Re: [eigen] cache-friendly matrix inverse

[ Thread Index | Date Index | More Archives ]

Christian Mayer wrote:
> Benoit Jacob schrieb:
>> Hi,
>> Just for reference, i post that here. There are 2 approaches to
>> cache-friendly matrix inversion, 
> We were taught: don't do a matrix inversion in numerics!

Ah, but for some numerics isn't the goal, but just a means to an end.

For example in Quantum Mechanics you can calculate a Green's function by
inverting a matrix (usually a Hamiltonian plus a few terms that describe
boundary conditions), and you can learn things about the system from
that inverted matrix.

And if the precision isn't good enough for the ambitious physicist, he
uses a higher precision scalar type, and if that's to slow he buys
faster computers ;-)

(It's true that most operations can be done by solving equation systems
rather than inverting matrices, but if you write libraries it's not
yours to decide how that library will be used)


Moritz Lenz | |

Mail converted by MHonArc 2.6.19+