> On Nov 15, 2017, at 9:57 PM, Mark Lohry wrote:
>
> What are the limitations of ILU in parallel you're referring to? Does
> Schwarz+local ILU typically fare better?
If ILU works fine for scalably in parallel that is great. Most of the PETSc
team has an explicit bias
What are the limitations of ILU in parallel you're referring to? Does
Schwarz+local ILU typically fare better?
On Nov 15, 2017 10:50 PM, "Smith, Barry F." wrote:
>
>
> > On Nov 15, 2017, at 9:40 PM, Jed Brown wrote:
> >
> > "Smith, Barry F."
> On Nov 15, 2017, at 9:40 PM, Jed Brown wrote:
>
> "Smith, Barry F." writes:
>
>>> On Nov 15, 2017, at 6:38 AM, Mark Lohry wrote:
>>>
>>> I've found ILU(0) or (1) to be working well for my problem, but the petsc
>>> implementation
"Smith, Barry F." writes:
>> On Nov 15, 2017, at 6:38 AM, Mark Lohry wrote:
>>
>> I've found ILU(0) or (1) to be working well for my problem, but the petsc
>> implementation is serial only. Running with -pc_type hypre -pc_hypre_type
>> pilut with default
I've debugged into the ISGlobalToLocalMappingApplyBlock() function and
it seems to me the bounds checking in there is not correct when the
blocksize is > 1.
It checks against the same bounds, scaled up by the blocksize, in both
the block and non-block versions of the function. I think for the
Thanks, Barry,
On Wed, Nov 15, 2017 at 4:04 PM, Smith, Barry F. wrote:
>
> Do the ASM runs for thousands of time-steps produce the same final
> "physical results" as the MUMPS run for thousands of timesteps? While with
> SuperLU you get a very different "physical
Do the ASM runs for thousands of time-steps produce the same final "physical
results" as the MUMPS run for thousands of timesteps? While with SuperLU you
get a very different "physical results"?
Barry
> On Nov 15, 2017, at 4:52 PM, Kong, Fande wrote:
>
>
>
> On
On Wed, Nov 15, 2017 at 3:35 PM, Smith, Barry F. wrote:
>
> Since the convergence labeled linear does not converge to 14 digits in
> one iteration I am assuming you are using lagged preconditioning and or
> lagged Jacobian?
>
We are using Jacobian-free Newton. So Jacobian
I actually attached the wrong test program last time- I've attached the
right one here, which is much simpler. It test global indices 0, 1, ... 9.
If I run on 2 processes, the local indices it returns are:
rank 0: 0, 1, 2, 3, 4, 0, 0, 0, -253701943, 0
rank 1: -1, -1, -1, -1, -1, -1, -1, -1,
There isn't an IDR in PETSc, but there is BCGSL which usually performs
similarly. Contributions welcome.
Evan Um writes:
> Dear PETSC users,
>
> I was wondering if anyone already tried/developed an induced dimension
> reduction (IDR) solver for PETSC? I think that it is a
Since the convergence labeled linear does not converge to 14 digits in one
iteration I am assuming you are using lagged preconditioning and or lagged
Jacobian?
What happens if you do no lagging and solve each linear solve with a new LU
factorization?
Barry
> On Nov 15, 2017, at
On Wed, Nov 15, 2017 at 2:52 PM, Smith, Barry F. wrote:
>
>
> > On Nov 15, 2017, at 3:36 PM, Kong, Fande wrote:
> >
> > Hi Barry,
> >
> > Thanks for your reply. I was wondering why this happens only when we use
> superlu_dist. I am trying to understand
To be clear: these differences completely go away with MUMPS?
Can you valgrind this? We have seen some valgrind warning from MUMPS from
BLAS routines. It could be that your BLAS is buggy (and SuperLU uses some
BLAS routines that MUMPS does not). I think SuperLU does more/different
pivoting than
hi Dave,
On 15/11/17 21:34, Dave May wrote:
Or am I wrong to expect this to give the same results regardless of
blocksize?
