I am using the direct solver routine pardiso in a fortran program of mine, for cases where the coefficient array is real. When I run the code for an unsymmetric array (mtype = 11), the routine works normally.
When I try to run the same code for a symmetric coefficient array (mtype = 1) or for a symmetric, positive-definite coefficient array (mtype = 2), the program aborts abnormally. I need to mention that, in all cases, I call pardiso and pass a coefficient array (in CSR3 format) that contains all the nonzero elements.
I was wondering whether the abnormal termination is potentially caused by the input arguments that I pass. Thus, I wanted to ask if there is any difference in what the input arguments of pardiso should contain, depending on whether I have a symmetric or unsymmetric coefficient array (for example, do I only need to pass the upper/lower triangular part of a symmetric coefficient array, etc.)?
As the documentation says, you need to pass only the upper triangular part of the matrix for symmetric matrix types. See https://software.intel.com/en-us/mkl-developer-reference-fortran-pardiso, section for ja prameter.
Also, whenever you are not sure about the input format, you can use the matrix checker (iparm(27)=1), it can provide additional output which can help.
Hope this is useful!