[lammps-users] Coupling lammps with finite element analysis?

Hi, on lammps website I saw that a group at Sandia is working at coupling LAMMPS to a continuum finite-element (FE) solver for doing stress-strain models of solids where a portion of the domain is treated atomistically and the remained by FE. As far as I know, right now lammps is atomistic-scale simulation but FE is based on continuum scale. For instance, the concepts of stress and strain in continuum mechanics are linked by the stiffness tensor of the material, however, even the definitions of stress/strain of atomistic level are not quite clear yet, am I right? I’m really curious how those genius will make it, although I don’t know why they divide the domain with part of atomistic portion and rest of FE portion.

Anyone could suggest some reference materials(paper, book…) on coupling MD simulations with FE analysis? It would be appealing to be able to bridge atomistic scale to continuum scale model.

Thanks first,

Wei

Wei:

We did it back in 1997. See e.g. our paper: NIMB 121 (1997) 44.
I can send you a copy.

Z. Insepov

Wei YE wrote:

I don't know if he's a genius, but the person to talk to about this
effort is Reese Jones (rjones at sandia.gov). They have a working,
finished version of their coupling package, which will become a "package"
in LAMMPS, invoked thru a fix. We hope to release this in the
next couple weeks. I believe the FE portion is solved in serial,
i.e. every processor owns the entire FE grid, but Reese can give
you details.

Steve

Thank you, Steve. In fact, I’m more interested in how they come up with the idea of bridging up such multi-scale method. At first, I just can’t figure out the whole framework.

Hope this feature will be released soon.

Wei

As I'm part of the ATC (Atomistic-to-Continuum) package team, I can provide
some information. The package is designed to implement a coupled FE/MD
simulation and/or on-the-fly estimation of continuum fields. A good
reference for the former application is the 2008 Comput. Methods Appl. Mech.
Engrg. article by Wagner, Jones, Templeton and Parks. A good reference for
the latter application is the 2008 Math. Mech. Solids article by Webb,
Zimmerman and Seel.

More to come once it gets the seal of approval.

Jon Zimmerman

From: Steve Plimpton <[email protected]>
Date: Wed, 26 Aug 2009 11:26:48 -0600
To: Wei YE <[email protected]>
Cc: <[email protected]>
Subject: Re: [lammps-users] Coupling lammps with finite element analysis?

I don't know if he's a genius, but the person to talk to about this
effort is Reese Jones (rjones at sandia.gov). They have a working,
finished version of their coupling package, which will become a "package"
in LAMMPS, invoked thru a fix. We hope to release this in the
next couple weeks. I believe the FE portion is solved in serial,
i.e. every processor owns the entire FE grid, but Reese can give
you details.

Steve

Hi, on lammps website I saw that a group at Sandia is working at coupling
LAMMPS to a continuum finite-element (FE) solver for doing stress-strain
models of solids where a portion of the domain is treated atomistically and
the remained by FE. As far as I know, right now lammps is atomistic-scale
simulation but FE is based on continuum scale. For instance, the concepts of
stress and strain in continuum mechanics are linked by the stiffness tensor
of the material, however, even the definitions of stress/strain of atomistic
level are not quite clear yet, am I right? I'm really curious how those
genius will make it, although I don't know why they divide the domain with
part of atomistic portion and rest of FE portion.

Anyone could suggest some reference materials(paper, book...) on coupling MD
simulations with FE analysis? It would be appealing to be able to bridge
atomistic scale to continuum scale model.

Thanks first,

Wei

----------------------------------------------------------------------------->>

Thanks first for that, although I can’t find the second paper(probably that’s one reason I need to work from a searcher to be a Researcher). Right now I have no idea about the first type of implementation, but the second implementation seems to be some kind of post-processing of MD result to get continuum property(eg: effective stiffness of material). I had read a bunch of papers on turning MD results into continuum properties, although they are not fulfilled in terms of FE.

Wei