MPI too many communicators?

Question

The idea of the code is to open a file in parallel and then each core reads different part of that big file. ir_start and ir_stop are therefore different for each core.

I have the following code (part of a much bigger code):

DO iq = 1, nqf
  IF (MOD(iq,100) == 0) THEN
      WRITE(stdout, '(a,i10,a,i10)' ) '     Progression iq (fine) = ',iq,'/',nqf
  ENDIF
  CALL MPI_FILE_OPEN (world_comm,filint,MPI_MODE_RDONLY,MPI_INFO_NULL,iunepmatwp2,ierr)
  IF( ierr /= 0 ) CALL errore( 'xxxx', 'error in MPI_FILE_OPEN',1 )

  DO ir = ir_start, ir_stop
    CALL MPI_FILE_SEEK(iunepmatwp2,lrepmatw,MPI_SEEK_SET,ierr)
    CALL MPI_FILE_READ(iunepmatwp2, aux, lrepmatw2, MPI_DOUBLE_PRECISION, MPI_STATUS_IGNORE,ierr)
     loops:
        eptmp ( ibnd, jbnd, ip, imode ) = aux (i)
     endloops
     CALL mp_sum(eptmp, world_comm)
  ENDDO
  something is then done with eptmp
ENDDO

The problem that the code systematically crash at:

 ...
 Progression iq (fine) =      32000/     50000
 Progression iq (fine) =      32100/     50000
 Progression iq (fine) =      32200/     50000
 Progression iq (fine) =      32300/     50000
 Progression iq (fine) =      32400/     50000
 Progression iq (fine) =      32500/     50000
 Progression iq (fine) =      32600/     50000
 error in MPI_FILE_OPEN

It look to me (I might be totally wrong) that there might be too many communicators open.

The problem is that I cannot do something like CALL MPI_Comm_free(world_comm) in the q-loop because it kills the code.

Is there a way for each core to use the same world_comm without creating new one?