An executable experiment in finite-support Schrödinger transport. Edge widths show the mass that the numerical solver assigns between atoms. Requested endpoints are compared with the mass actually returned.
A continuous view of weighted endpoints
Each moving disk carries one pair’s exact coupling weight; disk area shows mass. Straight paths are drawing geometry (C). The slider selects elapsed time t. Drawing progress is s(t) = (1−exp(−t))/(1−exp(−H)), the normalized probability of at least one event in A’s rate-one Poisson clock, with H = 1. This is an explicit display time warp. The A comparison uses the new jump generator T − I; its law is shown separately. Neither drawing paths nor equal endpoint laws imply stationary intermediate bridge marginals.
Exact pair weights and current drawing coordinates
Numerical evidence; no formal run certificate. B explicitly uses Q = K − I, H = 1, f(t) = f(0) exp(tQ), g(t) = exp((H−t)Q)v, p(t) = f(t)g(t). Factors are evaluated directly when seeking. Download continuous artifact and provenance. The original coupling inspector follows below.
Hover an edge to inspect its source, target and mass. Filtering changes visibility only.
Requested → observed
Atom
Source requested
Target requested
Target observed
Absolute error
What this establishes
The Lean library proves conditional finite-support statements: under explicit positivity, balance and symmetry assumptions, the constructed kernel has the stated stationarity or reversibility properties. Floating-point scaling is not a Lean proof of those premises. General continuous bridge existence, scaling-factor existence, KL optimality and solver correctness are outside the formal claim.