Generalized inverse
≈ 13 min read
Names in other languages
No names in other languages are held for this entry yet.
Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.
In mathematics, and in particular, algebra, a generalized inverse (or, g-inverse) of an element x is an element y that has some properties of an inverse element but not necessarily all of them. The purpose of constructing a generalized inverse of a matrix is to obtain a matrix that can serve as an inverse in some sense for a wider class of matrices than invertible matrices.
Generalized inverses can be defined in any mathematical structure that involves associative multiplication, that is, in a semigroup. This article describes generalized inverses of a matrix A . A matrix A^\mathrm{g} \in \mathbb{R}^{n \times m} is a generalized inverse of a matrix A \in \mathbb{R}^{m \times n} if AA^\mathrm{g}A = A. A generalized inverse exists for an arbitrary matrix, and when a matrix has a regular inverse, this inverse is its unique generalized inverse.
Motivation
Consider the linear system : Ax = y where A is an m \times n matrix and y \in \mathcal C(A), the column space of A . If m = n and A is nonsingular then x = A^{-1}y will be the solution of the system. Note that, if A is nonsingular, then : AA^{-1}A = A. Now suppose A is rectangular ( m \neq n ), or square and singular.
Then we need a right candidate G of order n \times m such that for all y \in \mathcal C(A), : AGy = y. That is, x=Gy is a solution of the linear system Ax = y .…
In mathematics, and in particular, algebra, a generalized inverse (or, g-inverse) of an element x is an element y that has some properties of an inverse element but not necessarily all of them. The purpose of constructing a generalized inverse of a matrix is to obtain a matrix that can serve as an inverse in some sense for a wider class of matrices than invertible matrices.
Generalized inverses can be defined in any mathematical structure that involves associative multiplication, that is, in a semigroup. This article describes generalized inverses of a matrix A . A matrix A^\mathrm{g} \in \mathbb{R}^{n \times m} is a generalized inverse of a matrix A \in \mathbb{R}^{m \times n} if AA^\mathrm{g}A = A. A generalized inverse exists for an arbitrary matrix, and when a matrix has a regular inverse, this inverse is its unique generalized inverse.
Ax = y where A is an m \times n matrix and y \in \mathcal C(A), the column space of A . If m = n and A is nonsingular then x = A^{-1}y will be the solution of the system. Note that, if A is nonsingular, then AA^{-1}A = A. Now suppose A is rectangular ( m \neq n ), or square and singular.
Then we need a right candidate G of order n \times m such that for all y \in \mathcal C(A), AGy = y. That is, x=Gy is a solution of the linear system Ax = y . Equivalently, we need a matrix G of order n\times m such that AGA = A. Hence we can define the generalized inverse as follows: Given an m \times n matrix A , an n \times m matrix G is said to be a generalized inverse of A if AGA = A. The matrix A^{-1} has been termed a regular inverse of A by some authors. (column space, im A) to a complement subspace of its kernel (null space, ker A) bijectively, and maps a complement subspace of A's image all to zero.]]
The problem is how to choose an x as the output of G for every y when the map x \mapsto y = Ax is not bijective. If A is not surjective, then not all y 's in its codomain have corresponding x 's via A . To circumvent it, we just let G map those y 's to arbitrary values.
For example, decompose the codomain of A as the direct sum of the column space \mathcal C(A) and a complement subspace, and construct G as follows. For y 's in the former subspace, let G map back to the corresponding x 's. For y 's in the latter subspace, let G map them all to zero (as there're no corresponding x 's). For other y 's, decompose them as the sum of the above two components, apply G respectively, then take the sum.
If A is not injective, then some y 's correspond to multiple x 's via A . To circumvent it, we let G map every y to one of the x 's according an algorithm. For example, decompose the domain of A as the direct sum of \ker A and a complement subspace. For every possible y , its preimage must be parallel to \ker A and intersect the chosen complement subspace at a single point.
Let G map the y to this point. If A is neither surjective nor injective, we combine the above two tricks. The picture on the right is an example.
Right inverse: If the matrix A has dimensions m \times n and \textrm{rank} (A) = m , then there exists an n \times m matrix A_{\mathrm{R}}^{-1} called the right inverse of A such that A A_{\mathrm{R}}^{-1} = I_m , where I_m is the m \times m identity matrix. Left inverse: If the matrix A has dimensions m \times n and \textrm{rank} (A) = n , then there exists an n \times m matrix A_{\mathrm{L}}^{-1} called the left inverse of A such that A_{\mathrm{L}}^{-1} A = I_n , where I_n is the n \times n identity matrix.
