MomentBoostedKernel #
Ported from scottnarmstrong/Superdiffusion at 7f20476385d4af418f110303085ad0740c9eba46. Module imports, support namespace and local option directives adapted for Lean Pool.
Internal moment-boosted tail kernel for stream-increment masses #
This module is internal probability infrastructure for the proof of
e.kl.bounds.large (ABK26). It is not a source-facing replacement for that
node and does not state a large-cube conclusion.
The usual conversion from Gamma_sigma moment growth to a weak-tail witness
throws away the factor sigma⁻¹ available when the Markov exponent is chosen
to be t^sigma. Retaining that factor gives the one-sided tail
exp (t^sigma / (2 sigma))⁻¹.
For sigma = 2 / p, its scale is a fixed base to the power p / 2, rather
than the forbidden p^12 inflation from the generic Gamma_sigma triangle
route. The later finite-range colored truncated-MGF argument consumes both
the strengthened tail and the genuine centered second-moment estimate proved
here.
No frozen declaration is imported.
References:
- ABK26 and Proposition
p.concentration.
The strengthened one-sided tail naturally supplied by all-real
Gamma_sigma moment growth. It is an internal kernel, not the source's
Gamma_sigma tail function.
Equations
Instances For
A strengthened kernel tail controls the ordinary Gamma_sigma tail at a
fixed factor-four enlargement whenever 0 < sigma ≤ 1. This conversion is
used only for the bounded partition-window branch; the finite-family route
retains the sharper kernel until after coloring.
All-real Gamma_sigma moment growth gives the strengthened one-sided
tail at scale 2^(1/sigma) M.
The actual local stream-increment mass has the strengthened centered
one-sided tail, uniformly in the observation scale l.
The same local mass has a genuine centered second-moment scale. This is the variance input to the finite-range truncated-MGF proof, not an extra premise of the source-facing large-cube statement.