derivation.md

back to table · edit · history · where entries came from · files · download

2584 bytes, as of the version from 2026-10-31 12:30 (current). Recorded here, not run.

# Formula and endpoint notes for T637

The shared convention is the conservative limit c -> 1 with integer spatial
dimension d, independent exponential free paths of mean one, and isotropic
scattering. The d=1 rod is defined by the two directions, not by extending the
Gamma-density formula to its singular endpoint.

In d'Eon--McCormick (2019), the dispersion equation has the expansion
Lambda(z) = 1-c-c/(d z^2)+O(z^-4). Consequently the positive discrete root
satisfies nu_0 sqrt(1-c) -> 1/sqrt(d). Equation (68) therefore yields
H_d(z) = sqrt(d)(1+z) exp[-integral_0^1 theta_d(t)/(t+z) dt/pi].
The first term in equation (191), nu_0 log((nu_0+1)/(nu_0-1))/2,
tends to one. Its integral tends to integral_0^1 theta_d(t) dt/pi,
which gives the generator's distance formula. The phase starts at zero and
stays in [0,pi]. atan2 with a nonnegative imaginary part retains that branch.

For 0<t<1, the imaginary part is positive. The phase extends continuously
to pi/2 at t=1 when d=2, and to pi when d>=3. These endpoint values are
used only when the numerical argument reaches the endpoint. The integrals
are evaluated over the full finite interval with tanh-sinh quadrature;
there is no manually cut-off tail. None of this supplies a rigorous bound on
the quadrature discretization error. Agreement at higher working precision
and with a separately transformed integral is the stated accuracy evidence.

For the independent equation (176), put t=tan(u), 0<=u<=pi/2, and write
F=2F1(1/2,1;d/2;-t^2), A=2F1(3/2,1;d/2+1;-t^2), and
B=2F1(5/2,1;d/2+2;-t^2). The hypergeometric series gives
1-F=t^2 A/d and A-1=-3t^2 B/(d+2). Thus the apparently singular factor
3+d/t^2-1/(1-F) equals 3-3d B/((d+2)A). The identity first follows near
zero and continues along the negative real hypergeometric argument, which
meets no branch cut. The rewritten factor tends to 6/(d+2) at t=0 and
to 2 at infinity. This cancellation is performed algebraically, before any
floating-point evaluation.

The separate angular-moment check uses mu=sin(u). Then
G_d(mu) dmu = [2 Gamma(d/2)/(sqrt(pi) Gamma((d-1)/2))] cos(u)^(d-2) du.
This removes the integrable square-root singularity in Flatland. The
nonzero controls alpha_0=2 and alpha_1=2/sqrt(d) precede comparison of
sqrt(d) alpha_2/2 with each distance. This quadrature uses float64 and its
measured discrepancies are only normalization/identity checks; they are not
claimed as support for the 100 stored digits.

Instruction manifest: ../20261031T110316Z-build-skills/manifest.json.
The two archived instruction digests are recorded in skill-evidence.json.