Make your paper interactiveCreate

Text from the interactive explanation

Simulation design for subluminal velocity-controlled spatio-temporal drivers in laser–plasma wakefields

Chiara Badiali, Rafael Almeida, Thales Silva and Jorge Vieira

Interactive paper · Badiali, Almeida, Silva & Vieira · Plasma Phys. Control. Fusion 68, 075016 (2026)

Every plane wave in this picture moves at c . Where they all line up, a bright focus forms, and it drifts at a velocity vf that you choose. Change vf, or switch the waves off one by one.

Every wave travels at c; their crossing point travels at vf.

Computed live as a sum of exact vacuum plane waves (2D slab geometry) in a window that moves with the envelope at c. The red curve is where the modes must sit; how far a mode is from the purple plane (above or below) sets its weight through the envelope's Fourier transform.

Faster than light: the superluminal case

The same recipe also builds a focus that outruns light. Only the tilt of the spectral plane changes.

A focal velocity above c tilts the plane ω = vfkz + ck₀(1 − vf/c) more steeply than the light cone. It then cuts the cone in an open hyperbola , the spectrum of X-waves, with infinite spectral extent. Below c the plane cuts through the inside of the cone, and the spectrum closes into an ellipse (Fig. 1). Every mode is still an exact vacuum plane wave travelling at c. Only the place where they line up moves faster than light.

Both pulses below share one envelope, one spot size (k₀w₀ = 10) and one window moving at c. The slider keeps them the same distance from c, so their foci slip at the same speed |c − vf| in opposite directions. The superluminal focus runs from the trailing edge of the envelope towards the leading edge, and the subluminal one runs the other way (Sec. 3.2). The paper concentrates on the subluminal regime, but notes that its spectral construction and simulation guidelines also apply to vf > c schemes, which aim to mitigate dephasing in laser–wakefield acceleration.

Superluminal vf = 1.50c

Subluminal vf = 0.50c

Field in a window 140/k₀ long moving at c (laser units). Lower plots: the (kz, kx) plane with the exact red conic, the sampled modes (brighter = stronger), the dotted circle |k| = k₀ and the central wavevector k₀ (white dot). The shaded strip kz < 0 is left out to keep the pulses forward-propagating. Superluminal modes sit at kz > k₀ and carry higher frequencies, subluminal ones sit at kz < k₀. Lsurv is given in units of the envelope length L: for the same slip speed, the faster focus covers more ground before it leaves the envelope.

Ghosts in a discrete spectrum

A simulation cannot add up a continuum of plane waves. It samples the spectrum on a grid, and any sampled Fourier sum repeats itself.

In the paper's OSIRIS implementation the continuous spectral weight is mapped onto modes spaced Δk apart and summed over the whole domain at t = 0. Copies of the pulse then reappear every Trep = 2π/Δk (eq. 7). They are numerical artefacts, and they can overlap the physical pulse inside the simulation window. Two geometric rules keep them out:

The beam radius grows away from the focus as w(ξ) = w₀[1 + (2ξ / k₀w₀²(1 − vf/c))²]½ (eq. 9), so the widest point inside the window sets the transverse rule. Finer sampling removes the ghosts, but the number of modes grows like 1/(ΔkxΔkz), and every mode costs memory and time.

Our lab: vf = 0.8c, k₀w₀ = 10; a simulation window 200/k₀ long with the focus at its centre; the envelope fills the window (160/k₀ flat top, 20/k₀ sin² ramps). The dashed box is the window, cyan lines are ±w(ξ), rings mark where replicas sit (±Trep,z along the axis, ±Trep,x sideways), and the white trace is the on-axis |E|. Each change re-samples and re-sums the spectrum.

–

Challenge: the cheapest clean sampling

Make both checks pass with as few modes as possible.

Tune the wake to the focus

For an ordinary laser pulse, spot size and duration are separate knobs. For a spatio-temporal driver they are tied together.

