Number tomography

A cute problem that my students got in a competition

This problem has appeared in many math competitions. It is quite interesting and connects to a lot of interesting mathematics.

For a collection (multiset) of numbers $A=\{\{a_1\leq a_2\leq\ldots\leq a_n\}\}$ one can compute the multiset of all the pairwise sums $a_i+a_j$ with $i\neq j$ of its elements $S=\{\{s_1\leq s_2\leq\ldots\leq s_m\}\}$, with $m=\frac{n(n-1)}{2}$. The main question is whether one can uniquely recover the multiset $A$ from the multiset $S$.

The process of obtaining $S$ from $A$ is like a tomography. For any choice of direction $i,j$, the superposition $a_i+a_j$ of the multiset $A$ along this direction is computed. The problem consists in reconstructing the original multiset $A$ from its tomography $S$. See, Radon transform and X-ray transform for reconstruction problems that are similar in spirit.

Claim 1:
When $n$ is not a power of $2$, then one can uniquely recover $A$ from $S$.

Proof of claim 1 [Symmetric polynomials]:1

Assume that $n$ is not a perfect power of $2$. Denote

\[\begin{align} P_{k} &= \sum_{i=1}^{m}s_i^k\\ p_k&=\sum_{i=1}^{n}a_i^k \end{align}\]

for $k=1,2,3,\ldots$ These are the power-sum symmetric polynomials in the variables $s_i$ and $a_i$, respectively.

Observe that the $s_i$ uniquely determine the $P_k$ and the $p_k$ uniquely determine the elementary symmetric polynomials in the $a_i$ via Girard-Newton formulae. In turn, the elementary symmetric polynomials in the $a_i$ uniquely determine these via Vieta’s formulas.

If we show that the $P_k$, $k=1,2,3,\ldots$, uniquely determine the $p_k$, then we are done.

\[\begin{align} P_k&=\sum_{i=1}^{m}s_i^k\\ &=\sum_{1\leq i<j\leq n}(a_i+a_j)^k\\ &=\frac{1}{2}\sum_{\substack{i,j=1\\i\neq j}}^{n}(a_i+a_j)^k\\ &=\frac{1}{2}\left(\sum_{i,j=1}^{n}(a_i+a_j)^k-\sum_{i=1}^{n}(2a_i)^k\right)\\ &=\frac{1}{2}\left(\sum_{i=0}^{k}\binom{k}{i}p_ip_{k-i}-2^kp_k\right)\\ &=\frac{1}{2}\left(2n-2^k\right)p_k - \frac{1}{2}\sum_{i=1}^{k-1}\binom{k}{i}p_ip_{k-i} \end{align}\]

these equations, for $k=1,2,3,\ldots$ determine $p_k$ from $P_k$ and the $p_i$, for $1\leq i\leq k-1$ since the coefficients $2n-2^k$ are never zero.

Proof of claim 1 [Generating polynomials]:2

Assume that $n$ that $B=\{\{b_1,b_2,\ldots,b_n\}\}$ is another multiset producing the same multiset of pairwise sums $S$. Define

\[\begin{align} f(x)&=x^{a_1}+x^{a_2}+\ldots+ x^{a_n}\\ g(x)&=x^{b_1}+x^{b_2}+\ldots+ x^{b_n} \end{align}\]

The condition on the pairwise sums implies that

\[\begin{align} f^2(x)-g^2(x)&=\left(\sum_{i=1}^{n}x^{2a_i}+2\sum_{1\leq i<j\leq n}x^{a_i+a_j}\right)-\left(\sum_{i=1}^{n}x^{2b_i}+2\sum_{1\leq i<j\leq n}x^{b_i+b_j}\right)\\ &=f(x^2)-g(x^2) \end{align}\]

Since $f(1)=g(1)=n$, then $x=1$ is a root of $f-g$. Observe that $f-g$ is a nonzero polynomial because $a_1,a_2,\ldots,a_n$ and $b_1,b_2,\ldots,b_n$ are distinct sequences. Let $f(x)-g(x)=(x-1)^kh(x)$, for $k\geq 1$ and $h(1)\neq0$. Then

\[\begin{align} f(x)+g(x)&=\frac{f^2(x)-g^2(x)}{f(x)-g(x)}\\ &=\frac{f(x^2)-g(x^2)}{f(x)-g(x)}\\ &=\frac{(x^2-1)h(x^2)}{(x-1)^kh(x)}\\ &=(x+1)^k\frac{h(x^2)}{h(x)} \end{align}\]

Setting $x=1$ we get that

\[2n=f(1)+g(1)=(1+1)^k\frac{h(1^2)}{h(1)}=2^k\]

Therefore, $n$ must be a power of $2$.

Claim 2:
For $n=2^k$, there are multisets $S=\{\{s_1,s_2,\ldots,s_{m}\}\}$, with $m=\frac{n(n-1)}{2}$ that are the pairwise sums of more than one multiset $A=\{\{a_1,a_2,\ldots,a_n\}\}$.

Proof of Claim 2 [Doubling the size of the example]1

Assume that the two multisets $\{\{x_1,x_2,…,x_n\}\}$ and $\{\{y_1,y_2,\ldots,y_n\}\}$ produce the same multiset of pairwise sums $S$. Let $a\neq0$. Then the multisets

\[\begin{align} X'&=\{\{x_1,x_2,...,x_n,y_1+a,y_2+a,\ldots,y_n+a\}\}\\ Y'&=\{\{x_1+a,x_2+a,\ldots,x_n+a,y_1,y_2,\ldots,y_n\}\} \end{align}\]

produce the same multiset of pairwise sums.

Applying this transformation to a small example, like $X=\{\{1,4\}\}$, $Y=\{\{2,3\}\}$, we can double its size multiple times and obtain examples with size any power of $2$.

Proof of Claim 2 [Thue-Morse sequence]:
The Thue-Morse sequence is obtained by the following procedure:

We start with $0$ as its first element. At every step we repeat the elements of the sequence that have been generated so far, but replace each occurrence of $0$ with a $1$ and each occurrence of $1$ with a $0$. In the first 3 steps we obtain

  • $0$
  • $0,1$
  • $0,1,1,0$
  • $0,1,1,0,1,0,0,1$

If $n=2^k$, for some integer $k$, we can write the numbers $0,1,2,\ldots,2^{k+1}-1$ above the elements of the Thue-Morse sequence. Let $A$ be the set of all numbers above a $0$ from the Thue-Morse sequence and $B$ the set of those above a $1$. The sets $A$ and $B$ are each of size $2^k$ and have the property of producing the same multiset of pairwise sums.

A presentation for my young students.

  1. Selfridge and Straus, On the determination of numbers by their sums of a fixed order, 1958.  2

  2. Andreescu and Savchev, Mathematical Miniatures, 2003.