How to view the PETScMatrix

General
1

Hello everyone!
I’m working with ubuntu 18.04 and Fenics to solve the 2D Possion equation, and the code is as follows:
import dolfin as df
import matplotlib.pyplot as plt
import numpy as np

Create mesh and define function space

mesh = df.UnitSquareMesh(6, 4)
V = df.FunctionSpace(mesh, “Lagrange”, 1)

Define boundary conditions

u0 = df.Expression(“1 + x[0]x[0] + 2x[1]*x[1]”, degree =2)
def u0_boundary(x, on_boundary):
return on_boundary
bc = df.DirichletBC(V, u0, u0_boundary)

Define variational problem

u = df.TrialFunction(V)
v = df.TestFunction(V)
f = df.Constant(-6.0)
a = df.inner(df.nabla_grad(u), df.nabla_grad(v))* df.dx
L = fvdf.dx

A = df.PETScMatrix()
b = df.PETScVector()
A = df.assemble(a)
b = df.assemble(L)
bc.apply(A, b)

u =df. Function(V)
U = u.vector()
#df.solve(A, U, b, ‘cg’, ‘ilu’)

solver = df.KrylovSolver(“cg”, “ilu”)
solver.parameters[“absolute_tolerance”] = 1e-17
solver.parameters[“relative_tolerance”] = 1e-4
solver.parameters[“maximum_iterations”] = 1000
solver.parameters[“error_on_nonconvergence”] = True
solver.parameters[“report”] = True

solver.solve(A, U, b)

Compute solution

file = df.File(“poisson-1.pvd”)
file << u

and now I want to output the PETScMatrix A in the form like the figure, but when I tried the function MatView in PETSc library, but it seems the function isn’t existing. Can anyone tell me how could I do to achieve the goal?
image

2

You need to call A.mat() to get the petsc4py matrix object with its functionality.

Please note that for further questions, please use markdown format to encapsulate your code i.e.

```python
import dolfin as df

def some_function(x):
    return x
# Replace this with code
# ...

```
3

Hello! Based on the given advice, I tried:

from petsc4py import PETSc
A.mat().MatView()

but it reminded me that AttributeError: ‘dolfin.cpp.la.Matrix’ object has no attribute ‘mat’, then what should I do for this condition

4

Do as_backend(A) first.

When you get error messages please supply a minimal example that reproduces the error. Without a minimal example it becomes educated guessing from other people what’s gone wrong.

5

The code is changed as:

from petsc4py import PETSc
PetSc.MatView(A.mat())

After running the code above, it reminded me that "AttritbuteError: module ‘petsc4py.PETSc’ has no attribute ‘MatView’ ". The output error is the same when using:

import petsc4py
petsc4py.MatView(A.mat())

or

import dolfin as df
df.MatView(A.mat())
6

Again, please make a minimal reproducible example.
In the snippets above, you have not defined A, so I cannot help you reproducing your issue.

MatView is a function from PETSc (petsc4py,src/binding/petsc4py/src/PETSc/Mat.pyx · main · PETSc / petsc · GitLab), working in a PETSc matrix. Thus you would need something like:

from dolfin import PETScMatrix, as_backend_type
A = PETScMatrix()
A_petsc = as_backend_type(A).mat()
A_petsc.assemble()
A_petsc.view()
7

Hello! I retried the process based on your suggestion, and the code is as follows:

import dolfin as df
import matplotlib.pyplot as plt
import numpy as np
from dolfin import PETScMatrix, as_backend_type

mesh = df.UnitSquareMesh(6, 4)
V = df.FunctionSpace(mesh, "Lagrange", 1)

# Define boundary conditions
u0 = df.Expression("1 + x[0]*x[0] + 2*x[1]*x[1]", degree =2)
def u0_boundary(x, on_boundary):
    return on_boundary
bc = df.DirichletBC(V, u0, u0_boundary)

# Define variational problem
u = df.TrialFunction(V)
v = df.TestFunction(V)
f = df.Constant(-6.0)
a = df.inner(df.nabla_grad(u), df.nabla_grad(v))* df.dx
L = f*v*df.dx

A = PETScMatrix()
A_petsc = as_backend_type(A.mat())
A_petsc.assemble(a)
A_petsc.view()

While it only output an error as follows, instead in the form of the figure in the below.

TypeError                                 Traceback (most recent call last)
/tmp/ipykernel_27380/3677198767.py in <cell line: 3>()
      1 A = PETScMatrix()
      2 A_petsc = as_backend_type(A.mat())
----> 3 A_petsc.assemble(a)
      4 A_petsc.view()

PETSc/Mat.pyx in petsc4py.PETSc.Mat.assemble()

PETSc/petscmat.pxi in petsc4py.PETSc.assemblytype()

TypeError: an integer is required

image

8

I never said you should add a here.
This whole step should be replaced by your normal assembly.

9

Actually, I haven’t truly understand what you mean since I know too less. But I also tried without a in the function, and it returns a message that:

Mat Object: 1 MPI processes
  type not yet set

instead the matrix content I needed in the form of the figure.