Odds are you *can't* find the global minimum with any of these methods for
any particularly large system - it's like looking for an electron in an
atom in a needle in a haystack, only worse. That said, these methods stand
a chance of at least finding the needle for cases like molecules...
If someone tells you that you should use simulated annealing to perform
this task specifically because it is guaranteed to converge, back away from
them slowly and never ever ask them for advice again - it's only guaranteed
to converge in *infinite* time and the only reason that's the case is
because it is ergodic which means it will theoretically eventually sample
all possible values of all possible variables - long after the sun has
turned to a red giant and swallowed the earth whole. And point them to
http://en.wikipedia.org/wiki/No_free_lunch_in_search_and_optimization if
they're persistent.
That said, sampling schemes that avoid stupid and impossible configurations
like MD IMO outperform blind sampling in finite time if you come up with a
reasonably randomized, independent and large set of starting
configurations. But that's only because they avoid large and mostly
disjoint regions of impossible configurations at the expense of less
efficient overall sampling. All that said, the above still holds for
finding the *global minimum*.
On Sun, Sep 2, 2012 at 4:05 AM, jit mukherjee <jitiitkgp.gmail.com> wrote:
> Dear Amber Users,
> I want to find out the global minimum of a system for which quantum
> mechanical methods will be very costly and nearly impossible. I want to use
> MD simulation for doing that. Is simulated annealing is a suitable way to
> find global minimum?
> Regards,
> jit
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Sep 02 2012 - 08:30:03 PDT
