---
title: Additive Basis Conjecture in Finite Fields
url: https://www.emergentmind.com/topics/additive-basis-conjecture
type: topic
---

# Additive Basis Conjecture in Finite Fields

The Additive Basis Conjecture, in the sense of Jaeger–Linial–Payan–Tarsi, asks whether for each prime \(p\) there exists a constant \(c(p)\) such that in every vector space \(V\) over \(\mathrm{GF}(p)\), the multiset-union of any \(c(p)\) linear bases is an additive basis, meaning that its \(0\)–\(1\) span equals \(V\). The conjecture belongs to a line of work linking additive combinatorics over finite fields, permanent-rank phenomena, and graph-flow theory. A recent milestone is Yang Yu’s proof that in \(\mathrm{GF}(3)^n\) the union of any four linear bases is an additive basis, establishing the \(p=3\) case with \(c(3)=4\) and yielding an alternative proof of the weak \(3\)-flow conjecture [2510.01300].

## 1. Definition and formal formulation

Let \(V=\mathbb{F}_p^n\), where \(p\) is prime. A multiset \(B\subseteq \mathbb{F}_p^n\) is called an additive basis if every \(w\in \mathbb{F}_p^n\) can be written in the form
\[
w=\sum_{v\in B}\epsilon_v v,\qquad \epsilon_v\in\{0,1\}.
\]
Equivalently, the \(0\)–\(1\) span of \(B\) is the whole space. In the conjectural setup, one starts with several linear bases \(B_1,\dots,B_t\) of \(\mathbb{F}_p^n\) and asks whether their multiset-union is automatically an additive basis once \(t\) is large enough but depends only on \(p\), not on \(n\) [2510.01300].

The conjecture was formulated in this form by Jaeger–Linial–Payan–Tarsi and was already associated earlier to Alon–Linial–Meshulam. Its qualitative content is that repeated linear spanning should force much stronger \(0\)–\(1\) spanning after a bounded number of repetitions. A related weakening, proposed by Szegedy, allows coefficients in \(\mathbb{F}_p^*\) rather than \(\{0,1\}\): for every odd prime \(p\), one asks whether there is a constant \(c_2(p)\) such that whenever \(B\subseteq \mathbb{F}_p^n\) is the union of \(c_2(p)\) linear bases, every \(w\in\mathbb{F}_p^n\) can be written as \(w=\sum_{v\in B} a_v v\) with all \(a_v\in\mathbb{F}_p^*\) [2111.13658].

## 2. Historical bounds and partial progress

Before the constant-bound question was settled for any specific prime, the main general progress was logarithmic in the dimension. Alon–Linial–Meshulam proved that if \(B_1,\dots,B_k\subseteq \mathbb{F}_p^n\) are linear bases and \(B=B_1\cup\cdots\cup B_k\), then \(S(B)=\mathbb{F}_p^n\) provided
\[
k\le (p-1)\log_2 n + (p-2),
\]
so \(k_p(n)=O(\log n)\) for fixed \(p\). Hatami and de Quehen later generalized the paradigm to arbitrary finite abelian groups: if \(A_1,\dots,A_k\subseteq G\) satisfy \(mA_i=G\) for each \(i\), then \(A_1+\cdots+A_k=G\) once
\[
k \ge \frac{2\,\log_2(\log_2 N)}{\log_2(m/(m-1))},
\qquad N=|G|.
\]
In the special case \(G=\mathbb{F}_p^n\), this again yields an \(O(\log n)\) regime. The same exposition notes that simple examples show \(k\ge p\) is necessary in the vector-space problem [1607.00563].

A different partial resolution concerns sparse-support bases. Esperet, de Joannis de Verclos, Le, and Thomassé proved the conjecture in the regime where every vector has support size at most \(2\). More precisely, if \(B_1,\dots,B_t\) are bases of \(\mathbb{Z}_p^n\), each vector in the union has support size at most \(2\), and there are at most \(\ell\) distinct shadows of size \(2\), then
\[
t \ge 8\,\ell\,(3p-4)+(p-2)
\]
suffices for the union to be an additive basis. This produced a universal bound in the support-\(\le 2\) case and connected the conjecture to modulo-orientation theorems and highly edge-connected graph flows [1701.03366].

For the weak version, Nagy, Pál, and Tomon proved a strong form. For every prime \(p\ge 5\), there exists a set \(A\subset \mathbb{F}_p^*\) of size \(2\lfloor \log_2 p\rfloor\) such that whenever \(B\subseteq \mathbb{F}_p^n\) is the union of \(p\) linear bases, the dilated set
\[
A\cdot B=\{a\cdot v:a\in A,\ v\in B\}
\]
is an additive basis of \(\mathbb{F}_p^n\). The same paper also shows that for \(p\ge 11\), the union of only \(3\) bases already suffices if arbitrary nonzero coefficients are allowed [2111.13658].

