Erdős–Tetali theorem

In additive number theory, an area of mathematics, the Erdős–Tetali theorem is an existence theorem concerning economical additive bases of every order. More specifically, it states that for every fixed integer ${\displaystyle h\geq 2}$, there exists a subset of the natural numbers ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ satisfying

$${\displaystyle r_{{\mathcal {B}},h}(n)\asymp \log n,}$$ where ${\displaystyle r_{{\mathcal {B}},h}(n)}$ denotes the number of ways that a natural number n can be expressed as the sum of h elements of B.^[1]

The theorem is named after Paul Erdős and Prasad V. Tetali, who published it in 1990.

The original motivation for this result is attributed to a problem posed by S. Sidon in 1932 on economical bases. An additive basis ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ is called economical^[2] (or sometimes thin^[3]) when it is an additive basis of order h and

${\displaystyle r_{{\mathcal {B}},h}(n)\ll _{\varepsilon }n^{\varepsilon }}$

for every ${\displaystyle \varepsilon >0}$. In other words, these are additive bases that use as few numbers as possible to represent a given n, and yet represent every natural number. Related concepts include ${\displaystyle B_{h}[g]}$-sequences^[4] and the Erdős–Turán conjecture on additive bases.

Sidon's question was whether an economical basis of order 2 exists. A positive answer was given by P. Erdős in 1956,^[5] settling the case h = 2 of the theorem. Although the general version was believed to be true, no complete proof appeared in the literature before the paper by Erdős and Tetali.^[6]

The proof is an instance of the probabilistic method, and can be divided into three main steps. First, one starts by defining a random sequence ${\displaystyle \omega \subseteq \mathbb {N} }$ by

${\displaystyle \Pr(n\in \omega )=C\cdot n^{{\frac {1}{h}}-1}(\log n)^{\frac {1}{h}},}$

where ${\displaystyle C>0}$ is some large real constant, ${\displaystyle h\geq 2}$ is a fixed integer and n is sufficiently large so that the above formula is well-defined. A detailed discussion on the probability space associated with this type of construction may be found on Halberstam & Roth (1983).^[7] Secondly, one then shows that the expected value of the random variable ${\displaystyle r_{\omega ,h}(n)}$ has the order of log. That is,

${\displaystyle \mathbb {E} (r_{\omega ,h}(n))\asymp \log n.}$

Finally, one shows that ${\displaystyle r_{\omega ,h}(n)}$ almost surely concentrates around its mean. More explicitly:

${\displaystyle \Pr {\big (}\exists c_{1},c_{2}>0~|~{\text{for all large }}n\in \mathbb {N} ,~c_{1}\mathbb {E} (r_{\omega ,h}(n))\leq r_{\omega ,h}(n)\leq c_{2}\mathbb {E} (r_{\omega ,h}(n)){\big )}=1}$

This is the critical step of the proof. Originally it was dealt with by means of Janson's inequality, a type of concentration inequality for multivariate polynomials. Tao & Vu (2006)^[8] present this proof with a more sophisticated two-sided concentration inequality by V. Vu (2000),^[9] thus relatively simplifying this step. Alon & Spencer (2016) classify this proof as an instance of the Poisson paradigm.^[10]

Unsolved problem in mathematics:

Let ${\textstyle h\geq 2}$ be an integer. If ${\textstyle {\mathcal {B}}\subseteq \mathbb {N} }$ is an infinite set such that ${\textstyle r_{{\mathcal {B}},h}(n)>0}$ for every n, does this imply that:

${\displaystyle \limsup _{n\to \infty }{\frac {r_{{\mathcal {B}},h}(n)}{\log n}}>0}$?

