[BUG]: Union gives seemingly wrongful error in simple setup

[Lomholy]

Describe the bug
I have made a simple instrument that places a cylinder inside a box, and sends neutron through it. The geometry is made with Union.

When running with 1e8 neutron rays, I get ~9 errors like:

ERROR, propagated out of current volume instead of to a point within!
length_to_scattering_specified = 0.00000000351831686216
length propagated = 0.00000000351831686216
length_to_boundary = 0.00004661509106687154
current_position = (-0.000681,0.000481,63.999977)
Debug info: Volumes[current_volume]->name = vert_out, but now inside volume number 2 named container.
Ray absorbed

It seems something might be off with my geometry, but to my mind, it seems that there is no overlap between the 2 geometries, and the cylinder is a complete child of the box.

The instrument is attached her:

fuel_cell.txt

OS and McStas/McXtrace version (please complete the following information):

• OS: MacOS Tahoe 26.4.1
• McCode variant: McStas
• Version: 3.6.21

[mads-bertelsen]

Thanks for the report! I have experienced similar before in these conditions, and luckily there is an easy workaround.

What I believe is happening: The Union_master receive the rays 64 m from the actual geometries, and then calculates intersection times with the geometries. That becomes a relatively large number due to the long distance, and the difference between the two intersections with the tiny cylinder are on the order of the numerical accuracy of the intersection algorithm. This can create a situation where the position the ray is propagated to is slightly off, and not even in the right volume, which is caught and discarded.

The simple workaround: Insert a slit before the sample, and it can be very large diameter, the important thing is the ray is propagated to it. That way the master receive the ray much closer to the geometries, we get smaller intersection times where there is more "floating point room" to distinguish between tiny time differences in the two sides of the tiny cylinder. Doing this dropped the number of errors from from ~10 in 1E8 to 0 for me.

Perhaps we can think of a way to detect this situation in the master and fix it. If the first ray sees only very large intersection times, it could be propagated closer and all intersection times recalculated from the new point. That is some additional complexity, buy may be worth it.

[Lomholy]

Thank you very much for the swift answer!

This indeed seems problematic.
Perhaps we could calculate a bounding box/sphere for the entire geometry and then move the neutron to the edge of that bounding box/sphere (If the neutron starts outside of the bounding box/sphere, that is). This has the added bonus of increased efficiency when there are many neutrons that don't hit the union environment.

But for now placing the slit also fixes it for me! Thank you!