Picongpu: Pusher: Structure-preserving, E-Field Compensating

[ax3l]

Just found this new pusher from Higuera and Cary (2017): structure-preserving (phase-space volume, energy) like Boris' method and E-B-drift conserving (like Vay's method) for relativistic beams (in the lab frame).

Properties of the new pusher (all three 2nd order accurate):

• Boris: energy conserving, NOT E + UxB
• Vay: NOT energy conserving, but E + UxB
• new: energy conserving AND E + UxB

Changes to Boris' method?
One line:

https://github.com/ComputationalRadiationPhysics/picongpu/blob/9b996254c4f6a667c91d119aa5ee9715b21d0912/include/picongpu/particles/pusher/particlePusherBoris.hpp#L66-L67

Reference:
https://arxiv.org/abs/1701.05605 (DOI:10.1063/1.4979989)

[sbastrakov]

The paper seems interesting. However, it seems unclear to me that the additional FLOPs required vs. the standard Boris are necessarily justified.

[ax3l]

Flops for already loaded data are for free :)
(Yes, sqrt is not ideal, but hey.)

[sbastrakov]

Was glancing through recent papers in arXiv and found this one
As far as I can see, picongpu only has the "standard" Boris scheme with approximate rotation, called Boris-B in the paper, and does not have a Boris scheme with exact rotation, also known as Boris with gyrophase correction (Boris-A in the paper). The paper also provides an alternative formulation for the latter rotating pusher and shows that both rotating schemes are volume-preserving. They also experimentally show their scheme is a little faster than Boris-A (though for me it is not clear why), and both should be just marginally slower than Boris-B in realistic scenarios.

Also two interesting papers in the references of this one:

• Implicit pusher
• Experimental evaluation of popular pushers in several scenarios

[steindev]

Can be closed. Implemented with #3280.