# Parameters in FEniCS 2018/2019 equivalent to LUSolver('mumps') from 2017
**URL:** https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924
**Category:** Uncategorized
**Created:** [May 30, 2019, 1:03pm UTC](https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924 "2019-05-30T13:03:15Z")
**Posts on this page:** 5
**Page:** 1
### Author:  [@bcsj](https://fenicsproject.discourse.group/u/bcsj)
#### Post date: [May 30, 2019, 1:03pm UTC](https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924/1 "2019-05-30T13:03:15Z")
I have been using DOLFIN 2017.2.0 for a long while and it seems I might have to switch to one of the later releases, however, my code breaks when I run it under the later versions. It is kind of hard to tell how much of it breaks, since I will have to fix each part to go on, but here i my first problem:
I used to create a solver
```
solver = LUSolver('mumps')
```
which now breaks.
What is/are the equivalent statement(s) in DOLFIN 2018.1.0 and/or 2019.1.0?
---
### Author:  [@nice-butt](https://fenicsproject.discourse.group/u/nice-butt)
#### Post date: [May 30, 2019, 4:43pm UTC](https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924/2 "2019-05-30T16:43:44Z")
If you look into the dolfin C++ code in dolfin-2018.1.0/dolfin/la/PETScLUSolver.h on Line 52, it reads as:
```auto
PETScLUSolver(std::string method="default")
```
And in the Pybind11 interface file (dolfin-2018.1.0/python/src/la.cpp) on Line 852, it reads as:
```auto
.def(py::init(), py::arg("method")="default")
```
as part of the available python interface for PETScLUSolver for Python.
I’ve not dived into dolfin.LUSolver. But if you are using PETSc as the backend, maybe give PETScLUSolver directly a try. The only thing to note is it seems like there’s no interface for `set_operator` for `PETScLUSolver`, so you might need to do
```auto
solver = dolfin.PETScLUSolver(A, 'mumps')
```
as indicated by Lines 855~856 of Pybind11 interface code.
Hope this help.
Victor
---
### Author:  [@plugged](https://fenicsproject.discourse.group/u/plugged)
#### Post date: [June 3, 2019, 11:40am UTC](https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924/3 "2019-06-03T11:40:01Z")
If you want to have more specific control over petsc, take a look at my answer here:
> [@Preconditioner not working or bug?](https://fenicsproject.discourse.group/t/preconditioner-not-working-or-bug/358/4):
>
> The results are “correct” since fenics / petsc uses the preconditioned residual in order to check the convergence criterion. Also, the petsc faq says that this discrepancy can happen, e.g., for saddle point problems: [https://www.mcs.anl.gov/petsc/documentation/faq.html#kspdiverged](https://www.mcs.anl.gov/petsc/documentation/faq.html#kspdiverged) You could circumvent this by applying the preconditioner from the right hand side, since you will then always have an unpreconditioned norm in your convergence check. A MWE could look like this: from fenics import \* …
There it is shown how you can use petsc directly, and how you can define which solver is chosen using “command line options” for petsc.
---
### Author:  [@Karl\_Banana](https://fenicsproject.discourse.group/u/Karl_Banana)
#### Post date: [September 27, 2019, 4:18am UTC](https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924/4 "2019-09-27T04:18:29Z")
Hi, Victor,
Would you mind to share more about PETScLUSolver? I used to use PETScLUSolver.set\_operator(A), A is la.GenericMatrix. However, I found PETScLUSolver(A) requires an input of PETScMatrix. Do you have any idea how la.GenericMatrix can be convert to PETScMatrix?
---
### Author:  [@bcsj](https://fenicsproject.discourse.group/u/bcsj)
#### Post date: [September 27, 2019, 9:08am UTC](https://fenicsproject.discourse.group/t/parameters-in-fenics-2018-2019-equivalent-to-lusolver-mumps-from-2017/924/5 "2019-09-27T09:08:32Z")
Now that some time has passed and I have used Victor’s advice I believe I have some qualification to help shed some more light on this too. At least I can show you how I have approached this. Everything here strictly operates on the native DOLFIN matrix type, typically by converting it to a PETScMatrix. Note that this is just copy-pasta from one of my project files.
```auto
from dolfin import __version__ as DOLFINversion
from dolfin import as_backend_type
from dolfin import Matrix, PETScMatrix
from dolfin import Vector
if DOLFINversion == '2017.2.0':
from dolfin import mpi_comm_self
else:
from dolfin import MPI
from scipy.sparse import csr_matrix
from petsc4py import PETSc
def transpose(M):
"""Transpose for matrix of DOLFIN type
"""
M_ = as_backend_type(M).mat()
Q = M_.copy()
Q.transpose()
return Matrix(PETScMatrix(Q))
def zeroVector(n):
"""Zero-vector of DOLFIN type
"""
if DOLFINversion == '2017.2.0':
return Vector(mpi_comm_self(),n)
else:
return Vector(MPI.comm_self,n)
# use MPI.comm_self() in 2018.1.0 and later
def eye(n):
"""Identity Matrix of DOLFIN type
"""
li = list(range(n+1))
I = PETSc.Mat().createAIJWithArrays((n,n),(li,li[:-1],[1 for _ in li[1:]]))
return Matrix(PETScMatrix(I))
def ToScipy(M):
"""Creates a sparse Scipy matrix from a DOLFIN type matrix
"""
M_ = as_backend_type(M).mat()
return csr_matrix(M_.getValuesCSR()[::-1],shape=M_.size)
def FromScipy(M):
"""Creates a DOLFIN type matrix from a sparse Scipy matrix
"""
M = M.tocsr()
IM = M.indptr
JM = M.indices
DM = M.data
shape = M.shape
M_ = PETSc.Mat().createAIJWithArrays(shape,(IM,JM,DM))
return Matrix(PETScMatrix(M_))
def Solver(A):
s = PETScLUSolver(as_backend_type(A),'mumps')
return s
```