Unequal multipliers let sparse collections of numbers reach every positive integer
A modular digit construction shows that weighting the slots in a sum prevents representations from multiplying endlessly, resolving an open problem in additive number theory.

When mathematicians build whole numbers by adding together elements from a single chosen list, they face an elusive balancing act. If the list contains too few numbers, vast stretches of the number line are missed entirely. If the list contains enough numbers to guarantee that every large integer can be assembled, the combinations typically begin to pile up, producing certain target totals dozens, hundreds, or infinitely many times over. For more than eight decades, number theorists have suspected that this pileup is unavoidable whenever two numbers from the set are added directly together. Yet altering the problem by assigning unequal weights to each slot in the sum changes the behavior entirely.
Instead of considering simple additions such as a plus b, researchers can investigate weighted combinations such as two times a plus three times b. In that asymmetric setting, the structural traffic jams that force representations to multiply can be completely untangled. A preprint posted to the arXiv repository by Christian Táfula, a mathematician at the University of São Paulo, demonstrates that for almost any pair of positive whole numbers used as multipliers, one can construct an infinite set of integers that forms every sufficiently large number without ever producing more than a strictly capped number of duplicate answers.12
The work addresses a central tension in additive number theory, the branch of mathematics that explores how numbers decompose under addition. In 1941, the Hungarian mathematicians Paul Erdős and Pál Turán formulated a famous conjecture concerning what they called additive bases.1 An asymptotic basis is an infinite collection of non-negative integers chosen so that every whole number from some point onward can be expressed as a sum of a fixed number of elements from the collection.1 Erdős and Turán proposed that if an infinite set can form every sufficiently large integer as the sum of two elements, the number of distinct ways to represent those totals cannot remain bounded.1 As the target totals grow toward infinity, the conjecture asserts, the maximum number of representations must also climb without limit.
The intuition behind the Erdős–Turán conjecture rests on how symmetric sums behave across the integers. If a single set must provide elements to reach every target value, the density of the set must remain substantial. But because addition is commutative, meaning the order of the terms does not change the result, two elements chosen from the set generate identical sums regardless of which variable is picked first. Over an infinite span of integers, attempts to prune the set to eliminate surplus solutions while maintaining uninterrupted coverage have continually failed, and after more than eighty years the binary Erdős–Turán conjecture remains unsolved.12
Why does weighting the sum change the outcome?
Unequal multipliers eliminate the symmetry that forces ordinary additions to cluster, allowing mathematicians to assign distinct roles to different components of a number. When evaluating an expression such as two times a plus three times b, swapping the values assigned to the two variables produces entirely different totals. This fundamental asymmetry allows an integer to be partitioned into separate numerical compartments, much like the digits in a positional numeral system. By designing the underlying set so that each variable controls a separate slice of the target number, one variable can fill the lower orders of magnitude while the other handles the higher orders.
This mechanism was first exploited in 1962 by the Canadian mathematician Leo Moser, who examined the linear form where one variable is unweighted and the second is multiplied by an integer m that is at least two.1 Moser proved that it is possible to construct an infinite set of non-negative integers such that every non-negative integer has exactly one representation as the first element plus m times the second element.1 In 2007, Melvyn Nathanson of the City University of New York extended Moser's idea to sums of three or more variables, showing that forms using successive powers of m can likewise achieve unique representation for every single integer.12 In these specialized cases, the representation function, which counts the number of valid solutions for each target value, equals exactly one everywhere.
Such perfectly unique representations, however, are exceptionally rigid. In 2025, Sándor Kiss and Csaba Sándor classified all strictly increasing vectors of coefficients and all sets of integers that admit exactly one representation for every non-negative target, confirming that Moser's and Nathanson's power-based designs form the standard backbone of exact systems.1 Meanwhile, other researchers investigated whether representation functions could settle into a steady, constant value above one. In 2009, Javier Cilleruelo and Juanjo Rué settled that question for two variables, demonstrating that aside from Moser's forms, an asymptotic basis whose representation count becomes strictly constant for all sufficiently large integers is mathematically impossible.1 In 2020, Rué and Christoph Spiegel extended that impossibility to broader families involving three or more variables.12
How far can bounded representations reach?
