An extension of the Prouhet-Tarry-Escott problem to non-integer powers

David Treeby, Edward Wang · 2025

Read An extension of the Prouhet-Tarry-Escott problem to non-integer powers by David Treeby, Edward Wang (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-2503-19919.txt.

Introduction

David Treeby and Edward Wang take up a classic number puzzle in their 2025 paper and stretch it in a new way. The old puzzle asks for two sets of whole numbers whose sums match when each number is raised to a run of powers. The pair show that signed sums of such powers can be made as close to zero as one likes, even when the powers are not whole numbers.

Catalog record

CREATORDavid Treeby, Edward Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE6 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-19919

Browse the full library · About the Tower of Babel

Full text

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

An extension of the Prouhet-Tarry-Escott problem to non-integer powers
David Treeby, Edward Wang

Abstract
. We present an extension of the Prouhet-Tarry-Escott problem by demonstrating that signed sums of non-integer powers of consecutive integers can be made arbitrarily close to zero.



1.
Introduction

The Prouhet–Tarry–Escott problem
[
1
,
2
]
seeks to divide a set
S
of integers into two disjoint subsets
S_{1}
and
S_{2}
so that the equation

\sum_{x\in S_{1}}x^{j}=\sum_{x\in S_{2}}x^{j}

holds true for all integers
j
from
0
to
k
.
It is known (but insufficiently well-known) that this condition can be satisfied if
S
is any sequence of
2^{k+1}
consecutive integers. For example, if
x_{1},x_{2},\dots,x_{16}
is any sequence of sixteen consecutive integers,

\begin{array}[]{cccccccccc}+&x_{1}^{j}&-&x_{2}^{j}&-&x_{3}^{j}&+&x_{4}^{j}\\
-&x_{5}^{j}&+&x_{6}^{j}&+&x_{7}^{j}&-&x_{8}^{j}\\
-&x_{9}^{j}&+&x_{10}^{j}&+&x_{11}^{j}&-&x_{12}^{n}\\
+&x_{13}^{j}&-&x_{14}^{j}&-&x_{15}^{j}&+&x_{16}^{j}&=&0\end{array}

when
j
is equal to
0,1,2
or
3
.
This identity leads to the following remarkable fact. Given any set of sixteen cubes whose side lengths are consecutive integers, we can partition the cubes into two sets so that in each set:

(1)

the total number of cubes is the same,

(2)

the sum of their side lengths is the same,

(3)

the sum of their surface areas is the same,

(4)

the sum of their volumes is the same.

This beautiful result follows from a simple argument that we recreate below. Following this, we show how the result can be generalized to (certain) non-integer values of
j
. Indeed, such signed sums can be made as close to zero as desired so long as the smallest of these consecutive integers is sufficiently large. For example,

\displaystyle+\sqrt{2009}-\sqrt{2010}-\sqrt{2011}+\sqrt{2012}

\displaystyle-\sqrt{2013}+\sqrt{2014}+\sqrt{2015}-\sqrt{2016}

\displaystyle-\sqrt{2017}+\sqrt{2018}+\sqrt{2019}-\sqrt{2020}

\displaystyle+\sqrt{2021}-\sqrt{2022}-\sqrt{2023}+\sqrt{2024}

\displaystyle\approx-0.000000000163.

2.
Non-negative Integer powers

The results in this note stem from one simple insight: if
f
is a polynomial of degree
n
, then
f(x+c)-f(x)
is a polynomial of degree
n-1
, for any
c\neq 0
. This allows us to define a sequences of polynomials with diminishing degrees. Specifically, for the non-negative integer
j
we define
f^{j}_{n}(x)=f^{j}_{n-1}(x)-f^{j}_{n-1}(x+2^{n-1})

with initial term
f^{j}_{0}(x)=(x+1)^{j}
. To illustrate this, the next two terms are

\displaystyle f^{j}_{1}(x)=f^{j}_{0}(x)-f^{j}_{0}(x+1)

\displaystyle=+\,(x+1)^{j}-(x+2)^{j},

\displaystyle f^{j}_{2}(x)=f^{j}_{1}(x)-f^{j}_{1}(x+2)

\displaystyle=+\,(x+1)^{j}-(x+2)^{j}-(x+3)^{j}+(x+4)^{j}.

In general, we find that

f^{j}_{n}(x)=\sum_{k=1}^{2^{n}}{s_{k}(x+k)^{j}}

where
(s_{k})_{k=1}^{\infty}=(1,-1,-1,1,-1,1,1,-1,\dots)
is the
Thue-Morse
sequence on the integers
1
and
-1

[
3
]
. Also note that the degrees of the polynomials in the sequence diminish:

\displaystyle\deg f_{0}^{j}

\displaystyle=j,

\displaystyle\deg f_{1}^{j}

\displaystyle=j-1,

\displaystyle\deg f_{2}^{j}

\displaystyle=j-2,

\displaystyle\mathmakebox[\widthof{{}={}}][c]{\vdots}

\displaystyle\deg f_{j}^{j}

\displaystyle=0.

