Rota's Basis Conjecture

• Rota's Basis Conjecture is a problem in matroid theory and linear algebra asserting that n bases of a rank-n matroid or vector space can be rearranged into n disjoint transversal bases.
• The conjecture admits several equivalent formulations using linear algebra, combinatorial disjointness, and matroid intersection, with key techniques including parity arguments and determinantal identities.
• Recent progress has yielded quantitative bounds from Θ(n/log n) to (1-o(1))n transversal bases and even constructive polynomial-time algorithms for asymptotic versions.

Searching arXiv for recent and foundational papers on Rota's Basis Conjecture. Rota’s Basis Conjecture (RBC) is a conjecture in matroid theory and linear algebra asserting that $n$ bases of rank $n$ can be repacked into $n$ disjoint transversal bases. In vector-space form, if $B_1,\dots,B_n$ are bases of an $n$-dimensional vector space, then after suitable independent permutations of the rows of the associated $n\times n$ array, every column should also be a basis; in matroid form, if $B_1,\dots,B_n$ are bases of a rank-$n$ matroid, then their union should decompose into $n$ disjoint bases each containing exactly one element from each $B_i$ (Bittner et al., 2013). The conjecture remains open in full generality, including for representable matroids, but a substantial body of work now describes its parity-theoretic, probabilistic, spectral, asymptotic, and algorithmic structure (Montgomery et al., 7 Aug 2025).

1. Statements and equivalent formulations

Let $n$0 be an $n$1-dimensional vector space over a field, and let

$n$2

be $n$3 bases of $n$4. Arranging the $n$5 as rows of an $n$6 array, RBC asks for permutations $n$7 such that after permuting row $n$8 by $n$9, each resulting column is also a basis of $n$0 (Bittner et al., 2013). Equivalently, the $n$1 vectors can be repacked into an $n$2 grid whose rows and columns are all bases (Bittner et al., 2013).

In matroid language, if $n$3 is a rank-$n$4 matroid and $n$5, then RBC asks whether one can partition $n$6 into $n$7 disjoint bases $n$8 such that each $n$9 contains exactly one element from each $B_1,\dots,B_n$0 (Bittner et al., 2013). In the colored-matroid viewpoint, each $B_1,\dots,B_n$1 is a color class, a rainbow basis is a basis using exactly one element of each color, and RBC becomes a decomposition problem into $B_1,\dots,B_n$2 disjoint rainbow bases (Montgomery et al., 7 Aug 2025).

The transversal language is standard. A transversal is a set containing exactly one element from each $B_1,\dots,B_n$3; in the array model it is a choice of one entry from each row. Two transversals are disjoint if they share no entry. RBC therefore predicts the existence of $B_1,\dots,B_n$4 pairwise disjoint transversals that are all bases (Bittner et al., 2013). This rephrasing isolates two logically distinct ingredients: the combinatorics of disjointness and the linear or matroidal condition of basishood.

A particularly useful reformulation treats the problem as a special case of matroid intersection coloring. In the RBC setting, one matroid is the original rank-$B_1,\dots,B_n$5 matroid $B_1,\dots,B_n$6, and the second is the partition matroid $B_1,\dots,B_n$7 whose parts are the given bases $B_1,\dots,B_n$8 with capacity $B_1,\dots,B_n$9 in each part. A rainbow basis is then a common independent set of size $n$0 in $n$1, and RBC asks for a partition of the ground set into $n$2 such common bases (Arndt et al., 4 Apr 2026).

2. Latin squares, parity, and determinantal methods

One major line of attack connects RBC to Latin squares and parity. For a Latin square $n$3 of order $n$4, the sign is the product of the signs of its row permutations and column permutations. Writing $n$5 and $n$6 for the numbers of even and odd Latin squares, the classical Alon–Tarsi Latin Square Conjecture asserts that for even $n$7,

$n$8

For even $n$9, Huang–Rota and Onn observed that this conjecture implies RBC over fields whose characteristic does not divide $n\times n$0, via Onn’s colorful determinantal identity (Aharoni et al., 2011).

Onn’s identity is a signed sum of products of determinants indexed by row permutations. If the right-hand side is nonzero, then at least one summand is nonzero, and that nonzero summand encodes a full set of transversal bases (Aharoni et al., 2011). This viewpoint converts the existence of rainbow bases into a nonvanishing statement about a structured determinant-like polynomial. Malkoun later placed Onn’s identity into a broader skew-symmetrization framework that also subsumes Svrtan’s $n\times n$1 formula, showing that these determinant-and-choice constructions are instances of a more general multilinear identity (Malkoun, 2018).