Bounded representations succeed across all primitive binary pairs because allowing a flexible ceiling circumvents the rigid algebraic barriers that rule out strictly constant counts. A vector of coefficients is called primitive when the greatest common divisor of its components equals one, a baseline condition required to ensure that the linear combination can represent every integer without leaving periodic gaps.1 In his new six-page preprint, Táfula proved that every primitive binary vector other than the unweighted pair (1,1) admits an infinite set of integers whose representation count never drops to zero and never exceeds a fixed finite bound for any target integer greater than or equal to the product of the two coefficients.13
Specifically, when the coefficients are b1 and b2, Táfula proved that there exists a set of non-negative integers such that every integer n greater than or equal to the product b1 times b2 has at least one representation and at most (b1 + 1)(b2 + 1) representations.12 For example, if the multipliers are three and five, the formula three times a plus five times b can generate every integer beyond 15 with no fewer than one combination and no more than 24 combinations from the set, where 24 represents the product of four and six. The upper bound is entirely independent of the target number n, meaning that even as n extends to millions, billions, or beyond, the number of distinct representations never exceeds that fixed threshold.
Christian, the study author, explained that the starting threshold does not reflect a fundamental barrier in the arithmetic. Told Primary about the lower bound, Christian stated that "the cutoff at b1*b2 is just a convenient feature of the construction" and noted that by the classical Frobenius coin problem, every integer larger than b1*b2 - b1 - b2 can be represented non-negatively. Christian added that "the remaining b1 + b2 - 1 integers below b1*b2 could in principle be handled separately, without affecting boundedness."

To build this set, Táfula turned to the p-adic valuations of the multipliers, which measure the highest power of a chosen prime number that divides an integer.1 Taking a prime factor p that divides the first coefficient b1, the construction isolates the base-p digit expansions of numbers.1 By grouping digits into alternating blocks of size e, where e is the highest power of p dividing b1, Táfula defined a base set whose members have non-zero digits only in every other block of positions.1 Because the digits occupy alternating positional windows, any integer can be broken into two components whose digit blocks do not overlap.
By shifting this sparse digit set by a finite collection of modular residues derived from the Chinese remainder theorem, Táfula ensured that both variables could be drawn from a single unified set of integers.1 The non-overlapping digit structure guarantees an injectivity property: whenever the two variables are selected from particular shifted components, the linear combination cannot produce the same target integer twice.12 In response to questions from Primary, Christian explained that "once we fix the two components containing the variables, the shifts are just constants" so that subtracting two identical representations cancels the shifts and leaves an equation in alternating base-p blocks whose injectivity forces the representations to coincide. Christian pointed out that collisions occur only across different shifted components, ensuring that the total count stays capped.
The total number of representations for any integer is therefore bounded by the number of component subsets that make up the unified collection, producing the clean ceiling of (b1 + 1)(b2 + 1).12 Christian noted that "the bound comes naturally from the way the set is constructed from the shifted parts" and acknowledged that "a more careful count could probably improve it."
What happens when more variables are added?
Higher-dimensional combinations also admit bounded representation functions whenever the prime-factor divisibilities of the coefficients can tile a finite clock-like number system. Moving from two variables to three or more introduces complex interactions, because the variables can collide in many more configurations. In Theorem 1.2 of the paper, Táfula showed that if there exists a prime p and a modulus H such that the residues of the p-adic valuations of the coefficients are distinct and form a direct sum tiling of the integers modulo H with an auxiliary set T, then a bounded asymptotic basis exists for that collection of coefficients.1
A direct sum tiling means that every residue class modulo H can be written in exactly one way as the sum of a valuation residue and an element of T. Under this condition, the base-p digit positions are cleanly divided among all h variables, allowing the same separation of variables that worked in the two-variable setting. In response to questions from Primary, Christian explained the geometric idea: "In the binary case, the base-p digit positions are split into two classes, so every digit belongs to exactly one of the two variables. For h variables, we need an analogous way to divide all digit positions among them. The tiling condition does exactly this: the p-adic valuations of the coefficients, together with the set T, partition the positions modulo H with no gaps and no overlaps."
