Fact-checked by Grok 2 weeks ago

Generalized inverse

In linear algebra, a generalized inverse (also known as a g-inverse) of an m \times n A is any n \times m G that satisfies the equation A G A = A. This condition extends the notion of a standard to non-square or singular matrices, where no true exists, and allows for solutions to linear systems A x = b even when b is not in the column space of A. Generalized inverses always exist for any but are generally not unique unless additional constraints are imposed. Generalized inverses are categorized by the subset of the four Penrose equations they satisfy, introduced by in 1955: (1) A G A = A, (2) G A G = G, (3) (A G)^* = A G, and (4) (G A)^* = G A, where * denotes the conjugate transpose (or transpose for real matrices). A {1}-inverse satisfies only the first equation, while a {1,2}-inverse (reflexive generalized inverse) also satisfies the second. The Moore–Penrose pseudoinverse, denoted A^\dagger, satisfies all four equations and is unique for every matrix; it provides the minimum-norm least-squares solution to A x = b and projects orthogonally onto the column space of A. The concept traces its origins to E. H. Moore's 1920 work on the "general reciprocal" of algebraic matrices, which laid foundational ideas for handling non-invertible cases, though published as an abstract. Penrose formalized the four equations in his seminal paper, establishing the pseudoinverse as a choice. The broader theory of generalized inverses, including classifications and applications, was systematically developed by C. Radhakrishna Rao and Mitra in their 1971 monograph, which emphasized statistical uses such as linear and testing. These tools are essential in fields like statistics, , , and for solving underdetermined or overdetermined systems.

Fundamentals

Definition and Motivation

In algebraic structures such as semigroups or rings, a generalized of an element A is an element g (often denoted A^g) that satisfies the equation A g A = A. This condition represents a minimal form of partial invertibility, where g acts as an for A restricted to the image of A, without requiring full invertibility or the existence of a two-sided . Weaker variants include one-sided generalized inverses, such as those satisfying only A g A = A (a left-ish ) or g A g = g (a right-ish ), which capture asymmetric notions of partial reversal in non-commutative settings. In the context of linear algebra, the concept applies to matrices over fields like the real or numbers, where A is an arbitrary m \times n and g (or A^-) is an n \times m satisfying A g A = A. This extends the classical inverse, which exists only for square nonsingular matrices (where \det(A) \neq 0), to handle singular square matrices or rectangular ones, where the is undefined or the dimensions preclude a two-sided . The equation A g A = A ensures that g A is a onto the column space of A, providing a way to "undo" the action of A where possible. The primary motivation for generalized inverses arises in solving linear systems A x = b, where A may be singular or rectangular, rendering standard inversion impossible. Here, if the system is consistent (i.e., b lies in the column space of A), a generalized inverse g yields a particular solution x = g b, while the full general solution is given by x = g b + (I_n - g A) z for arbitrary z \in \mathbb{R}^n (or \mathbb{C}^n), capturing the null space contributions. To derive this, note that substituting x = g b + (I_n - g A) z into A x gives A g b + A (I_n - g A) z = A g b + (A - A g A) z = A g b + (A - A) z = A g b = b, confirming consistency preservation; the term (I_n - g A) projects onto the kernel of A, parameterizing all solutions. This framework addresses the failure of regular inverses in underdetermined or overdetermined systems, enabling systematic treatment of ill-posed problems. As a foundational concept, the generalized inverse establishes the limitations of classical invertibility and sets the stage for exploring specialized types, such as the Moore-Penrose inverse, which refines the basic definition with additional symmetry and orthogonality properties.

