In February 2026, Huy Tuan Pham and Lisa Sauermann (Pham and Sauermann 2026) completed, together with earlier work, a proof of Graham’s rearrangement conjecture for all sufficiently large primes. In October, Peiru Kuang and Yan Wang (Kuang and Wang 2026) extended the conclusion to sufficiently large subsets of arbitrary finite abelian groups.
The problem is easy to explain using arithmetic modulo a prime. Working modulo means that we identify integers whose difference is divisible by . For example, modulo , we have .
Choose a set of distinct nonzero residues. Can we arrange them so that their running totals are all different?
For example, take modulo . The ordering gives partial sums so a value repeats. However, the ordering gives which are all distinct.
An ordering of a set is called valid if its nonempty partial sums are pairwise distinct. In 1971, Ronald Graham (Graham 1971, 36) conjectured that every subset of the nonzero residues modulo a prime has a valid ordering.
The same question makes sense in any finite abelian group. Such a group is a finite set with an associative and commutative addition, a zero element, and an additive inverse for every element. Examples include the integers modulo any positive integer, and binary vectors with coordinatewise addition modulo .
The two new results can be summarized as follows.
Part (a) covers subsets of every size once the prime is sufficiently large. Part (b) covers every finite abelian group, with one universal lower bound on the size of the subset. The same works regardless of the size or structure of the group.
There is an important convention in this statement: the empty partial sum is not included among the sums required to be distinct. In particular, the total sum of is allowed to be zero. For example, the ordering modulo has partial sums and is valid.
Progress towards Graham’s conjecture initially came from opposite ends of the size range. In a 2024 preprint, Bedert and Kravitz (Bedert and Kravitz 2026) treated small sets, proving the conjecture when In 2025, Bedert, Bucić, Kravitz, Montgomery and Müyesser (Bedert et al. 2025) treated large sets, including all those with for an absolute constant .
Pham and Sauermann filled the intermediate range. More precisely, for every fixed , they proved the result whenever where depends only on . Choosing and combining the three ranges gives part (a) for all sufficiently large primes.
Why is finding a valid ordering difficult? If two partial sums agree, say then cancellation gives Thus a collision occurs exactly when a consecutive block starting after the first term sums to zero.
A greedy approach has room to work early on. After steps, each previously reached position forbids at most one choice for the next step. When many elements remain unused, one can avoid these forbidden choices. Near the end, however, the remaining elements may leave no such freedom. An early choice can create a difficulty that only becomes visible much later.
Pham and Sauermann use randomness to overcome this obstacle. Their proof begins with a random ordering and makes local adjustments to eliminate zero-sum segments, swapping an endpoint with a suitable later element. The essential estimates control how likely a random subset of a given size is to have any particular sum. These are called anticoncentration estimates: they prevent too much probability from accumulating at one value and provide the flexibility needed for the adjustments.
Kuang and Wang’s extension combines probabilistic ideas with the additive structure of general abelian groups. An important ingredient is a dissociated subset, meaning that all its subset sums are different, including the sum of the empty subset. If itself has this property, every ordering is valid, because different prefixes correspond to different subsets.
Their argument uses a sufficiently large dissociated subset to help organize the remaining elements while avoiding collisions. Sets without such a large subset require a different analysis, using their concentration in translates of subgroups. Together, these approaches yield the universal threshold in part (b).
The theorem does not require the elements of to have few additive relations. A set may contain many subsets summing to zero, yet its elements can still be threaded into an ordering whose nonempty running totals are all different.