Hi, can ELPA only get half of the symmetric matrix and compute the eigenvalue and eigenvector? Just like ScaLAPACK pdsyevd?
There is a symmetric matrix that will be transformed before computing its eigenvalue and eigenvector. But after transformed it may become slightly unsymmetric due to the numeric error. Since this matrix is stored in parallel, remaking it symmetry will cost much communication. But if we just use the transformed matrix, the ELPA will give inconsistent results with ScaLAPACK.
So can ELPA just use half of the input matrix and assume it's a symmetric matrix? Thank you!
Hi, can ELPA only get half of the symmetric matrix and compute the eigenvalue and eigenvector? Just like ScaLAPACK pdsyevd?
There is a symmetric matrix that will be transformed before computing its eigenvalue and eigenvector. But after transformed it may become slightly unsymmetric due to the numeric error. Since this matrix is stored in parallel, remaking it symmetry will cost much communication. But if we just use the transformed matrix, the ELPA will give inconsistent results with ScaLAPACK.
So can ELPA just use half of the input matrix and assume it's a symmetric matrix? Thank you!