Bivariate Flow Polynomial in Signed Graphs

• Bivariate flow polynomial is a two-variable invariant that counts nowhere-zero flows in signed graphs by incorporating A and A[2] parameters.
• It employs chain groups, boundary operators, and homological methods to reconcile complexities introduced by negative cycles and outer-edges.
• Deletion–contraction recurrences and expansion formulas illustrate deep connections with Tutte polynomial concepts and computational challenges.

Searching arXiv for the specified paper and closely related signed-graph/Tutte work mentioned in the source data. {} The bivariate flow polynomial of a signed graph is the unique polynomial $F_\Sigma(x,y)$ such that, for every finite abelian group $A$, the number of nowhere-zero $A$-flows on $\Sigma$ is $F_\Sigma(|A|,|A[2]|)$, where $A[2]=\{a\in A:2a=0\}$ is the $2$-torsion subgroup. In the formulation developed for signed graphs with possible outer-edges, the invariant arises from a chain-group and homological framework in which flows are $\ker \partial$, tensions are defined through orthogonality to integer-valued flows, and the frame matroid appears via minimal supports of nonzero flows. The second variable is forced by the signed setting: negative circles constrain flow values to $A[2]$, so nowhere-zero flow counts are not governed solely by $|A|$ (Chen, 15 Aug 2025).

1. Signed graphs, chain groups, and incidence structure

A signed graph is a triple $A$0, where $A$1 is a finite vertex set, $A$2 is a finite edge set, and $A$3 assigns a sign to each edge. The framework allows outer-edges as well as loops and links. An outer-edge has one end attached to a vertex and the other end “outside”; the paper treats these edges as carrying a sign, while emphasizing that their role is topological through noncompactness and algebraic through incidence (Chen, 15 Aug 2025).

Each edge has two ends and admits a bi-directional orientation. If an edge is positive, the two arrows point in the same direction along the edge; if it is negative, they point in opposite directions. For a link or loop $A$4, the orientations are:

• positive link or loop: $A$5 and $A$6,
• negative link or loop: $A$7 and $A$8.

For an outer-edge $A$9, the orientations are:

• positive outer-edge: $A$0 and $A$1,
• negative outer-edge: $A$2 and $A$3.

The signed incidence function is defined on $A$4 by

$A$5

and extended by $A$6. After fixing an orientation $A$7, one obtains the signed incidence matrix $A$8 with $A$9 for $\Sigma$0 (Chen, 15 Aug 2025).

With coefficients in an abelian group $\Sigma$1, the $\Sigma$2-chain group $\Sigma$3 consists of functions $\Sigma$4 satisfying $\Sigma$5, and the $\Sigma$6-chain group $\Sigma$7 consists of functions $\Sigma$8. The boundary operator

$\Sigma$9

has the equivalent edgewise description

• $F_\Sigma(|A|,|A[2]|)$0 for a positive link or loop $F_\Sigma(|A|,|A[2]|)$1,
• $F_\Sigma(|A|,|A[2]|)$2 for a negative link or loop $F_\Sigma(|A|,|A[2]|)$3,
• $F_\Sigma(|A|,|A[2]|)$4 for an outer-edge arc $F_\Sigma(|A|,|A[2]|)$5.

The coboundary operator

$F_\Sigma(|A|,|A[2]|)$6

is equivalently given by

• $F_\Sigma(|A|,|A[2]|)$7 for $F_\Sigma(|A|,|A[2]|)$8,
• $F_\Sigma(|A|,|A[2]|)$9 for $A[2]=\{a\in A:2a=0\}$0,
• $A[2]=\{a\in A:2a=0\}$1 for $A[2]=\{a\in A:2a=0\}$2, with $A[2]=\{a\in A:2a=0\}$3 (Chen, 15 Aug 2025).

This setup standardizes how the sign of an edge affects conservation. In particular, negative edges replace differences by sums, and negative loops contribute $A[2]=\{a\in A:2a=0\}$4 at their incident vertex.

2. Flows, tensions, and homological structure

The flow group is

$A[2]=\{a\in A:2a=0\}$5

A nowhere-zero $A[2]=\{a\in A:2a=0\}$6-flow is a flow $A[2]=\{a\in A:2a=0\}$7 such that $A[2]=\{a\in A:2a=0\}$8 for all edges $A[2]=\{a\in A:2a=0\}$9. The dependence on signed incidence is explicit: on negative edges the conservation law uses $2$0 rather than $2$1, and on a negative loop the local contribution is doubled (Chen, 15 Aug 2025).

The tension group is defined through the canonical bilinear pairing

$2$2

where

$2$3

An element $2$4 is a tension if $2$5 for all $2$6. The paper states that $2$7 and that the adjoint identity

$2$8

holds (Chen, 15 Aug 2025).

