Virtual collisions with surface tracking

[ChasingNeutrons]

I have added a very Serpent-like feature: one can specify a minimum average flight distance before collision. The collision is then rejection sampled and can be real or virtual. This is mostly useful because it allows using surface tracking with the collision estimator to score in low density regions.

The operation is quite simple and comes mostly from the database. An 'avgDist' is input, which creates a cross section. During tracking, the maximum of the majorant/total and this cross section is used for tracking. This is then stored in cache and use for tallying as well. Beyond that, there are a few small changes to surface tracking routines.

[valeriaRaffuzzi]

It looks very good overall! It doesn't look like anything should break or be biased anywhere :) Just a suggestion on a different logic that avoids some repetition and avoids having NaN distances.

[valeriaRaffuzzi]

I don't like that you're inviting a division by ZERO and a NaN distance in the common case of vacuum material and unspecified collisionXS. What about this modified logic? If we're not bothered by the trackXS == ZERO. We could also have a tolerance and the check becomes trackXS < TOL where TOL can be 1.0e-12 or so?

      trackXS = self % xsData % getTrackingXS(p, p % matIdx(), MATERIAL_XS)

      if (trackXS == ZERO) then

          dist = INFINITY
          invSigmaTrack = INFINITY
          sigmaT = ZERO

      else

          invSigmaTrack = ONE / trackXS
          dist = -log( p % pRNG % get()) * invSigmaTrack

          ! Should never happen! Catches NaN distances
          if (dist /= dist) call fatalError(Here, "Distance is NaN")

          ! Obtain the local cross-section
          sigmaT = self % xsData % getTrackMatXS(p, p % matIdx())

      end if

[ChasingNeutrons]

I agree, I shouldn't invite the NaN. It is maybe a bit complicated because it isn't clear whether the check you propose is identical to checking for a void material. This might then imply adding a change in logic to getTrackMatXS, which is a bit messy.

[ChasingNeutrons]

Yes, to be more clear: if trackXS /= ZERO in a void (as we have imposed a minimum average distance) this requires getting sigmaT for a void. This implies changing updateTrackMatXS

[valeriaRaffuzzi]

Fair, I hadn't thought of this last case! You can still have a if VOID sigmaT = ZERO in the tracking logic.

[valeriaRaffuzzi]

This looks better now, but I am not against adding an extra VOID check here to be extra safe!