Helicity selection rules

[ohl]

Replace the numerical heuristics for helicity selection rules by statically computed ones.

[Juergen Reuter]

Highest priority for 2.0_final now!!!

[ohl]

Let's be realistic: TO and WK have to coordinate this. Let's concentrate on getting a tested (but slow) RC1 this weekend and implement helicity selection next week for RC2.

Moving the ticket to v2.0-rc2.

[Juergen Reuter]

According to Wolfgang, the W1 method for helicities has to be implemented before release of rc1.

[ohl]

Re-implemented the old heuristics in r1629. It's not yet validated, but it compiles.

Algorithm (in omegalib):

pure subroutine omega_update_helicity_selection (amp, mask, zeros, threshold, cutoff)
  complex(kind=default), dimension(:,:,:), intent(in) :: amp
  logical, dimension(:), intent(inout) :: mask
  integer, dimension(:), intent(inout) :: zeros
  real(kind=default), intent(in) :: threshold
  integer, intent(in) :: cutoff
  integer :: h
  real(kind=default) :: avg
  if (threshold .gt. 0) then
     avg = sum (abs (amp)) / size (amp)
     do h = lbound (amp, 2), ubound (amp, 2)
        if (mask(h)) then
           if (all (abs (amp(:,h,:)) .lt. threshold * epsilon (avg) * avg )) then
              zeros(h) = zeros(h) + 1
              if (zeros(h) .gt. cutoff) then
                 mask(h) = .false.
              end if
           end if
        end if
     end do
  end if
end subroutine omega_update_helicity_selection

new interface function

  subroutine reset_helicity_selection (threshold, cutoff)
    real(kind=default), intent(in) :: threshold
    integer, intent(in) :: cutoff
    hel_is_allowed = .true.
    hel_zeros = 0
    hel_threshold = threshold
    hel_cutoff = cutoff
  end subroutine reset_helicity_selection

where threshold = O(100) and cutoff = O(1000).

Obviously, a negative threshold will disable the helicity selection.

[ohl]

Performed some basic tests. Appears to work correctly and to be efficient. Needs tuning of THRESHOLD and CUTOFF. Access from SINDARIN would be nice.

Downgrading the priority and severity.

[kilian]

The SINDARIN interface is done. There are three variables:

• ?helicity_selection_active (default true) as overall switch
• real helicity_selection_threshold (default 100)
• int helicity_selection_cutoff (default 100)

Please check.

[ohl]

Passed some more tests. Further tests of the heuristics belong to validation.

Moving this ticket to 2.0final for further optimizations and (maybe later) for an implementation of an analytical approach.

[ohl]

Back to RC1 for an improvement of the logic and a test case. Neither is a big deal and could be moved to RC2.

[ohl]

... no longer defunct ...

[ohl]

It probably can wait until RC2, but here's the situation. The current version (checkpointed in

svn+ssh://login.hepforge.org/hepforge/svn/whizard/tags/ohl/helicity_tuning/agressive_2010_01_31

(sic!) is very aggressive and kills a helicity after CUTOFF zeros anywhere in the sequence. This appears to work, but is dangerous in very long simulations. A more conservative version

svn+ssh://login.hepforge.org/hepforge/svn/whizard/tags/ohl/helicity_tuning/conservative_2010_02_02

requires CUTOFF zeros at the beginning. To my surprise, it is significantly slower (keeps more helicities), but produces the same results. I have to investigate

• where the conservative version is tricked into believing that a vanishing helicity combination in nonvanishing and
• whether the aggressive version is safe.

Since it can be switched off, the aggressive version is probably OK for RC1.

[ohl]

Implemented a much more robust heuristic in r1655: comparing the maximum with the average of the first CUTOFF evaluations. A good new default for THRESHOLD is 1E10.

Further tuning can go to RC2 and later.

[Juergen Reuter]

As TO has two major tasks before this, moving to 2.1

[Juergen Reuter]

Milestone renamed

[Juergen Reuter]

Milestone renamed

[Juergen Reuter]

Transferred to Gitlab issue 328.