Minor (linear algebra)

In linear algebra, a minor of a matrix A is the determinant of some smaller square matrix generated from A by removing one or more of its rows and columns. Minors obtained by removing just one row and one column from square matrices (first minors) are required for calculating matrix cofactors, which are useful for computing both the determinant and inverse of square matrices. The requirement that the square matrix be smaller than the original matrix is often omitted in the definition.

Definition and illustration

First minors

If A is a square matrix, then the minor of the entry in the Template:Mvar-th row and Template:Mvar-th column (also called the (i, j) minor, or a first minor) is the determinant of the submatrix formed by deleting the Template:Mvar-th row and Template:Mvar-th column. This number is often denoted M_{i, j}. The (i, j) cofactor is obtained by multiplying the minor by (−1)^{i + j}.

To illustrate these definitions, consider the following 3 × 3 matrix,

$$\left[\begin{array}{ccc}1& 4& 7\\ 3& 0& 5\\ -1& 9& 11\\ \end{array}\right]$$

To compute the minor M_2,3 and the cofactor C_2,3, we find the determinant of the above matrix with row 2 and column 3 removed.

$$M_{2,3}=\det \left[\begin{array}{ccc}1& 4& ◻\\ ◻& ◻& ◻\\ -1& 9& ◻\\ \end{array}\right]=\det \left[\begin{array}{cc}1& 4\\ -1& 9\\ \end{array}\right]=9-(-4)=13$$

So the cofactor of the (2,3) entry is

$$C_{2,3}=(-1{)}^{2+3}(M_{2,3})=-13.$$

General definition

Let A be an m × n matrix and Template:Mvar an integer with 0 < k ≤ m, and k ≤ n. A k × k minor of A, also called minor determinant of order Template:Mvar of A or, if m = n, the (n − k)th minor determinant of A (the word "determinant" is often omitted, and the word "degree" is sometimes used instead of "order") is the determinant of a k × k matrix obtained from A by deleting m − k rows and n − k columns. Sometimes the term is used to refer to the k × k matrix obtained from A as above (by deleting m − k rows and n − k columns), but this matrix should be referred to as a (square) submatrix of A, leaving the term "minor" to refer to the determinant of this matrix. For a matrix A as above, there are a total of $(\genfrac{}{}{0ex}{}{m}{k})\cdot (\genfrac{}{}{0ex}{}{n}{k})$ minors of size k × k. The minor of order zero is often defined to be 1. For a square matrix, the zeroth minor is just the determinant of the matrix.

Let $$\begin{array}{rl}I& =1\le i_1<i_2<\cdots <i_k\le m,\\ J& =1\le j_1<j_2<\cdots <j_k\le n,\end{array}$$ be ordered sequences (in natural order, as it is always assumed when talking about minors unless otherwise stated) of indexes. The minor $\det (({\mathbf{𝐀}}_{i_p,j_q}{)}_{p,q=1,\dots ,k})$ corresponding to these choices of indexes is denoted $\underset{I,J}{\det }A$ or $\det {\mathbf{𝐀}}_{I,J}$ or $[\mathbf{𝐀}{]}_{I,J}$ or $M_{I,J}$ or $M_{i_1,i_2,\dots ,i_k,j_1,j_2,\dots ,j_k}$ or $M_{(i),(j)}$ (where the (i) denotes the sequence of indexes Template:Mvar, etc.), depending on the source. Also, there are two types of denotations in use in literature: by the minor associated to ordered sequences of indexes Template:Mvar and Template:Mvar, some authors mean the determinant of the matrix that is formed as above, by taking the elements of the original matrix from the rows whose indexes are in Template:Mvar and columns whose indexes are in Template:Mvar, whereas some other authors mean by a minor associated to Template:Mvar and Template:Mvar the determinant of the matrix formed from the original matrix by deleting the rows in Template:Mvar and columns in Template:Mvar; which notation is used should always be checked. In this article, we use the inclusive definition of choosing the elements from rows of Template:Mvar and columns of Template:Mvar. The exceptional case is the case of the first minor or the (i, j)-minor described above; in that case, the exclusive meaning $M_{i,j}=\det \left({\left({\mathbf{𝐀}}_{p,q}\right)}_{p\ne i,q\ne j}\right)$ is standard everywhere in the literature and is used in this article also.

Complement

The complement B_{ijk..., pqr...} of a minor M_{ijk..., pqr...} of a square matrix, A, is formed by the determinant of the matrix A from which all the rows (Template:Mvar) and columns (Template:Mvar) associated with M_{ijk..., pqr...} have been removed. The complement of the first minor of an element Template:Mvar is merely that element.

