1. Introduction
A Tutte–Grothendieck invariant
is a mapping from a class
of matroids to a commutative ring with the property that there are elements
from the ring such that for a matroid
on the ground set
E, we have
(Notice that
is an isthmus of
M if it is contained in each base of
M, and
is a loop of
M if it belongs to no bases of
M.) The best known Tutte–Grothendieck invariant is the Tutte polynomial:
which maps matroids to the ring of integral polynomials with two variables,
x and
y. This invariant was introduced by Tutte in [
1] for graphs and encodes many properties of graphs and matroids. Applications of the Tutte polynomial in combinatorics, knot theory, statistical physics, and coding theory are surveyed in [
2,
3,
4,
5].
We consider classes of matroids whose ground sets contain a fixed subset
B and study functions from the matroids to finite sets. For each of the matroids, consider the cardinality of a set of functions with fixed values on
B. We show that if the cardinalities satisfy contraction–deletion rules, then there exist relations among these numbers that can be expressed in terms of linear algebra. In this way, we study numbers of regular chain groups, nowhere-zero flows and tensions on graphs, and acyclic and totally cyclic orientations of oriented matroids and graphs. These results generalize the approach that we introduced in [
6,
7].
2. Matroids and -Classes
Throughout this paper, we consider finite matroids on finite sets. The ground set of a matroid M we denote by .
Let
B be a finite set. A class of matroids
is called a
B-class if
There exist only finitely many matroids on
B. Thus, there exists a finite set
consisting of pairwise nonisomorphic matroids on
B belonging to
. (For example, if
is a class of matroids closed under contraction and deletion, then it is an ∅-class and
.)
The collection of mappings from
E to a finite set
S is denoted by
. Assume that
M is a matroid from
with the ground set
. By an
S-function on
M, we mean any function
. Then,
denotes the restriction of
f to
B (i.e.,
so that
for each
). Let
be a class of
S-functions on matroids from
. If
, then
denotes the set of
S-functions on
M belonging to
. For every
and
, let
If
and
, then
M is called
-trivial; otherwise, it is called
-nontrivial. In this paper, we denote by
an ordered
n-tuple
of all
-nontrivial elements of
. For each
and
, let
(
) if
, (
). Let
.
Let denote the standard basis vector of and let denote the zero vector of . Vectors from are considered row vectors. If , then the dot product can be expressed as a matrix multiplication .
Assume that there exist rational numbers
such that for each
, each
, and each
, we have
In this case, we say that
is
-regular. Since the cardinalities of sets are nonnegative integers,
,
must be nonnegative, but only one of
,
can be negative (but not both).
Theorem 1. Let be a B-class of matroids, with B finite, be an -regular class of S-functions on matroids from , with S finite, and . Then, for each , there exists a vector such that for every , , i.e., . Furthermore, if are integers, then can be chosen to be an integral vector.
Proof. We apply induction on . Let and . If M is -trivial, then , and we can set . If M is -nontrivial, then , where , and . Thus, satisfies the assumptions.
If , then there exists . By the induction hypothesis, there are vectors and such that for every , and .
If
e is an isthmus, then from the first row of (
1),
, where the vector
satisfies the assumptions.
If
e is a loop, then from the second row of (
1),
, where the vector
satisfies the assumptions.
If
e is neither an isthmus nor a loop, then from (
1),
, where the vector
satisfies the assumptions.
If are integers, then all vectors considered in the proof are integral. This proves the statement. □
Let , , be an ordered basis of the linear hull of . Denote such that . For example, if , then we can choose and then .
Theorem 2. Let be a B-class of matroids, with B finite, be an -regular class of S-functions on matroids from , with S finite, , and , , be an ordered basis of the linear hull of . Then, for each , there exists a unique vector such that for every , . Furthermore:
If and M is trivial, then ;
If and , , then such that is the i-th coordinate of , ;
If , then satisfies the following recursive rules:
Finally, if are integers and are integral vectors, then is an integral vector for each M from .