Bott–Duffin inverse Drazin inverse Moore–Penrose inverse Some generalized inverses are defined and classified based on the Penrose conditions: A A^\mathrm{g} A = A A^\mathrm{g} A A^\mathrm{g}= A^\mathrm{g} (A A^\mathrm{g})^* = A A^\mathrm{g} (A^\mathrm{g} A)^* = A^\mathrm{g} A, where {}^* denotes conjugate transpose. If A^\mathrm{g} satisfies the first condition, then it is a generalized inverse of A . If it satisfies the first two conditions, then it is a reflexive generalized inverse of A . If it satisfies all four conditions, then it is the pseudoinverse of A , which is denoted by A^+ and also known as the Moore–Penrose inverse, after the pioneering works by E. H. Moore and Roger Penrose.
It is convenient to define an I -inverse of A as an inverse that satisfies the subset I \subset \{1, 2, 3, 4\} of the Penrose conditions listed above. Relations, such as A^{(1, 4)} A A^{(1, 3)} = A^+ , can be established between these different classes of I -inverses. When A is non-singular, any generalized inverse A^\mathrm{g} is equal to A^{-1} and is therefore unique.
For a singular A , some generalized inverses, such as the Drazin inverse and the Moore–Penrose inverse, are unique, while others are not necessarily uniquely defined. A = \begin{bmatrix} 1 & 0 & 3 \\ 2 & 0 & 6 \\ 0 & 0 & 0 \end{bmatrix}, \quad G = \begin{bmatrix} 1 & 0 & 0 \\ 0 & 0 & 1 \\ 0 & 0 & 0 \end{bmatrix}. Obviously, A is singular.
A and G satisfy Penrose conditions (1), but not the other there. Hence, G is a non-reflexive generalized inverse of A . The first column of A spans \operatorname{im} A , and G maps it to (1,0,0) , which does not lie in \ker A . Additionally, G maps (0,0,1) to (0,1,0) , which lies in \ker A . The relationship is summarized in the picture on the right.
A = \begin{bmatrix} 1 & 2 & 3 \\ 4 & 5 & 6 \\ 7 & 8 & 9 \end{bmatrix}, \quad G = \begin{bmatrix} -\frac{5}{3} & \frac{2}{3} & 0 \\[4pt] \frac{4}{3} & -\frac{1}{3} & 0 \\[4pt] 0 & 0 & 0 \end{bmatrix}. Since \det(A) = 0 , A is singular and has no regular inverse. However, A and G satisfy Penrose conditions (1) and (2), but not (3) or (4).
Hence, G is a reflexive generalized inverse of A . A = \begin{bmatrix} 1 & 2 & 3 \\ 4 & 5 & 6 \end{bmatrix}, \quad A_\mathrm{R}^{-1} = \begin{bmatrix} -\frac{17}{18} & \frac{8}{18} \\[4pt] -\frac{2}{18} & \frac{2}{18} \\[4pt] \frac{13}{18} & -\frac{4}{18} \end{bmatrix}. Since A is not square, A has no regular inverse.
However, A_\mathrm{R}^{-1} is a right inverse of A . The matrix A has no left inverse. File:Generalized inverse subspaces-left.svg| X is a left inverse of A ( X A = I_U ) File:Generalized inverse subspaces-right.svg| X is a right inverse of A ( A X = I_V ) The element b is a generalized inverse of an element a if and only if a \cdot b \cdot a = a , in any semigroup (or ring, since the multiplication function in any ring is a semigroup).
