- petsc-users - mailman3.cels.anl.gov

Re: [petsc-users] petsc-users Digest, Vol 205, Issue 2
by Michael Whitten 16 Jan '26

16 Jan '26

I hope that I am replying in the correct manner to respond to the "Having trouble with basic Fortran-PETSc interoperability" thread I previously started. If not, please correct me. Thank you everyone for your replies, they were very helpful. I have what appears to be a working example code using the suggested updates, specifically the VecGetArray() and VecRestoreArray() and getting the sequential vector back from PETSc using the information from the FAQ. <<Example code attached>> I have a question about this example code to make sure I am writing reasonably efficient code. It seems like I have to create an additional PETSc vector 'out_seq' which will essentially be a copy of the PETSc vector 'v1p' which is not the most efficient use of memory. It also seems to me like there isn't a way around this additional 'out_seq' vector because there needs a place to aggregate the data from the various processes. Is this a reasonable use of PETSc or is there a more efficient way? Note, I am trying to interface my existing code base with PETSc to use the solvers and this may be the performance trade-off for not developing my program fully within the PETSc ecosystem. I have another question only tangentially related to this topic. Should I ask it as part of this thread or create a new topic? Michael ________________________________ From: petsc-users <petsc-users-bounces(a)mcs.anl.gov> on behalf of petsc-users-request(a)mcs.anl.gov <petsc-users-request(a)mcs.anl.gov> Sent: Saturday, January 3, 2026 1:00 PM To: petsc-users(a)mcs.anl.gov <petsc-users(a)mcs.anl.gov> Subject: petsc-users Digest, Vol 205, Issue 2 External Email - Use Caution Send petsc-users mailing list submissions to petsc-users(a)mcs.anl.gov To subscribe or unsubscribe via the World Wide Web, visit https://urldefense.us/v3/__https://nam02.safelinks.protection.outlook.com/?… <https://lists.mcs.anl.gov/mailman/listinfo/petsc-users > or, via email, send a message with subject or body 'help' to petsc-users-request(a)mcs.anl.gov You can reach the person managing the list at petsc-users-owner(a)mcs.anl.gov When replying, please edit your Subject line so it is more specific than "Re: Contents of petsc-users digest..." Today's Topics: 1. Re: Having trouble with basic Fortran-PETSc interoperability (Barry Smith) 2. Re: Having trouble with basic Fortran-PETSc interoperability (Matthew Knepley) ---------------------------------------------------------------------- Message: 1 Date: Fri, 2 Jan 2026 16:33:27 -0500 From: Barry Smith <bsmith(a)petsc.dev> To: Michael Whitten <mwhitte6(a)jhu.edu> Cc: "petsc-users(a)mcs.anl.gov" <petsc-users(a)mcs.anl.gov> Subject: Re: [petsc-users] Having trouble with basic Fortran-PETSc interoperability Message-ID: <3B11AFD2-F767-4D6A-9CB6-D8D153BEA27E(a)petsc.dev> Content-Type: text/plain; charset="utf-8" VecGetValues() uses 0 based indexing in both Fortran and C. You don't want to use VecGetValues() and VecSetValues() usually since they result in two copies of the arrays and copying entire arrays back and forth. You can avoid copying between PETSc vectors and your arrays by using VecGetArray(), VecGetArrayWrite(), and VecGetArrayRead(). You can also use VecCreateMPIWithArray() to create a PETSc vector using your array; for example for input to the right hand side of a KSP. These arrays start their indexing with the Fortran default of 1. Barry > On Jan 2, 2026, at 2:42?PM, Michael Whitten via petsc-users <petsc-users(a)mcs.anl.gov> wrote: > > Hi PETSc mailing list users, > > I have managed to install PETSc and run some PETSc examples and little test codes of my own in Fortran. I am now trying to make PETSc work with my existing Fortran code. I have tried to build little test examples of the functionality that I can then incorporate into my larger code base. However, I am having trouble just passing vectors back and forth between PETSc and Fortran. > > I have attached a minimum semi-working example that can be compiled with the standard 'Makefile.user'. It throws an error when I try to copy the PETSc vector back to a Fortran vector using VecGetValues(). I get that it can only access values of the array on the local process but how do I fix this? Is this even the right approach? > > In the final implementation I want to be able to assemble my matrix and vector, convert them to PETSc data structures, use PETSc to solve, and then convert the solution vector back to Fortran and return. I want to be able to do this with both the linear and nonlinear solvers. It seems like this is what PETSc is, in part, built to do. Is this a reasonable expectation to achieve? Is this a reasonable use case for PETSc? > > Thanks in advance for any help you can offer. > > best, > Michael > <test.F90>

