October 2022 - petsc-users - mailman3.cels.anl.gov

Does flagging a matrix as symmetric improving performances?
by Edoardo alinovi 21 Oct '22

21 Oct '22

Hi PETSc friends, As per object, do you think that flagging a matrix as symmetric might improve setup times of the preconditioner? Thank you as always.

2 3

An issue of Interior-Point Methods in TAO
by 赵刚 20 Oct '22

20 Oct '22

Dear PETSc/TAO team, I am using an interior-point method in TAO to solve a quadratic programming problem with bound constraints. I noticed that TAO includes three interior-point methods, they are Mehrotra Predictor-Corrector Method (bqpip), Primal-Dual Interior-Point Method (pdipm) and "ipm", respectively. I'd like to ask what is the interior-point method implemented by "-tao_type ipm", thank you. Best Regards, Gang

2 2

DMPlex adding boundary without PETSc discretization
by Gong Yujie 20 Oct '22

20 Oct '22

Dear development team, I'm trying to write a code to solve a partial differential equation. I didn't use PETSc's discretization. I have a question about the implementation of the boundary condition (Dirichlet boundary condition). When implementing the Dirichlet boundary condition. I need to set the corresponding rows in the Jacobian matrix diagonal 1 and others 0. I created the matrix use DMCreateMatrix. Can I add the Dirichlet boundary condition to the DMPlex mesh so that the Jacobian matrix won't contain these Dirichlet boundary rows? If yes, is there a function that can set the Dirichlet boundary points' value? Best Regards, Jerry

2 1

how to reuse Mumps factorization
by 袁煕 19 Oct '22

19 Oct '22

Hello, I am using Mumps to solve a problem with multiple time steps. The matrix structure does not change but its value may or may not change during those steps. That means I should reuse the symbolic factorization but recall numeric factorization when needed. I have found the following anwser of a similar question https://lists.mcs.anl.gov/pipermail/petsc-users/2020-August/041846.html which says "it automatically uses the same factorization", but I don't know if it includes numerical factorization also. My question is : 1. Does numeric factorization do automatically? If not 2. Could I control when numeric factorization should be done and how to do it? Much thanks YUAN

3 3

Some questions about KSP type and PC type selection
by Jackie Chan 19 Oct '22

19 Oct '22

Dear All， I hope you're having a nice day. In finite element problems, the stiffness matrix and load vector are constructed to calculate the displacement vector using DMCreateMatrix and DMCreateGlobalVector, respectively. For some reason, I need to make sure that the displacements of nodes on the opposite edges of two-dimensional structured grid domain satisfy relative displacement condition. For example, the displacements of two points with natural coordinates (x,Ymin) and (x,Ymax), i.e. the points are located on the upper and lower edges of 2D grid domain and have the same x-coordinates, are equal respectively. To achieve this, I need to add particularly big numbers to specific entries in stiffness matrix. In this way, the positions of big numbers are usually far away from each other and belong to different processes. I have tried many KSP types and PC settings to solve displacement vector. However, the final results are mostly incorrect. The best solver type I have tried is cg, but it still has some problems like excessive time consuming and convergence steps. So, for this problem, what kind of KSP type and PC type are suitable? Or is there a way to speed up the calculation process? Thanks, Jackie Chan

2 1

Question about Sequential & Parallel part
by 김성익 19 Oct '22

19 Oct '22

Dear PETSc users, I have a question about structure of programming. My blueprint of Finite Element programming with PETSc solver is below. [image: image.png] The blue box is whole loop for my FE program. There is a loop B that performs a convergence iteration (KU=F) inside the loop A in which the load increment increases. I want to proceed in rank=0 sequentially except for the kspsolve part that solves KU=F. How can I do this?? Best regards, Hyung Kim

2 1

Initializing a large sparse matrix
by fujisan 17 Oct '22

17 Oct '22

Hi everyone, I initialize a large sparse matrix (50000x20000) using MatCreate() and then filling it with MatSetValues() line by line but it takes a bit more than an hour on 80 cores to fill in the matrix. Is there a way to optimize this initialization? Fuji

2 3

Issue with single precision complex numbers in petsc4py
by Peng Sun 14 Oct '22

14 Oct '22

Dear PETSc community, I have a question regarding the single precision complex numbers of petsc4py. I configured PETSc with the “--with-scalar-type=complex --with-precision=single" option before compiling, but all the DA structures I created with petsc4py had double precision. Here is a minimum test code on Python 3.8/PETSc 3.12/petsc4py 3.12: both print commands show data type of complex128. Could anybody please help me? Thanks! import petsc4py import sys petsc4py.init(sys.argv) from petsc4py import PETSc da=PETSc.DA().create(sizes=[2,2,2],dof=1,stencil_type=0,stencil_width=1,boundary_type=1) da_1 = da.createGlobalVec() print(petsc4py.PETSc.ComplexType) print(da_1.getArray().dtype) Best regards, Peng Sun

5 14

compiling with Intel 2022.3
by Martin Diehl 14 Oct '22

14 Oct '22

FYI: Classic Intel compilers/MPI wrappers give a warning that confuses configure.py. The solution is to add "-diag-disable=10441" when invoking the compiler. source: https://community.intel.com/t5/Intel-C-Compiler/undefined-reference-for-202… -- KU Leuven Department of Computer Science Department of Materials Engineering Celestijnenlaan 200a 3001 Leuven, Belgium

2 1

Laplace Equation preconditioner
by Alfredo J Duarte Gomez 13 Oct '22

13 Oct '22

Good morning PETSC users, I have a current solver that requires the solution of a Laplace equation to be reused for all future time steps. The configuration is axisymmetric with Dirichlet BCs at the top and bottom boundaries, and Zero Neumman conditions at the axis and far field. The grid is curvilinear and structured. So far I have been using PCHYPRE boomeramg as the preconditioner, which often works well, but I have also experienced DIVERGED_BREAKDOWN on many occasions. When I use direct solver PCMUMPS it always produces a satisfactory answer, which gives me confidence that the solution exists for the given grid in all these configurations where boomeramg fails. I am looking for recommendations on other preconditioners to try in this problem that can produce a solution faster than PCMUMPS, or recommendations on which parameters to adjust on boomeramg. Thank you, -- Alfredo Duarte Graduate Research Assistant The University of Texas at Austin

3 3