Newbie: Pardiso Tutorial

[ustatic]

Recently I've purschased Intel VisualFortran Compiler with MKL and i am interested in solve sparse systems of linear equations by pardiso solver. The system is ill conditioned from finite element (bars and shells). Is there any basic fortran f90 tutorial to easy link libraries to a windows executable with some information to configure visual studio 2005 paths etc...

Thanks in advanced

[ustatic]

After applying compiler directives as shown in mkl document 347461.pdf i could compile the pardiso_sym_f.f example of mkl but linker raises error LNK2019: unresolved external symbol _mkl_solver_pardiso referenced in fuction _PARDISO in file mkl_intel_c.lib

Any idea?

Thanks

[Todd_R_Intel]

You mention mkl_intel_c.lib so you must have that added to your project. That is the interface library. You must also add a threading layer (e.g., mkl_intel_thread.lib), mkl_core.lib which contains the main functionality, and then the threading runtime library libiomp5mt.lib. The user guide in the doc directory will have more information on these choices.

-Todd

[ustatic]

Added in Additional Dependencies of Linker|input tab mkl_intel_c.lib and the problem persists

[ustatic]

Problem solved: Also added mkl_solver.lib !!

[ustatic]

I solved a huge problem os structural beams ( A beam with one end fixed discretized in 5000 parts) but laks of precision.

Is there any solution to fix this lack of precision for ill conditioned matrix?? Ansys gives slightly bad solution about 10000 + parts

Thanks

[Alemdar__Bulent]

If it is ill conditioned, the solution may not be stable (or the results may not be reliable). In fact, I was expecting that PARDISO fails to solve the system if it is positive-definite symmetric ill-conditioned matrix.

I just wonder why you solve a cantilever beam with thousand of elements?
bulent

[ustatic]

Despite of all ANSYS gives right solution upon 10000 parts...

Because i wish to solve geometrically non linear analysis and check stability of ill conditioned matrix