Re: [petsc-users] changing solver options at runtime
TSSetOptionsPrefix() or SNESSetOptionsPrefix() or KSPSetOptionsPrefix(); you can call them at any time and then any call to TS/SNES/KSPSetFromOptions() on that object after that will use the command line options associated with that prefix. Check the manual pages. Barry On Feb 26, 2015, at 11:01 PM, Manav Bhatia bhatiama...@gmail.com wrote: Hi, Is there a way to change, between two consecutive linear solves, the command line options that petsc uses to initialize the solver (using xxxSetFromOptions)? I am attempting a multidisciplinary simulation, such that each discipline has its own linear system of equations to solve (perhaps repeatedly), and I wish to set separate options for each disciplinary solve. Passing the options at command line will set the same values for each discipline, which is what I wish to change. Of course, this can be done by writing code to set each option, but the convenience of doing it through command line options is very attractive. Any help will be greatly appreciated. Thanks, Manav
Re: [petsc-users] Memory leak in PetscRandom?
On 26 February 2015 at 01:09, Jed Brown j...@jedbrown.org wrote: Barry Smith bsm...@mcs.anl.gov writes: Thanks for the report. I ran the example with those options under Linux with valgrind and found no memory leaks, I cannot run valgrind on my Mac. When are you going to get a real operating system? In the mean time, several people have reported that this works: http://ranf.tl/2014/11/28/valgrind-on-mac-os-x-10-10-yosemite/ When is valgrind going to use a real VCS? ;-) -- Lisandro Dalcin Research Scientist Computer, Electrical and Mathematical Sciences Engineering (CEMSE) Numerical Porous Media Center (NumPor) King Abdullah University of Science and Technology (KAUST) http://numpor.kaust.edu.sa/ 4700 King Abdullah University of Science and Technology al-Khawarizmi Bldg (Bldg 1), Office # 4332 Thuwal 23955-6900, Kingdom of Saudi Arabia http://www.kaust.edu.sa Office Phone: +966 12 808-0459
Re: [petsc-users] Memory leak in PetscRandom?
Lisandro Dalcin dalc...@gmail.com writes: When is valgrind going to use a real VCS? ;-) Good question. signature.asc Description: PGP signature
[petsc-users] Partitioning in DMNetwork
Hi In DMNetwork we have edges and vertices and we can add an arbitrary number of variables to both edges and vertices. When the partitioning is carried out, is this done according to the number of edges and vertices or the variables? Say for example, that we assigned a very large number of variables to the first edge, a number much greater than the total number of edges or vertices. After the partitioning, the processor that contains that edge with many variables would have a very large portion of the global vector, wouldn't it? This case is hypothetical and something to avoid, but, could it happen? Would the partitioning be made in some other way to avoid this load unbalance? Thanks Miguel -- *Miguel Angel Salazar de Troya* Graduate Research Assistant Department of Mechanical Science and Engineering University of Illinois at Urbana-Champaign (217) 550-2360 salaz...@illinois.edu
Re: [petsc-users] Partitioning in DMNetwork
On Thu, Feb 26, 2015 at 11:33 AM, Miguel Angel Salazar de Troya salazardetr...@gmail.com wrote: Hi In DMNetwork we have edges and vertices and we can add an arbitrary number of variables to both edges and vertices. When the partitioning is carried out, is this done according to the number of edges and vertices or the variables? Say for example, that we assigned a very large number of variables to the first edge, a number much greater than the total number of edges or vertices. After the partitioning, the processor that contains that edge with many variables would have a very large portion of the global vector, wouldn't it? This case is hypothetical and something to avoid, but, could it happen? Would the partitioning be made in some other way to avoid this load unbalance? We can do weighted partitioning. We have not done it yet because in my experience the partitioners are quite fragile with respect to the weights and also the weighting has to be very coarse-grained. However, if you have something that really needs it, we can do it. Thanks, Matt Thanks Miguel -- *Miguel Angel Salazar de Troya* Graduate Research Assistant Department of Mechanical Science and Engineering University of Illinois at Urbana-Champaign (217) 550-2360 salaz...@illinois.edu -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener
[petsc-users] GMRES stability
Hi I tried to solve Ax=b with my own Gauss-Seidel code and Petsc's GMRES. With my GS, for a steady state problem I can set CFL=40 and for unsteady case can set dt=0.1. However, for GMRES I can't set CFL more than 5 and for unsteady case dt more than 0.1. I need GMRES for parallel computations so I cannot use GS for this purpose. Is there a way to improve the stability of GMRES?
[petsc-users] changing solver options at runtime
Hi, Is there a way to change, between two consecutive linear solves, the command line options that petsc uses to initialize the solver (using xxxSetFromOptions)? I am attempting a multidisciplinary simulation, such that each discipline has its own linear system of equations to solve (perhaps repeatedly), and I wish to set separate options for each disciplinary solve. Passing the options at command line will set the same values for each discipline, which is what I wish to change. Of course, this can be done by writing code to set each option, but the convenience of doing it through command line options is very attractive. Any help will be greatly appreciated. Thanks, Manav
Re: [petsc-users] GMRES stability
By stability I assume you mean the the GMRES does not converge (or converges too slowly)? The way to improve GMRES convergence is with a preconditioner better suited to your problem. By default PETSc uses GMRES with a block Jacobi preconditioner with one block per process and ILU(0) on each block. For some problems this is fine, but for many problems it will give bad convergence. What do you get for -ksp_view (are you using the default?) Are you running yet in parallel? As a test on one process you can use GS in PETSc as the preconditioner and make sure you get similar convergence to your code. For example -ksp_richardson -pc_type sor on one processor will give you a GS solver. Once we know a bit more about your problem we can suggest better preconditioners. Barry On Feb 26, 2015, at 10:25 PM, Orxan Shibliyev orxan.shi...@gmail.com wrote: Hi I tried to solve Ax=b with my own Gauss-Seidel code and Petsc's GMRES. With my GS, for a steady state problem I can set CFL=40 and for unsteady case can set dt=0.1. However, for GMRES I can't set CFL more than 5 and for unsteady case dt more than 0.1. I need GMRES for parallel computations so I cannot use GS for this purpose. Is there a way to improve the stability of GMRES?
