. UHF or limited (or extensive) CI may be required if this happens. [ Side note: ALWAYS use DERINU keyword within AMPAC with CI. ] 5. Check the geometry for any wacky things, like atoms too close. AMPAC will warn you about this within limits. 6. Draw the final geometry that is dumped and see what has happened to the geometry. This may give you a clue about what is causing the optimization problem. A few more ideas. Users of AMPAC (yes, Virginia there is a MOPAC as well) MUST check final geometries by computing frequencies at the level of theory used to determine the geometry. This requires either use of FORCE (force constants in Cartesian coordinates) or LTRD (AMPAC only, force constants in internal coordinates). There should be no negative force constants for a ground state, one for a transition state and so forth (see page 13 of the AMPAC 2.1 manual). Before this can be done, however, the gnorm must be reduced. You must check this at the end of your optimization. I always use the PRECISE keyword when doing any AMPAC calculation. This requires better refinement of the wave function and the gnorms. It is almost required when doing FORCE analysis. For large systems, the geometry may need to be preminimized prior to application of PRECISE. Andy =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= DR. ANDREW HOLDER Assistant Professor of Computational/Organic Chemistry Department of Chemistry || BITNET Addr: AHOLDER {*at*} UMKCVAX1 University of Missouri - Kansas City || Internet Addr: aholder[ AT ]vax1.umkc.edu Spencer Chemistry, Room 502 || Phone Number: (816) 235-2293 Kansas City, Missouri 64110 || FAX Number: (816) 235-1717 =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=