[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
## Re: [Help-gsl] Question on Cholesky decomposition error - bspline penalt

**From**: |
Foivos Diakogiannis |

**Subject**: |
Re: [Help-gsl] Question on Cholesky decomposition error - bspline penalty matrix |

**Date**: |
Wed, 19 Nov 2014 13:37:59 +0800 |

Hi Rhys,
again thanks for sending code and comments - much appreciated!!
It seems that this issue is reported by others and it must be a problem of
finite precision arithmetic, check this link :
http://mathoverflow.net/questions/108700/quantifying-the-failure-of-the-cholesky-factorization-test-for-indefinite-matric
So I'll have to trust the theoretical prediction of positive definiteness
for Omega_{ij} and work around it with exception handling (my results are
ok with this, but still need to do more tests).
Cheers,
Foivos
On Wed, Nov 19, 2014 at 11:57 AM, Rhys Ulerich <address@hidden>
wrote:
>* >> \Omega_{ij} = \int_0^1 { d^2B_i(x)/dx^2 * d^2 B_j(x)/dx^2 } dx*
>
>* > *p += wl * gsl_pow_2(dB_i * dB_j);*
>
>* Ha. I apparently cheated. Remove that gsl_pow_2 and I encounter the*
>* same issues.*
>
>* *shrugs**
>
>* - Rhys*
>