search

acts-project / detray

Test library for detector surface intersection
Mozilla Public License 2.0
10 stars 18 forks source link

Develop Helix Stepper for Fast Propagation in CKF #852

Open beomki-yeo opened 2 days ago

beomki-yeo commented 2 days ago

In this talk (Slide 8) of the FTAG workshop, CMS speaker implies that constant magnetic field is used in track finding. We can also follow the same strategy.

In case of using the constant field in the track propagation inside vacuum, we can use the Helix propagation instead of the current RKN stepper. Fortunately, everything we need is included in https://github.com/acts-project/detray/blob/main/core/include/detray/navigation/detail/helix.hpp

asalzburger commented 2 days ago

Hi @beomki-yeo - this is case dependent, for ATLAS ITk offline chain we are aiming to not refit the tracks after track finding and thus save the time of this; in that case, you want the Runge-Kutta propagator to be used there.

The original ATLAS RungeKutta propagator had a flag to fall back onto single step, which turns it effectively into a helix propagator. (Actually, in ATLAS we had a fast field and a precise field, one for pattern, one for refit ... not sure if this is still in use).

Having the option to do so, is good, making it the standard not so in my mind.

beomki-yeo commented 2 days ago

I agree that the standard always should be the RKN with an inhomogeneous field. Helix one is definitely for special cases - I started to think that CMS folks are using the helix propagator for track finding after reading the portability paper you shared recently. I also think that helix calculation will still be cheaper than RKN even with a single step as the RKN requires calculation at four points.

Time calculation of helix and RKN stepper is going to be the same, so this may not be a big issue