Historical Background

The concept of the generalized inverse emerged in the early as a means to extend the notion of inversion beyond nonsingular square matrices, particularly to address systems in linear equations. In 1920, introduced the idea in his abstract "On the Reciprocal of the General Algebraic ," where he described a generalized for arbitrary algebraic matrices, laying foundational groundwork for handling singular and rectangular cases. This work, later elaborated in a publication, motivated solutions to inconsistent linear systems by generalizing the inverse to encompass projections onto relevant subspaces. Mid-20th-century advancements formalized and expanded these ideas into broader algebraic structures. Roger Penrose's 1955 paper "A Generalized Inverse for Matrices" provided a rigorous definition through four axiomatic conditions, establishing the unique pseudoinverse now known as the Moore-Penrose inverse, which satisfies symmetry, idempotence, and orthogonality properties for real or complex matrices. An explicit formula for the Moore-Penrose inverse via full-rank factorization of matrices was first pointed out by C. C. MacDuffee in private communications, bridging linear algebra with ring theory. Parallel developments in semigroup theory introduced partial inverses; the algebraic framework of inverse semigroups, which model partial symmetries through unique idempotent inverses, was pioneered in the 1950s by Gordon B. Preston and Viktor V. Wagner, extending generalized inversion to non-invertible transformations. In the late , specialized types of generalized inverses proliferated to address singular operators and elements. Michael P. Drazin introduced the Drazin inverse in 1958 for elements in associative , defined via a power condition that captures the invertible part of perturbations, proving useful for differential equations and Markov chains. The group inverse, applicable to index-1 elements where the and align appropriately, was developed in and contexts in the , emphasizing reflexive properties in partial algebraic structures. Extensions into the have refined these concepts with new characterizations and broader applicability. The core inverse, introduced by Oskar M. Baksalary and Götz Trenkler in 2010 as an alternative to the group inverse for index-1 matrices, combines outer inverse properties with range conditions to preserve core-EP structures. Recent work includes a 2023 geometric characterization of the Moore-Penrose inverse using polar decompositions of operator perturbations in Hilbert spaces, enhancing . Ongoing refinements, such as the W-weighted m-weak core inverse proposed in 2024, continue to extend these inverses to rectangular matrices without introducing major paradigm shifts.

Types

One-Sided and Reflexive Inverses

In the context of generalized s within s and matrix algebras, one-sided inverses provide a minimal extension of classical left and right inverses to non-invertible elements. A right one-sided inverse g of an element A satisfies A g A = A, which captures a partial left-inversion property without requiring full invertibility. This condition holds for a strong right inverse when A g = I, applicable to surjective linear maps or matrices where the number of rows m is at most the number of columns n with full row . Similarly, a left one-sided inverse g satisfies g A g = g, representing a partial right-inversion, and aligns with the strong left inverse g A = I for injective maps or matrices with m \geq n and full column . These definitions arise naturally in theory, where they characterize elements within Green's L- and R-classes relative to the variety of inverses V(A) = \{g \mid A g A = A, g A g = g\}. A reflexive generalized inverse combines both one-sided conditions, satisfying A g A = A and g A g = g simultaneously, thereby acting as a two-sided partial that preserves A under with g from either side. This makes g a von Neumann regular in terms, ensuring A is regular (i.e., A \in A S A for some S). Unlike one-sided inverses, which apply to rectangular matrices or asymmetric elements (e.g., enabling solutions in over- or under-determined systems), reflexive inverses exhibit square-like behavior, requiring compatible dimensions or class structures where left and right properties align. The Moore-Penrose inverse represents a special reflexive type augmented with conditions for . Key properties of these inverses include their non-uniqueness—multiple g may satisfy the equations for a given A—and a close to idempotents: for a right one-sided , A g is idempotent since (A g)^2 = A g A g = A g, projecting onto the image of A; analogously, g A is idempotent for a left one-sided inverse. In finite semigroups, reflexive generalized inverses exist for every element, as finiteness implies regularity (every a admits g with a g a = a and g a g = g), though one-sided versions may exist more broadly via class decompositions like V_l(A) = \{f A^{-1} \mid f \in E(L_A)\} for left inverses, where E(L_A) denotes idempotents in the left principal ideal. These structures facilitate applications in solving inconsistent equations or analyzing partial orderings in algebraic settings without full invertibility.

