Basis Partitions in Mathematics

• Basis partitions are uniquely minimal partitions defined by a prescribed successive rank vector, ensuring optimal representation in partition theory.
• They play a pivotal role in Boolean-function minimization by decomposing prime implicants into essential subsets and independent covering blocks.
• In infinite matroid theory, basis partitions enable simultaneous base covering and packing, establishing a Cantor–Bernstein-type theorem for common base decompositions.

Basis partitions is a polysemous term in current mathematical literature. In partition theory it denotes the unique minimal partition attached to a prescribed successive rank vector, together with refinements such as signature and completeness (Alladi, 19 Jul 2025). In the theory of separable integer partitions it denotes the minimal, or “basal,” members of the type $B$ class arising from billiard partitions (Dragović et al., 2024). In Boolean-function minimization, a basis is a minimum-cost sum-of-primes representation, and the relevant partition is a canonical decomposition of the full prime set into essential, unnecessary, and independent covering blocks (Feit, 2013). In infinite matroid theory, base partition refers to a family of bases that simultaneously covers and packs a common ground set, and mixed finitary/cofinitary families satisfy a Cantor–Bernstein-type existence theorem for such a partition (Erde et al., 2019).

1. Minimal partitions for prescribed successive ranks

For an ordinary partition

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$

with conjugate $\pi^*$ having column lengths $c_j$, the Durfee square $D(\pi)$ has size $k=\max\{i:b_i\ge i\}$. The successive rank vector is

$r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$

Fixing an integer vector $r$ of length $k$, there is a unique partition $\pi_0$ of minimal total size $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$0 among all partitions having successive rank vector $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$1; Gupta called $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$2 the basis partition of rank-vector $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$3 (Alladi, 19 Jul 2025).

The structural characterization is given in terms of the decomposition of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$4 into its Durfee square $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$5, the right-piece $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$6, and the bottom-piece $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$7. A partition is minimal exactly when no column-length in $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$8 equals any row-length in $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$9. This criterion isolates the minimal representative in the class of all partitions sharing the same successive rank vector.

The generating series for basis partitions of fixed Durfee size $\pi^*$0 is

$\pi^*$1

Here $\pi^*$2 is the number of basis partitions of $\pi^*$3 with Durfee size $\pi^*$4. This formula is obtained by comparing the standard Durfee-square generating function for ordinary partitions with the factor contributed by inserting matching rows and columns of even total size while preserving the successive rank vector.

A second description proceeds through Rogers–Ramanujan partitions. A partition is called primary when $\pi^*$5 is empty. Primary partitions into $\pi^*$6 parts are equinumerous with Rogers–Ramanujan partitions into exactly $\pi^*$7 parts. If $\pi^*$8 is a Rogers–Ramanujan partition with $\pi^*$9 strict gaps, then the weighted identity

$c_j$0

expresses the number of basis partitions of $c_j$1 and Durfee size $c_j$2 as a weighted count of Rogers–Ramanujan partitions. Combinatorially, each strict gap gives a binary choice in the sliding procedure that moves a block from the right-piece to the bottom-piece without changing the successive rank vector.

2. Signature, completeness, and the $c_j$3-modular analogue

A direct construction of the basis partition for a specified successive rank vector uses the differences

$c_j$4

Working from $c_j$5 down to $c_j$6, one attaches $c_j$7 columns of length $c_j$8 to the right of the Durfee square if $c_j$9, $D(\pi)$0 rows of length $D(\pi)$1 below the Durfee square if $D(\pi)$2, and does nothing if $D(\pi)$3. The resulting Ferrers graph has successive ranks $D(\pi)$4 and is minimal in the sense above (Alladi, 19 Jul 2025).

The signature $D(\pi)$5 of a basis partition is the number of negative $D(\pi)$6: $D(\pi)$7 Equivalently, it is the number of distinct row-lengths in $D(\pi)$8. If $D(\pi)$9 counts basis partitions of $k=\max\{i:b_i\ge i\}$0 with Durfee size $k=\max\{i:b_i\ge i\}$1 and signature $k=\max\{i:b_i\ge i\}$2, then

$k=\max\{i:b_i\ge i\}$3

Setting $k=\max\{i:b_i\ge i\}$4 yields the theta-series identity

$k=\max\{i:b_i\ge i\}$5

and hence the parity theorem: for each $k=\max\{i:b_i\ge i\}$6, the number of basis partitions of $k=\max\{i:b_i\ge i\}$7 with even signature minus the number with odd signature is $k=\max\{i:b_i\ge i\}$8 if $k=\max\{i:b_i\ge i\}$9 is a perfect square and $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$0 otherwise.

