- Contemporary messages sorted: [ by date ] [ by thread ] [ by subject ] [ by author ] [ by messages with attachments ]

From: Jason Swails <jason.swails.gmail.com>

Date: Wed, 14 Jul 2010 06:56:34 -0400

Hello,

This is because there are a lot more forces involved with a larger molecule,

so there are a lot more computations that have to be made per time step. If

you're simulating with a time step of 0.002 ps (2 femtoseconds) then a 2 ns

simulation will require 1e6 steps.

MD simulations scale as n^2 formally, due to the pairwise force terms.

Larger systems have more pairs between which forces have to be evaluated.

For clarity, let's consider 2 systems, one of which has 10 atoms and the

other of which has 100 atoms. In the 10 atom system, there are 10*10=100

total pairs of atoms between which the electrostatic and van der waals

forces/energies need to be evaluated. For the 100 atom system, however,

there is 100*100=10 000 total pairs. It will clearly take the CPU longer to

compute 10 000 quantities rather than just 100. There are tricks played so

that not all pairs are computed (a distance cutoff), but the general trend

remains the same (using a cutoff leads to more linear scaling than quadratic

scaling). Another minor effect is the fact that there are typically more

bonds, angles, and dihedral terms to evaluate in a 100 atom system than a 10

atom system, though the majority of the time spent by a machine in force

evaluations is done on the non-bonded terms mentioned above (since there are

far more of those than any other term in typically sized systems).

Hope this helps,

Jason

On Wed, Jul 14, 2010 at 3:40 AM, <monica.imtech.res.in> wrote:

*> hi all
*

*>
*

*> my question may seem stupid but i want to know wat is the reason that a
*

*> small
*

*> peptide (say) simulations complete earlier and the protein takes longer
*

*> time
*

*> although we run both for same 2 ns (say).
*

*> since 2ns is same time for both . i just want to know how system applies
*

*> force
*

*> after all its calculations with variable time for the above cases. means
*

*> 2ns is
*

*> 2ns only then why some hours for the first and some days for the second
*

*> one.
*

*>
*

*> i hope i made myself clear and hope for the reply.
*

*>
*

*> thanx
*

*> monica
*

*>
*

*> _______________________________________________
*

*> AMBER mailing list
*

*> AMBER.ambermd.org
*

*> http://lists.ambermd.org/mailman/listinfo/amber
*

*>
*

Date: Wed, 14 Jul 2010 06:56:34 -0400

Hello,

This is because there are a lot more forces involved with a larger molecule,

so there are a lot more computations that have to be made per time step. If

you're simulating with a time step of 0.002 ps (2 femtoseconds) then a 2 ns

simulation will require 1e6 steps.

MD simulations scale as n^2 formally, due to the pairwise force terms.

Larger systems have more pairs between which forces have to be evaluated.

For clarity, let's consider 2 systems, one of which has 10 atoms and the

other of which has 100 atoms. In the 10 atom system, there are 10*10=100

total pairs of atoms between which the electrostatic and van der waals

forces/energies need to be evaluated. For the 100 atom system, however,

there is 100*100=10 000 total pairs. It will clearly take the CPU longer to

compute 10 000 quantities rather than just 100. There are tricks played so

that not all pairs are computed (a distance cutoff), but the general trend

remains the same (using a cutoff leads to more linear scaling than quadratic

scaling). Another minor effect is the fact that there are typically more

bonds, angles, and dihedral terms to evaluate in a 100 atom system than a 10

atom system, though the majority of the time spent by a machine in force

evaluations is done on the non-bonded terms mentioned above (since there are

far more of those than any other term in typically sized systems).

Hope this helps,

Jason

On Wed, Jul 14, 2010 at 3:40 AM, <monica.imtech.res.in> wrote:

-- Jason M. Swails Quantum Theory Project, University of Florida Ph.D. Graduate Student 352-392-4032 _______________________________________________ AMBER mailing list AMBER.ambermd.org http://lists.ambermd.org/mailman/listinfo/amberReceived on Wed Jul 14 2010 - 04:00:04 PDT

Custom Search