AMBER Archive (2009)

Subject: Re: [AMBER] Ewald error estimated

From: Robert Duke (
Date: Fri Sep 25 2009 - 10:50:18 CDT

Well, I would say instead that error estimates in the range of 0.1E-03 to
0.1E-05 are to be expected. This is an estimate based on a known
relationship between the exact electrostatic energy and the virial trace
(see Essman, Perera, Berkowitz, Darden, Lee and Pedersen, JCP 103, 8577
(1995)). This value is basically sensitive to dsum_tol (a measure of what
you lose in the "tail" of the erf function in direct space) and the grid
density and interpolation order of the fft grids in reciprocal space. You
will see this number vary for a given simulation depending on the steepness
of the electrostatic force gradients - basically pme will not capture the
electrostatics as accurately if you have really steep gradients caused by
hotspots in your simulation. But numbers in the range above are no cause
for concern, and actually trying to measure electrostatics to a higher
degree of precision than you get with the default ewald parameters is
typically a waste of time - you are getting higher accuracy electrostatics
probably than anything else (this is a little bit of a lie, in that by
tweaking dsum_tol you can get even better accuracy for no cost, and I think
that we will change the default in a future version - but it is not a really
big deal, and I don't recommend that users dink with the default ewald
parameters as a general rule - you are more apt to shoot yourself in the
foot than anything...).
Regards - Bob Duke
Oh, by the way - what I noted about the setup here is that you are using a
0.2 fsec timestep (dt=0.0002); this is probably on the order of 5x to 10x
shorter than required if you are using shake, and will make your simulations
5x to 10x slower (and thus more computationally expensive) than they need

----- Original Message -----
From: "case" <>
To: "AMBER Mailing List" <>
Sent: Friday, September 25, 2009 11:15 AM
Subject: Re: [AMBER] Ewald error estimated

> On Fri, Sep 25, 2009, xuemeiwang1103 wrote:
>> I want to konw the meaning of "Ewald error estimate: 0.1017E-03"
> This is not a error in the usual sense. It is an estimate of the
> uncertainties arising from finite grid resolution in the PME calculation.
> You can ignore these messages.
>> My
>> &cntrl
>> imin=0,irest=0,ntx=1,
>> nstlim=10000,dt=0.0002,
>> ntc=2,ntf=2,
>> ntt=0,
> Since you set ntt=0, you don't have any temperature regulation!
> _______________________________________________
> AMBER mailing list

AMBER mailing list