A basis partition of Durfee size $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$1 is called complete if, for each $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$2, exactly one of the two occurs: a block of length $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$3 in $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$4 or a block of length $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$5 in $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$6. Equivalently, its successive-rank vector satisfies $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$7 for $r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$8. The generating function of complete basis partitions is

$r=(r_1,r_2,\dots,r_k),\qquad r_i=b_i-c_i.$9

and for all $r$0, $r$1 is even.

The theory has a parallel form inside the set $r$2 of partitions whose odd parts are all distinct. Using $r$3-modular Ferrers graphs, one defines successive ranks by row-sums minus column-sums of the $r$4-modular diagram. For each integer vector $r$5 of length $r$6 there exists a unique minimal basis partition $r$7 with those successive ranks. The characterization again uses incompatibility between row-sums in $r$8 and column-sums in $r$9, with the additional condition that if $k$0 then the Durfee-corner is $k$1. The corresponding parity theorems identify exactly the values $k$2 and $k$3 at which the even–odd signature imbalance is nonzero.

3. Type $k$4 basis partitions and billiard partitions

A different notion of basis partition arises from Euclidean billiard partitions. These are partitions

$k$5

into distinct parts such that $k$6 is even and no two consecutive parts $k$7 are both odd. The associated basis partitions form the minimal subset $k$8 consisting of those partitions that cannot be further shifted down by $k$9 in any part while remaining in the billiard class (Dragović et al., 2024).

Equivalently, a strict partition

$\pi_0$0

belongs to $\pi_0$1 if and only if it satisfies three axioms: $\pi_0$2; no two consecutive parts are both odd; and every gap satisfies $\pi_0$3. The subset $\pi_0$4 of basis partitions with $\pi_0$5 parts is therefore determined by a local parity condition together with a bounded-gap condition.

The enumeration is Fibonacci. With $\pi_0$6, $\pi_0$7, and $\pi_0$8, one has $\pi_0$9. More precisely, among these $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$00 have largest part even and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$01 have largest part odd. Refining by largest part $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$02, if

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$03

then in the billiard case

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$04

The billiard class is the $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$05 case of a general type $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$06 SIP class. For modulus $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$07 and odd-like residue count $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$08, the counts $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$09 satisfy

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$10

The specializations $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$11, $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$12, and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$13 yield the Fibonacci, Pell-type, and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$14 recurrences, respectively.

Section $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$15 of the same work relates the generating series of even-part basis partitions to the quiver generating series of the symmetric two-vertex quiver with adjacency matrix

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$16

after the specialization $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$17, $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$18. The same quiver occurs in the knots–quivers correspondence for the $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$19-framed unknot, and a quotient of its generating series gives, up to a power of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$20, the generating function of weighted Schröder paths. In the general SIP framework, every partition in the full billiard class $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$21 uniquely decomposes as a sum of a basis element and an even partition.

4. Canonical basis partitions in Boolean-function minimization

For a Boolean function $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$22, a prime implicant is an implicant from which no literal can be dropped without losing the implicant property. A basis for $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$23 is a set of primes whose disjunction is equivalent to $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$24, and the optimization problem is to find a basis of minimum total cost under a positive additive cost function (Feit, 2013).

Feit’s canonical partition splits the complete set of primes into three disjoint parts: Essential Primes, Unnecessary Primes, and disjoint subsets $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$25, each with its own covering table $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$26. An essential prime is one that no other set of primes covers; all essential primes must appear in every basis. After removing the essentials, some primes cover only products that the essentials already cover; these free non-essentials, or surplus primes, cannot appear in any minimum-cost basis. The remaining primes are organized into the disjoint blocks $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$27, and the overall minimization reduces to independent subproblems on these blocks plus the essentials.

The key organizing notion is that of Ancestor Sets. By graph reachability on a table of covering triples, one identifies Independent Ancestor Sets $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$28. Each $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$29 contains precisely the primes $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$30 that must be treated as a block, with no interaction across distinct blocks. The Ancestor Theorem states that if $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$31 is an Ancestor Set, $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$32, $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$33 is the set of essentials, and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$34 is the set of all other primes not in $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$35, then every minimum-cost basis has the unique decomposition

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$36

where $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$37 is a minimum-cost cover of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$38 and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$39 is a minimum-cost cover of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$40.

