RelayFinite
Finite-size CV-MDI: estimated coordinates, the four widths, and Eq. (23)
11/11 passed in 2ms
| Test | What it does | Result |
|---|---|---|
test_excess_bridges | The (v_q, v_p) of cvmdi_excess rebuild chi to 1e-12, and pure loss puts both at zero. | ✅ pass |
test_finite_classes | The coherent and collective classes are refused by name and an unrecognised one is a ValueError. | ✅ pass |
test_finite_converges | cvmdi_finite climbs with the block size to within 2e-3 of the key rounds' share of the asymptotic rate at N = 1e14. | ✅ pass |
test_finite_extremal | The paper's two-mode optimum lowers both v_q and v_p, and a thousandth past it is not a state. | ✅ pass |
test_finite_headline | At the paper's stated parameters the block sizes it names each give above 1e-2 bit, a comparison against a sentence and not against a published number. | ✅ pass |
test_finite_pessimistic | Over 243 combinations of modulation, arm split and correlation sign the key never exceeds the asymptotic rate times the key share. | ✅ pass |
test_finite_unclamped | A block that cannot distil returns a finite negative number, not zero. | ✅ pass |
test_finite_worst | Eq. (23) takes tau_a, tau_b 6.5 sigma low and v_q, v_p 6.5 sigma high, less the two-term Delta(n) over the key rounds. | ✅ pass |
test_gauss_closed | cvmdi_gauss matches the published Eqs. (24) to (26), conditioned and written out here, to 1e-11. | ✅ pass |
test_gauss_matches | cvmdi_gauss reproduces cvmdi_point to 2e-9 across helping and harming correlations. | ✅ pass |
test_width_matches | cvmdi_width matches the published Eqs. (14) to (20) to 1e-13 and falls as 1/sqrt(m). | ✅ pass |