GitHub

Matroids

We presume familiarity with matroids. See the What is a Matroid? section of this documentation.

Creating Matroids

In this implementation of matroids, the ground set, S, is always of the form {1,2,...,m} where m is a nonnegative integer.

Matroid from a Matrix

Given a matrix A, use Matroid(A) to create a matroid based on the column vectors in A.

Matroid from a Graph or Multigraph

Given a graph g, use Matroid(g) to create the cycle matroid of g. Here, g is an undirected graph from the Graphs module. The graph may have loops, but multiple edges are not supported by Graphs.

We also provide a basic implementation of multigraphs, EasyMultiGraph, that allows multiple edges and loops. [See the Multigraphs section of this documentation.] If g is an EasyMultiGraph, then Matroid(g) creates its cycle matroid.

Uniform Matroids

Use UniformMatroid(m,k) to create a matroid whose ground set is {1,2,...,m} in which all sets of size k or smaller are independent.

Matroid Properties

Display Format

A matroid is printed (e.g, using println) in the form {m, r} matroid where m is the number of elements in the matroid and r is its rank. Example:

julia> using Matroids, Graphs

julia> Matroid(complete_graph(5))
{10, 4} matroid

Basic Properties

Let M be a matroid.

  • ne(M) returns the number of elements in the ground set of M.

  • rank(M) returns the rank of the matroid M.

  • rank(M,X) returns the rank of the set X in the matroid M.

  • isindependent(M,X) checks if X is an independent subset of the elements of M.

  • iscircuit(M,X) checks if X is a circuit of M.

  • isloop(M,e) checks if e is a loop element of M.

  • iscoloop(M,e) checks if e is a coloop element of M.

Bases

A basis of a matroid is a maximum-size independent set. To find a basis of a matroid M, use basis(M). Note that a matroid typically has many bases. This function returns one of them with no guarantee as to which.

Given weights wt (specified as a Dict) for the elements of a matroid M, use min_weight_basis(M, wt) to return a basis the sum of whose weights is smallest.

The function random_basis(M) returns a random basis of M by the following algorithm: Assign random weights to the elements of M and then apply min_weight_basis.

Finally, all_bases(M) returns an iterator that generates all the bases of M. Note that the number of bases may be enormous.

Closure

The closure of a set of elements $X$ of a matroid is a superset of $X$ that includes all elements $x$ such that the rank of $X \cup \{x\}$ equals the rank of $X$. A set of elements $X$ is called flat if it is equal to its closure.

  • closure(M,X) computes the closure of the set X in the matroid M.
  • isflat(M,X) determines if X is a flat in M.

Equality Testing

Randomized Equality Test

fuzzy_equal performs a randomized equality check on a pair of matroids. Two matroids are equal if their ground sets are equal and, for any subset X of the ground set, the rank of X is the same in both matroids.

If two matroids have, say, 20 elements each, testing that the rank functions give identical results would entail calculating the ranks of over a million subsets.

The function fuzzy_equal tests equality by first checking that the two matroids have the same number of elements and the same rank. It then repeatedly generates a random subset X of the ground set and checks that the rank of X is the same in both matroids. Then it repeatedly creates minimum weight bases of the two matroids using (the same) random weights and checks to be sure they are the same.

To use this function, simply call fuzzy_equal(M1,M2). One thousand random sets X will be generated and their ranks compared. If the function returns false, the matroids are definitely not equal. If the function returns true, they probably are equal.

Options

  • The number of tests can be modified by calling fuzzy_equal(M1,M2,reps) with a different value for reps.
  • A random subset of the ground set is created by choosing each element of the ground set with probability r/m where r is the rank of the matroid and m is the number of elements. A different probability may be used by calling fuzzy_equal(M1,M2,reps,p) and providing the desired value for p.

Exact Equality Test

Equality of matroids can be tested using ==. Except in the case of small matroids, this can be terribly slow. Equality is tested by comparing the sets of all bases of the two matroids. See all_bases.

Operations

These operations create new matroids from previously created matroids.

Matroids are immutable; operations do not modify existing matroids.

Dual: $M^*$

dual(M) creates the dual of M.

The resulting matroid has the same ground set as M and the labels in the new matroid are the same as the labels in M.