Proof. We prove the existence of by induction on . Let and . If M is -trivial, then , and we can set . If M is -nontrivial and , , then from the proof of Theorem 1, . Let A be an -matrix with the j-th row equal to , . Using matrix multiplication, we can express and , where . Thus, , where such that is the i-th coordinate of , .
If , then there exists . By the induction hypothesis, there are integral vectors and such that for every , and .
If
e is an isthmus, then from the first row of (
1),
, where
.
If
e is a loop, then from the second row of (
1),
, where
.
If
e is neither an isthmus nor a loop, then from (
1),
, where the vector
satisfies the assumptions. This proves (
2).
The uniqueness of follows from the fact that Z is a basis of the linear hull of .
Furthermore, if
are integral vectors, then
is integral for each
such that
. If
are also integers, then from (
2),
are integral vectors for each
M from
. □
We apply Theorem 1 for various S-functions of B-classes of matroids. Analogously, we can apply Theorem 2.
3. Regular Chain Groups
If R is a ring, the elements of are considered vectors indexed by E, and we will use the notation , , and for , and . A chain on E (over R, or simply an R-chain) is , and the support of f is . The zero chain has null support. Given and , define such that for each .
A matroid M on E of rank is regular if there exists an (, ) totally unimodular matrix D (called a representative matrix of M) such that independent sets of M correspond to independent sets of columns of D.
We recall properties of regular matroids presented in [
1,
8,
9,
10,
11]). For any basis
B of
M,
D can be transformed to a form
such that
corresponds to
B and
U is totally unimodular. The dual of
M is a regular matroid
with a representative matrix
(where
corresponds to
).
By a regular chain group N on E (associated with D), we mean a set of chains on E over that are orthogonal to each row of D (i.e., are integral combinations of rows of a representative matrix of ). The set of chains orthogonal to every chain of N is a chain group called orthogonal to N and denoted by (clearly, is the set of integral combinations of rows of D). By the rank of N, we mean . Then, . We always assume that a regular chain group N is associated with a matrix representing a matroid .
We have and . Clearly, arises from after deleting the columns corresponding to X. Furthermore, and .
A chain
f of
N is
elementary if there is no nonzero
of
N such that
. An elementary chain
f is called a
primitive chain of
N if the coefficients of
f are restricted to the values 0, 1, and
. (Notice that the set of supports of primitive chains of
N is the set of circuits of
.) We say that a chain
gconforms to a chain
f if
and
are nonzero and have the same sign for each
such that
. From [
1] (Section 6.1),
Let
A be an Abelian group with additive notation. We shall consider
A as a (right)
-module such that the scalar multiplication
of
by
is equal to 0 if
,
if
, and
if
. Similarly, if
and
, then define
so that
for each
. If
N is a regular chain group on
E, define
Notice that
if
. From [
8] (Proposition 1),
Suppose that is a B-class of regular matroids, with B finite. Denote by the class of -functions on matroids from such that for each . In other words, is the class of where . We claim that is -regular.
Lemma 1. For each , , and , we have Proof. Notice that
e is a loop (isthmus) of
e if
(
). Thus, if
e is an isthmus of
M, then from (
5), each
satisfies
, where
.
Given and , let be defined so that and .
If
e is a loop of
M, then from (
5), for each
and
,
. Similarly, if
, then
. Thus,
.
If
e is neither an isthmus nor a loop of
, then there exists
such that
and
. From (
3), for any
, there exists
such that
. From (
5),
must be orthogonal to
, where
a is unique. Furthermore, if
(resp.
), then from (
3),
(resp.
); i.e.,
is a bijection from
to the disjoint union of
and
. This implies the last row of (
6). □
Corollary 1. Suppose that is a B-class of regular matroids, with B finite, and let be the class of where . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. It follows from (
6) and Theorem 1. □
4. Nowhere-Zero Flows and Tensions on Graphs
We deal with finite undirected graphs with multiple edges and loops. If G is a graph, then and denote its vertex and edge sets, respectively. Every edge e of G determines two opposite arcs arising from it after endowing e with two distinct orientations. All arcs obtained in this way are called arcs of G, and the set of them is called the arc set of G and denoted by . Clearly, . If x is an arc of G, then denote by the second arc arising from the same edge. Clearly, and for every arc x of G. If , then let denote . For any vertex v of G, denote by the set of arcs from directed out of v. If A is an Abelian group, then a nowhere-zero A-chain in G is a mapping such that for every .