Moore-Penrose Inverse

The Moore-Penrose inverse, also known as the pseudoinverse, of a A is a unique A^+ that generalizes the concept of the inverse for non-square or singular matrices, satisfying a specific set of four conditions introduced by . These conditions ensure that A^+ provides a way to solve linear systems in Hilbert spaces, particularly for matrices. The notion traces back to E. H. Moore's earlier work on the "general reciprocal" of matrices, which laid foundational ideas for handling divisors of zero in algebraic structures. The four Penrose conditions defining A^+ are:
  1. A X A = A
  2. X A X = X
  3. (A X)^* = A X
  4. (X A)^* = X A
where ^* denotes the (). These equations capture the essential properties of an while incorporating symmetry to ensure uniqueness in the Euclidean structure of vector spaces. For any matrix A \in \mathbb{C}^{m \times n}, there exists a unique X = A^+ satisfying all four conditions simultaneously. Geometrically, the Moore-Penrose inverse provides the minimum-norm least-squares solution to the A x = b. Specifically, for a given b \in \mathbb{C}^m, the vector x = A^+ b minimizes \|A x - b\|_2 among all least-squares solutions and, among those, has the smallest norm \|x\|_2. This interpretation arises from the requirement of orthogonal projections in Hilbert spaces, where the solution projects b onto the range of A in a way that respects the inner product structure. To derive these conditions from the least-squares minimization framework, consider the problem of solving A x = b where A may not have full or b may not lie in the of A. The least-squares solutions satisfy the normal equations A^* A x = A^* b, but this may have multiple solutions if A^* A is singular. To select the minimum-norm solution, impose the that x is orthogonal to the null space of A, i.e., x \in \operatorname{range}(A^*). Let P = A A^+ denote the orthogonal onto \operatorname{[range](/page/Range)}(A). Then, the least-squares condition requires A x - b \perp \operatorname{[range](/page/Range)}(A), or equivalently, P b = A x. Substituting x = A^+ b yields P b = A A^+ b, confirming that A A^+ is indeed the projection operator. To connect this to the Penrose conditions, assume x = A^+ b satisfies the minimization: first, A x is the projection of b onto \operatorname{[range](/page/Range)}(A), so A (A^+ b) = P b, and applying A again gives A A A^+ b = A P b = A b' where b' = P b, satisfying condition 1: A A^+ A = A. For condition 2, A^+ A A^+ = A^+, note that A^+ b lies in \operatorname{range}(A^*), and the idempotence follows from the projection properties: A^+ (A A^+ b) = A^+ P b = A^+ b, since P b is the closest point. The symmetry conditions 3 and 4 arise from the self-adjoint nature of orthogonal projections: A A^+ is self-adjoint because it projects orthogonally, so (A A^+)^* = A A^+, and similarly for A^+ A, which projects onto \operatorname{range}(A^*). Thus, (A A^+)^* = (A A)^+ A^* = A A^+ implies the Hermitian symmetry. This derivation shows how the conditions encode the variational principles of and minimum norm in Hilbert spaces. The relation to orthogonal projections is explicit: P = A A^+ is the orthogonal projection onto \operatorname{range}(A), and Q = A^+ A is the orthogonal projection onto \operatorname{range}(A^*). These projectors satisfy P^2 = P, Q^2 = Q, and P^* = P, Q^* = Q, directly following from the Penrose conditions. For example, from condition 1 and 3, (A A^+) ^2 = A A^+ A A^+ = A (A^+ A) A^+ = A A^+, confirming and self-adjointness. This framework underscores the Moore-Penrose inverse's role in decomposing spaces into orthogonal complements, essential for applications in linear algebra over Hilbert spaces.

Drazin and Group Inverses

The index of a square matrix A, denoted \operatorname{ind}(A), is defined as the smallest nonnegative integer k such that \ker(A^k) = \ker(A^{k+1}), or equivalently, \operatorname{rank}(A^k) = \operatorname{rank}(A^{k+1}). This index measures the "singularity depth" of A and is finite if and only if the ascent (dimension of the generalized kernel growth) stabilizes. The Drazin inverse of A, denoted A^D, is a generalized inverse that exists if and only if \operatorname{ind}(A) is finite. It is the unique matrix satisfying the conditions A^{k+1} A^D = A^k, \quad A^D A A^D = A^D, \quad A A^D = A^D A, where k = \operatorname{ind}(A). The first equation ensures that A^D "inverts" A on the range of A^k, while the latter two impose idempotence and commutativity. The Drazin inverse commutes with A and is idempotent on the core subspace, with A A^D being the spectral idempotent projecting onto the range of A^k. When \operatorname{ind}(A) = 0, A is invertible and A^D = A^{-1}, which is a reflexive . A special case arises when \operatorname{ind}(A) = 1, in which the Drazin inverse is called the group inverse, denoted A^\#. It satisfies A A^\# A = A, \quad A^\# A A^\# = A^\#, \quad A A^\# = A^\# A. These equations characterize A^\# uniquely when it exists, and it arises naturally in the study of power-regular elements in semigroups where the index is at most 1. For square matrices over the complex numbers, the Drazin inverse relates closely to the canonical form of A. Specifically, if A = P J P^{-1} where J is the form, then A^D = P J^D P^{-1}, with J^D obtained by replacing each block for a nonzero eigenvalue \lambda with the corresponding block of \lambda^{-1} (adjusted for the part via the finite ), and setting blocks for eigenvalue 0 to zero except for the core structure aligned with the . This construction inverts the semisimple part while annihilating the component beyond the . An equivalent characterization of the Drazin inverse involves the core polynomial equation A^{k+1} (A A^D - I) = 0, which highlights that A A^D acts as an identity on the image of A^{k+1}. This equation, along with commutativity and idempotence, ensures uniqueness and ties the inverse to the minimal polynomial of A restricted to the non-nilpotent part.