Covering is encoded by the Cascade operation. A covering table consists of triples $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$41 meaning that once primes $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$42 and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$43 are chosen, prime $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$44 is automatically covered, together with the one-prime case $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$45. For a set $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$46, the iterative process $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$47 removes covered triples and propagates newly covered products until stabilization. One has

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$48

if and only if $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$49 covers $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$50. This converts each block into a finite covering-table optimization problem.

The block structure can often be simplified further. The Span-Basis Theorem states that if some $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$51 has $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$52 and is cheapest among the equivalent primes with the same span, then $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$53 is the unique minimum-cost cover of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$54. Even when no such single-prime cover exists, Independent-Prime Decomposition may split a block into smaller independent subblocks that can be minimized separately.

5. Base partitions in mixed families of infinite matroids

Let $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$55 be a family of matroids on a common ground set $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$56. A choice of bases $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$57 is a base covering if $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$58, a base packing if the $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$59 are pairwise disjoint, and a base partitioning if it is both a covering and a packing (Erde et al., 2019).

The main theorem is a Cantor–Bernstein-type result. If every $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$60 is either finitary or cofinitary, then $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$61 admits a base partitioning if and only if it admits both a base covering and a base packing. Thus, for mixed families of finitary and cofinitary matroids, the simultaneous existence of a covering by bases and a packing by bases is sufficient to obtain a genuine partition of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$62 by bases.

The proof has two stages. In the countable-ground-set or finite-index case, one recursively constructs feasible partial assignments

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$63

with $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$64, each $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$65 independent, each $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$66 spanning, the $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$67 pairwise disjoint, and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$68. The construction ensures continued extendibility to both a covering and a packing. After $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$69 steps one sets

$\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$70

Finitarity and cofinitarity ensure that each $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$71 is both independent and spanning in the appropriate matroid, while the recursive invariants ensure disjointness and covering. For large $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$72 or $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$73, the argument reduces to small subproblems via a continuous chain of subsets $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$74 of size $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$75, where $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$76, and stitches the local partitions together.

The theorem is best possible in the sense made explicit by a consistency result. Without the finitary/cofinitary hypothesis, and under the Continuum Hypothesis, there exists a countable uniform matroid $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$77 on $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$78 such that two copies of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$79 admit a packing and a covering but no partitioning. The construction uses a refined $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$80-system argument and the fact that under CH the reaping number $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$81.

Several special cases recover known themes. If each $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$82 is finite-rank and $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$83 is finite, the theorem recovers Edmonds’ matroid-union theorem and its dual. For a countable graph $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$84, taking the finite-cycle matroid $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$85 yields the statement that if there is both a covering by $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$86 spanning trees and a packing of $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$87 edge-disjoint spanning trees, then $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$88 can be partitioned into $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$89 spanning trees. When $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$90, the theorem gives a matroid Cantor–Bernstein criterion for the existence of a common base. The work also leaves open whether the partition property extends to tame matroids or nearly $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$91finitary matroids, and whether a failure can be proved in ZFC for arbitrary matroid families.

6. Comparative scope of the term

The cited usages suggest three recurrent patterns. In partition theory, a basis partition is a minimal representative attached to a fixed invariant, such as a successive rank vector or a type $\pi: b_1\ge b_2\ge \cdots \ge b_\nu>0,$92 separable-partition class (Alladi, 19 Jul 2025, Dragović et al., 2024). In Boolean minimization, the term appears in the optimization problem of finding a minimum-cost basis, and the partition is a canonical decomposition of the prime set into independent covering blocks (Feit, 2013). In matroid theory, the partition is literal: the ground set is partitioned by bases belonging to different matroids (Erde et al., 2019).

A common misconception is to read these constructions as instances of a single standard notion. The literature instead uses the same phrase for mathematically different objects: Ferrers-graph minima, separable integer partitions, prime-implicant decompositions, and base packings/coverings in infinite matroids. What they share is not a common formal definition, but the repeated combination of two ideas: a distinguished family of generators or bases, and a partition operation that either isolates irreducible components or realizes a global decomposition.

This suggests that “basis partition” is best treated as a context-dependent technical term. In some areas it identifies canonical minimal objects; in others it identifies a decomposition of a search space into independent subproblems; and in infinite matroid theory it expresses a precise covering-packing equivalence theorem.