By an orientation of G, we mean any such that and . In other words, an orientation of G can be considered a directed graph arising from G after endowing each edge with an orientation.
Let A be an Abelian group with additive notation. A nowhere-zero A-chain in G is called a nowhere-zero A-flow if for every vertex v of G. Considering as a mapping on an arbitrary but fixed orientation of G, we obtain the usual definition of nowhere-zero A-flows. Such nowhere-zero A-flows on G coincide with , where N is the regular chain group associated with , the cycle matroid of G (edge sets of subforests of G forming independent sets of ).
By a B-class of graphs, we mean a class such that for each , , and for each , . Then, the class of cycle matroids of graphs from is a B-class of matroids . Denote by the class of nowhere-zero A-flows on a graph from . Clearly, coincides with the class associated with described in the previous section. Analogously, we write instead of , where is the cycle matroid of for .
Lemma 2. For each , , and , we have Proof. Apply Lemma 1 for a class of cycle matroids of graphs from . □
Corollary 2. Suppose that is a B-class of graphs, with B finite, and let be the class of nowhere-zero A-flows on graphs from . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. Apply Corollary 1 for a class of cycle matroids of graphs from . □
We applied the idea of Corollary 2 in [
6,
7,
12,
13] and proved that the smallest counterexample to the 5-flow conjecture of Tutte (that every bridgeless graph has a nowhere-zero 5-flow) must be cyclically 6-edge-connected and has a girth of at least 11.
A circuit C of G is a connected 2-regular subgraph of G (notice that the loop is a circuit of order 1). By a directed circuit of G, we mean an orientation X of C such that for each vertex v of C.
A nowhere-zero A-chain in G is called a nowhere-zero A-tension if for every directed circuit X of G. Considering as a mapping on an arbitrary but fixed orientation of G, we obtain nowhere-zero A-tensions on G that coincide with such that is the bond matroid of G (dual of the cycle matroid of G). Denote by the class of nowhere-zero A-tensions on graphs from . Clearly, coincides with the class associated with the class of bond matroids of graphs from . Therefore, is -regular.
Lemma 3. For each , , and , we have Proof. Apply Lemma 1 for the class of bond matroids of graphs from . □
Corollary 3. Suppose that is a B-class of graphs, with B finite, and let be the class of nowhere-zero A-tensions on graphs from . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. Apply Corollary 1 for the class of bond matroids of graphs from . □
5. Orientations in Oriented Matroids
In this section, we use notation and results from [
14,
15] (see also [
9,
16,
17]). We define a signed set
X to be a set
, called the set
underlying X, and the mapping
, called the
signature of
X. Let
X be a signed set. Then,
is partitioned into two distinguished subsets:
and
. The
opposite of
X is defined by
and
. If
is a subset of
E, then
X will be called a
signed subset of
E, and if
, then we write
.
An
oriented matroid M on
E is a couple
, where
is a collection of signed sets satisfying
Signet sets from
are called
signed circuits of
M. Let
. Then,
is a collection of circuits of a matroid
on
E. The circuits of the dual matroid
(i.e., the cocircuits of
) can be oriented in a unique way such that the
of signed cocircuits of
M satisfies the
orthogonality property: for all
and
such that
, both
and
are non-empty. Then,
satisfies (
7)–(
9) and defines an oriented matroid
, the
dual of
M. The orthogonality property holds for all
and
such that
. We have
. Thus, the class of oriented matroids is a minor and dual closed class of matroids.
A circuit
is
positive if
. We say that
is
totally cyclic if each
is contained in a positive circuit
and that
is
acyclic if no
is positive. From [
14] (Theorem 2.2),
For any , denote by the oriented matroid obtained from M by reversing signs on Z, i.e., , where , supposing that satisfies and . Set such that if and if . If X is a directed circuit of with the signature , then the signature of () satisfies for each . Thus, uniquely determines .