(more unsolved problems in mathematics)

The original Erdős–Turán conjecture on additive bases states, in its most general form, that if ${\textstyle {\mathcal {B}}\subseteq \mathbb {N} }$ is an additive basis of order h then

${\displaystyle \limsup _{n\to \infty }r_{{\mathcal {B}},h}(n)=\infty ;}$

that is, ${\textstyle r_{{\mathcal {B}},h}(n)}$ cannot be bounded. In his 1956 paper, P. Erdős^[5] asked whether it could be the case that

${\displaystyle \limsup _{n\to \infty }{\frac {r_{{\mathcal {B}},2}(n)}{\log n}}>0}$

whenever ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ is an additive basis of order 2. In other words, this is saying that ${\textstyle r_{{\mathcal {B}},2}(n)}$ is not only unbounded, but that no function smaller than log can dominate ${\textstyle r_{{\mathcal {B}},2}(n)}$. The question naturally extends to ${\displaystyle h\geq 3}$, making it a stronger form of the Erdős–Turán conjecture on additive bases. In a sense, what is being conjectured is that there are no additive bases substantially more economical than those guaranteed to exist by the Erdős–Tetali theorem.

All the known proofs of Erdős–Tetali theorem are, by the nature of the infinite probability space used, non-constructive proofs. However, Kolountzakis (1995)^[11] showed the existence of a recursive set ${\displaystyle {\mathcal {R}}\subseteq \mathbb {N} }$ satisfying ${\displaystyle r_{{\mathcal {R}},2}(n)\asymp \log n}$ such that ${\displaystyle {\mathcal {R}}\cap \{0,1,\ldots ,n\}}$ takes polynomial time in n to be computed. The question for ${\displaystyle h\geq 3}$ remains open.

Given an arbitrary additive basis ${\displaystyle {\mathcal {A}}\subseteq \mathbb {N} }$, one can ask whether there exists ${\displaystyle {\mathcal {B}}\subseteq {\mathcal {A}}}$ such that ${\displaystyle {\mathcal {B}}}$ is an economical basis. V. Vu (2000)^[12] showed that this is the case for Waring bases ${\displaystyle \mathbb {N} ^{\wedge }k:=\{0^{k},1^{k},2^{k},\ldots \}}$, where for every fixed k there are economical subbases of ${\displaystyle \mathbb {N} ^{\wedge }k}$ of order ${\displaystyle s}$ for every ${\displaystyle s\geq s_{k}}$, for some large computable constant ${\displaystyle s_{k}}$.

Another possible question is whether similar results apply for functions other than log. That is, fixing an integer ${\displaystyle h\geq 2}$, for which functions f can we find a subset of the natural numbers ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ satisfying ${\displaystyle r_{{\mathcal {B}},h}(n)\asymp f(n)}$? It follows from a result of C. Táfula (2019)^[13] that if f is a locally integrable, positive real function satisfying

• ${\displaystyle {\frac {1}{x}}\int _{1}^{x}f(t)\,\mathrm {d} t\asymp f(x)}$, and
• ${\displaystyle \log x\ll f(x)\ll x^{\frac {1}{h-1}}(\log x)^{-\varepsilon }}$ for some ${\displaystyle \varepsilon >0}$,

then there exists an additive basis ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ of order h which satisfies ${\displaystyle r_{{\mathcal {B}},h}(n)\asymp f(n)}$. The minimal case f(x) = log x recovers Erdős–Tetali's theorem.

• Erdős–Fuchs theorem: For any non-zero ${\displaystyle C\in \mathbb {R} }$, there is no set ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ which satisfies ${\displaystyle \textstyle {\sum _{n\leq x}r_{{\mathcal {B}},2}(n)=Cx+o\left(x^{1/4}\log(x)^{-1/2}\right)}}$.
• Erdős–Turán conjecture on additive bases: If ${\displaystyle {\mathcal {B}}\subseteq \mathbb {N} }$ is an additive basis of order 2, then ${\displaystyle r_{{\mathcal {B}},2}(n)\neq O(1)}$.
• Waring's problem, the problem of representing numbers as sums of k-powers, for fixed ${\displaystyle k\geq 2}$.

• Erdős, P.; Tetali, P. (1990). "Representations of integers as the sum of k terms". Random Structures & Algorithms. 1 (3): 245–261. doi:10.1002/rsa.3240010302.
• Halberstam, H.; Roth, K. F. (1983). Sequences. Springer New York. ISBN 978-1-4613-8227-0. OCLC 840282845.
• Alon, N.; Spencer, J. (2016). The probabilistic method (4th ed.). Wiley. ISBN 978-1-1190-6195-3. OCLC 910535517.
• Tao, T.; Vu, V. (2006). Additive combinatorics. Cambridge University Press. ISBN 0521853869. OCLC 71262684.