Applications of minors and cofactors

Cofactor expansion of the determinant

The cofactors feature prominently in Laplace's formula for the expansion of determinants, which is a method of computing larger determinants in terms of smaller ones. Given an n × n matrix A = (a_ij), the determinant of A, denoted det(A), can be written as the sum of the cofactors of any row or column of the matrix multiplied by the entries that generated them. In other words, defining $C_{ij}=(-1{)}^{i+j}{M}_{ij}$ then the cofactor expansion along the Template:Mvar-th column gives:

$$\begin{array}{rl}\det (\mathbf{𝐀})& =a_{1j}{C}_{1j}+a_{2j}{C}_{2j}+a_{3j}{C}_{3j}+\cdots +a_{nj}{C}_{nj}\\ & =\sum _{i=1}^n{a}_{ij}{C}_{ij}\\ & =\sum _{i=1}^n{a}_{ij}(-1{)}^{i+j}{M}_{ij}\end{array}$$

The cofactor expansion along the Template:Mvar-th row gives:

$$\begin{array}{rl}\det (\mathbf{𝐀})& =a_{i1}{C}_{i1}+a_{i2}{C}_{i2}+a_{i3}{C}_{i3}+\cdots +a_{in}{C}_{in}\\ & =\sum _{j=1}^n{a}_{ij}{C}_{ij}\\ & =\sum _{j=1}^n{a}_{ij}(-1{)}^{i+j}{M}_{ij}\end{array}$$

Inverse of a matrix

One can write down the inverse of an invertible matrix by computing its cofactors by using Cramer's rule, as follows. The matrix formed by all of the cofactors of a square matrix A is called the cofactor matrix (also called the matrix of cofactors or, sometimes, comatrix):

$$\mathbf{𝐂}=\left[\begin{array}{cccc}{C}_{11}& C_{12}& \cdots & C_{1n}\\ C_{21}& C_{22}& \cdots & C_{2n}\\ ⋮& ⋮& \ddots & ⋮\\ C_{n1}& C_{n2}& \cdots & C_{nn}\end{array}\right]$$

Then the inverse of A is the transpose of the cofactor matrix times the reciprocal of the determinant of A:

$${\mathbf{𝐀}}^{-1}=\frac{1}{\det (\mathbf{𝐀})}{\mathbf{𝐂}}^{\mathsf{T}}.$$

The transpose of the cofactor matrix is called the adjugate matrix (also called the classical adjoint) of A.

The above formula can be generalized as follows: Let $$\begin{array}{rl}I& =1\le i_1<i_2<\dots <i_k\le n,\\ J& =1\le j_1<j_2<\dots <j_k\le n,\end{array}$$ be ordered sequences (in natural order) of indexes (here A is an n × n matrix). Then

$$[{\mathbf{𝐀}}^{-1}{]}_{I,J}=\pm \frac{[\mathbf{𝐀}{]}_{J^{\prime },I^{\prime }}}{\det \mathbf{𝐀}},$$

where I′, J′ denote the ordered sequences of indices (the indices are in natural order of magnitude, as above) complementary to I, J, so that every index 1, ..., n appears exactly once in either Template:Mvar or Template:Mvar, but not in both (similarly for the Template:Mvar and Template:Mvar) and [A]_{I, J} denotes the determinant of the submatrix of A formed by choosing the rows of the index set Template:Mvar and columns of index set Template:Mvar. Also, $[\mathbf{𝐀}{]}_{I,J}=\det ((A_{i_p,j_q}{)}_{p,q=1,\dots ,k}).$ A simple proof can be given using wedge product. Indeed,

$$[{\mathbf{𝐀}}^{-1}{]}_{I,J}(e_1\wedge \dots \wedge e_n)=\pm ({\mathbf{𝐀}}^{-1}{e}_{j_1})\wedge \dots \wedge ({\mathbf{𝐀}}^{-1}{e}_{j_k})\wedge e_{i{\text{'}}_1}\wedge \dots \wedge e_{i{\text{'}}_{n-k}},$$

where $e_1,\dots ,e_n$ are the basis vectors. Acting by A on both sides, one gets

$$\begin{array}{rl}& [{\mathbf{𝐀}}^{-1}{]}_{I,J}\det \mathbf{𝐀}(e_1\wedge \dots \wedge e_n)\\ =& \pm (e_{j_1})\wedge \dots \wedge (e_{j_k})\wedge (\mathbf{𝐀}{e}_{i{\text{'}}_1})\wedge \dots \wedge (\mathbf{𝐀}{e}_{i{\text{'}}_{n-k}})\\ =& \pm [\mathbf{𝐀}{]}_{J^{\prime },I^{\prime }}(e_1\wedge \dots \wedge e_n).\end{array}$$

The sign can be worked out to be $$(-1{)}^{(\sum _{s=1}^k{i}_s-\sum _{s=1}^k{j}_s)},$$ so the sign is determined by the sums of elements in Template:Mvar and Template:Mvar.