## 3. The \(p=3\) theorem

Yu’s theorem for \(\mathrm{GF}(3)\) gives the first constant-value resolution of the original conjecture for a specific prime:
\[
\text{If }V=\mathrm{GF}(3)^n\text{ and }B_1,B_2,B_3,B_4\text{ are any four linear bases of }V,\text{ then }B_1+B_2+B_3+B_4=V.
\]
In the formulation used in the paper, this implies that the multiset-union of any four bases is an additive basis, and hence \(c(3)=4\) [2510.01300].

The significance of this result is twofold. First, it converts the previously known logarithmic dependence on \(n\) into an exact constant for \(p=3\). Second, it closes a graph-theoretic loop already emphasized by Jaeger–Linial–Payan–Tarsi: the \(p=3\) additive-basis statement implies the weak \(3\)-flow conjecture for graphs. Yu’s argument therefore supplies an alternative proof of Thomassen’s 2012 theorem on weak \(3\)-flows, but now through a purely algebraic route rather than a graph-theoretic one [2510.01300].

The theorem is also structurally notable because its proof does not proceed by direct sumset expansion or Fourier-analytic estimates. Instead, it passes through a permanent-rank criterion for a block matrix and then through an induction in a truncated polynomial algebra. That proof architecture is unusually algebraic even by the standards of additive combinatorics over finite fields.

## 4. Proof architecture: permanent rank and truncated polynomial algebra

The first stage is a reduction via the Combinatorial Nullstellensatz. Let \(P,R,S,T\) be nonsingular \(n\times n\) matrices over a field of characteristic \(3\), and form the \(2n\times 4n\) block matrix
\[
M=
\begin{pmatrix}
P & R & S & T\\[6pt]
P & R & S & T
\end{pmatrix}.
\]
Yu shows that if \(M\) has full permanent rank, meaning that some \(2n\times 2n\) submatrix has nonzero permanent, then the union of the four row-bases \(P,R,S,T\) is an additive basis of \(\mathrm{GF}(3)^n\). The core algebraic statement is therefore the block-matrix theorem asserting that \(M\) has full permanent rank whenever \(P,R,S,T\) are nonsingular [2510.01300].

The second stage takes place in the truncated polynomial ring
\[
A=\mathrm{GF}(3)[x_1,\dots,x_n]/(x_1^2,\dots,x_n^2),
\]
graded by degree. If \(U\subseteq A_1\) is a linear-form space, one defines \(\operatorname{supp}(u)\) for \(u\in U\), and for \(1\le i\le \dim U\),
\[
ms_i(U)=\min\{|\operatorname{supp}(V)|:V\subseteq U,\ \dim V=i\}.
\]
The proof introduces division and remainder operators by a linear form, together with a generalized remainder \(R_{(u,x)}\), in order to control the ideals \(\mathrm{Im}(U)\) generated by \(U\) and the annihilator spaces
\[
\mathrm{Ker}_k(U)=\{f\in A_k:f\cdot u=0\ \forall u\in U\}.
\]

Two preparatory results drive the induction. Theorem 5 identifies kernels and images for a single linear form \(u\) under support hypotheses:
\[
\mathrm{Ker}_k(u)=\mathrm{Im}_k(u^2),\qquad
\mathrm{Ker}_k(u^2)=\mathrm{Im}_k(u),
\]
provided \(|\operatorname{supp}(u)|\) is at least \(2k+1\) or \(2k+2\), respectively. Lemma 6 is a dimension/support splitting statement: if \(U\) covers an increasing sequence \((a_1,\dots,a_n)\) in the sense that \(ms_i(U)\ge a_i\), then for each \(k\) one can choose a \(k\)-dimensional subspace \(U_k\subseteq U\) covering the tail \((a_{n+1-k},\dots,a_n)\) [2510.01300].

These ingredients feed Theorem 7, a kernel–image duality statement involving the subspace \(U^{2n}\subseteq A_{2n}\) spanned by all \(2n\)-fold products of elements of \(U\). Under the support lower bounds
\[
ms_i(U)\ge 4i-2+2k
\quad\text{or}\quad
ms_i(U)\ge 4i-3+2k,
\]
the theorem gives
\[
\mathrm{Ker}_k(U^{2n})=\mathrm{Im}_k(U)
\]
and a companion equality in high degree. Taking \(k=0\) yields \(U^{2n}\neq 0\), hence a nonzero permanent and full permanent rank. Since the row-forms associated with \(P,R,S,T\) satisfy \(ms_i(U)\ge 4i\), the criterion applies and closes the proof [2510.01300].

