[CPMD-list] Large-scale DFT calculations?

[Alessandro Curioni]

Hello.

I  really think that both the question and  the answer are not clear enough
and deserves some comments.

First, the answer depends a lot on what you want to  simulate and which
properties you want to extract.
CPMD is a very well parallelized plane waves based code   that allow to
make from  simple geometry optimization
to  ab-initio path integral MD calculation, on isolated molecules, liquid
and solids  at the DFT(pseudo) accuracy without any  further approximation;
you can study from insulating materials to metallic surfaces thank to the
free energy functional implementation, at the same controlled accuracy.

On a updated wks or a Linux cluster can you handle these days in a
routinely way solid state systems (with no vacuum) of some hundred of atoms
(300-400)
as you can easily see from the publications of most the experienced CPMD
users. Please check the  cpmd web site.

 On  100 GFlops machines with ~100GByte of RAM  you can easily simulate
~1000   atoms (geometry optimization) and
on an updated server  as a 32 processor  pwr4  a system with 256 water in
the liquid phase  and ab-initio MD are  routine calculations  these days.

On the other hands,  on a powerful supercomputer (power4 based) we have
recently demonstrated good scalability up to 880 processors and  ~3000 of
atoms.


Moreover, in the next (very soon) release of CPMD  a linear scaling (and
robust) geometry optimizer will be included,  which  will solve most of the
instabilities
and scaling problems of the already implemented optimizers and which will
be able also to locate (in a linear scaling fashion)  classical transition
states.

Regards,



Alessandro CURIONI, PhD
Research Staff Member
Computational Biochemistry and Material Science group
IBM Research Division - Zurich Research Laboratory
Saumerstrasse 4
8003 Rueschlikon - Switzerland
e-mail: cur at zurich.ibm.com
www:    www.zurich.ibm.com
Tel: +41-1-7248633
Fax: +41-1-7248958



|---------+----------------------------+----------------------------------------------------------------------------------------------->
|         |           HW Sheng         |                                                                                               |
|         |           <hwsheng at jhu.edu>|       To:       cpmd-list at cpmd.org                                                            |
|         |           Sent by:         |       cc:                                                                                     |
|         |           cpmd-list-admin at c|       Subject:  Re: [CPMD-list] Large-scale DFT calculations?                                 |
|         |           pmd.org          |                                                                                               |
|         |                            |                                                                                               |
|         |                            |                                                                                               |
|         |           11/26/2002 06:42 |                                                                                               |
|         |           PM               |                                                                                               |
|---------+----------------------------+----------------------------------------------------------------------------------------------->
  >------------------------|
  |                        |
  >------------------------|




Hi, Ichiro, I think order-N method is desired. But cpmd does not has such
functionality (order-N algorithm) for now, i guess. Prof Hutter et al has
developed the GAPW method by making use of the locality principles,
unfortunately, it's not implemented in CPMD.

There are other order-N methods available such as SIESTA and CONQUEST.

Howard

----- Original Message -----
From: "Ichiro Nagano" <nagano at atrc.mhi.co.jp>
To: <cpmd-list at cpmd.org>
Sent: Monday, November 25, 2002 8:50 PM
Subject: [CPMD-list] Large-scale DFT calculations?


> Dear CPMD users,
>
> I would like to try more large-scale crystals (100-300 atoms in a unit
> cell).
> If it is possible cost effectively, could you give me an advice?
>
> Alternatively, should I use order-N method?
>
> Thanks for any help,
>
> Ichiro Nagano
>
> ********************************************************************
> Advanced Technology Research Center, Mitsubishi Heavy Industries, Ltd.
> 8-1, Sachiura, 1-Chome, Kanazawa-ku, Yokohama, 236-8515, Japan
> Ichiro Nagano
> E-mail: <nagano at atrc.mhi.co.jp>
> TEL: + 81 (45) 771-1222
> FAX: + 81 (45) 771-3879
>
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://www.cpmd.org/mailman/listinfo/cpmd-list
>

_______________________________________________
CPMD-list mailing list
CPMD-list at cpmd.org
http://www.cpmd.org/mailman/listinfo/cpmd-list