In my case, I want to solve a underdetermined equation system, A*λ = b, with the JacobiSVD solver from Eigen (3.0.12).
The linear equation system has following structure in my C++ program:
Coefficient matrix (A):
0.6 5.68434e-20 -0.2
5.68434e-20 7.06819e-39 -4.26326e-20
-0.2 -4.26326e-20 0.4
RHS (b):
-1.962
2.78817e-19
-5.886
Lambda:
-9.81
3.94467e+19 <---------- error (Where does this come from?)
-19.62
- The rank of matrix A is - Rank(A) = 2
- So the matrix has No Full rank. Therefore, A is singular and not invertible.
- The condition is - cond(A) = Inf
- For solving A*λ = b, I used the SVD decomposition method from Eigen (JacobiSVD)
I also verified this with MATLAB: http://www.pictureupload.us/image-172220092351c5ae0c1706e.htm
On begin, the first simulation steps are approximatley correct. But there is a very small numerical error, which is increasing during solving A*λ = b.
Then the system is crashing and my results are not anymore correct and I get NaN results.
Here the code:
/******** SVD ********/
JacobiSVD<TMatrixX> svd(A, ComputeThinU | ComputeThinV);
lambda = svd.solve(b);
What have I done wrong?
