Continuous particle injection in 2D3V (circular source in center)

[kolotinsky1998]

Hello WarpX developers,

I have a question regarding continuous particle injection.

I am working in 2D3V (x–z plane, velocities in all 3 directions) and would like to implement a continuous source of electrons from a circular region in the center of the domain, defined by (x^2 + z^2 < R_inj^2). The injected electrons should have an initial velocity along the y axis (perpendicular to the simulation plane).

I tried the following approach using do_continuous_injection = 1, but it only injects particles at the first iteration and no new particles are added afterwards:

# --- Electrons ---
electrons.charge = -q_e
electrons.mass   = m_e
electrons.injection_style = NRandomPerCell
electrons.num_particles_per_cell = 8

# bounding box (rectangle enclosing the circle)
electrons.xmin = -Rinj
electrons.xmax =  Rinj
electrons.zmin = -Rinj
electrons.zmax =  Rinj

# density profile: circle of radius Rinj
electrons.profile = parse_density_function
electrons.density_function(x,y,z) = "n0 * ((x*x + z*z) < (Rinj*Rinj))"
electrons.density_min = 1.0

electrons.momentum_distribution_type = constant
electrons.ux = 0.0
electrons.uy = 0.01978746   # ~100 eV along +y
electrons.uz = 0.0

electrons.do_continuous_injection = 1

Here is the animation of the simulated system:

And the full script for this example.
What is the recommended way in WarpX to implement such a continuous central source in 2D3V — e.g. a circular emission region with constant injection of particles every timestep?

Thank you very much for your help!

Best regards,
Daniil Kolotinskii

[RemiLehe]

Thanks for this question @kolotinsky1998, and for your interest in WarpX!

We unfortunately do not have this capability in WarpX currently (but you can potentially try to reproduce it with Python callback function).

There are several capabilities that are close:

• do_continuous_injection is specifically related to the moving window functionality of WarpX. It initializes particles at each timestep in the new cells that moving window moved into.
• Flux injection (see NFluxPerCell in the documentation) injects particles from a surface (for 3D simulation) / a line (for 2D simulation), but it somewhat different from what you want here, which is injecting from a surface in a 2D simulation.

[kolotinsky1998]

Thanks a lot for the clarification @RemiLehe!

1. Now I understand better what do_continuous_injection does — it is indeed tied to the moving window and just initializes new particles in the freshly created cells. I saw it mentioned in a discussion thread as a potential solution for my case, which is why I tried to look in that direction.
2. I also experimented with NFluxPerCell. As you mentioned, in 2D it indeed allows me to implement continuous injection from a line, but not from a surface, which is what I actually need.

A follow-up question: regarding the Python callback approach — how efficient is it in terms of performance? This is critical for me, since I plan to run large-scale simulations using either MPI+OpenMP or GPU accelerators. Would it still be possible to achieve high performance with Python callbacks, or will the overhead be too significant in such cases?