Other applications

Given an m × n matrix with real entries (or entries from any other field) and rank Template:Mvar, then there exists at least one non-zero r × r minor, while all larger minors are zero.

We will use the following notation for minors: if A is an m × n matrix, Template:Mvar is a subset of {1, ..., m} with Template:Mvar elements, and Template:Mvar is a subset of {1, ..., n} with Template:Mvar elements, then we write [A]_{I, J} for the k × k minor of A that corresponds to the rows with index in Template:Mvar and the columns with index in Template:Mvar.

• If I = J, then [A]_{I, J} is called a principal minor.
• If the matrix that corresponds to a principal minor is a square upper-left submatrix of the larger matrix (i.e., it consists of matrix elements in rows and columns from 1 to Template:Mvar, also known as a leading principal submatrix), then the principal minor is called a leading principal minor (of order Template:Mvar) or corner (principal) minor (of order Template:Mvar). For an n × n square matrix, there are Template:Mvar leading principal minors.
• A basic minor of a matrix is the determinant of a square submatrix that is of maximal size with nonzero determinant.
• For Hermitian matrices, the leading principal minors can be used to test for positive definiteness and the principal minors can be used to test for positive semidefiniteness. See Sylvester's criterion for more details.

Both the formula for ordinary matrix multiplication and the Cauchy–Binet formula for the determinant of the product of two matrices are special cases of the following general statement about the minors of a product of two matrices. Suppose that A is an m × n matrix, B is an n × p matrix, Template:Mvar is a subset of {1, ..., m} with Template:Mvar elements and Template:Mvar is a subset of {1, ..., p} with Template:Mvar elements. Then $$[\mathbf{𝐀}\mathbf{𝐁}{]}_{I,J}=\sum _K[\mathbf{𝐀}{]}_{I,K}[\mathbf{𝐁}{]}_{K,J}$$ where the sum extends over all subsets Template:Mvar of {1, ..., n} with Template:Mvar elements. This formula is a straightforward extension of the Cauchy–Binet formula.

Multilinear algebra approach

A more systematic, algebraic treatment of minors is given in multilinear algebra, using the wedge product: the Template:Mvar-minors of a matrix are the entries in the Template:Mvar-th exterior power map.

If the columns of a matrix are wedged together Template:Mvar at a time, the k × k minors appear as the components of the resulting Template:Mvar-vectors. For example, the 2 × 2 minors of the matrix $$\left(\begin{array}{cc}1& 4\\ 3& -1\\ 2& 1\\ \end{array}\right)$$ are −13 (from the first two rows), −7 (from the first and last row), and 5 (from the last two rows). Now consider the wedge product $$({\mathbf{𝐞}}_1+3{\mathbf{𝐞}}_2+2{\mathbf{𝐞}}_3)\wedge (4{\mathbf{𝐞}}_1-{\mathbf{𝐞}}_2+{\mathbf{𝐞}}_3)$$ where the two expressions correspond to the two columns of our matrix. Using the properties of the wedge product, namely that it is bilinear and alternating, $${\mathbf{𝐞}}_i\wedge {\mathbf{𝐞}}_i=0,$$ and antisymmetric, $${\mathbf{𝐞}}_i\wedge {\mathbf{𝐞}}_j=-{\mathbf{𝐞}}_j\wedge {\mathbf{𝐞}}_i,$$ we can simplify this expression to $$-13{\mathbf{𝐞}}_1\wedge {\mathbf{𝐞}}_2-7{\mathbf{𝐞}}_1\wedge {\mathbf{𝐞}}_3+5{\mathbf{𝐞}}_2\wedge {\mathbf{𝐞}}_3$$ where the coefficients agree with the minors computed earlier.

A remark about different notation

In some books, instead of cofactor the term adjunct is used. Moreover, it is denoted as A_ij and defined in the same way as cofactor: $${\mathbf{𝐀}}_{ij}=(-1{)}^{i+j}{\mathbf{𝐌}}_{ij}$$

Using this notation the inverse matrix is written this way: $${\mathbf{𝐌}}^{-1}=\frac{1}{\det (M)}\left[\begin{array}{cccc}{A}_{11}& A_{21}& \cdots & A_{n1}\\ A_{12}& A_{22}& \cdots & A_{n2}\\ ⋮& ⋮& \ddots & ⋮\\ A_{1n}& A_{2n}& \cdots & A_{nn}\end{array}\right]$$

Keep in mind that adjunct is not adjugate or adjoint. In modern terminology, the "adjoint" of a matrix most often refers to the corresponding adjoint operator.

See also

• Submatrix
• Compound matrix

External links

• MIT Linear Algebra Lecture on Cofactors at Google Video, from MIT OpenCourseWare
• Cofactor Expansion at PlanetMath.