Other Types

The core inverse, applicable to square matrices of index at most one, is defined as the unique matrix A^c satisfying the equations A A^c A = A, A^c A A^c = A^c, and \mathcal{R}(A^c) \subseteq \mathcal{R}(A). This inverse can be explicitly expressed for such matrices as A^c = (A^* A)^\# A^*, where \# denotes the group inverse. It serves as an intermediate between the group inverse and the Moore-Penrose inverse, particularly useful for matrices where the index condition holds, and extends the Drazin inverse for higher indices in a specialized manner. The Bott-Duffin inverse, originally developed for analyzing electrical networks, provides a generalized {1,3}-inverse for square matrices that minimizes a specific in representations involving . For positive operators, it arises in contexts where an operator A is represented as A = P A P + (I - P) for a suitable P, yielding the as the minimizer over such decompositions. This construction ensures invertibility in constrained subspaces and has niche applications in optimization problems requiring bounded representations. Recent extensions include the extended core inverse, introduced in 2024, which for square complex matrices combines the sum and difference of the , , and to form a unique inner inverse satisfying specific matrix equations. This variant reduces to the standard for index-one matrices and addresses limitations in prior extensions by preserving inner inverse properties. In 2025, the was defined in Banach *-algebras as an extension of the , characterized via right core decompositions and quasi-nilpotent parts, with polar-like properties that facilitate algebraic manipulations in non-commutative settings.

Constructions

For Matrices

One practical method for constructing a generalized inverse of a finite-dimensional A \in \mathbb{R}^{m \times n} with r relies on its rank factorization A = BC, where B \in \mathbb{R}^{m \times r} has full column rank and C \in \mathbb{R}^{r \times n} has full row rank. To obtain a reflexive generalized (satisfying both AGA = A and GAG = G), compute the Moore-Penrose inverses B^+ and C^+ explicitly using their full-rank properties: B^+ = (B^T B)^{-1} B^T and C^+ = C^T (C C^T)^{-1}. Then set G = C^+ B^+. This G satisfies A G A = B C C^+ B^+ B C = B (C C^+) (B^+ B) C = B I_r I_r C = BC = A, confirming the {1}-inverse property; the reflexive property follows similarly from G A G = C^+ B^+ B C C^+ B^+ = C^+ (B^+ B) (C C^+) B^+ = C^+ I_r I_r B^+ = C^+ B^+ = G. The Moore-Penrose inverse A^+, a specific reflexive generalized inverse satisfying all four Penrose conditions, can be constructed via the singular value decomposition (SVD) of A. Compute the SVD A = U \Sigma V^H, where U \in \mathbb{R}^{m \times m} and V \in \mathbb{R}^{n \times n} are unitary matrices, \Sigma \in \mathbb{R}^{m \times n} is diagonal with nonnegative singular values \sigma_1 \geq \sigma_2 \geq \cdots \geq \sigma_r > 0 on the main diagonal (and zeros elsewhere), and r = \rank(A). Form \Sigma^+ as the n \times m matrix with diagonal entries $1/\sigma_i for i = 1, \dots, r and zeros otherwise. Then A^+ = V \Sigma^+ U^H. This construction satisfies the Penrose conditions. The following steps outline the computation of A^+ using SVD in practice:
function A_plus = moore_penrose(A, m, n)
    [U, Sigma_diag, V] = [svd](/page/SVD)(A);  // Compute SVD: A = U * diag(Sigma_diag) * V^H
    // Sigma_diag is m x n diagonal matrix with singular values on diagonal
    r = rank(Sigma_diag);  // Number of nonzero singular values
    Sigma_plus = zeros(n, m);
    for i = 1 to r
        Sigma_plus(i, i) = 1 / Sigma_diag(i, i);
    end
    A_plus = V * Sigma_plus * U';  // For real matrices, H = transpose
end
This algorithm leverages standard SVD routines available in numerical libraries, with computational complexity dominated by the SVD step, typically O(\min(m n^2, m^2 n)). Another approach for a reflexive generalized inverse uses a maximal nonsingular submatrix. Select index sets I \subset \{1,\dots,m\} and J \subset \{1,\dots,n\} with |I| = |J| = r such that the submatrix A_{I J} is nonsingular (invertible). Permute rows and columns of A so that A_{I J} occupies the leading r \times r block. A reflexive generalized inverse G is then obtained by placing (A_{I J})^{-1} in the leading r \times r block of G (corresponding to the permuted positions) and setting all other entries to zero; adjustments via permutation matrices ensure compatibility with the original indexing. This method reduces computation to inverting an r \times r nonsingular matrix after rank-revealing permutation.

