Hi all,

First I have to say that i apreciate using Eigen and everything worked fine until now. The problem:

I have to solve a linear problem with a squared sparse matrix of variable size (from 20 up to 100 000 rows/col). RHS is a sparse vector (no problem to set it to dense) and the solution would be a dense vector. Matrices and vectors have been created with Eigen.
I'm working on Windows 7 64 Bit version and using visual studio 2010.

Since there is no sparse solver included in Eigen, which solver would you recommend to use? I'd like a rather fast, stable and mainly easy to implement solver for this issue (I guess everyone wishes that...^^).
I tried the UmfPack und the SuperLu so far, because of the support modules for them included in Eigen. I downloaded all needed packages, read tons of documentations but I still have no clue how "install" this packages and make them work with Eigen.

I'm grateful for any help.

Update: I included all needed files now (includepath), so that the compiler stops complaining about missing files (UFconfig, AMD...) I tried your example:

Code: Select all

#define EIGEN_YES_I_KNOW_SPARSE_MODULE_IS_NOT_STABLE_YET // if with this

#include <iostream>
/*#include<string>
#include <vector>
#include<iterator>*/


//#include<Eigen/Eigen>
//#include <Eigen/Sparse>
#include <unsupported/Eigen/UmfPackSupport>
//#include <unsupported/Eigen/SparseExtra>

using namespace std;
using namespace Eigen;



int main() {


DynamicSparseMatrix<double>  A(5,5);
   A.insert(0,0) = -1.3;
   A.insert(0,4) =  1.75;
   A.insert(1,1) =  5.28;
   A.insert(2,0) = -1.1;
   A.insert(2,3) = 23.0;
   A.insert(3,3) = 4.8;
   A.insert(4,0) = -2.3;
   A.insert(4,2) = 7.54;
   A.insert(4,4) = 2.0;

   A.finalize();

   cout << A << endl;

   VectorXd b(5), x(5);

   b << 0.0 , 1.3, 0.0, -2.5, 1.1;
   

   cout << b << endl;
   cout << x <<endl;

   // solve Ax = b using UmfPack:
SparseLU<SparseMatrix<double>,UmfPack> lu_of_A(A);
if(!lu_of_A.succeeded()) {
  // decomposiiton failed
  return 0;
}
if(!lu_of_A.solve(b,&x)) {
  // solving failed
  return 0;
}


return 0;
}


But I get compiling errors from the linker:
"1>main.obj : error LNK2001: Nicht aufgelöstes externes Symbol "_umfpack_di_free_numeric".
1>main.obj : error LNK2001: Nicht aufgelöstes externes Symbol "_umfpack_di_free_symbolic".
1>main.obj : error LNK2001: Nicht aufgelöstes externes Symbol "_umfpack_di_symbolic".
1>main.obj : error LNK2001: Nicht aufgelöstes externes Symbol "_umfpack_di_numeric".
1>main.obj : error LNK2001: Nicht aufgelöstes externes Symbol "_umfpack_di_solve".
1>H:\Visual Studio 2010\Projects\ACID3D_Prototype\Debug\ACID3D_Prototype.exe : fatal error LNK1120: 5 nicht aufgelöste externe Verweise."
Which means that this links can not be resolved.

Additional Question:
When this finaly works, I want to improve speed by using ATLAS instead of the standard implemented BLAS. How do I tell Eigen, or UMFPack to do so?

thanks, medic

look at the devel branch, there are a conjugate gradient, a bicgstab, and a direct cholesky solver. To use umfpack you have to link your application against the umfpack library and its dependencies that depends on how your umfpack copy has been compiled...