Take the 2-minute tour ×
Computational Science Stack Exchange is a question and answer site for scientists using computers to solve scientific problems. It's 100% free, no registration required.

The use of DDM in FEA makes parallel solution of the whole analysis e.g. assembly, solver etc possible. DDM splits the model in domains and runs them in parallel. Since there are interconnected nodes between domains, Will DDM give unreliable result? Also, is there any other factor that can affect the correct solution result?

The application is in structural dynamics FEA and the platform is SMP (single machine with 2 processing cores).

share|improve this question

1 Answer 1

up vote 4 down vote accepted

Domain decomposition methods do not discard coupling, they just isolate it to enable parallel computation. For explicit dynamic simulations, this just amounts to exchanging ghost values, and the (distributed) solution is identical to a serial implementation. For implicit problems, DD is an iterative method and the convergence rate will depend on the decomposition (number and shapes of subdomains) for any particular method. Methods in which the rate is robust to the number of subdomains must have more than one level (they need a "coarse grid" to produce global communication on each iteration). Continuity between subdomains will be enforced either by definition of the state space (a non-overlapping vertex partition) or by Lagrange multipliers (the FETI family).

share|improve this answer
    
I am doing implicit dynamic structural simulation on a single machine (2 cores) with DDM. The FE solution result differs from the serial run by a big margin. So does it mean that I have to try and find optimum number of sub-domains by trying on different SMP setup? –  M. G. Rashed Dec 28 '12 at 19:52
    
It means either (a) your implementation is incorrect or (b) you are not using it correctly (e.g., solver tolerances). DD has a rigorous theoretical foundation and at any moment in time, occupies millions of cores in both science and industry. Your problem is fixable. –  Jed Brown Dec 28 '12 at 20:07
1  
You'll have to be a lot more specific. You're welcome to ask more specific questions on this site, but "my DD code doesn't work, can you guess what's wrong (without seeing it)?" certainly won't fly. You may want to start with these general guidelines for iterative linear solver convergence. –  Jed Brown Dec 31 '12 at 0:26

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.