The boundary group and the homology and cohomology groups are

$2$9

A central structural point is that this chain-group formalism is not merely notational. It aligns the combinatorics of signed graphs with algebraic-topological operators and makes flows, tensions, homology, and cohomology available uniformly even in the presence of outer-edges. This suggests a deliberate replacement of ad hoc signed-graph case analysis by a systematic linear-algebraic and homological treatment.

3. Frame matroid interpretation

The paper’s guiding principle is the correspondence between representable matroids over $\ker \partial$0 on a ground set $\ker \partial$1 and subspaces of the vector space of real-valued chains on the same ground set. In this setting, the frame matroid of a signed graph emerges by defining circuits as minimal supports of nonzero flows rather than by listing circuit patterns independently (Chen, 15 Aug 2025).

Let $\ker \partial$2 be the real-valued flow space, with the standard inner product on $\ker \partial$3. An elementary flow is a nonzero flow whose support is minimal under inclusion. The supports of elementary flows form the circuit family of a matroid $\ker \partial$4 on $\ker \partial$5, and this matroid is the frame matroid $\ker \partial$6. The signed-graph circuits are precisely the following five types:

• (C1) Open line $\ker \partial$7.
• (C2) Positive circle $\ker \partial$8.
• (C3) Two negative circles sharing a unique common vertex.
• (C4) A negative circle together with a half-line meeting at the endpoint of the half-line.
• (C5) Two disjoint negative circles connected by a line segment.

For a chosen direction $\ker \partial$9 of such a circuit, the principal integer-valued circular flow is defined by

$A[2]$0

Dually, tensions span a subspace $A[2]$1, and elementary tensions yield the circuits of the dual matroid $A[2]$2. These cocircuits are precisely the bonds of the signed graph: minimal edge sets $A[2]$3 such that

$A[2]$4

and

$A[2]$5

Equivalently, a bond is a minimal support of a nonzero tension (Chen, 15 Aug 2025).

Deletion and contraction follow signed-graphic edge operations with special care for negative loops and outer-edges. Positive-link contraction merges endpoints; a bridge or an outer-edge co-loop yields no additional flows; a positive loop contributes a direct $A[2]$6-summand to $A[2]$7; and a negative loop that is also a co-loop contributes an $A[2]$8-summand. This is the matroidal mechanism behind the later recurrence formulas.

4. Definition, expansion, and the necessity of two variables

For a signed graph $A[2]$9, the bivariate flow polynomial $|A|$0 is characterized by

$|A|$1

for every finite abelian group $|A|$2. The paper also uses $|A|$3 with $|A|$4 and $|A|$5, and sets $|A|$6 (Chen, 15 Aug 2025).

Its main expansion theorem is

$|A|$7

where $|A|$8 is the number of compact unbalanced components of the spanning subgraph $|A|$9, and $A$00 is its cycle rank, equivalently the matroid corank

$A$01

The theorem implies that the number of nowhere-zero $A$02-flows depends only on $A$03 and $A$04 (Chen, 15 Aug 2025).

The reason a univariate invariant fails in general is explicit. For unsigned or balanced graphs, nowhere-zero flow counts depend only on $A$05, producing the ordinary univariate flow polynomial. In signed graphs, however, every flow on a directed negative circle $A$06 has the form $A$07 with $A$08. Thus the cardinality of the $A$09-torsion subgroup directly affects enumeration. For $A$10, one has $A$11 when $A$12 is odd and $A$13 when $A$14 is even, so the counting functions differ by parity class (Chen, 15 Aug 2025).

The simplest examples already exhibit the dichotomy:

• a single negative loop has $A$15,
• a single positive loop has $A$16,
• a single positive link has $A$17.

A plausible implication is that the second variable $A$18 is not a refinement added for formal symmetry, but the minimal additional parameter needed to record the obstruction created by negative cycles.

5. Structural properties, recurrences, and degrees

The polynomial satisfies deletion–contraction recurrences that mirror the edge taxonomy of signed graphs. If $A$19, then (Chen, 15 Aug 2025):

• if $A$20 is a positive loop,

$A$21

• if $A$22 is a negative loop and also a co-loop,

$A$23

• if $A$24 is a positive link,

$A$25

The same source states that $A$26 if and only if $A$27 contains a co-loop that is either a bridge co-loop or an outer-edge co-loop. The base cases are:

• edgeless signed graph: $A$28,
• single positive loop: $A$29,
• single negative loop: $A$30,
• single outer-edge: $A$31,
• single positive link: $A$32.

Two invariance properties are singled out. First, $A$33 is multiplicative on disjoint unions: $A$34 when $A$35. Second, it is switching invariant, because switching preserves flows, tensions, and the signed incidence data up to isomorphism (Chen, 15 Aug 2025).

