Strange behavior with l_mperp=T although l_noco=F

Summary

Two calculations, which only differ by the switch l_mperp=F/T, yield completely different results:

l_mperp=F starts with a reasonable charge distance after the generation of a starting density and converges to the correct magnetic moment of 1.9 bohr
l_mperp=T starts at a super high distance and converges to a non-magnetic state.

Strangely, the flag l_mperp is only supposed to affect the calculation if l_noco=T. However, l_noco=F in both cases.

Input and output file:

inp.xml_bug out.xml_bug out_bug

inp.xml_correct out.xml_correct out_correct

Compute environment

************************Summary***************************
Compiler   :  /usr/local/intel/compilers_and_libraries_2019.3.199/linux/mpi/intel64/bin/mpiifort
Compiler ID:  Intel
Flags      :   -mkl  -mkl -mkl -qopenmp -assume byterecl -r8
Libraries  :  -lxml2;-lmkl_scalapack_lp64;-lmkl_blacs_intelmpi_lp64;-mt_mpi

These Libraries are required:
  XML Library found       :     TRUE
  LAPACK Library found    :     TRUE
These Libraries are optional:
  ELPA (one node) Library found :     FALSE
  FFT from MKL found            :     TRUE
  LibXC Library found           :     FALSE
  HDF5 Library found            :     TRUE
  Wannier90 1.2 Library found   :     FALSE
  Wannier90-4   Library found   :     
  Wannier90-5   Library found   :     
  MAGMA Library found           :     FALSE
  MPI Library found             :     TRUE
    SCALAPACK Library found     :     TRUE
    ELPA Library found          :     FALSE
    ChASE Library found         :     FALSE
  Compile GPU version           :     FALSE


Compile serial version        :     FALSE
Compile parallel version      :     TRUE


Git describe  : snapshot-02.19-612-gbb20fec
Git hash      : bb20fec99f000d0a6e0016724afbb69c5d838714
Doxygen found : YES
************************-------**************************

This is BUG because:

Cobalt is magnetic