In Corollary 1.3, Táfula noted that this condition is automatically met whenever the p-adic valuations of the coefficients form an arithmetic progression modulo hd, where h is the number of variables and d is a positive integer.1 In particular, this covers any coefficient vector where the weights scale as powers of a prime multiplied by factors not divisible by that prime.12 For instance, the four-variable vector with coefficients one, p, p to the fourth power, and p to the fifth power yields valuations of zero, one, four, and five.1 Modulo eight, this four-element set tiles the cyclic group when paired with the two-element set containing zero and two, because every number from zero through seven can be expressed uniquely as one of the four valuations plus either zero or two.1 For every such family, Táfula proved that the representation function for all sufficiently large integers is bounded above by a constant depending only on the chosen coefficients.1
What the mathematical proofs do not settle
These findings provide existence proofs rather than practical counting recipes for all possible equations. The paper constructs explicit sets of integers that satisfy the bounded representation conditions, but it does not claim that these bounds are minimal, nor does it establish that every primitive weighted form with three or more variables admits such a basis. The paper is an unrefereed preprint posted to the arXiv repository and has not yet completed peer review.14
The underlying framework depends strictly on the prime-power divisibility pattern of the coefficients aligning with cyclic group tilings. If the coefficients of a linear form do not possess valuations that tile a cyclic group under some prime, the construction cannot be applied. Táfula explicitly highlighted two simple three-variable test cases that fall entirely outside the reach of the current method: the vector (1,1,2), which contains repeated coefficients, and the vector (1,2,5), whose coefficients have distinct subset sums but lack the valuation structure required for the modular tiling argument.1 Asked about this barrier, Christian told Primary that "for (1,2,5), the problem is that no single prime separates all three coefficients" and observed that overcoming this obstacle would require separating variables with more than just the valuation at a single prime.
The result also sharpens the boundary around the original Erdős–Turán conjecture itself. In earlier work published in 2026 in the journal Random Structures & Algorithms, Táfula had conjectured a general quantitative lower bound, proposing that for any asymptotic basis of a linear form, the limit superior of the representation count divided by the logarithm of n must be at least one.1 While Moser's and Nathanson's power forms were already known to violate that bound by having representation counts identically equal to one, Táfula's new theorem proves that among all primitive binary forms, the logarithmic growth bound can only possibly hold for the unweighted form x1 + x2.1 Discussing how that remaining unweighted case might be resolved, Christian said that beating the threshold for the pair (1,1) would require a structured construction or a finer random model, because separating the roles of variables through unequal coefficients disappears in the symmetric setting.
The open challenge in additive number theory is now to determine whether bounded asymptotic bases exist for every primitive linear form other than the all-ones vector. If every asymmetric form admits a bounded basis, the Erdős–Turán phenomenon would be revealed as an exclusive property of total symmetry, isolating the classical conjecture as a singular anomaly of unweighted addition.
This piece was prepared from the arXiv preprint and public records together with answers from Christian to six questions from the Primary news team, completed September 2026.1
The record
31 sentences trace to 4 sources.
- 1 Bounded asymptotic bases for linear forms Preprint · may not have been peer reviewed See the source
- 2 Bounded asymptotic bases for linear forms Preprint · may not have been peer reviewed See the source
- 3 Mathematics Preprint · may not have been peer reviewed See the source
- 4 Bounded asymptotic bases for linear forms Preprint · may not have been peer reviewed See the source