The generalized inverses of the element 3 in the ring \mathbb{Z}/12\mathbb{Z} are 3, 7, and 11, since in the ring \mathbb{Z}/12\mathbb{Z} : 3 \cdot 3 \cdot 3 = 3 3 \cdot 7 \cdot 3 = 3 3 \cdot 11 \cdot 3 = 3 The generalized inverses of the element 4 in the ring \mathbb{Z}/12\mathbb{Z} are 1, 4, 7, and 10, since in the ring \mathbb{Z}/12\mathbb{Z} : 4 \cdot 1 \cdot 4 = 4 4 \cdot 4 \cdot 4 = 4 4 \cdot 7 \cdot 4 = 4 4 \cdot 10 \cdot 4 = 4 If an element a in a semigroup (or ring) has an inverse, the inverse must be the only generalized inverse of this element, like the elements 1, 5, 7, and 11 in the ring \mathbb{Z}/12\mathbb{Z} . In the ring \mathbb{Z}/12\mathbb{Z} any element is a generalized inverse of 0; however 2 has no generalized inverse, since there is no b in \mathbb{Z}/12\mathbb{Z} such that 2 \cdot b \cdot 2 = 2 .
A right inverse of a non-square matrix A is given by A_\mathrm{R}^{-1} = A^{\intercal} \left( A A^{\intercal} \right)^{-1} , provided A has full row rank. A left inverse of a non-square matrix A is given by A_\mathrm{L}^{-1} = \left(A^{\intercal} A \right)^{-1} A^{\intercal} , provided A has full column rank.
If A = BC is a rank factorization, then G = C_\mathrm{R}^{-1} B_\mathrm{L}^{-1} is a g-inverse of A , where C_\mathrm{R}^{-1} is a right inverse of C and B_\mathrm{L}^{-1} is left inverse of B . If A = P \begin{bmatrix}I_r & 0 \\ 0 & 0 \end{bmatrix} Q for any non-singular matrices P and Q , then G = Q^{-1} \begin{bmatrix}I_r & U \\ W & V \end{bmatrix} P^{-1} is a generalized inverse of A for arbitrary U, V and W . Let A be of rank r . Without loss of generality, let A = \begin{bmatrix}B & C\\ D & E\end{bmatrix}, where B_{r \times r} is the non-singular submatrix of A . Then, G = \begin{bmatrix} B^{-1} & 0\\ 0 & 0 \end{bmatrix} is a generalized inverse of A if and only if E=DB^{-1}C . Ax = b, with vector x of unknowns and vector b of constants, all solutions are given by x = A^\mathrm{g}b + \left[I - A^\mathrm{g}A\right]w, parametric on the arbitrary vector w , where A^\mathrm{g} is any generalized inverse of A . Solutions exist if and only if A^\mathrm{g}b is a solution, that is, if and only if AA^\mathrm{g}b = b . If A has full column rank, the bracketed expression in this equation is the zero matrix and so the solution is unique. A = U \begin{bmatrix} \Sigma_1 & 0 \\ 0 & 0 \end{bmatrix} V^\operatorname{T} be its singular-value decomposition.
Then for any generalized inverse A^g , there exist matrices X , Y , and Z such that A^g = V \begin{bmatrix} \Sigma_1^{-1} & X \\ Y & Z \end{bmatrix} U^\operatorname{T}. Conversely, any choice of X , Y , and Z for matrix of this form is a generalized inverse of A . The \{1,2\} -inverses are exactly those for which Z = Y \Sigma_1 X , the \{1,3\} -inverses are exactly those for which X = 0 , and the \{1,4\} -inverses are exactly those for which Y = 0 .
In particular, the pseudoinverse is given by X = Y = Z = 0 : A^+ = V \begin{bmatrix} \Sigma_1^{-1} & 0 \\ 0 & 0 \end{bmatrix} U^\operatorname{T}. (UAV)^+ = V^* A^+ U^* . The Drazin inverse, A^\mathrm{D} satisfies the following definition of consistency with respect to similarity transformations involving a nonsingular matrix S: \left(SAS^{-1}\right)^\mathrm{D} = S A^\mathrm{D} S^{-1} .
The unit-consistent (UC) inverse, A^\mathrm{U}, satisfies the following definition of consistency with respect to transformations involving nonsingular diagonal matrices D and E: (DAE)^\mathrm{U} = E^{-1} A^\mathrm{U} D^{-1} . The fact that the Moore–Penrose inverse provides consistency with respect to rotations (which are orthonormal transformations) explains its widespread use in physics and other applications in which Euclidean distances must be preserved.
The UC inverse, by contrast, is applicable when system behavior is expected to be invariant with respect to the choice of units on different state variables, e.g., miles versus kilometers.
Text derived from https://en.wikipedia.org/w/index.php?oldid=1352747379 under CC BY-SA 4.0; modifications have been made.