The paper also ties polynomial degrees to signed-graphic rank parameters. Writing

$A$36

one has $A$37, where $A$38 is the negative cycle-rank. More specifically:

• if $A$39 has no negative circles, then $A$40 is monic of degree $A$41;
• if $A$42 has outer-edges, then $A$43 is monic of degree $A$44;
• if $A$45 has negative circles and no outer-edges, then $A$46 and $A$47 may vanish;
• if $A$48 has negative circles and no outer-edges, then $A$49 is monic of degree $A$50 (Chen, 15 Aug 2025).

These statements place the $A$51- and $A$52-degrees under matroidal control, while separating the role of total cycle rank from the contribution of negative cycles.

6. Classical specializations, coefficient interpretations, and examples

When $A$53 is balanced and contains no outer-edges, the bivariate polynomial specializes to the ordinary graph flow polynomial: $A$54 In this case $A$55, and the signed theory collapses to the unsigned one. In particular, for cycles $A$56, $A$57, and for general graphs $A$58, $A$59 counts nowhere-zero $A$60-flows (Chen, 15 Aug 2025).

The relation to tensions and Tutte-type invariants is stated in dual language. Tensions are dual to flows through the orthogonal decomposition $A$61. For balanced graphs, the tension polynomial is dual to the flow polynomial through the Tutte polynomial $A$62. For signed graphs and the frame matroid $A$63, $A$64 is described as a switching-invariant bivariate invariant and as an inclusion–exclusion evaluation weighted simultaneously by cycle rank and unbalanced compact components. The paper notes that a signed-graphic dichromate due to Goodall–Litjens–Regts–Vena extends Tutte’s universality, while a full Tutte-type universality for $A$65 itself requires multivariate parameters for balanced and unbalanced components (Chen, 15 Aug 2025).

The coefficient interpretation follows directly from the expansion: $A$66 Thus the coefficient of $A$67 is the alternating sum over subsets $A$68 with $A$69 and $A$70. Here $A$71 counts independent cycles, while $A$72 counts unbalanced compact components. The maximal exponent of $A$73 equals $A$74, described as the negative cycle-rank (Chen, 15 Aug 2025).

Several concrete examples are given:

• a single negative link between two vertices has $A$75;
• a balanced cycle $A$76 has $A$77;
• an unbalanced cycle with one negative edge has $A$78;
• two loops at a vertex, one positive and one negative, give

$A$79

• a triangle with all positive edges gives

$A$80

• a triangle with one negative edge gives

$A$81

For the family $A$82 with one vertex, $A$83 outer-edges, and $A$84 negative loops, the paper records the base values

$A$85

the recurrences

$A$86

$A$87

and the closed forms

$A$88

$A$89

$A$90

The paper also describes a parity phenomenon. For $A$91, one has

$A$92

Accordingly, the counting function in $A$93 is a quasi-polynomial of degree $A$94 with period $A$95 when negative cycles are present, and a genuine polynomial when $A$96 is balanced (Chen, 15 Aug 2025).

7. Computation, complexity, and broader consequences

Two general computation methods are identified. The first is the inclusion–exclusion expansion over all $A$97, which requires determining $A$98 and $A$99 for each subset and is therefore exponential in $A$00. The second is deletion–contraction based on the recurrences for positive loops, negative loop co-loops, and positive links, again exponential in general (Chen, 15 Aug 2025).

The paper further states that dynamic programming yields polynomial-time computation on special classes, specifically cactus graphs, series–parallel signed graphs, and graphs with bounded treewidth, because on those subclasses $A$01 and $A$02 can be computed efficiently. It also notes that computing $A$03 generalizes computation of the Tutte polynomial and of the ordinary flow polynomial for unsigned graphs, and is therefore expected to be $A$04-hard in general (Chen, 15 Aug 2025).

The broader significance attributed to the invariant is structural. The chain-group framework introduces boundary and coboundary operators intrinsically for signed graphs with outer-edges, harmonizing flows and tensions with algebraic topology and matroid theory. Within this framework:

• flows are $A$05,
• tensions are $A$06 and, more generally, orthogonal complements annihilating integer flows,
• circuits and bonds emerge as minimal supports of elementary flows and tensions,
• the frame matroid is identified through elementary chains,
• the bivariate flow polynomial resolves the “mystery” of the nonexistence of a univariate flow polynomial for signed graphs by isolating the essential dependence on $A$07 (Chen, 15 Aug 2025).

In that sense, the bivariate flow polynomial occupies the role that the ordinary flow polynomial plays for balanced graphs, but in a form adapted to signed incidence, negative cycles, and outer-edges. It is simultaneously a counting invariant, a deletion–contraction invariant, and a matroidally interpretable invariant derived from homological data.