Let
M be an oriented matroid on
E and
. From [
15] (Lemma 3.1.1),
Suppose that is a B-class of oriented matroids. For any and , denote by the set of subsets Z of such that is acyclic. Since is uniquely determined by , can be considered a set of -functions corresponding to acyclic orientations. Denote by the union of , where M runs through and Y runs through the subsets of B. We claim that is -regular.
Lemma 4. For any , , and , Proof. The statement is obvious if
e is an isthmus or a loop of
M. Let
be neither an isthmus nor a loop of
M. For a subset
Z of
, set
if
is not acyclic and
if
is acyclic. We have
If
is not a loop of
M and
Z is a subset of
, then from (
11)–(
13), we have
Now,
. Summing up for all subsets
Z of
, we obtain
as required. □
Considering as the class of -functions on matroids from corresponding to acyclic orientations, any coincides with such that . Thus, we can write and instead of and , respectively. We apply this notation in the following corollary of Theorem 1.
Corollary 4. Suppose that is a B-class of oriented matroids, with B finite, and let be the class of acyclic orientations of oriented matroids from . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. It follows from Lemma 4 and Theorem 1. □
For any , denote by the set of subsets Z of such that is totally cyclic. Since is uniquely determined by , can also be considered set of -functions corresponding to totally cyclic orientations. Denote by the union of , where M runs through and Y runs through the subsets of B. We claim that is -regular.
Lemma 5. For any , , and , Proof. It follows from Lemma 4 and (
10). □
Similar to the above, can be considered the class of -functions on matroids from corresponding to totally cyclic orientations. Any coincides with such that , and we can write and instead of and , respectively.
Corollary 5. Suppose that is a B-class of oriented matroids, with B finite, and let be the class of totally cyclic orientations of oriented matroids from . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. It follows from Corollaries 4 and (
10). □
Let
M be a regular matroid on
E associated with a totally unimodular matrix
D and
N be the regular chain group associated with
D. The set of circuits of
M coincides with the set of supports of primitive chains of
N. If fact, each circuit
of
M corresponds to exactly one primitive function
of
N such that
. The set of primitive functions forms a set of oriented circuits of an oriented matroid (see [
15]). Thus, we can apply Lemmas 4 and 5 and Corollaries 4 and 5 for any
B-class of regular matroids.
6. Orientations of Graphs
Consider a fixed orientation D of a graph G. Each circuit C in G indicates two directed circuits; we denote one of them by Q and the other one by . The edges of C and Q indicate a signed set X such that , consists of the edges having the same orientation in D and Q, and consists of the edges having different orientations in D and Q. Then, indicates in an analogous way. Applying this process for each circuit of G, we generate a set such that is an oriented matroid M on , and the underlying matroid is the cycle matroid of G; i.e., is the set of circuits of G.
If , then denote by the orientation of G arising from D after changing the orientation of edges from Z. Clearly, corresponds to . Analogously, an orientation D of G is totally cyclic if each edge of G is covered by a directed circuit and is acyclic if no edge of G is covered by a directed circuit.
Recall that a B-class of graphs is a class such that for each , , and for each , . The class (resp. ) of acyclic (resp. totally cyclic) orientations of digraphs from is the class of acyclic (resp. totally cyclic) orientations of matroids from the class of cyclic matroids of graphs from . Similarly, we write (resp. ) instead of (resp. ), supposing that M denotes the cyclic matroid of G. Analogously, we write instead of , where is the cycle matroid of for .
Lemma 6. For each , , and , Proof. It follows from Lemma 4. □
Corollary 6. Suppose that is a B-class of graphs, with B finite, and let be the class of acyclic orientations of graphs from . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. It follows from Corollary 4. □
Lemma 7. For each , , and , Proof. It follows from Lemma 5. □
Corollary 7. Suppose that is a B-class of graphs, with B finite, and let be the class of totally cyclic orientations of graphs from . Assume that . Then, for each , there exists an integral vector such that for every , .
Proof. It follows from Corollary 5. □