guess_axis scans only half the toroidal planes when lasym = T

[CharlesCNorton]

guess_axis computes ivmax for the two symmetry cases,

if (.not.lasym) then
  ivmax = nzeta/2+1
else
  ivmax = nzeta
end if

and then never uses it as the loop bound: the plane loop is planes: DO iv = 1, nzeta/2+1 (guess_axis.f90:72). For lasym = F the remaining planes are filled after the loop by the stellarator-symmetry mirror

do iv = nzeta/2+2, nzeta
    rcom(iv) = rcom(nzeta+1-(iv-1))
    zcom(iv) =-zcom(nzeta+1-(iv-1))
end do

which is correctly skipped for lasym = T. Nothing else writes rcom and zcom, so for lasym = T the entries nzeta/2+2 .. nzeta reach the axis Fourier transform

raxis_cc(n) = dzeta*SUM(cosnv(:,n)*rcom(:))/nscale(n)

never having been assigned. Both arrays are allocatable, so no bounds or initialization check catches it.

PARVMEC guess_axis.f:132 runs its loop over all nzeta planes and uses the mirror as an in-loop shortcut for .not.lasym instead:

planes: DO iv = 1, nzeta
   IF (.not.lasym .and. iv .gt. nzeta/2 + 1) THEN
      rcom(iv) = rcom(nzeta + 2 - iv)
      zcom(iv) =-zcom(nzeta + 2 - iv)

input.cma reaches guess_axis in both symmetry modes. With LASYM = F the reguess succeeds and the run goes on to ns = 51. With LASYM = T and every asymmetric coefficient left at zero, so the same equilibrium, it fails:

  NS =   25 NO. FOURIER MODES =   59 FTOLV =  1.000E-06 NITER =   1000
  INITIAL JACOBIAN CHANGED SIGN!
  TRYING TO IMPROVE INITIAL MAGNETIC AXIS GUESS

 INITIAL JACOBIAN CHANGED SIGN (IMPROVE INITIAL GUESS)

 The jacobian was non-definite!