Baricz–Pogány's asymptotic-coefficient positivity conjecture

What Lean checks

Original factorial products, formal division, all-degree strict positivity and the explicit lower bound are proved. The source coefficient recurrence uniquely determines the same sequence; its inequality is now an explicit endpoint. Strict compilation, an all-owned audit of 65 declarations/55 theorem declarations and all seven required endpoints, and stock-kernel replay passed. Local receipt; combined repository gate pending. Source-access and priority limits are unchanged.

Status and source

This is a complete proof of the unnumbered conjecture in Section 4.4 of Á. Baricz and T. K. Pogány, Probabilistic and Analytical Aspects of the Symmetric and Generalized Kaiser–Bessel Window Function, Constructive Approximation 58 (2023), 713–783, DOI 10.1007/s00365-023-09627-3. Independent mathematical reviews have passed. The bounded prior-resolution audit and its access limits are recorded in NOVELTY.md.

The source defines coefficients by the large-argument expansion

tIν(t)Iν+1(t)∼t∑n=0∞αn(ν)t−n,t→+∞,\frac{tI_\nu(t)}{I_{\nu+1}(t)} \sim t\sum_{n=0}^{\infty}\alpha_n(\nu)t^{-n}, \qquad t\to+\infty,

with α0(ν)=1\alpha_0(\nu)=1 and α1(ν)=ν+1/2\alpha_1(\nu)=\nu+1/2. It conjectures, for every integer n≥2n\geq2 and every real |ν|<1/2|\nu|<1/2, that

πn(ν):=αn(ν)−(ν+1/2)n>0.\pi_n(\nu):=\alpha_n(\nu)-(\nu+1/2)^n>0. \tag{1}

The proof below establishes the entire assertion. No finite-degree or fixed-parameter restriction is imposed.

1. The formal coefficient series

For real μ\mu, the source's equations (3.5)–(3.6) use

Iμ(t)∼et2πt∑n=0∞(−1)nan(μ)t−n,an(μ)=∏k=1n(4μ2−(2k−1)2)8nn!,a0(μ)=1.I_\mu(t)\sim\frac{e^t}{\sqrt{2\pi t}} \sum_{n=0}^{\infty}(-1)^n a_n(\mu)t^{-n}, \qquad a_n(\mu)=\frac{\prod_{k=1}^{n}(4\mu^2-(2k-1)^2)}{8^n n!}, \quad a_0(\mu)=1. \tag{2}

Put a=ν+1/2a=\nu+1/2, so 0<a<10<a<1, and define formal power series over the real numbers by

N(z)=∑n≥0cnzn,cn=(−1)nan(ν),D(z)=∑n≥0dnzn,dn=(−1)nan(ν+1).N(z)=\sum_{n\geq0}c_nz^n, \quad c_n=(-1)^na_n(\nu), \qquad D(z)=\sum_{n\geq0}d_nz^n, \quad d_n=(-1)^na_n(\nu+1). \tag{3}

Both constant coefficients are one. The source's recurrence in Section 4.4 is exactly formal division:

A(z):=∑n≥0αn(ν)zn=N(z)D(z).A(z):=\sum_{n\geq0}\alpha_n(\nu)z^n=\frac{N(z)}{D(z)}. \tag{4}

Indeed, comparing the coefficient of degree nn in DA=NDA=N gives

αn(ν)=cn−∑m=0n−1αm(ν)dn−m,\alpha_n(\nu)=c_n- \sum_{m=0}^{n-1}\alpha_m(\nu)d_{n-m}, \tag{5}

which is the defining recurrence printed in the source. Since d0=1d_0=1, this uniquely determines the coefficient sequence. Thus no analytic convergence assumption about the asymptotic series is needed to identify the coefficients in this proof.

2. Coefficient signs

Factoring the product in (2) gives

cn=∏k=1n(k−a)(k+a−1)2nn!,cn=(n−a)(n+a−1)2ncn−1(n≥1).c_n=\frac{\prod_{k=1}^{n}(k-a)(k+a-1)}{2^n n!}, \qquad c_n=\frac{(n-a)(n+a-1)}{2n}c_{n-1}\quad(n\geq1). \tag{6}

Every factor in the first product is strictly positive because 0<a<10<a<1. Hence cn>0c_n>0 for every n≥0n\geq0.

The analogous product for the denominator is

dn=∏k=1n(k−a−1)(k+a)2nn!.d_n=\frac{\prod_{k=1}^{n}(k-a-1)(k+a)}{2^n n!}.