Yep.
Maybe I am not using this function correctly then.
The man page says it "Provides the local block numbering for a list of
integers specified with a
On Wed, Nov 15, 2017 at 4:36 PM, Kong, Fande wrote:
> Hi Barry,
>
> Thanks for your reply. I was wondering why this happens only when we use
> superlu_dist. I am trying to understand the algorithm in superlu_dist. If
> we use ASM or MUMPS, we do not produce these differences.
> On Nov 15, 2017, at 3:36 PM, Kong, Fande wrote:
>
> Hi Barry,
>
> Thanks for your reply. I was wondering why this happens only when we use
> superlu_dist. I am trying to understand the algorithm in superlu_dist. If we
> use ASM or MUMPS, we do not produce these
Hi Barry,
Thanks for your reply. I was wondering why this happens only when we use
superlu_dist. I am trying to understand the algorithm in superlu_dist. If
we use ASM or MUMPS, we do not produce these differences.
The differences actually are NOT meaningless. In fact, we have a real
transient
Meaningless differences
> On Nov 15, 2017, at 2:26 PM, Kong, Fande wrote:
>
> Hi,
>
> There is a heat conduction problem. When superlu_dist is used as a
> preconditioner, we have random results from different runs. Is there a random
> algorithm in superlu_dist? If we
Hi,
There is a heat conduction problem. When superlu_dist is used as a
preconditioner, we have random results from different runs. Is there a
random algorithm in superlu_dist? If we use ASM or MUMPS as the
preconditioner, we then don't have this issue.
run 1:
0 Nonlinear |R| = 9.447423e+03
Dear PETSC users,
I was wondering if anyone already tried/developed an induced dimension
reduction (IDR) solver for PETSC? I think that it is a useful one but I
couldn't find its example with PETSC. If you have any idea about IDR
routines for PETSC, please let me know. Thanks!
Best,
Evan
>
>
> > Partially unrelated, PC block-jacobi fails with MFFD type not supported,
> but additive schwarz with 0 overlap, which I think is identical, works
> fine. Is this a bug?
>
>Huh, is this related to hypre, or plan PETSc? Please send all
> information, command line options etc that
I've found ILU(0) or (1) to be working well for my problem, but the petsc
implementation is serial only. Running with -pc_type hypre -pc_hypre_type
pilut with default settings has considerably worse convergence. I've tried
using -pc_hypre_pilut_factorrowsize (number of actual elements in row) to
On Wed, Nov 15, 2017 at 6:09 AM, Matteo Semplice
wrote:
> On 15/11/2017 11:39, Matthew Knepley wrote:
>
> On Wed, Nov 15, 2017 at 3:11 AM, Matteo Semplice > wrote:
>
>> Hi.
>>
>> I am struggling with indices into matrices associated to a
On 15/11/2017 11:39, Matthew Knepley wrote:
On Wed, Nov 15, 2017 at 3:11 AM, Matteo Semplice
> wrote:
Hi.
I am struggling with indices into matrices associated to a DMPLex
mesh. I can explain my problem to the following
On Wed, Nov 15, 2017 at 3:11 AM, Matteo Semplice
wrote:
> Hi.
>
> I am struggling with indices into matrices associated to a DMPLex mesh. I
> can explain my problem to the following minimal example.
>
> Let's say I want to assemble the matrix to solve an equation (say
On Wed, 15 Nov 2017 at 05:55, Adrian Croucher
wrote:
> hi
>
> I'm trying to use ISGlobalToLocalMappingApplyBlock() and am a bit
> puzzled about the results it's giving.
>
> I've attached a small test to illustrate. It just sets up a
> local-to-global mapping with 10
Hi.
I am struggling with indices into matrices associated to a DMPLex mesh.
I can explain my problem to the following minimal example.
Let's say I want to assemble the matrix to solve an equation (say
Laplace) with data attached to cells and the finite volume method. In
principle I
- loop
27 matches
Mail list logo