Andreas,
> Here my question:
> I wanted to play around with scfconv, tight_p_conv and grms_tol and
> turned on verbosity. For the ash test I get for the standard non-verbose
> output
> ...
> sqm energy: 50 -179.5406 0.001885
> sqm energy: 60 -179.5406 0.000357
> Final SCF energy is -179.540648949995
>
> Turning on verbosity shows that there is *many more* calls to the
> energy/gradient routines than what is apparent from sqm's output. I
> printed the number of force calls in the do while loop in xmin.f and it
> counts 870 calls... this means that the TNCG optimizer does not need 60+
> steps but rather almost 900... if this is correct there is still
> something wrong with the optimizer, interface, or gradients. So what
> does the variable xmin_iter actually count? Or did I overlook something?
> Switching back to LBFGS results in 846 force calls.
When you run TNCG, it uses gradient evaluations for computing
Hessian-vector matrix products in the inner conjugate gradient loop.
This is where the numerous calls come from. For some systems the total
number of function calls, which is ultimately the rate limiting
calculation, turns out to be less with LBFGS than with TNCG even
though the latter uses a lot less outer minimization steps (counted by
xmin_iter). For most systems, however, especially when you set the
minimization convergence criterion really tight, TNCG wins out.
Istvan
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Dec 18 2009 - 08:30:02 PST
