I can't get the sparse linearsolver to work robustly...
my matrix/vector are:
Eigen::SparseMatrix<double> m1(150,150);
Eigen::VectorXd vecEig(150);

I tried:
Eigen::SimplicialLLT<Eigen::SparseMatrix<double> > solver(m1);
Eigen::VectorXd resultEig2 = solver.compute(m1);

but over time it introduces a small error that explodes over 30 iterations of some algorithm.

If i convert it to a dense matrix it works fine:
Eigen::MatrixXd dummy(m1);
Eigen::LLT<Eigen::MatrixXd> solver2(dummy);
Eigen::VectorXd resultEig = solver2.solve(vecEig);

For debuging i tried initializing SimplicialLLT before using the dense solver...
Eigen::SimplicialLLT<Eigen::SparseMatrix<double> > solver;
Eigen::MatrixXd dummy(m1);
Eigen::LLT<Eigen::MatrixXd> solver2(dummy);
Eigen::VectorXd resultEig = solver2.solve(vecEig);
When I run this code I get exactly the same result as for the sparse solver
which seem pretty strange to me...

Is there some initialization this line breaks? :
Eigen::SimplicialLLT<Eigen::SparseMatrix<double> > solver;

and is it easy to fix, so I can get proper (and faster) results with the sparse solver?

hm, this looks very strange to me as Eigen::SimplicialLLT<Eigen::SparseMatrix<double> > solver; is doing nothing. Which compiler are you using? Do you have some fast-math mode enabled?

hmm, I guess your right...
at least after a computer restart I am getting the same results with LLT and SimplicialLLT.

So I guess it was some temporary error in my build files.