Bernoulli Partitions

Thomas Curtright · 2025

Read Bernoulli Partitions by Thomas Curtright (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-09633.txt.

Introduction

Thomas Curtright shows that the Bernoulli numbers, a famous sequence in mathematics, can each be split into a finite sum of same-signed rational numbers that march in one direction. The paper appeared in two thousand twenty-five. The hint comes from scale invariant scattering, an unexpected physical source. Some properties of the parts are discussed, especially for large indices.

Catalog record

CREATORThomas Curtright
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE17 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-09633

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bernoulli Partitions Thomas Curtright Abstract Scale invariant scattering suggests that all Bernoulli numbers B 2 ​ n B_{2n} can be naturally partitioned, i.e., written as particular finite sums of same-signed, monotonic, rational numbers. Some properties of these rational numbers are discussed here, especially in the limit of large n n . In tribute to Luca Mezincescu (1946-2025) In this note I extend some earlier observations [ 1 ] about Bernoulli numbers as obtained in the context of computing the scattering amplitude for a scale invariant potential [ 2 , 3 ] . In two spatial dimensions (2D), when computed using quantum mechanics, non-relativistic scattering by a repulsive inverse square potential V=\kappa/r^{2} ( \kappa>0 ) results in a simple form for the integrated cross section, \sigma=\int_{0}^{2\pi}\left(\frac{d\sigma}{d\theta}\right)d\theta . For a mono-energetic beam of mass m particles the 2D result is \sigma=2\pi^{2}m\kappa/\left(\hbar^{2}k\right) where the incident energy is E=\hbar^{2}k^{2}/\left(2m\right) . This result follows from a straightforward application of 2D phase-shift analysis for the potential V=\kappa/r^{2} upon realizing a remarkable identity involving the sinc function, \operatorname{sinc}\left(z\right)\equiv\sin\left(z\right)/z . A succinct form of the identity in question is [ 4 ] 1=\frac{\sin\left(\pi x\right)}{\pi x}+2\sum_{l=1}^{\infty}\frac{\left(-1\right)^{l}\sin\left(\pi\sqrt{l^{2}+x^{2}}\right)}{\pi\sqrt{l^{2}+x^{2}}} (1) All higher powers of x cancel when terms on the RHS are expanded as series in x^{2} , as a consequence of familiar \zeta\left(2n\right) exact values for integer n>0 , as shown in [ 2 ] . Upon expressing the sinc function in terms of spherical Bessel functions, and then using series representations for Bessel functions [ 5 ] in terms of Bernoulli numbers B_{n} , the identity ( 1 ) leads directly to a set of finite sum identities involving those numbers. Namely, 1=\left(-1\right)^{n+1}\left(4n+2\right)\sum_{k=0}^{n}\frac{\left(2n\right)!}{n!k!\left(n-k\right)!}\left(\frac{B_{n+k+1}}{n+k+1}\right)\text{ \ \ for integer }n\geq 1\text{.} (2) This finite sum involves trinomial coefficients as well as divided Bernoulli numbers , \beta_{m}\equiv B_{m}/m . It is not difficult to check the validity of ( 2 ) using various expressions of the Bernoulli numbers as finite, alternating sums [ 6 ] . Recall the usual phases are given by B_{2n}=\left(-1\right)^{n+1}\left|B_{2n}\right| as well as B_{2n+1}=0 for n=1,2,\cdots . Also recall the well-known relation \left|B_{2n}\right|=\frac{2\left(2n\right)!