In Algebraic Structures

In semigroups, a generalized inverse of an element A is defined as an element g satisfying A g A = A. This notion is analyzed through Green's relations, which partition the semigroup into equivalence classes: the \mathcal{R}-class for elements generating the same right , the \mathcal{L}-class for left principal ideals, and the \mathcal{H}-class as their . To construct such a g, one identifies an "inverse along an element d" where g belongs to the \mathcal{H}-class of d and satisfies g A d = d = d A g, ensuring g \in d S \cap S d. Existence relies on the presence of idempotents in these classes; specifically, an inverse along d exists and is unique if there is an idempotent e \in \mathcal{R}_d \cap E(S) such that A e d and d e A form trace products in the semigroup, or equivalently if A d \mathcal{L} d and the \mathcal{H}-class of A d is a group. In rings, generalized inverses are particularly well-developed for regular elements, where an element A is regular if there exists x such that A = A x A. In this case, a reflexive generalized inverse is given by g = x A x, satisfying both A g A = A and g A g = g. regular rings, in which every element is regular, admit such inverses for all elements. The facilitates the construction by decomposing the ring relative to the idempotents e = A x and f = x A: the ring R splits into orthogonal components e R e, e R (1-e), (1-e) R e, and (1-e) R (1-e), allowing explicit representation of A and g in matrix-like form over these corner rings and enabling study of uniqueness and absorption properties. In Banach algebras, constructions of Drazin-like generalized inverses extend these ideas to infinite-dimensional settings, often via perturbation theory for elements with isolated spectral points. For an element A whose spectrum \sigma(A) has 0 as an isolated point of finite algebraic multiplicity m, the Drazin inverse A^D satisfies A^{m+1} A^D = A^m = (A^D)^{m+1} A^m and A A^D = A^D A. This is constructed using the holomorphic functional calculus: if \Gamma is a contour enclosing \sigma(A) excluding the isolated point 0, then A^D = \frac{1}{2\pi i} \int_\Gamma \lambda^{-1} R(\lambda, A) d\lambda, where R(\lambda, A) = (\lambda I - A)^{-1} is the resolvent, effectively projecting onto the generalized eigenspace for nonzero eigenvalues while handling the pole at 0 perturbatively. Perturbation methods bound the stability of A^D under small changes E, yielding estimates like \| (A+E)^D - A^D \| \leq C \|E\| for sufficiently small \|E\| when the index remains finite. The matrix case exemplifies these constructions as a special instance of finite von Neumann regular rings.