We collect these observations in the following theorem.

Theorem 1
.

The degree of
f^{j}_{n}
is
j-n
. In particular,
f^{j}_{j}
is a non-zero constant polynomial, and
f^{j}_{n}
is the zero polynomial provided
j<n
.

The result given in the introduction then follows from Theorem
1
by letting
n=4
and noting that
f_{4}^{0},f_{4}^{1},f_{4}^{2}
and
f_{4}^{3}
are all equal to the zero polynomial.

3.
Non-integer powers

The previous theorem shows that
f^{j}_{n}(x)=0
for all
x
given non-negative integers
n
and
j
satisfying
n>j
. Our aim is to generalise Theorem
1
, yielding a result that is applicable when
j
is not an integer. The desired generalization is stated in Theorem
3
. To prove this, we will require the following famous result, the proof of which can be found in most analysis textbooks
[
4
]
.

Theorem 2

(Newton’s generalized binomial theorem)
.

If
j\geq 0
is not an integer and
-1<x<1
, then

(1+x)^{j}=\sum_{i=0}^{\infty}\binom{j}{i}x^{i}.

Any proof of this theorem will essentially rest on the fact that the difference between
(1+x)^{j}
and the partial sum for the series can be estimated:

(1)

(1+x)^{j}-\sum_{i=0}^{n}\binom{j}{i}x^{i}=\sum_{i=n+1}^{\infty}\binom{j}{i}x^{i}=O(x^{n+1}).

Theorem 3
.

If
n
is a non-negative integer and
j\in[0,n)
, then
f^{j}_{n}(x)\to 0
as
x\to\infty
.

Proof.

Suppose
j\in[0,n)
is a real number. Using the binomial theorem, we find that

\displaystyle f^{j}_{n}(x)

\displaystyle=\sum_{k=1}^{2^{n}}{s_{k}(x+k)^{j}}

\displaystyle=x^{j}\sum_{k=1}^{2^{n}}{s_{k}\left(1+\frac{k}{x}\right)^{j}}

\displaystyle=x^{j}\sum_{k=1}^{2^{n}}s_{k}\left[\sum_{i=0}^{\infty}{\binom{j}{i}\left(\frac{k}{x}\right)^{i}}\right].

As one of the summations is finite, we can reverse the order of summation without need for caution. Moreover,
f^{i}_{n}(0)=\sum_{k=1}^{2^{n}}{s_{k}k^{i}}
. Therefore

\displaystyle f^{j}_{n}(x)

\displaystyle=x^{j}\sum_{i=0}^{\infty}{\frac{1}{x^{i}}\binom{j}{i}\left[\sum_{k=1}^{2^{n}}s_{k}k^{i}\right]}

\displaystyle=x^{j}\sum_{i=0}^{\infty}{\frac{1}{x^{i}}\binom{j}{i}f^{i}_{n}(0).}

Crucially, Theorem
1
ensures that
f^{i}_{n}(0)=0
for all integers
i<n
. Using this fact, and then reversing the order of summation once more gives

\displaystyle f^{j}_{n}(x)

\displaystyle=x^{j}\sum_{i=n}^{\infty}{\frac{1}{x^{i}}\binom{j}{i}f^{i}_{n}(0)}

\displaystyle=x^{j}\sum_{i=n}^{\infty}{\binom{j}{i}\frac{1}{x^{i}}\sum_{k=1}^{2^{n}}{s_{k}k^{i}}}

\displaystyle=x^{j}\sum_{k=1}^{2^{n}}s_{k}\left[\sum_{i=n}^{\infty}{\binom{j}{i}\left(\frac{k}{x}\right)^{i}}\right].

The term in parentheses is the remainder term in a (generalized) binomial expansion as seen in (
1
). Moreover, as the outer summation is finite we conclude that

f^{j}_{n}(x)=x^{j}O(1/x^{n})=O(x^{j-n}).

Therefore, this shows that
f^{j}_{n}(x)\to 0
as
x\to\infty
provided
n>j
.
∎

In Figure
1
we can observe the behavior of
f^{j}_{3}(x)
for
j\in[0,3]
.

j
0
1
2
3
f^{j}_{3}(x)
x=1
x=2
x=3
x=4

Figure 1.
f^{3}_{j}(x)\to 0
as

Read the complete text