Actually, all of the input files provided in the APBS examples/ directory demonstrate this in a variety of ways. Some of the examples remove self-energies by calculating solvation energies while other examples remove self-energies by using exactly the same grid positions for all atoms in each part of the calculation.

Hope this helps,

Nathan

--

Nathan Baker (http://bakergroup.wustl.edu)

Associate Professor, Dept. of Biochemistry and Molecular Biophysics

Director, Computational and Molecular Biophysics Graduate Program

Center for Computational Biology, Washington University in St. Louis

On Fri, Dec 11, 2009 at 9:23 AM, Gatti, Domenico <dgatti@med.wayne.edu> wrote:

Hi All,

Could we post an EXAMPLE SCRIPT of how the corrections suggested by

Gernot and Nathan for Dziedzic's simple calculation of two point charges

should be implemented? Does this mean that the grid artifact subtraction

should be carried out always in all the APBS calculations? Perhaps, this

occurs already by default, and I did not realize it.

Best,

Domenico

Domenico Gatti

Biochemistry & Mol. Biology

Wayne State University School of Medicine

540 E. Canfield Avenue

Detroit, MI 48201

Tel: 313-577-0620 or 313-993-4238

Fax: 313-577-2765

dgatti@med.wayne.edu

Hi All --

Gernot is absolutely correct. I would also add that, after correcting the

issues Gernot raised below, you should also examine the sensitivity of your

results on "chgm spl0" vs. "chgm spl2" since charge discretization can

affect these types of calculations as well.

Thanks,

Nathan

On Dec 10, 2009, at 9:31 AM, Gernot Kieseritzky wrote:

> Hi!

>

> On Wed, 2009-12-09 at 17:49 +0000, J.Dziedzic wrote:

>> Hi!

>>

>> I am confused about the result of a very simple calculation when done

>> with APBS. Consider a trivial system of two point charges with unit

>> charges at a separation of 1 Bohr length, in vacuum. The Coulombic

>> energy of this system is exactly 1 Hartree, that is 2625.5 kJ/mole.

>> ...

>> ... APBS yields 31740 kJ/mole, which off by a factor of 12. Making the

>> grid finer only makes things worse, the results being:

>>

>> dime Energy (kJ/mole)

>> 65 1.209E04

>> 129 2.222E04

>> 193 3.174E04

>> 257 4.058E04

>> 289 4.539E04

>>

>> which are all way off, by a factor of 5-20, from the correct value of

>> 2625.5 kJ/mole.

>>

>> I understand that normally one is interested in energy differences

>> between a system in vacuo and a solvated system, and any discretization

>> errors introduced are canceled if the grid is the same in both

>> calculations. Yet, with a system so trivial, without any dielectric

>> at all and, thus, without the arbitrariness of the cavity, what is the

>> underlying reason for the calculation being so off from the mark?

>

> Two words: grid artefact! Basically, the deviation is not due to

> numerical problems, rather the high energy values you observe are the

> result of the self-interaction of the grid points. That's why the

> deviation is increasing with higher resolution as the grid points are

> getting closer. What you have to do to get the total electrostatic

> energy of your system without self-energies:

>

> 1) Compute the Coulomb energy in the homogeneous continuum.

>

> 2) Compute the solvation energy of the same charge distribution using

> APBS. The grid artefact cancels as you calculate an energy difference

> here. This, of course, requires that you use the same grid setup in both

> APBS runs.

>

> 3) Add the values together.

>

> Best regards,

> Gernot Kieseritzky

------------------------------------------------------------------------------

Return on Information:

Google Enterprise Search pays you back

Get the facts.

http://p.sf.net/sfu/google-dev2dev

_______________________________________________

apbs-users mailing list

apbs-users@lists.sourceforge.net

https://lists.sourceforge.net/lists/listinfo/apbs-users

--

Nathan Baker (http://bakergroup.wustl.edu)

Associate Professor, Dept. of Biochemistry and Molecular Biophysics

Director, Computational and Molecular Biophysics Graduate Program

Center for Computational Biology, Washington University in St. Louis