CCL: QM/MM computational cost?

 Sent to CCL by: "Ross Walker" [ross|*|rosswalker.co.uk]
 Hi Nathan,
 > Greetings CCLers.  Can anyone point me to a reference discussing the
 > computational cost of QM/MM simulations and how it scales
 > with size of the
 > components?  I have some experience with quantum chemistry
 > simulations, as
 > well as molecular dynamics simulations, but I'm new to QM/MM
 > and am being
 > asked as part of a proposal I'm writing to discuss the
 > computational cost o=
 > f
 > the calculations I want to do.  Right now the plan is to
 > solvate a simple 6
 > atom quantum mechanical system with approximately 1000 TIP4P water
 > molecules.
 This really depends on the type of QM calculation you plan on doing which
 you don't state. Semi-empirical is significantly cheaper than full ab
 initio. For what you suggest above if you wanted to do semi-empirical you
 can use Amber 9 and the computational cost will be about 6% more than doing
 the calculation classically. You can probably go to about 90 QM atoms or so
 before it even starts to become expensive. Plus you can do a simulation with
 a full treatment of electrostatics and periodic boundaries for both the QM
 and MM regions using PME. Without a PME treatment you will need to use an
 infinite (i.e. no) cutoff and probably add an implicit solvent treatment for
 the gas phase region of your system to get meaningful results. Using a cut
 off with gas phase / water cap simulations introduces serious artefacts into
 your simulation and can completely invalidate your results. Hence often it
 can be quicker to do a full PME treatment than to run the gas phase
 non-periodic simulation.
 A lot of work has been put into making Amber 9's QM support as fast as
 possible while preserving accuracy in the gradients. It is around 10 times
 quicker than Amber 8 and Charmm c32. The Amber 9 QMMM implementation is
 currently being added to Charmm and will I believe be released shortly.
 As a quick guide here are the timings I got for two water molecules treated
 with PM3 and then solvated with a rectangular box consisting of 332 TIP4P
 water molecules. PME was run with an 8 angstrom MM and QM direct space cut
 off, shake was used for both the MM and QM region, langevin thermostating
 was used at 300K and a timestep of 2fs was used. For the gas phase
 simulation I used an infinite MM and QM cut off. The default SCF convergence
 of 1.0x10-8 KCal/mol was used.
 This is for a pentium-D 3.4GHz 2MB L2 cache machine. Time is for 2000 steps
 = 4ps.
                      1 cpu               2 cpu's
                  time(s) ns/day      time(s) ns/day
 PME classical =   39.99  8.64        23.35   14.80
 PME QMMM      =   62.15  5.56        40.15    8.83
 Gas phase clas=  123.67  2.79        64.48    5.36
 Gas phase QMMM=  130.99  2.64        70.12    4.93
 So for the system you want to study it really won't cost you much at all.
 However, if you want to do a full abinitio treatment then it could get very
 expensive.
 All the best
 Ross
 /\
 \/
 |\oss Walker
 | HPC Consultant and Staff Scientist |
 | San Diego Supercomputer Center |
 | Tel: +1 858 822 0854 | EMail:- ross[]rosswalker.co.uk |
 | http://www.rosswalker.co.uk | PGP Key available on request |
 Note: Electronic Mail is not secure, has no guarantee of delivery, may not
 be read every day, and should not be used for urgent or sensitive issues.