This is a 2D-native pair to the 3D orbital conjunction simulator, computing the identical formulas a different way. Instead of rendering one static B-plane geometry, the top panel draws random samples from the actual 2D Gaussian position-uncertainty distribution the covariance describes, and counts what fraction land inside the combined hard-body circle — a direct empirical (Monte Carlo) estimate of Pc that should converge to the closed-form Alfano/Foster value as more samples accumulate. The bottom panel sweeps the same closed-form formula across the full 1–72 h horizon range at the current d₀/σ₀/ΔV to show the trade a single slider position can't: waiting longer grows the tracking uncertainty σ(t) (which by itself lowers Pc) but also gives a constant-velocity burn more time to open the miss distance (which lowers it further) — so the curve is monotonically informative in a way the 3D view's fixed-instant readout is not.
σ(t) = σ0 · √(1 + (t/24h)²) — OD uncertainty grows with propagation time
Δd = ΔV · t — a lead-time burn shifts the miss distance
Pc ≈ R² / (2σ²) · exp(−d²/2σ²) — Alfano/Foster small-Pc approximation
Pc_MC = hits / trials, (dx,dy) ~ N(d_eff, σ) × N(0, σ) — independent empirical check
- Top panel — each frame draws a batch of new (dx, dy) samples from the relative-position Gaussian, colors them by hit/miss against the R = 20 m hard-body circle, and accumulates a running hit-rate. Sliders reset the running sample so the estimate always reflects the current inputs.
- Bottom panel — a log-scale Pc vs. horizon curve, current horizon marked with a dot, operational 10⁻⁴ maneuver threshold marked with a dashed line.
- Monte Carlo needs roughly 1/Pc trials before its estimate is statistically meaningful — for very small Pc the sampler will correctly read 0 hits for a while; that itself is informationally consistent with a vanishingly small collision probability.