A matrix \( A \in \{-1,0,+1\}^{m \times n} \) is called series-parallel if it can be obtained from a \( 0 \)-by- \( 0 \) matrix by successively adjoining

a zero row/column vector,
a (negated) standard unit row/column vector, or
a (negated) copy of an existing row/column.

The removal of such a row/column is called an SP-reduction. A matroid is called series-parallel if it is represented by a series-parallel matrix. This is equivalent to being the graphic matroid of a series-parallel graph.

Theorem. A matrix \( A \in \{-1,0,1\}^{m \times n} \) is either series-parallel or it contains, up to scaling of rows/columns with \( -1 \),

a \( 2 \)-by- \( 2 \) submatrix \( M_2 := \begin{pmatrix} -1 & 1 \\ 1 & 1 \end{pmatrix} \),
a \( 3 \)-by- \( 3 \) submatrix \( M_3' := \begin{pmatrix} 1 & 1 & 0 \\ 1 & 1 & 1 \\ 0 & 1 & 1 \end{pmatrix} \), or
a \( k \)-by- \( k \)-submatrix \( M_k := \begin{pmatrix} 1 & 0 & 0 & 0 & \dotsb & 0 & 1 \\ 1 & 1 & 0 & 0 & \dotsb & 0 & 0 \\ 0 & 1 & 1 & 0 & \dotsb & 0 & 0 \\ 0 & 0 & 1 & 1 & \ddots & 0 & 0 \\ \vdots & \vdots & \vdots & \ddots & \ddots & \vdots & \vdots \\ 0 & 0 & 0 & 0 & \dotsb & 1 & 0 \\ 0 & 0 & 0 & 0 & \dotsb & 1 & 1 \end{pmatrix} \) for \( k \geq 3 \).

The latter two matrices are called wheel matrices since their represented matroids are the graphic matroids of wheel graphs.

Recognizing Series-Parallel Matrices

The command

cmr-series-parallel IN-MAT [OPTION...]

determines whether the matrix given in file IN-MAT is series-parallel. If this is not the case, then a maximal number of SP-reductions is carried out, leading to the reduced matrix. Moreover, one can ask for one of the minimal non-series-parallel submatrices above.

Options:

-i FORMAT Format of file IN-MAT; default: dense.
-S OUT-SP Write the list of series-parallel reductions to file OUT-SP; default: skip computation.
-R OUT-REDUCED Write the reduced submatrix to file OUT-REDUCED; default: skip computation.
-N NON-SUB Write a minimal non-series-parallel submatrix to file NON-SUB; default: skip computation.
-b Test for being binary series-parallel; default: ternary.

Advanced options:

--stats Print statistics about the computation to stderr.
--time-limit LIMIT Allow at most LIMIT seconds for the computation.

Formats for matrices: dense, sparse If IN-MAT is - then the matrix is read from stdin. If OUT-SP, OUT-REDUCED or NON-SUB is - then the list of reductions (resp. the submatrix) is written to stdout.

Algorithm

The implemented algorithm is not yet published. For a matrix \( A \in \{0,1\}^{m \times n}\) with \( k \) (sorted) nonzeros it runs in \( \mathcal{O}( m + n + k ) \) time assuming no hashtable collisions.

C Interface

The corresponding functions in the library are

CMRspTestTernary() tests a binary matrix for being series-parallel.
CMRspTestBinary() tests a binary matrix for being series-parallel.
CMRspDecomposeBinary() tests a binary matrix for being series-parallel, but may also terminate early, returning a 2-separation of \( A \).
CMRspDecomposeTernary() tests a ternary matrix for being series-parallel, but may also terminate early, returning a 2-separation of \( A \).

and are defined in series_parallel.h.