I probably found a bug:

Vector2=something;
Vector1 = Vector2;

the solution of
vec = sparse_mat.transpose() * Vector1 is right.

vec = sparse_mat.transpose() * (Vector1+Vector2)*0.5 is wrong.

do you have a complete test code because I cannot reproduce the bug

ggael wrote:do you have a complete test code because I cannot reproduce the bug

Yes!

Code: Select all

// result looks reasonable:
   TVector tmp1 = global_stiffness_matrix * constraintVectorRhs();
   constraintVectorRhsLoad() = constraint_matrix.transpose() * tmp1;
// should be the same as stffness is symmetric in the test case:
   TVector tmp1 = global_stiffness_matrix * constraintVectorRhs();
    TVector tmp2 = global_stiffness_matrix.transpose() * constraintVectorRhs();
    constraintVectorRhsLoad() = constraint_matrix.transpose() * (tmp1+tmp2)*0.5;


sorry, but I cannot reproduce the issue. Actually your example is not helping much, what is TVector ? how is declared global_stiffness_matrix ? etc.

The data types are default VectorXd and SparseMatrix . But I found the origin: The input vector contained uninitialized values. I did not really discover the location of the programming error, but will look into that later. Funny that there are two different solutions. But no wonder that they were wrong...