As was suggested at the G4 workshop, this is now a formal bug-report:

When using G4IntersectionSolid the navigator behaves real strange.

I have put a compressed tar-ball of this example at:

http://trshare.triumf.ca/~gum/makonyik.tar.Z

The problem is that the G4Navigator sometimes fails to make a zero-length step
(in the next volume beyond the boundary) upon reflection. Instead, the step is
much longer and after that the tracking stops all together. G4OpBoundary has
reflected the track direction though the volume identified is that of what's
beyond the surface.

An example from test.out in the tar-ball:

*********************************************************************************************************
* G4Track Information:   Particle = opticalphoton,   Track ID = 1,   Parent ID = 0
*********************************************************************************************************


Step#      X         Y         Z        KineE    dEStep   StepLeng  TrakLeng
Volume     Process
   0      0 fm      0 fm     -0 fm      3 eV      0 eV      0 fm      0 fm
  guide    initStep
Photon at Boundary!
Old Momentum Direction: (-0.76093104,-0.09179635,-0.6423063)
Old Polarization:       (-0.99500417,-0.099833417,0)
theLocalNormal: (0.9949054,0.10081294,-0)
New Momentum Direction: (0.7638782,0.062711314,-0.6423063)
New Polarization:       (-0.99480567,-0.10179237,-0)
*** SpikeReflection ***
   1  -22.2 cm  -2.68 cm  -18.7 cm      3 eV      0 eV   29.2 cm   29.2 cm
  guide  Transportation
   2  -14.1 cm  -2.01 cm  -25.6 cm      3 eV      0 eV   10.6 cm   39.8 cm
  World  Transportation
Track (trackID 1, parentID 0) is processed with stopping code 2

Besides the line that prints 'theLocalNormal' the above printout is the default
from G4OpBoundaryProcess verbose=1. You see that G4OpBoundaryProcess calculates
a new direction from the old direction and the normal. You can verify that the
angle of incidence with the normal is the same as the angle on reflection and
the directions are opposite, thus we have 'SpikeReflection'. In this case, the
next step should be in the 'World' but with zero length, instead it is 10.6cm
and then the track just stops.

There are also cases in test.out where reflection behaves normally:

*********************************************************************************************************
* G4Track Information:   Particle = opticalphoton,   Track ID = 1,   Parent ID = 0
*********************************************************************************************************


Step#      X         Y         Z        KineE    dEStep   StepLeng  TrakLeng
Volume     Process
   0      0 fm      0 fm     -0 fm      3 eV      0 eV      0 fm      0 fm
  guide    initStep
Photon at Boundary!
Old Momentum Direction: (0.17897827,-0.98157815,0.066866421)
Old Polarization:       (-0.99500417,-0.099833417,0)
theLocalNormal: (-0.15602987,0.85572124,-0.49335163)
New Momentum Direction: (-0.10214757,0.56021228,-0.82202681)
New Polarization:       (0.97321591,0.21932769,-0.068892405)
*** SpikeReflection ***
   1   4.41 cm  -24.2 cm   1.65 cm      3 eV      0 eV   24.6 cm   24.6 cm
  guide  Transportation
   2   4.41 cm  -24.2 cm   1.65 cm      3 eV      0 eV      0 fm   24.6 cm
  World  Transportation

A picture of the solid can be found at the URL below (thanks to Karoly Makonyi).

You can run the program with:

/bin/Linux-g++/LG_Main optPhoton.mac > test.out &

Best regards, Peter Gumplinger