CCL:G: when NOT to use guess=read?

 Sent to CCL by: Cory Pye [cpye#,#crux.smu.ca]
 Hello Antonio,
 I would hazard a "guess" that guess=read would perform poorly if
 (1) there is a large change in geometric structure, say when you do a
 wide-swath 2D fixed PES scan, eg. varying the two O-H distances in H2O
 independently, maybe.
 #N hf/6-311++G** scan
 O
 H1 O H1O
 H2 O H2O H1 H2OH
 H1O = 1.0     19    0.1
 H2O = 1.0     19    0.1
 H2OH = 105.0
 Which would produce values such as
 H1O  H2O
 1.0  1.0
 1.0  1.1
 ...
 1.0  3.0
 1.1  1.0 <---
 1.1  1.1
 The arrow indicates where you might have SCF problems becasue of the large
 sudden change in H2O.
 (2) If you are restarting an SCF that is already diverging (error is getting
 large or oscillating), guess=read would just read in the crappy guess.
 (3) I have found that it is always a good idea to use the MO guess (and also
 geometry and force constants) from an optimized lower-level theory with
 frequency calculation to verify signature of Hessian) to start a higher level
 (e.g guess=read, geom=allcheck, Opt=ReadFC) unless the optimized geometry is a
 poor choice.
 On Thu, 21 Aug 2008, Antonio G. De Crisci antonio.decrisci(a)utoronto.ca wrote:
 > Hello All,
 >
 > In Gaussian 03, when is it NOT a good idea to use the guess=read option?
 For example, when performing several calculations on the same molecule, which
 may including tweaking of the grids, changing the threshold criteria, level of
 theory or from check files of a failed/crashed/incomplete calculation. I guess
 im just wondering how smart that guess=read option is when you use it and
 whether you should start the guess from scratch instead (no puns intended).
 >
 > Thanks in advance for your time.
 > Antonio De Crisci
 > University of Toronto
 >