I know this question is stupid. I want to find (only) the lowest M eigenvalues and eigenvectors of a complex hermitian sparse matrix. Intel MKL can clearly do this, but I am simply unable to find the manual for the latest version of Intel MKL.
The official Intel guidelines tell me to look at the file path-to-mkl/examples/solvers_eec/source/dexample_extremal_svd_c.c
That file does not exist, and as far as I can tell this guideline is from 2018 and does not conform to the latest version of mkl, which is from 2023. (I am using the Arch-Linux package intel-oneapi-mkl version 2023.0.0_25398-1)
I believe that the function I am looking for is zfeast_hcsrev as per this possibly newer guide. I even found an example program in the library which literally claims to demonstrate finding the first M eigenvalues ($MKLROOT/examples/c/sparse_eigsolvers/source/zfeast_sparse.c, depending on installation this may need to be built manually first). This includes this passage:
EDIT: Copyright issue, I am literally not allowed to post the original example, (even though it is a free example), so instead I have rewritten this, to avoid copyright infringement.
//Set the array fpm to default values, I don't think that should include the number of eigenvalues
feastinit(fpm);
//z (complex double) feast _ hcsr (complex sparse hermitian) ev (solve Ax = ex for e and x).
zfeast_hcsrev
(
&UPLO, // Is this the full or upper triangular matrix
&N, // matrix size
val, //CSR value, rows and cols
rows,
cols,
fpm, //parameter array
&rel_error, //program writes relative error found
&loop, //number of loops used
&Emin, //lower and upper bound of eigenvalues
&Emax,
&M0, //The example calls this: "initial guess for subspace dimension to be used", I don't know what that means
E, //Claimed to be the first M eigenvalues
X, //Claimed to the first M Eigenvectors as a matrix
&M, // M, the number I want, but this only seems to be an output
res_err, //error on residual vectors
&error_code //error code
);
The full example is included below (EDIT: NO, BECAUSE COPYRIGHT, even though it can be downloaded for free). This gives me 4 eigenvalues inside this interval.
The claim is that this should give the first M entries, the example program has. So I simply try to set M=2.
Obviously, this does not work. I still get 4 eigenvalues. So clearly I don't understand what the input parameters really do, and I can not find any official guidelines for this. (I have tried looking in $MKL_INCLUDE_DIR/mkl_solvers_ee.h where the function is defined, but that obviously doesn't include any comments telling me what the parameters actually do).
I even tried asking chatGPT, but it could not help, because the newest version of intel MKL is newer than 2021. It suggested I use a function which does not exist anymore.
Besides, I don't even think this is the right function to use, as it only searches in a particular pre-defined interval. I would prefer if I could just get the lowest eigenvalues everywhere.
So my stupid question is: how do I actually find only the lowest M eigenvalues and eigenvectors of a complex hermitian sparse matrix using Intel MKL? Please do not suggest that I switch over to the armadillo library, I just switched away from it because it is not thread-safe. (Edit: Normally armadillo is thread-safe, but NOT in the particular case of finding eigenvalues of a complex sparse matrix using eigs_gen which relies on ARPACK, as per my previous question)
EDIT: I am not allowed to publish the full example code, due to license issues, (kind of weird, seeing that I can install it for free but ok)