}{\left(2\pi\right)^{2n}}~\zeta\left(2n\right)\text{ \ \ for \ \ }n=1,2,\cdots (3) Thus, given an identity for \left|B_{2n}\right| a similar identity for \zeta\left(2n\right) follows immediately. As quipped previously [ 1 ] , if encountered as graffiti on the stones of a bridge (say, in Ireland) either ( 1 ) or its companion identity ( 2 ) might cause nothing more than a raised eyebrow in passing. Perhaps justifiably so. However, upon inverting the linear relations in ( 2 ) to obtain expressions for each individual Bernoulli number, the results are more striking: The Bernoulli numbers B_{2n} for n\geq 2 are naturally partitioned . That is to say, each B_{2n} is given by an interesting sum of n-1 monotonic, same-signed, rational numbers. Unlike many well-known representations of B_{2n} , here the terms in the finite sums do not alternate in sign [ 7 ] . For example, the first ten partitions are given by \left(\begin{array}[c]{c}\left|B_{2}\right|\\ \left|B_{4}\right|\\ \left|B_{6}\right|\\ \left|B_{8}\right|\\ \left|B_{10}\right|\\ \left|B_{12}\right|\\ \left|B_{14}\right|\\ \left|B_{16}\right|\\ \left|B_{18}\right|\\ \left|B_{20}\right|\end{array}\right)=\left(\begin{array}[c]{c}\frac{1}{6}\\ \frac{1}{30}\\ \frac{1}{42}\\ \frac{1}{30}\\ \frac{5}{66}\\ \frac{691}{2730}\\ \frac{7}{6}\\ \frac{3617}{510}\\ \frac{43\,867}{798}\\ \frac{174\,611}{330}\end{array}\right)=\left(\begin{array}[c]{l}\frac{1}{6}\\ \frac{1}{30}\\ \frac{1}{60}+\frac{1}{140}\\ \frac{1}{45}+\frac{1}{105}+\frac{1}{630}\\ \frac{1}{20}+\frac{3}{140}+\frac{1}{252}+\frac{1}{2772}\\ \frac{1}{6}+\frac{1}{14}+\frac{17}{1260}+\frac{1}{693}+\frac{1}{12\,012}\\ \frac{691}{900}+\frac{691}{2100}+\frac{59}{945}+\frac{41}{5940}+\frac{5}{10\,296}+\frac{1}{51\,480}\\ \frac{14}{3}+\ 2\ +\frac{359}{945}+\frac{8}{189}+\frac{4}{1287}+\frac{1}{6435}+\frac{1}{218\,790}\\ \frac{3617}{100}+\frac{10\,851}{700}+\frac{1237}{420}+\frac{217}{660}+\frac{293}{12\,012}+\frac{1}{780}+\frac{7}{145\,860}+\frac{1}{923\,780}\\ \frac{43\,867}{126}+\frac{43\,867}{294}+\frac{750\,167}{26\,460}+\frac{6583}{2079}+\frac{943}{4004}+\frac{1129}{90\,090}+\frac{217}{437\,580}+\frac{2}{138\,567}+\frac{1}{3879\,876}\end{array}\right) In the finite sequence of terms that sum to give \left|B_{2n}\right| for n>2 , as displayed above, obviously the second number in the sequence is just 3/7 times the first. Less obviously, each term in the sequence for \left|B_{2n}\right| is greater than the subtotal of all the smaller terms in that same sequence. The general result for \left|B_{2n}\right| is obtained by writing ( 2 ) as an infinite matrix equation, \boldsymbol{1}=\boldsymbol{M\cdot B} , where \boldsymbol{B} is an infinite column of the even index Bernoulli numbers, \boldsymbol{1} is an infinite column of 1 s, and \boldsymbol{M}_{m,n}=2\left(-1\right)^{m+1}\binom{2n-1}{m}\binom{2m+1}{2n} . Computing the inverse for the triangular matrix \boldsymbol{M} then