Incompatibility between Y and Z reflective field BC

[mccoys]

Description

In 3Dcartesian using both Y and Z with reflective EM boundary conditions makes the corners in the z=0 plane unstable and eventually creates infinite values. This is true both on CPU/GPU, with or without particles. The problem is fixed if either Y or Z are switched to periodic.

Steps to reproduce the problem

Namelist that shows the bug

from numpy import pi, sqrt, sin, cos, exp, s_, array

# normalisations in space & time
l0 = 2.*pi
t0 = 2.*pi
um = l0 / 0.8
fs = t0 * 3e8 / 0.8e-6 * 1e-15

npatch = [48,1,1]

# box size & sim.time
L = [40.*um, 4.*um, 4.*um]
Tsim = 100.*fs
d   = l0 / 64.
ncell = [4*np*( int( l / d ) // (4*np) ) for l,np in zip(L,npatch)]

Main(
    geometry = '3Dcartesian',
    patch_arrangement = "linearized_XYZ",
    interpolation_order = 2,
    timestep_over_CFL = 0.6,
    simulation_time = Tsim,
    grid_length = L,
    number_of_cells = ncell,
    number_of_patches = npatch,
    EM_boundary_conditions = [['silver-muller'],['reflective'],['reflective']],
    print_every = 20,
    solve_poisson = False,
    reference_angular_frequency_SI = 2 * pi * 3e8 / 0.8e-6,
)

LaserGaussian3D(
    box_side        = "xmin",
    a0              = 10.,
    omega           = 1.,
    focus           = [l/2 for l in L],
    waist           = 10000000.,
    polarization_phi = 0.,
    ellipticity     = 0.,
    time_envelope   = tsin2plateau(start=0., fwhm=25*fs, plateau=0) #, slope2=100.*t0
)

DiagFields(
    every = 100,
    time_average = 2,
    fields = [],
    subgrid = s_[::4, ::4, ::4]
)

Parameters

• Smilei version : 5.1-366-g35cfab5-develop
• Computing resources
  • tornado skylake (384 cores) / adastra mi250 (48 GPUs)

[mccoys]

Can be reproduced on a very quick 1-core simulation:

from numpy import pi, sqrt, sin, cos, exp, s_, array

# normalisations in space & time
l0 = 2.*pi
t0 = 2.*pi
um = l0 / 0.8
fs = t0 * 3e8 / 0.8e-6 * 1e-15

npatch = [1,1,1]

# box size & sim.time
L = [4.*um, 4.*um, 4.*um]
Tsim = 100.*fs
d   = l0 / 32.
ncell = [4*np*( int( l / d ) // (4*np) ) for l,np in zip(L,npatch)]

Main(
    geometry = '3Dcartesian',
    patch_arrangement = "linearized_XYZ",
    interpolation_order = 2,
    timestep_over_CFL = 0.9,
    simulation_time = Tsim,
    grid_length = L,
    number_of_cells = ncell,
    number_of_patches = npatch,
    EM_boundary_conditions = [['silver-muller'],['reflective'],['reflective']],
    print_every = 20,
    solve_poisson = False,
    reference_angular_frequency_SI = 2 * pi * 3e8 / 0.8e-6,
)

LaserGaussian3D(
    box_side        = "xmin",
    a0              = 10.,
    omega           = 1.,
    focus           = [l/2 for l in L],
    waist           = 10000000.,
    polarization_phi = 0.,
    ellipticity     = 0.,
    time_envelope   = tsin2plateau(start=0., fwhm=25*fs, plateau=0) #, slope2=100.*t0
)

DiagFields(
    every = 100,
    time_average = 2,
    fields = [],
    subgrid = s_[::4, ::4, ::4]
)

Plotting:

S.Field(0,"Ez",average={"z":0},vsym=0.01).plot()

[mccoys]

The issue is fixed if the ordering of boundary conditions is changed from XYZ to ZYX in void ElectroMagn::boundaryConditions

I will test whether this is true for all combinations

[mccoys]

XYZ ordering of BCs:

• laser from xmin: fails from both (0,0,0) and (0,ymax,0)
• laser from xmax: fails from (xmax, ymax, 0) + diffraction appears at xmax corners
• laser from ymin: fails from (xmax, 0, 0)
• laser from ymax: fails from (xmax, ymax, 0)
• laser from zmin: seems ok at first but some oscillating emission appears from the (xmax, ..., 0) line. Maybe this is a different bug?
• laser from zmax: same thing but looks different (some pixelated defects)

ZYX ordering of BCs:

• laser from xmin: no issues
• laser from xmax: no issues
• laser from ymin: no issues
• laser from ymax: no issues
• laser from zmin: fails from (xmax, 0, 0) and (xmax, ymax, 0)
• laser from zmax: same

This does not seem symmetrical to me, so I would guess something is really off somewhere in BCs. As far as I can tell, the issue is always somewhere at the zmin boundary ...

EDIT: This asymmetry may be completely wrong: I forgot to update EM_boundary_conditions_k accordingly

[mccoys]

Strange effect: when the laser comes from zmax, the bug starts from zmin only, and it develops way before the laser reaches zmin. In fact it develops from the very beginning. I would thus expect that there is no influence from the laser that is at the other side. However, without laser there is no bug at all (zeros everywhere).

I thus suspect that the issue is related to the laser doing something wrong somewhere it shouldn't, likely in the corners (even opposite to where it is).

NOTE: when there is a laser, but its field is effectively 0, there is no instability