CCL: [Bad Mail]RE: OPT on large molecules

 Sent to CCL by: Alexander Kollias [ackollias::gmail.com]
 Sam,
 	There are several problems in attempting to carry out such calculations
 in QMC.  The first problem concerns the calculation of forces in QMC.
 This is due to the fact that the force estimator (by Hellmann-Feynman
 theorem) has an infinite variance.  Although improved estimators can be
 constructed, i.e. using Pulay corrections along with zero-variance
 techniques, that will result in a finite variance. Unfortunately these
 estimators do not provide sufficient improvement in the variance for
 geometry optimization, if the geometry is far from equilibrium. Second
 there is the problem of reoptimizing the MOs at each step in the
 geometry optimization. Both of these problems are areas of active
 research.  To date triatomics (XH2) are the largest molecules for which
 the force has be calculated while orbital optimization has some greater
 success. Finally, QMC calculations for large systems require hundreds of
 thousands of hours.  This is because although the variance of the local
 energy decreases as 1/sqrt(T) where T is computer time, the inherent
 variance associated with a trial function tends to grow with the
 magnitude of the total energy, i.e. the larger the total energy the
 larger the initial variance and associated error bar of the local
 energy.
 regards,
 	Sasha
 On Mon, 2006-01-30 at 19:47 -0500, Abrash, Samuel sabrash/arichmond.edu
 wrote:
 > Sent to CCL by: "Abrash, Samuel" [sabrash!=!richmond.edu]
 >
 > Could you use quantum monte carlo to do the optimization?
 >
 > Sam
 >
 >         -----Original Message-----
 >         From: Shobe, David dshobe[#]sud-chemieinc.com [mailto:owner-chemistry()
 ccl.net]
 >         Sent: Mon 1/30/2006 12:14 PM
 >         To: Abrash, Samuel
 >         Cc:
 >         Subject: CCL: OPT on large molecules
 >
 >
 >
 >         Sent to CCL by: "Shobe, David" [dshobe.()
 .sud-chemieinc.com]
 >         We still haven't answered the main question: will the lack of
 optimization
 > significantly effect the intramolecular charge transfer?  I'll assume here
 that the
 > effect in question does not involve vibrational frequencies.
 >
 >         Even if the use of an arbitrary geometry doesn't materially change
 the results of
 > the calculation, I would imagine you'd have some trouble convincing peer
 reviewers
 > of that.  But there are some options (besides just being patient :-).
 >
 >         1. You could use molecualr mechanics to do the optimization
 >         2. You could use an X-ray diffraction or other published structure,
 if one exists.
 >         3. You could repeat the calculation at several geometries to
 demonstrate that,
 > indeed, the calculated results do not have a significant dependence on
 molecular
 > geometry.
 >
 >         The idea of checking whether the effect occurs at all, using an
 arbitrary but
 > reasonable geometry, might not be a bad one.  After this investigational
 > calculation, you could then decide whetehr it's worthwhile to redo the
 calculation
 > with a proper geometry (and with all the other i's dotted and t's crossed,
 as it
 > were).
 >
 >         --David Shobe, Ph.D., M.L.S.
 >         Süd-Chemie, Inc.
 >         phone (502) 634-7409
 >         fax (502) 634-7724
 >
 >         Don't bother flaming me: I'm behind a firewall.
 >
 >
 >
 >         -----Original Message-----
 >         > From: owner-chemistry%a%ccl.net [mailto:owner-chemistry%a%ccl.net]
 >
 >         Sent: Saturday, January 28, 2006 3:48 PM
 >         To: Shobe, David
 >         Subject: CCL: OPT on large molecules
 >
 >         Sent to CCL by: "David M. Close" [closed(_)etsu.edu]
 >           CCLers:
 >           I have a question about the necessity of optimizations.  I have
 uncovered an
 > effect experimentally that involves changes in large molecules.  By large I
 mean
 > that even partial optimizations with modest basis sets take several days.
 The
 > effect involves charge migration from one region of the molecule to another
 > region.  I would like to examine this effect with computations.  After a
 weeks
 > work I still don't have an answer because of the times taken to determine
 the
 > optimized structures.
 >           I realize that the correct approach is to do calculations on the
 optimized
 > structures.  For example, it is necessary to do frequency calculations on
 the
 > optimized structures using the same basis sets in both calculations.  But
 what if
 > one didn't use the optimized structures?  Suppose one only used a good
 guess at
 > the optimized structure?  Would this cause the later calculations to be
 invalid?
 > Or might they be good enough to at least decide if the effect of interest
 is
 > present and that actual optimizations are worthwhile?
 >           Regards, Dave Close.
 >
 > [Edited by jkl() ccl.net to removed winmail.dat attachment]>
 >
 >