## 5. Related conjectures using the same or a similar label

The phrase “additive basis conjecture” is used for several distinct problems. In the finite-integer \(2\)-basis problem, one studies finite \(A\subset\mathbb{Z}_{\ge 0}\) such that \(A+A\) contains \(0,1,\dots,n\) but not \(n+1\). Here the conjectural asymptotic ceiling is
\[
\limsup_{k\to\infty}\frac{n(k)}{k^2}=\tfrac12,
\]
where \(n(k)\) is the maximum range of a \(2\)-basis of size \(k\). Kohonen improved the explicit lower bound to
\[
\liminf_{k\to\infty}\frac{n(k)}{k^2}\ge \frac{85}{294}>0.2891
\]
by a generalized Mrose construction [1606.04770].

In another direction, the Erdős–Turán additive-basis conjecture concerns infinite \(B\subset\mathbb{N}\) with \(r_B(n)>0\) for all sufficiently large \(n\), where
\[
r_B(n)=\#\{(a,b)\in B\times B:a+b=n\},
\]
and predicts
\[
\limsup_{n\to\infty} r_B(n)=\infty.
\]
Agama’s 2020 paper proves several variants under density and energy hypotheses, while the generalized circle-of-partition paper formulates Corollary 2.6 as a proof under the condition \(|A\cap[1,n]|>n^{1-\varepsilon}\) for some \(0<\varepsilon<1\) and all large \(n\) [2010.11857] [1707.05679].

A further variant arises for finite abelian groups through the invariant \(c_0(G)\), the least \(t\) such that every regular sequence \(S\) over \(G\) with \(|S|\ge t\) is an additive basis in the sense \(\Sigma(S)=G\). Gao and Peng conjectured that \(c_0(G)=m(G)\). For rank \(2\), this was confirmed in the generic case
\[
G=C_{n_1}\oplus C_{n_2},\quad n_1\mid n_2,\quad p\ge 3,\quad n_1\ge 2p,\quad n_1n_2\ge 72p^6,
\]
where the value is
\[
c_0(G)=m(G)=\frac{n_1n_2}{p}+p-2,
\]
and earlier exact progress included
\[
\mathsf c_0(C_3\oplus C_{3q})=3q+3
\quad (q\ge 5\text{ prime}).
\]
These are sequence-sumset problems rather than the finite-field basis-union problem of Jaeger–Linial–Payan–Tarsi [2112.02564] [2107.06976].

Finally, Bukh, van Hintum, and Keevash formulated a geometric additive-basis conjecture over \(\mathbb{Q}^n\) and its strengthened version: if \(S+S\subseteq A+B\) for a basis \(S\) and \(|A|=n-t\), then \(|B|\ge n+\binom{t+1}{2}\). Xu proved the full strengthened statement over \(\mathbb{R}^n\), for arbitrary bases \(S\), and showed that the bound is sharp for every \(S\) and every \(0\le t\le n-1\) [2605.10771].

## 6. Open directions

Despite the \(p=3\) theorem, the original constant-bound conjecture remains open for general prime \(p\). Yu explicitly notes that a conjecture on permanent rank of a \(p\)-block matrix would imply \(c(p)=p\) in general, and that his method already yields an \(O(\log n)\) bound for arbitrary \(p\). He also emphasizes that extending the constant-bound result beyond \(p=3\) appears to require new ideas in the truncated-polynomial framework, specifically analogues of the single-form kernel/image theorem and the dimension/support splitting lemma for larger \(p\) [2510.01300].

The weak theory suggests a complementary route. Nagy, Pál, and Tomon show that the weak conjecture can be solved with \(p\) bases plus a multiplier set \(A\subset\mathbb{F}_p^*\) of size \(O(\log p)\), and that for \(p\ge 11\) only \(3\) bases suffice with arbitrary nonzero coefficients. Their paper remarks that progress on arithmetic-like multiplier sets of constant size might translate into progress on the original conjecture [2111.13658].

In the graph-theoretic direction, the support-\(\le 2\) case already yields strong list-flow and antisymmetric-flow theorems in highly edge-connected graphs, but the unrestricted-support problem is still open. This suggests that the full conjecture may require either a higher-dimensional analogue of the graph-connectivity machinery used in the sparse-support regime or a further refinement of the polynomial and permanent-rank methods that proved decisive for \(p=3\) [1701.03366].

Source: https://www.emergentmind.com/topics/additive-basis-conjecture