Intel® Fortran Compiler
Build applications that can scale for the future with optimized code designed for Intel® Xeon® and compatible processors.

Linear System Solution problem



I have recently  installed intel's fortran compiler. i also have a valid IMSL licence.

I am trying to solve a linear system AX=B (A,B have complex-double values) that gives me fatal errors when i run it for big matricx A cause of RCOND.
i use imsl functions LIN_SOL_GEN or LIN_SOL_LSQ or LIN_SOL_SVD to solve the system
or DLINCG OR DL2NCG for matrix A inversion and i do A^(-1)*B later.

i had a thought about using quadruple precision complex variables in A and B , to be able to solve bigger systems cause of precision, but unfortunately these IMSL functions do not work with quad variables. 

i also have no experience with MKL library.. maybe there exists a function that could help me..

Any information or help would be very apreciated. 

Labels (1)
0 Kudos
2 Replies
Black Belt

Tell us a bit more about the problem. Is the matrix full or sparse? What is its size? Do you need to solve the equations just once, or thousands of times with the same A but different b (in A.x = b)? Does A remain constant? What value of R_cond did you obtain.

MKL does have Lapack as part of it, so you are not necessarily restricted to using IMSL.

Few libraries support quadruple precision.

Forming the inverse of A just to solve A.x = b is a bad idea, especially when A is ill-conditioned.

0 Kudos
Valued Contributor III

In another direction on matrix inversions. 

I really liked this book we used at UNI == 

I still use it as a reference when I am stumped on an algorithm, although it is pretty basic. 


0 Kudos