VlasovAmpereOneSpecies: total_energy jumps after the first step with the control variate (and is offset at t=0 with full-f weights) #767

Bug: with WeightsParameters(control_variate=True), the total_energy scalar of VlasovAmpereOneSpecies jumps after the first time step. A second, smaller offset at t=0 also appears with full-f weights.

1. Control variate. kinetic_energy (KineticEnergyPIC, models/scalars.py) sums weights * |v|^2. At t=0 it is computed with the full-f weights, and after the first step (once update_weights has run in VlasovAmpereCoupling) with the delta-f weights. The two values are energies of different quantities, so total_energy is not comparable across the first step, and energy conservation cannot be checked with the control variate.
2. Full-f weights. total_energy is conserved from step to step to about 1e-9 (Crank-Nicolson coupling, pcg tolerance 1e-12), but its value at t=0 differs from the later values by about 2e-4 relative. Not investigated yet; probably the initial diagnostic, not the time stepping.

Both were found in #734: its e2e test test_gpu_VlasovAmpereOneSpecies.py uses full-f weights for this reason, and checks the t=0 value only loosely (check_reference).

Expected: the kinetic energy diagnostic uses the same weights (full f, i.e. delta-f weights plus the background contribution) at every time, and total_energy at t=0 matches the conserved value.

🤖 Generated with Claude Code