Mopac: Summary: Structure of excited states

 My thanks to those who replied to my query on optimizing structures of excited
 states in Mopac.  This appears to be less straightforward than I expected. .
  .ah, well, fools rush in. . .
 Below are the replies I received, subjected to some (unmarked) snipping:
 >>Andreas Parusel, Vienna (andreas -AatT- majestix.msp.univie.ac.at):
 I read your posting in the ccl about excited states. I am working also with
 these problems in the VAMP package, but I do have the same problems.
 As fas as I know, you have to change the structure of the MOLECULE a bit ( for
 my large molecules for ex. -Ph --> -CH3) then it works better.
 Sometimes it works if you coose another converger ... PULAY or IIS ... you
 could try this.
 Or you have to optimize the excited state with GNORM=20 f. ex. and lowering the
 gradient step by step ...  this are all possibilities that work sometimes but
 not always!
 >>Dr. Jose Ignacio Garcia-Laureiro (jig -AatT- qorg.unizar.es):
 In general, geometric optimizations at the CI level are MUCH more time
 demanding than "normal" ones, but I realize that 10 cpu hours on an
 Indigo2 is too much... You may take a look to the following reference:
 M.J.S. Dewar, S. Olivella, J.J.P. Stewart
 J. Am. Chem. Soc., 108 (1986) 5771-5779
 These authors recommend to start with an UHF geometry optimization, which
 is as fast as an normal RHF one, but leads to a better starting point for
 a further CI optimization.
 >>Dr. Peter Gedeck (gedeck -AatT- organik.uni-erlangen.de):
 As far as I think SHIFT only damp SCF oscillations and does not change
 the behaviour of the EF geometry optimization. You should try setting DMAX
 to a smaller value. This restricts the maximum step size of the geometry
 optimization and might thus damp the oscillations. Also try different
 geometry optimizers.
 Another problem that can occur is a crossing of excited states during the
 course of the geometry optimization. If such a crossing occurs the
 calculated gradients suddenly belong to a different state than in the
 steps before and you can imagine that this way your calculation must
 fail. A possible way to reduce the problem here is to define the
 multiplicity of the state you want to optimize. E.g use:
    root=1 singlet    for the S0 state
    root=2 singlet    for the S1 state
    root=1 triplet    for the T1 state
 This way at least we get rid of the problem of crossing singlet and
 triplet states.
 It might be that some of the keywords are not correct because I usually
 use VAMP and not Mopac but there are definitly similar keywords.
 >>Lutfur Khundkar (LKHUNKR -AatT- NEU.EDU):
      I have tried a few excited state calculations using MOPAC 6 using OS/2 on a
 486D50. My molecules were relatively small, but what I would consider medium to
 large for semiempirical calculations, specially geometry optimizations. I have
 worked with about 16 heavy and 12  H atoms. There are some special keywords
 (SINGLET and TRIPLET) that you can use to restrict yuour basis set. However,
 geomtry optimizations do take a long time.From what I understand from talking to
 a few people, C.I.=2 is not really a useful choice --- unless you are interested
 in very approximate numbers. My own experience with a limited set of molecules
 has been that even the energies that you get seem to depend on the level of CI.
 I find that CI=4 (single geometry SCF calculations) and less gives results that
 are reasonably similar, but CI=5 makes a significant difference to the energies
 and relative ordering of states (I use dipole moments to "identify" my
 states
 --- its an approximate handle for me, since thats what I am interested in). I
 remember seeing a post to this list several months back about MECI in Mopac that
 discussed an alternative to using the preset CI keyword. You can define your
 microstates to be used in MECI by Mopac, although this does not address your
 initial concern about how long it takes to get a converged calcualtion.