2 1

PETSc GPU matrix creation
by Alberto Cattaneo 14 Jan '26

14 Jan '26

Greetings I hope this email reaches you all well. I was wondering whether it was possible to create PETSc mat objects directly from data that exists on the GPU in AIJ format without copying? For example, either via DLPack or just an assurance that the pointer provided to a creation method is in the needed AIJ format? Ideally, I'd like to be able to build a PETSc AIJCUSPARSE object out of data created by another program. I know there are a few builder methods and paradigms, but I'm a bit confused as to which would be ideal in this circumstance since in some sense the matrix is already created in memory, just not as a PETSc object. Thank you for your assistance, please let me know if I should provide more information. Respectfully: Alberto Cattaneo

4 4

Question on Matrix Product with DMDA Matrix
by Micol Bassanini 14 Jan '26

14 Jan '26

Dear User Support Team, I have a question regarding my project. I would like to compute a matrix product between a DMDA matrix that is distributed across multiple ranks and a dense matrix that is available on only a single processor. What is the recommended or standard approach for implementing this kind of operation? Thank you in advance for your help. Best regards, Micol

2 1

Confusion about MatSetValuesStencil batching over x–dimension with DMDA
by malaboeuf＠cines.fr 13 Jan '26

13 Jan '26

Hi all, First, I have to apologize for I lack some experience on this subject, I have tried looking around to solve my issue without much success. I am assembling a 3D Laplacian matrix on a DMDA using MatSetValuesStencil. The following code, which inserts one row at a time, works as expected: for (PetscInt iz = info.zs; iz < info.zs + info.zm; ++iz) { for (PetscInt iy = info.ys; iy < info.ys + info.ym; ++iy) { for (PetscInt ix = info.xs; ix < info.xs + info.xm; ++ix) { const std::array<MatStencil, 1> rows{MatStencil{iz, iy, ix, 0}}; const std::array<MatStencil, 7> cols{{ {iz, iy, ix, 0}, {iz, iy, ix - 1, 0}, {iz, iy, ix + 1, 0}, {iz, iy - 1, ix, 0}, {iz, iy + 1, ix, 0}, {iz - 1, iy, ix, 0}, {iz + 1, iy, ix, 0} }}; CHKERRQ(MatSetValuesStencil(matrix_A, rows.size(), rows.data(), cols.size(), cols.data(), laplacian.data(), INSERT_VALUES)); } } } How would you batch the whole x dimension as a single MatSetValuesStencil call ? Each time I try, I endup getting errors such as: Inserting a new nonzero at global row/column (..., ...) into matrix Regards, -- Etienne Malaboeuf Ingénieur de recherche HPC, Département Calcul Intensif (DCI) Centre Informatique National de l'Enseignement Supérieur (CINES) 950 rue de Saint Priest, 34097 Montpellier tel : (334) 67 14 14 02 web : https://urldefense.us/v3/__https://www.cines.fr__;!!G_uCfscf7eWS!dEs8i8p3qA… | https://urldefense.us/v3/__https://dci.dci-gitlab.cines.fr/webextranet__;!!…

1 0

Last Call for User Presentations: PETSc Online BoF (Feb. 11, 2026)
by Junchao Zhang 13 Jan '26

13 Jan '26

Dear PETSc/TAO Community, Thank you to those who have already signed up to present their work at the upcoming PETSc Birds-of-a-Feather (BoF) session. We are still seeking additional user presentations. The PETSc BoF will be held online via Zoom on *February 11, 2026, from 11:00 AM to 12:30 PM U.S. Eastern Time*. Please note that this session will not be recorded. We are seeking approximately five user presentations, each consisting of a 5-minute talk followed by 2 minutes of questions. We currently have four and can still accept one or two presentations. If you are interested, please contact petsc-maint(a)mcs.anl.gov with your talk title and a brief abstract. The BoF is hosted by the Consortium for the Advancement of Scientific Software (CASS <https://urldefense.us/v3/__https://cass.community/__;!!G_uCfscf7eWS!ccFV4rP… >) and led by the PESO <https://urldefense.us/v3/__https://pesoproject.org/__;!!G_uCfscf7eWS!ccFV4r… > (Partnering for Scientific Software Ecosystem Stewardship) project. The CASS BoF events will take place from February 10–12, 2026. In addition to user presentations, during the PETSc BoF, PETSc developers will highlight recent advances developed following the Exascale Computing Project, including the new PETSc Fortran bindings, PetscRegressor, TaoTerm, updates to PETSc GPU backends, mixed-precision support in PETSc/MUMPS, and integration with OpenFOAM, among other topics. The program will also include an open discussion of emerging PETSc research directions, such as leveraging agentic artificial intelligence to enhance and exploit the PETSc knowledge base. The BoF will provide insight into PETSc’s near-term development roadmap and offer a forum for user feedback on desired features and improvements. Active participation and questions from the audience are strongly encouraged, enabling the PETSc team to better align future development with community needs. The agenda and Zoom link will be shared once the program is finalized. Thank you, and we look forward to your participation. Junchao Zhang On behalf of the PETSc Team