Examples

Matrix Examples

A simple example of a reflexive generalized inverse arises with the singular matrix A = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}. Here, the matrix g = A itself serves as a reflexive generalized inverse, satisfying A g A = A and g A g = g. To verify, first compute A^2 = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix} = A, so A g A = A^2 A = A^3 = A (since A^2 = A and A^3 = A). Similarly, g A g = A^2 A = A. This example illustrates a , where the reflexive inverse coincides with the original due to . For a one-sided generalized inverse, consider the rectangular matrix A = \begin{pmatrix} 1 \\ 2 \end{pmatrix}, a 2×1 column of full column rank. A right generalized inverse g satisfies A g A = A. One such g is the row g = \begin{pmatrix} \frac{1}{5} & \frac{2}{5} \end{pmatrix}, corresponding to the minimal solution among possible choices. First, compute the scalar g A = \frac{1}{5} \cdot 1 + \frac{2}{5} \cdot 2 = \frac{1}{5} + \frac{4}{5} = 1. Then, A g A = A (g A) = A \cdot 1 = A, confirming the condition. The product A g = \begin{pmatrix} \frac{1}{5} & \frac{2}{5} \\ \frac{2}{5} & \frac{4}{5} \end{pmatrix} is the orthogonal onto the column of A, a rank-1 that is not the full but acts as a partial inverse in the range. Other choices, such as g = \begin{pmatrix} 1 & 0 \end{pmatrix}, also satisfy the condition since g A = 1 and A g A = A, but yield a different projector A g = \begin{pmatrix} 1 & 0 \\ 2 & 0 \end{pmatrix}. The Moore-Penrose inverse provides a unique generalized inverse satisfying additional and orthogonality conditions. Consider the rank-1 singular A = \begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}. Its Moore-Penrose inverse is A^+ = \frac{1}{4} \begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}. To compute this via (SVD), first find A A^T = \begin{pmatrix} 2 & 2 \\ 2 & 2 \end{pmatrix}, with eigenvalues 4 and 0 ( 4, 0). The nonzero is \sigma_1 = 2, with left singular vector u_1 = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 1 \end{pmatrix} and right singular vector v_1 = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 1 \end{pmatrix}. Thus, A = u_1 \sigma_1 v_1^T, and A^+ = v_1 \sigma_1^{-1} u_1^T = \frac{1}{2} \cdot \frac{1}{\sqrt{2}} \begin{pmatrix} 1 \\ 1 \end{pmatrix} \cdot \frac{1}{\sqrt{2}} \begin{pmatrix} 1 & 1 \end{pmatrix} = \frac{1}{4} A. shows A A^+ A = A, A^+ A A^+ = A^+, (A A^+)^T = A A^+, and (A^+ A)^T = A^+ A, confirming the four Penrose conditions. For the Drazin inverse, consider a Jordan such as N = \begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix}, with index of nilpotency 2 since N^2 = 0. The Drazin inverse is N^d = 0, satisfying the defining equations trivially as the core part is absent.

Non-Matrix Examples

In the \mathbb{Z}/6\mathbb{Z}, the 2 is , possessing a generalized inverse g = 2 that satisfies $2 \cdot 2 \cdot 2 \equiv 2 \pmod{6}, since $8 \equiv 2 \pmod{6}. Another generalized inverse for 2 in this ring is g = 5, as $2 \cdot 5 \cdot 2 = 20 \equiv 2 \pmod{6}. Similarly, in \mathbb{Z}/12\mathbb{Z}, the 4 is with generalized inverse g = 1, verifying $4 \cdot 1 \cdot 4 = 16 \equiv 4 \pmod{12}. Other inverses for 4 include g = 4, g = 7, and g = 10, each satisfying the equation $4 g 4 \equiv 4 \pmod{12}. In semigroups, generalized inverses appear in structures like the full transformation semigroup T_3 on the set \{1, 2, 3\}. For instance, consider the transformation a = (2\ 2\ 1), which maps 1 to 2, 2 to 2, and 3 to 1, and d = (2\ 3\ 2), mapping 1 to 2, 2 to 3, and 3 to 2; here, b = (3\ 2\ 3), mapping 1 to 3, 2 to 2, and 3 to 3, serves as an inner inverse along d, satisfying b a d = d = d a b with b in the appropriate Green's class. Another example in T_3 involves a' = (1\ 2\ 2), mapping 1 to 1, 2 to 2, and 3 to 2, which is invertible along the same d via b = (3\ 2\ 3), though not inner, as a' d and d a' are distinct trace products. These illustrate how generalized inverses extend to partial transformations, where reflexivity aligns with idempotent properties in the semigroup.