For n≥1n\geq1, division of these products telescopes:

dncn=∏k=1nk−a−1k−a∏k=1nk+ak+a−1=−an−an+aa=−n+an−a.\frac{d_n}{c_n} =\prod_{k=1}^{n}\frac{k-a-1}{k-a} \prod_{k=1}^{n}\frac{k+a}{k+a-1} =\frac{-a}{n-a}\frac{n+a}{a} =-\frac{n+a}{n-a}. \tag{7}

In particular, dn<0d_n<0 for every positive nn. Define

B(z)=∑n≥1bnzn,bn=−dn=n+an−acn>0.B(z)=\sum_{n\geq1}b_nz^n, \qquad b_n=-d_n=\frac{n+a}{n-a}c_n>0.

Then

D(z)=1−B(z),1D(z)=∑j≥0B(z)j.D(z)=1-B(z),\qquad \frac{1}{D(z)}=\sum_{j\geq0}B(z)^j. \tag{8}

The last equality is valid in the formal power-series ring: BB has zero constant coefficient, so each coefficient on the right is a finite sum. Every coefficient of D−1D^{-1} is nonnegative, and its constant coefficient is one.

3. The coefficient identity

Equations (6)–(7) imply, for n≥1n\geq1, that

dn=cn−(n+a−1)cn−1.d_n=c_n-(n+a-1)c_{n-1}. \tag{9}

For clarity, substituting (6) into the right side gives

((n−a)(n+a−1)2n−(n+a−1))cn−1=−(n+a)(n+a−1)2ncn−1=−n+an−acn.\left(\frac{(n-a)(n+a-1)}{2n}-(n+a-1)\right)c_{n-1} =-\frac{(n+a)(n+a-1)}{2n}c_{n-1} =-\frac{n+a}{n-a}c_n.

Comparing coefficients in (9) proves the formal identity

D(z)=(1−az)N(z)−z2N′(z).D(z)=(1-az)N(z)-z^2N'(z). \tag{10}

Here N′N' is the formal derivative. Equation (10) was proved directly from coefficient products. It does not differentiate an asymptotic remainder.

Combining (4) and (10) now gives

∑n≥0(αn(ν)−an)zn=N(z)D(z)−11−az=(1−az)N(z)−D(z)(1−az)D(z)=z2N′(z)(1−az)(1−B(z)).\begin{aligned} \sum_{n\geq0}\left(\alpha_n(\nu)-a^n\right)z^n &=\frac{N(z)}{D(z)}-\frac{1}{1-az}\\ &=\frac{(1-az)N(z)-D(z)}{(1-az)D(z)}\\ &=\frac{z^2N'(z)}{(1-az)(1-B(z))}. \end{aligned} \tag{11}

All denominators in (11) have constant coefficient one, so every division is well-defined formally.

4. Positivity in every degree

Since a>0a>0, the series

11−az=∑j≥0ajzj\frac{1}{1-az}=\sum_{j\geq0}a^jz^j

has nonnegative coefficients and constant coefficient one. Equation (8) gives the same facts for (1−B(z))−1(1-B(z))^{-1}.

The remaining factor in (11) is

z2N′(z)=∑n≥2(n−1)cn−1zn.z^2N'(z)=\sum_{n\geq2}(n-1)c_{n-1}z^n.

Its coefficient of degree nn is strictly positive for every n≥2n\geq2. Multiplication by the other two factors cannot decrease any coefficient, and their product has constant coefficient one. Therefore (11) proves the explicit lower bound

πn(ν)≥(n−1)cn−1=(n−1)∏k=1n−1(k−a)(k+a−1)2n−1(n−1)!>0(n≥2).\pi_n(\nu)\geq(n-1)c_{n-1} =\frac{(n-1)\prod_{k=1}^{n-1}(k-a)(k+a-1)}{2^{n-1}(n-1)!}>0 \qquad(n\geq2). \tag{12}

This proves the full conjecture for every permitted real parameter and every degree.

5. Scope and attribution

The argument is entirely coefficientwise. It does not assert convergence of the generally divergent asymptotic series or infer a pointwise inequality by summing that series. The full coefficient assertion (1) is the source's conjecture and is exactly what has been proved.

The large-argument Bessel expansion, formal division, and geometric-series identity are classical. No novelty is claimed for those tools. The source and prior-resolution audit are recorded separately in NOVELTY.md. No numerical computation or formal proof assistant is used in the proof.