Rational kernels
Abstract
Derivative certificates and estimates for rational kernels.
Theorem 1.1 (The rational derivative certificate).
Lean statement: D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_N
Proof. Machine-checked in Lean as D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_N (✓ std3). ∎
Source. Repository-derived.
Commentary.
For a positive scale a, the Nth derivative of t^(N+2)/(t^2+a^2) equals N! times one minus the real part of (1+i t/a)^(-(N+1)). The proof differentiates a complex rational function and takes its real part. Its recurrence separates a polynomial term from a simple pole.
Theorem 1.2 (Strict positivity on the positive ray).
Lean statement: D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_N_pos
Proof. Machine-checked in Lean as D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_N_pos (✓ std3). ∎
Source. Repository-derived.
Commentary.
When a and t are positive, the modulus of 1+i t/a is greater than one. Its negative power has real part strictly below one, so the Nth derivative in the certificate is strictly positive.
Theorem 1.3 (The phase and amplitude formula).
Lean statement: D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_succ
Proof. Machine-checked in Lean as D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_succ (✓ std3). ∎
Source. Repository-derived.
Commentary.
The derivative of order N+1 equals (N+1)!/a times (1+(t/a)^2)^(-(N+2)/2) times sin((N+2) arctan(t/a)). The scale factor is exactly a^(-1). This follows by differentiating the complex certificate and applying the polar form of a complex power.
Theorem 1.4 (A uniform negative phase).
Lean statement: D5/S3/Analytic/CothBernoulliDegree/Kernel.phase_bound
Proof. Machine-checked in Lean as D5/S3/Analytic/CothBernoulliDegree/Kernel.phase_bound (✓ std3). ∎
Source. Repository-derived.
Commentary.
For every real M at least five, the product of (1+(5/M)^2)^(-M/2) and sin(M arctan(5/M)) is at most -1/8. The arctangent lies above its chord on [0,1], while log(1+x^2) lies below its chord. These give a phase between 5 pi/4 and 5 and an amplitude at least sqrt(2)/8. The sine is at most -sqrt(2)/2 throughout that interval. Equality holds at M=5.
Theorem 1.5 (The reciprocal-power tail).
Lean statement: D5/S3/Analytic/CothBernoulliDegree/Kernel.tail_bound
Proof. Machine-checked in Lean as D5/S3/Analytic/CothBernoulliDegree/Kernel.tail_bound (✓ std3). ∎
Source. Repository-derived.
Commentary.
For every natural n at least two, the sum over natural k of 1/(k+2)^(2n+1) is at most 1/16. Each term is bounded by (1/(k+1)-1/(k+2))/16. The latter series telescopes; bounded nonnegative partial sums yield the asserted infinite-sum inequality.
References
- Truth anchor:
D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_N - Truth anchor:
D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_N_pos - Truth anchor:
D5/S3/Analytic/CothBernoulliDegree/Kernel.kernel_deriv_succ - Truth anchor:
D5/S3/Analytic/CothBernoulliDegree/Kernel.phase_bound - Truth anchor:
D5/S3/Analytic/CothBernoulliDegree/Kernel.tail_bound