1 0

Some questions about modifying the linear system in SNES
by 陈培群 08 Jan '26

08 Jan '26

$\quad$ Hello, I use `SNESComputeJacobianDefault` to build a preconditioning matrix for JFNK and use `MatGetRowSumAbs` to obtain the vector for scaling. When I modify the preconditioning matrix in formJacobian[`SNESSetJacobian`] and modify rhs and the solution variables in preSolve[`KSPSetPreSolve`] and postSolve[`KSPSetPostSolve`] as scaling, I encounter the following questions. 1. If I need to scale the solution variables, do I need to call `SNESGetSolutionUpdate` to scale the increment vector instead of the solution vector in `postSolve(KSP /*ksp*/, Vec rhs, Vec x, void * ctx)`? 2. Whether this can achieve the same scaling effect as `-pc_jacobi_type rowl1`, so that the built-in preconditioner scheme of petsc can be applied on the basis of the above modified linear system? Or whether this will affect the residual used to construct the matrix through the finite difference and then lead to the wrong scaling effect?

3 4

Question regarding SNES error about locked vectors
by David Knezevic 06 Jan '26

06 Jan '26

Hi all, I have a question about this error: > Vector 'Vec_0x84000005_0' (argument #2) was locked for read-only access in > unknown_function() at unknown file:0 (line numbers only accurate to > function begin) I'm encountering this error in an FE solve where there is an error encountered during the residual/jacobian assembly, and what we normally do in that situation is shrink the load step and continue, starting from the "last converged solution". However, in this case I'm running on 32 processes, and 5 of the processes report the error above about a "locked vector". We clear the SNES object (via SNESDestroy) before we reset the solution to the "last converged solution", and then we make a new SNES object subsequently. But it seems to me that somehow the solution vector is still marked as "locked" on 5 of the processes when we modify the solution vector, which leads to the error above. I was wondering if someone could advise on what the best way to handle this would be? I thought one option could be to add an MPI barrier call prior to updating the solution vector to "last converged solution", to make sure that the SNES object is destroyed on all procs (and hence the locks cleared) before editing the solution vector, but I'm unsure if that would make a difference. Any help would be most appreciated! Thanks, David

3 17

VecNorm() of VECNEST returns 0 while sub-vectors have nonzero norms
by Michael Wick 06 Jan '26

06 Jan '26

Hello PETSc users, I have a VECNEST norm issue and would appreciate help. I create a nest vector: VecCreateNest(PETSC_COMM_WORLD, 2, NULL, subG, &G); But: VecNorm(G, NORM_2, &nG) returns 0 While the sub-vectors are clearly nonzero: VecNestGetSubVec(G,0,&v0); VecNorm(v0,NORM_2,&n0); // n0 > 0 VecNestGetSubVec(G,1,&v1); VecNorm(v1,NORM_2,&n1); // n1 > 0 Can someone give a hint on what might go wrong? Thanks, M

2 1

Having trouble with basic Fortran-PETSc interoperability
by Michael Whitten 02 Jan '26

02 Jan '26

Hi PETSc mailing list users, I have managed to install PETSc and run some PETSc examples and little test codes of my own in Fortran. I am now trying to make PETSc work with my existing Fortran code. I have tried to build little test examples of the functionality that I can then incorporate into my larger code base. However, I am having trouble just passing vectors back and forth between PETSc and Fortran. I have attached a minimum semi-working example that can be compiled with the standard 'Makefile.user'. It throws an error when I try to copy the PETSc vector back to a Fortran vector using VecGetValues(). I get that it can only access values of the array on the local process but how do I fix this? Is this even the right approach? In the final implementation I want to be able to assemble my matrix and vector, convert them to PETSc data structures, use PETSc to solve, and then convert the solution vector back to Fortran and return. I want to be able to do this with both the linear and nonlinear solvers. It seems like this is what PETSc is, in part, built to do. Is this a reasonable expectation to achieve? Is this a reasonable use case for PETSc? Thanks in advance for any help you can offer. best, Michael

4 3

A partition of DMPlex mesh similar to what DMDA provides?
by liluo 22 Dec '25

22 Dec '25

Dear PETSc developers, I’m using DMPlex to manage an unstructured mesh. However, in my case, the input mesh is actually a structured tetrahedral mesh, and its geometric domain is just a simple box. Is there any PETSc functionality or recommended approach to obtain a partition similar to what DMDA provides—i.e., a simple Cartesian block partition—when working with such a mesh in DMPlex? Any guidance or best practices would be greatly appreciated. Thank you! Bests, Li Luo <https://urldefense.us/v3/__https://www.fst.um.edu.mo/personal/liluo/__;!!G_… >

2 2