Odd dimensions require different parity data, because for odd $n\times n$2 one has $n\times n$3. Aharoni and Kotlar replaced the classical parity imbalance by the difference

$n\times n$4

where $n\times n$5 and $n\times n$6 count reduced even and odd Latin squares. They proved a modified colorful determinantal identity in which one determinant is replaced by a permanent, and showed that if $n\times n$7 is odd and

$n\times n$8

then any $n\times n$9 bases in characteristic $B_1,\dots,B_n$0 admit $B_1,\dots,B_n$1 disjoint independent transversals (Aharoni et al., 2011). This is a weak odd-dimensional form of RBC: one transversal may fail to be a basis, but the other $B_1,\dots,B_n$2 are independent.

The parity approach therefore yields two distinct mechanisms. For even $B_1,\dots,B_n$3, classical Latin-square parity can imply the full conjecture in representable settings; for odd $B_1,\dots,B_n$4, reduced-Latin-square parity gives a conditional $B_1,\dots,B_n$5 theorem rather than the full $B_1,\dots,B_n$6-transversal conclusion (Aharoni et al., 2011). This division reflects a structural asymmetry already visible in the vanishing of $B_1,\dots,B_n$7 for odd $B_1,\dots,B_n$8.

3. Quantitative progress and asymptotic forms

A second line of work asks not for all $B_1,\dots,B_n$9 transversal bases, but for the largest number that can always be guaranteed. Dong and Geelen proved that if $n$0 are disjoint bases of a rank-$n$1 matroid, then there are at least

$n$2

pairwise disjoint transversal bases (Dong et al., 2017). Their proof uses Rado’s theorem, a worst-case reduction to a coupon-collector model, and a union bound. Combined with the earlier Geelen–Webb bound $n$3, this yields a clean overall lower bound of order $n$4 (Dong et al., 2017).

Bucić, Kwan, Pokrovskiy, and Sudakov then proved that for every $n$5 and sufficiently large $n$6, any collection of $n$7 bases of a rank-$n$8 matroid has at least

$n$9

disjoint transversal bases (Bucić et al., 2018). This was the first linear lower bound in complete generality. Their argument is based on iterative augmentations of disjoint rainbow independent sets by swap and cascade operations.

Pokrovskiy proved an asymptotic version of RBC in a weaker but still highly structured form: any $n$0 disjoint bases in a rank-$n$1 matroid contain

$n$2

disjoint rainbow independent sets of size

$n$3

and the proof yields the explicit quantitative form

$n$4

for both the number of sets and the size of each set, for a fixed large constant $n$5 (Pokrovskiy, 2020). These are not necessarily bases, but they cover $n$6 of the $n$7 colored elements.

Two 2021 papers established near-complete decompositions in structurally restricted regimes. McGuinness proved that if a rank-$n$8 matroid has $n$9 elements, then any sequence of $B_i$0 bases contains at least

$B_i$1

disjoint rainbow bases (McGuinness, 2021). Friedman and McGuinness proved that if the girth satisfies $B_i$2 with $B_i$3, and no element belongs to more than $B_i$4 bases, then

$B_i$5

hence $B_i$6 under high-girth and low-overlap hypotheses (Friedman et al., 2019).

The strongest current asymptotic packing and covering results are due to Montgomery and Sauermann. For every $B_i$7 and sufficiently large $B_i$8, any collection of $B_i$9 bases of a rank-$n$00 matroid has at least $n$01 disjoint transversal bases, and can be covered by at most $n$02 transversal bases (Montgomery et al., 7 Aug 2025). These theorems are asymptotically tight: RBC predicts the exact value $n$03, and the remaining discrepancy is additive rather than multiplicative (Montgomery et al., 7 Aug 2025).

4. Structural and algebraic frameworks

A distinct structural program studies the universal combinatorics of transversals independently of any particular vector configuration. Given an $n$04 array of distinct symbols, let $n$05 be the set of all transversals. The incidence matrix $n$06 of disjoint transversals is the $n$07 matrix indexed by $n$08, with

$n$09

if the two transversals are disjoint and $n$10 otherwise (Bittner et al., 2013). In this framework, each specific RBC instance corresponds to a subset $n$11 consisting of those transversals that are bases, and the conjecture asks whether the induced subgraph on $n$12 contains a clique of size $n$13 (Bittner et al., 2013).

Huang and Srinivasan computed the spectrum and Smith normal form of $n$14. The eigenvalues are

$n$15

with multiplicities

$n$16

and the invariant factors in the Smith normal form are

$n$17

with the same multiplicities (Bittner et al., 2013). These are global invariants of the disjointness relation; they do not solve RBC, but they isolate the universal linear-algebraic and arithmetic structure behind every instance.