Seen from the focus, in the co-moving coordinate ξ = z − vft, a plasma oscillating at ωp repeats every λwake = 2πvf/ωp (eq. 12). Resonant driving asks for a driver about half that long. In the paraxial model the on-axis intensity of the ST focus is a Lorentzian, I(0, ξ) ∝ [1 + (ξ/ξR)²]−1 (eq. 15), with ξR = k₀w₀²(1 − vf/c)/2 (eq. 14), and its 1/e² width is Lpulse = √(e² − 1)·2ξR (eq. 17). Setting Lpulse ≈ λwake/2 picks the spot size:

Wakefield theory also asks for a transverse match, kpw₀ ∼ 1. Both conditions act on the same w₀, so in general they cannot both be met. One works where they are compatible, or gives one up. Slower foci stretch the bright region, so lower vf needs a tighter spot.

Plasma units (c/ωp) with ω₀/ωp = 10 as in the paper. The colour gives the ratio Lpulse/(λwake/2) = (k₀w₀ / k₀w₀eq. 18)² for every focal velocity and spot size: green on the eq. (18) curve, ember above it (driver longer than half a wake period), blue below it (shorter). Dashed contours mark a ratio of 2 and ½. The white probe slides along the curve, and its two bars, Lpulse (orange) and λwake/2 (green), stay equal everywhere on it. Drag the orange dot, or use the sliders, to place your own driver; the dotted line drops it onto the curve, which is what “Apply eq. (18)” does. The purple line is the transverse match kpw₀ = 1: the two conditions meet only near vf ≈ 0.89c, where the curve crosses it (for ω₀/ωp = 10). Squares 4a–4d are the four runs of Fig. 4 and square 6 is the a₀ scan of Fig. 6.

Mini-game: match the wake

Get a random focal velocity, tune only k₀w₀, and get a score for how well the driver fits half a wake period.

All panels: ω₀/ωp = 10, a₀ = 2.2. Greyscale = electron density, blue–red = laser field. Eq. (18) predicts k₀w₀ ≈ 7.05 at 0.8c (paper: 7.1) and ≈ 4.32 at 0.6c (paper: 4.3).

Figure 6 (panels a–d) from Badiali et al. 2026, Plasma Phys. Control. Fusion 68 075016, CC BY 4.0 (cropped).

Turn up the driver: a₀ dial

Fixed driver vf = 0.96c, k₀w₀ = 17 (PIC). Upper part: electron density with the on-axis laser envelope (red); lower part: on-axis accelerating field.

The focus has a clock

The envelope rides at c, the focus at vf. So the focus slides through the envelope and eventually falls off its end.

In a moving window that follows the envelope at c, the envelope stays put while the intensity peak drifts at |c − vf|. It drifts backwards for subluminal foci and forwards for superluminal ones (Fig. 7). Starting at one edge of an envelope of length L, the focus survives for Δtsurv ≃ L/|c − vf| and covers

This holds in both regimes and diverges as vf → c, which recovers the usual moving-window picture. Far from c, a long interaction needs a long envelope, and therefore a long simulation window. In the waterfall below, time runs upward. Lines of constant wake phase move at vf, so in the c-window they lean backwards at 0.8c, stand almost upright near 0.99c and lean forwards at 1.4c, the trend of the paper's Fig. 5.

Top: the window moving at c, with the envelope (stationary here), the focus and the on-axis wake now. Bottom: the waterfall builds row by row. Dashed verticals are the envelope edges, the red line is the live focus, and the dashed horizontal marks Δtsurv. Wake exists only in plasma the live focus has crossed, and that plasma streams backward at c through the window. Brighter means larger Ez.

Design challenge: shortest envelope

Reach 400 c/ωp at vf = 0.80c with the shortest envelope.

Feed the focus from the walls

A focus slower than light needs waves arriving from steep angles, and those waves keep travelling sideways.

Subluminal ST pulses carry a broad angular spectrum, so the envelope spreads with an effective half-angle tan θdiv ≈ 2/[k₀w₀(1 − vf/c)] (eq. 20). To hold the expanding pulse, a quasi-3D box needs a radius

with zprop ∼ cΔtsurv as a conservative choice for propagation over the full survival time. That usually means a large transverse domain. The paper's alternative is continuous wall injection : at every time step the code evaluates the plane-wave superposition at the transverse boundaries and feeds it in. The wings that build the moving focus are then supplied from the walls, and the box only has to cover the interaction region. Each step costs about 1.5–2× more, but without injection the transverse domain can need to be ~10–100 times larger. Cost scales with N⊥ in quasi-3D and with N⊥² in full 3D, so the paper reports net savings of often one to two orders of magnitude.

