diff options
| author |  Benoit Jacob <jacob.benoit.1@gmail.com> | 2009-08-12 18:23:39 -0400 |
|---|---|---|
| committer | Benoit Jacob <jacob.benoit.1@gmail.com> | 2009-08-12 18:23:39 -0400 |
| commit | 2b618a2c1631548b00af3fe923e9db76aca71825 (patch) | |
| tree | 6cbe9347e5ba541c0f22346e376f96e1e2fce612 /Eigen/src/Jacobi | |
| parent | 309d540d4a71527a4dba778dc5221b81fd18f540 (diff) | |
make jacobi SVD more robust after experimenting with very nasty matrices...
it turns out to be better to repeat the jacobi steps on a given (p,q) pair until it
is diagonal to machine precision, before going to the next (p,q) pair. it's also
an optimization as experiments show that in a majority of cases this allows to find out
that the (p,q) pair is already diagonal to machine precision.
Diffstat (limited to 'Eigen/src/Jacobi')
| -rw-r--r-- | Eigen/src/Jacobi/Jacobi.h | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/Eigen/src/Jacobi/Jacobi.h b/Eigen/src/Jacobi/Jacobi.h index 5866ac44f..48195c1ec 100644 --- a/Eigen/src/Jacobi/Jacobi.h +++ b/Eigen/src/Jacobi/Jacobi.h @@ -50,7 +50,7 @@ void MatrixBase<Derived>::applyJacobiOnTheRight(int p, int q, Scalar c, Scalar s template<typename Scalar> bool ei_makeJacobi(Scalar x, Scalar y, Scalar z, Scalar *c, Scalar *s) { - if(ei_abs(y) < ei_abs(z-x) * 0.5 * machine_epsilon<Scalar>()) + if(ei_abs(y) <= ei_abs(z-x) * 0.5 * machine_epsilon<Scalar>()) { *c = Scalar(1); *s = Scalar(0); |