gives \boldsymbol{B=M}^{-1}\boldsymbol{\cdot 1} . All terms in a given row of the triangular matrix \boldsymbol{M}^{-1} have the same sign. The ordered terms in the sums above are just the unsigned entries in \boldsymbol{M}^{-1} for the n th row. For example, the first six rows and columns of the triangular matrices \boldsymbol{M} and \boldsymbol{M}^{-1} are given by \boldsymbol{M=}\left(\begin{array}[c]{cccccc}6&0&0&0&0&0\\ 0&-30&0&0&0&0\\ 0&70&140&0&0&0\\ 0&0&-840&-630&0&0\\ 0&0&924&6930&2772&0\\ 0&0&0&-18\,018&-48\,048&-12\,012\end{array}\right),\ \ \boldsymbol{M}^{-1}=\left(\begin{array}[c]{cccccc}\frac{1}{6}&0&0&0&0&0\\ 0&-\frac{1}{30}&0&0&0&0\\ 0&\frac{1}{60}&\frac{1}{140}&0&0&0\\ 0&-\frac{1}{45}&-\frac{1}{105}&-\frac{1}{630}&0&0\\ 0&\frac{1}{20}&\frac{3}{140}&\frac{1}{252}&\frac{1}{2772}&0\\ 0&-\frac{1}{6}&-\frac{1}{14}&-\frac{17}{1260}&-\frac{1}{693}&-\frac{1}{12\,012}\end{array}\right) hence the previous partitions of \left|B_{2n}\right| for n=1 to 6 . Various things can be said in general about the entries in \boldsymbol{M}^{-1} , e.g., the diagonal is \left|\boldsymbol{M}_{n,n}^{-1}\right|=\frac{\left(n!\right)^{2}}{\left(2n+1\right)!} , the first sub-diagonal is \left|\boldsymbol{M}_{n\geq 2,n-1}^{-1}\right|=\frac{1}{6}\left(n-2\right)\frac{n!\left(n-1\right)!}{\left(2n-1\right)!} , the second sub-diagonal is \left|\boldsymbol{M}_{n\geq 3,n-2}^{-1}\right|=\frac{7}{360}\left(n-\frac{8}{7}\right)\left(n-3\right)\frac{n!\left(n-2\right)!}{\left(2n-3\right)!} , etc. Since the first columns of \boldsymbol{M} and \boldsymbol{M}^{-1} are null except for the top entry, such that \left|B_{2}\right|=1/6 and \left|B_{4}\right|=1/30 are both without any partitioning, it makes sense to begin the set of partitions for \left|B_{2m}\right| with those for m=2 . One way to express the general result is then \left|B_{2m}\right|=\sum_{n=2}^{m}b_{m}\left(n\right)\ ,\ \ \ m\geq 2 (4) where b_{m}\left(n\right) is the non-zero n^{th} unsigned column entry in the m^{th} row of the aforementioned triangular matrix \boldsymbol{M}^{-1} . For example, b_{2}\left(2\right)=1/30 , b_{3}\left(2\right)=1/60 , and b_{3}\left(3\right)=1/140 , with no other non-zero entries in the 2nd and 3rd rows of \boldsymbol{M}^{-1} . So each \left|B_{2m}\right| is given as a sum of m-1 monotonically falling, positive rational numbers as tabulated above for 2\leq m\leq 10 . For larger values of m and n , as presented previously in [ 1 ] , b_{m}\left(n\right)=\frac{m!n!}{\left(2n+1\right)!}\left(n-1\right)q_{m-1-n}\left(m\right)\ ,\ \ \ m\geq 2 (5) where again m is the row index and n is the column index of \boldsymbol{M}^{-1} . Here q_{-1}\left(n\right)\equiv\frac{1}{n-1} while the q_{l}\left(n\right) , with l\geq 0 are l th order polynomials in powers of n . These polynomials may be obtained sequentially [ 1 ] by iteration of q_{l}\left(n\right)=\frac{\left(-1\right)^{l}\left(n-l-3\right)!}{\left(2l+3\right)!\left(n-2l-3\right)!}+\sum_{j=0}^{l-1}\frac{\left(-1\right)^{l+j+1}\left(n+j-l-1\right)!}{\left(2l+1-2j\right)!\left(n+2j-2l-1\right)!