Laser units (1/k₀). With injection, every mode is present inside the box. Without it, each tilted component e±ikxx keeps only the part of the box it filled at t = 0 and drifts sideways at c·kx/|k| until it leaves through a wall (smoothed edges). Large-angle waves leave first and the focus fades, as in the paper's Fig. 8(f). Dashed lines show θdiv from eq. (20).

Box-size calculator (eqs. 19–21)

Uses the vf and k₀w₀ set above, in plasma units with ω₀/ωp = 10 (so w₀ = 1.00 c/ωp) and zprop = cΔtsurv.

Scaling arithmetic only, using the paper's estimates. It ignores everything except the transverse cell count, so it can exceed what a real run gains. The real benefit depends on the pulse parameters and interaction length. The paper's benchmark: savings of often one to two orders of magnitude.

The recipe

The paper closes with a concise recipe for robust PIC modelling of velocity-controlled ST drivers.

Specify the driver through a Maxwell-consistent spectral decomposition: a superposition of exact vacuum solutions on the light cone ∩ spectral plane. Build one in the hero ↑

Choose the k-space discretisation so that the repetition lengths exceed the longitudinal and transverse window. Replica lab ↑

Select (w₀, vf) with the resonance scaling of eq. (18) for strong and stable wake excitation. Wake tuner ↑

For long subluminal propagation, use wall injection instead of holding the full laterally expanding envelope in the box. Survival clock · wall lab ↑

How the paper's simulations were set up

OSIRIS, azimuthal-mode decomposition in (r, z–φ) geometry (quasi-3D), keeping modes m = 0 and m = 1.

Uniform plasma density n₀ after a 4 c/ωp vacuum-to-plasma ramp; 32 particles per cell.

Moving window at the speed of light in all simulations.

Figs. 5–6 scans: domain 100 × 20 (c/ωp)² with 10000 × 400 cells (Δz = 0.01, Δr = 0.05 c/ωp); perfectly matched layers at the transverse edge.

Driver polarised along Ey; flat-top envelope with 5 c/ωp sin² ramps and a 90 c/ωp flat top.

All results in the paper are particle-in-cell simulations; the Fig. 4 and Fig. 8 studies use ω₀/ωp = 10.

Source & credits

Chiara Badiali, Rafael Almeida, Thales Silva and Jorge Vieira, “Simulation design for subluminal velocity-controlled spatio-temporal drivers in laser–plasma wakefields”, Plasma Physics and Controlled Fusion 68 , 075016 (2026). doi:10.1088/1361-6587/ae8056

The article is open access under the Creative Commons Attribution 4.0 licence. Figures 2, 3, 4, 5, 6, 7 and 8 are reproduced from it. The Fig. 4 and Fig. 6 panels are cropped, and nothing else in the figures is modified.

The interactive models on this page are simplified illustrations: the hero, the two-regime comparison, and the replica and wall labs are live 2D-slab sums of exact vacuum plane waves using the paper's spectral construction (our own sampling and parameters, stated with each lab). The wake tuner evaluates the paper's paraxial formulas exactly (the wake itself is shown only in the paper's PIC panels), the waterfall is a kinematic sketch, and wall truncation is a ray-optics sketch. Quantitative claims (eq. 18 validation, nonlinear self-modulation, wall-injection benchmarks) are the paper's PIC results.

Topics in this iPaper

Physics
Laser–plasma accelerators have emerged as a promising route towards compact sources of relativistic

Introduction, page 1 (PDF page 2)

laser wakefield acceleration
Efficient wakefield generation relies on matching the laser-pulse dimensions to the characteristic plasma

Section 3.1, page 6 (PDF page 7)

particle in cell simulation
All simulations are performed with the OSIRIS azimuthal-mode decomposition algorithm

Appendix, page 12 (PDF page 13)

spatio temporal pulse shaping
whose intensity peak travels at a prescribed velocity that is decoupled from the group velocity of the

Introduction, page 1 (PDF page 2)

structured light
Keywords: structured light, spatio-temporal pulses, particle-in-cell simulations, laser-wakefield acceleration optimisation

Keywords, page 1 (PDF page 2)

Simulation design for subluminal velocity-controlled spatio-temporal drivers in laser–plasma wakefields — Interactive explanation | iPaper.ai