A different algebraic framework proves a saturation form of RBC over $n$18. Derksen and Makam showed that for any $n$19 bases $n$20 of $n$21, there exists $n$22 and an $n$23 matrix such that in the $n$24-th row each element of $n$25 appears exactly $n$26 times and every column is a basis (Yeliussizov, 2021). Their proof uses Tao’s slice rank and geometric invariant theory: the Levi–Civita tensor $n$27 has full slice rank in all tensor powers, hence is semistable, which yields a nonzero invariant polynomial whose expansion encodes the desired saturated arrangement (Yeliussizov, 2021). This is strictly weaker than RBC, since it allows multiplicity $n$28, but it is a genuine positive result in a natural asymptotic enlargement of the problem.

Recent work has also made the asymptotic theory algorithmic. In the language of matroid-intersection coloring, RBC becomes a question about partitioning the ground set into common bases of a matroid and a partition matroid. A 2026 paper gives a polynomial-time $n$29-approximation for coloring the intersection of two general matroids, a $n$30 coloring for $n$31 matroids, and an FPRAS for coloring the intersection of two matroids when $n$32 is large (Arndt et al., 4 Apr 2026). In the RBC setting this yields the first polynomial-time constructive algorithm for an asymptotic variant of RBC, constructivizing the asymptotic packing theorem and extending it from representable settings to arbitrary matroids (Arndt et al., 4 Apr 2026).

5. Variants, strengthenings, and related conjectures

Several natural strengthenings and variants clarify which parts of RBC are genuinely difficult. Bollen and Draisma formulated an online version in which the permutation of row $n$33 must be fixed immediately after seeing that row, without knowledge of later rows. If the characteristic of the field does not divide $n$34, then the online conjecture holds (Bollen et al., 2013). By contrast, for any odd $n$35 and any field containing a primitive $n$36-th root of unity for every odd $n$37, the online version is false (Bollen et al., 2013). Thus the online problem exhibits a sharp even–odd dichotomy not known for the classical conjecture.

Kahn’s basis conjecture is a two-dimensional strengthening. Given an $n$38 array of bases $n$39, one seeks representatives $n$40 such that each row and each column of representatives is a basis. Rota’s conjecture is the special case $n$41 for fixed $n$42 (Bucić et al., 2018). A companion note to the “Halfway” paper shows that for every $n$43 and sufficiently large $n$44, one can realize this simultaneously on at least $n$45 rows in the general Kahn setting, by adapting the same cascade-based machinery (Bucić et al., 2018).

Another generalization replaces the square shape by an arbitrary Young diagram. The wide partition conjecture of Chow–Fan–Goemans–Vondrák generalizes RBC, and in the free matroid case it becomes the Latin Tableau Conjecture: a partition shape $n$46 and type $n$47 admit a Latin tableau precisely when the chromatic difference sequence $n$48 dominates $n$49 (Chow et al., 2024). Chow and Tiefenbruck proved that for every $n$50, the conjecture is correct for at least the first four parts of $n$51, and verified it computationally for all $n$52 contained in a $n$53 square (Chow et al., 2024). This does not prove RBC, but it develops the free-matroid combinatorics behind one influential generalization.

6. Status and outlook

RBC is now surrounded by a substantial body of exact, asymptotic, conditional, and algorithmic results. It is known in several special cases, including paving matroids, strongly base-orderable matroids, rank $n$54, and various real-representable cases via the Alon–Tarsi conjecture on Latin squares (Montgomery et al., 7 Aug 2025). The full conjecture nevertheless remains open in general, even for representable matroids (Montgomery et al., 7 Aug 2025).

At the quantitative level, the progression is now unusually sharp. The guaranteed number of disjoint transversal bases has moved from $n$55 to $n$56, then to $n$57, and finally to $n$58 (Dong et al., 2017). Covering bounds have likewise reached $n$59 (Montgomery et al., 7 Aug 2025). On the algebraic side, slice-rank and invariant-theoretic methods prove a saturated multiplicity version over $n$60 (Yeliussizov, 2021). On the algorithmic side, asymptotic RBC now has constructive polynomial-time realizations through matroid-intersection coloring (Arndt et al., 4 Apr 2026).

The remaining gap is therefore exact rather than asymptotic. The available results show that, in several precise senses, almost all of the conjectured structure can be forced: almost all required rainbow bases can be packed, all elements can be covered with almost the optimal number of transversal bases, and over $n$61 a multiplicity-$n$62 version always exists (Montgomery et al., 7 Aug 2025). This suggests that the decisive difficulty lies in eliminating the final additive slack and passing from approximate or saturated decompositions to an exact $n$63-by-$n$64 transversal-basis partition.