Deletion: $M \backslash X$

Given a matroid M and a subset X of the ground set of M, the function delete(M,X) forms a new matroid by deleting the members of X from M. Here X may be either a Set or a Vector of integer values. In addition, delete(M,x), where x is an integer, deletes the single element from M. In all cases, the \ operator may be used: M\X or M\x.

Recall our convention that the ground set of a Matroid must be of the form {1,2,...,m}. The implication of this is that an element of the new matroid may correspond to a higher number element of the original.

For example, define a Matroid using the following 2x7 matrix:

julia> A = [1 2 3 4 5 6 7; 8 9 10 11 12 13 14]
2×7 Matrix{Int64}:
 1  2   3   4   5   6   7
 8  9  10  11  12  13  14

julia> M = Matroid(A)
{7, 2} matroid

From this matroid, we delete elements 2 and 5. 

julia> MM = delete(M, [2,5])
{5, 2} matroid

The deletion of element 2 from M makes element 3 in M move to position 2 in MM. Likewise, element 4 moves to position 3 in MM. We skip element 5 (it has been deleted) and so element 6 goes to position 4 in MM. Likewise element 7 in M becomes element 5 in MM.

This can be illustrated by examining the labels. Consider element 3 of M which is now at index 2 in MM:

julia> get_label(M,3)
2-element Vector{Int64}:
  3
 10

julia> get_label(MM,2)
2-element Vector{Int64}:
  3
 10

Likewise, element 7 of M moves to position 5 in MM:

julia> get_label(M,7)
2-element Vector{Int64}:
  7
 14

julia> get_label(MM,5)
2-element Vector{Int64}:
  7
 14

Contraction: $M / X$

Given a matroid M and a subset of its ground set X, use contract(M,X) to produce a new matroid formed by contracting the elements in X. Here, X may be either a Set or a Vector of integer values. In addition, contract(M,x), where x is an integer, contracts the single element x. In all cases the / operator may be used: M/X or M/x.

As in the case of deletion, the elements of X are eliminated from the matroid by the contraction operation, and the remaining elements are renumbered so that the resulting ground set is of the usual form, {1,2,...,m}.

Labels carry forward from the original matroid to the contracted result.

Element contraction in a matroid corresponds to edge contraction in a graph. For example, if we delete an edge from a cycle, we get a path, whereas if we contract an edge in a cycle we get a smaller cycle. This is reflected in the corresponding matroids:

julia> g = cycle_graph(8)
{8, 8} undirected simple Int64 graph

julia> M = Matroid(g)
{8, 7} matroid

julia> delete(M,1)
{7, 7} matroid

julia> contract(M,1)
{7, 6} matroid

Disjoint Union: $M_1 + M_2$

Given matroids M1 and M2, the result of disjoint_union(M1,M2) is a new matroid defined as follows:

  • Let m1 and m2 be the number of elements of M1 and M2, respectively.
  • Form a copy of M2 (call it M2a) by shifting its elements from 1 to m2 to be from m1+1 to m1+m2.
  • Let S1 and S2a be the ground sets of M1 and M2a, respectively.
  • The rank of a set X is calculated as the sum of the rank (in M1) of X ∩ S1 and the rank (in M2a) of X ∩ S2a.

This is analogous to the direct sum of matrices and the disjoint union of graphs. For example, suppose $A_1 = \begin{bmatrix} 1&2&3\\4&5&6 \end{bmatrix}$ and $A_2 = \begin{bmatrix} 7 & 8 \\ 9& 10 \end{bmatrix}$ and let M1 and M2 be their corresponding matroids. Then the disjoint union of M1 and M2 is the matroid derived from this matrix: $A_1 \oplus A_2 = \begin{bmatrix} 1 & 2 & 3 & 0 & 0 \\ 4 & 5 & 6 & 0 & 0 \\ 0 & 0 & 0 & 7 & 8 \\ 0 & 0 & 0 & 9 & 10 \end{bmatrix}.$

Aside: The $\oplus$ operation is implemented as dcat in SimpleTools.

The function disjoint_union(M1, M2) may alternatively be invoked as M1 + M2.

Note: Labels in the disjoint union of M1 and M2 are set to be consecutive integers starting with 1. We do not carry labels forward from either M1 or M2.