}~q_{j}\left(n+j-l\right) (6) A few explicit examples are [ 8 ] q_{-1}\left(n\right)=\frac{1}{n-1}\ ,\ \ \ q_{0}\left(n\right)=\frac{1}{6}\ ,\ \ \ q_{1}\left(n\right)=\frac{7}{360}~n-\frac{1}{45}\ ,\ \ \ q_{2}\left(n\right)=\frac{31}{15\,120}~n^{2}-\frac{89}{15\,120}~n+\frac{1}{315}\ ,\ \ \ \text{etc.} (7) For Bernoulli partitions the q_{l}\left(n\right) polynomials are only needed for n\geq l+3 . Now it so happens, for any fixed column n , numerical calculations [ 9 ] reveal that b_{m}\left(n\right) grows without bound as m increases with asymptotic behavior proportional to that of \left|B_{2m}\right|\sim\frac{2\left(2m\right)!}{\left(2\pi\right)^{2m}} . Moreover, as the row index m increases, there is a monotonic approach to an asymptotic ratio \frac{b_{m}\left(n\right)}{\left|B_{2m}\right|}\underset{m\rightarrow\infty}{\sim}a\left(n\right) (8) where a\left(n\right) is a monotonically falling function of the column index for n\geq 2 . For example, numerically to 10 digits: a\left(2\right)=0.657\,973\,626\,7 , a\left(3\right)=0.281\,988\,697\,2 , a\left(4\right)=0.0535\,819\,841\,1 , a\left(5\right)=0.005\,985\,334\,466 , etc. When augmented by https://oeis.org/A182448 , these numerical revelations lead to the following. Conjecture 1 The asymptotic ratios are a\left(n\right)=p_{n}\left(\pi^{2}\right) for n\geq 2 where p_{n}\left(x\right) are polynomials: p_{n}\left(x\right)=\frac{1}{4n^{2}-1}\sum_{k=0}^{\left\lfloor n/2\right\rfloor-1}\frac{\left(-4\right)^{k}x^{k+1}}{\left(2k+1\right)!}\frac{\Gamma\left(n-1\right)\Gamma\left(2n-2-2k\right)}{\Gamma\left(n-1-2k\right)\Gamma\left(2n-2\right)} (9) with \left\lfloor\ldots\right\rfloor representing the floor function. 1 1 1 For n\geq 2 , note that \sum_{k=0}^{\left\lfloor n/2\right\rfloor-1}\frac{\left(-4\right)^{k}x^{k+1}}{\left(2k+1\right)!}\frac{\Gamma\left(n-1\right)\Gamma\left(2n-2-2k\right)}{\Gamma\left(n-1-2k\right)\Gamma\left(2n-2\right)}=\sum_{k=0}^{\left\lfloor\left(n-1\right)/2\right\rfloor}\frac{\left(-2\right)^{k}x^{k+1}}{\left(2k+1\right)!}\left({\displaystyle\prod_{l=1}^{k}}\frac{\left(n-k-1-l\right)}{\left(2n-1-2l\right)}\right) as used in earlier versions of this paper, with the empty product equal to 1 by convention. For example, numerically to 16 digits: \ p_{2}\left(\pi^{2}\right)=\frac{1}{15}\pi^{2}=0.657\,973\,626\,739\,290\,6\cdots , p_{3}\left(\pi^{2}\right)=\frac{1}{35}\pi^{2}=0.281\,988\,697\,173\,981\,7\cdots , p_{4}\left(\pi^{2}\right)=\frac{1}{63}\pi^{2}\left(1-\frac{1}{15}\pi^{2}\right)=0.0535\,819\,841\,082\,941\,2\cdots , and p_{5}\left(\pi^{2}\right)=\frac{1}{99}\pi^{2}\left(1-\frac{2}{21}\pi^{2}\right)=0.005\,985\,334\,466\,027\,733\cdots , in complete agreement with the previous numerical results. Remark 1 Here are suggestions for a possible proof of the conjecture. In the large m limit, keeping only leading terms, recast the iteration ( 6 ) as a homogeneous second-order difference equation in n for a_{n}\equiv p_{n}\left(x\right) , to obtain \left(2n+1\right)p_{n}\left(x\right)=\left(2n-3\right)p_{n-1}\left(x\right)-\frac{x}{\left(2n-1\right)}~p_{n-2}\left(x\right) (10) where x is a parameter related to the normalization of the solution. With chosen

Read the complete text