Applications

Linear Systems and Least Squares

Generalized inverses provide a framework for finding solutions to linear systems Ax = b, where A is an m \times n that may be rectangular, singular, or both, leading to either inconsistent systems (when b \notin \operatorname{range}(A)) or underdetermined systems (with infinitely many s). A reflexive generalized inverse A^g, satisfying the Penrose conditions A A^g A = A and A^g A A^g = A^g, yields a particular x_p = A^g b. If the system is consistent, this x_p satisfies A x_p = b; otherwise, it approximates the solution in a manner dependent on the type of generalized inverse used. The complete solution set for a consistent system is given by x = x_p + (I_n - A^g A) z, where z \in \mathbb{R}^n is arbitrary and I_n - A^g A is the projection onto the null space of A, capturing all homogeneous solutions. Among all possible generalized inverses, the Moore-Penrose pseudoinverse A^+ produces the particular solution x_p = A^+ b of minimum Euclidean norm \|x_p\|_2. This minimal-norm property arises from the additional conditions defining A^+, ensuring symmetry in the projections involved. In the context of least squares problems, where the goal is to minimize \|Ax - b\|_2 for inconsistent systems, the solution is x = A^+ b. This x satisfies the normal equations A^* A x = A^* b, with A^* denoting the conjugate transpose, and represents the minimum-norm least-squares solution. The corresponding residual error is \|b - A x_p\|_2 = \operatorname{dist}(b, \operatorname{range}(A)), the shortest distance from b to the column space of A. The Moore-Penrose inverse ensures the residual is orthogonal to \operatorname{range}(A).

Optimization and Statistics

In problems, such as formulated as minimizing \mathbf{x}^T Q \mathbf{x} + \mathbf{c}^T \mathbf{x} subject to A \mathbf{x} = \mathbf{b}, the generalized inverse A^+ facilitates onto the affine , enabling the computation of minimum-norm for underdetermined systems. This approach is particularly useful when the A is rank-deficient, as A^+ provides a least-squares to the consistency conditions while preserving the quadratic objective's structure. Tikhonov regularization extends this by defining a damped generalized inverse A_\lambda^+ = (A^* A + \lambda I)^{-1} A^*, where \lambda > 0 is a regularization , to mitigate ill-conditioning and amplify small singular values in the original pseudoinverse. This formulation balances fidelity to the data with stability, commonly applied in variants of quadratic optimization. In statistical estimation, generalized inverses underpin the best linear unbiased (BLUE) for parameters in linear models \mathbf{Y} = X \boldsymbol{\beta} + \boldsymbol{\epsilon}, where X may be singular due to or unbalanced designs. Specifically, the BLUE is \hat{\boldsymbol{\beta}} = (X^T X)^+ X^T \mathbf{Y}, which minimizes the variance among all linear unbiased even when X^T X lacks full rank. This extends the classical Gauss-Markov to singular design matrices and arbitrary nonnegative structures, confirming that the BLUE achieves the minimum for estimable linear combinations of \boldsymbol{\beta}. In analysis of variance (ANOVA) for unbalanced data, the generalized inverse resolves the singularity in the sums-of-squares-and-products matrix, yielding unbiased estimates of variance components and effects. Ridge regression, as a biased extension of the , incorporates Tikhonov regularization to shrink estimates toward zero, reducing in high-dimensional or collinear settings while maintaining computational tractability via the generalized inverse framework. solutions serve as a special unregularized case, but the generalized approach handles broader inconsistencies without assuming full .

Modern Uses

In , the Moore-Penrose pseudoinverse A^+ plays a central role in kernel methods, such as support vector machines (SVMs) and (). For least-squares SVMs, the pseudoinverse facilitates efficient computation of the dual solution by inverting the kernel matrix, enabling classification in high-dimensional spaces. In , is achieved through the () of the data matrix, where the pseudoinverse of the reduced singular value matrix projects data onto principal components, preserving variance while mitigating the curse of dimensionality. An emerging application appears in neural networks, where the Moore-Penrose inverse initializes weights for singular or rank-deficient layers, ensuring stable training by providing a minimum-norm solution to the formed by hidden-layer outputs. This approach enhances convergence in deep linear networks solving inverse problems, such as image reconstruction. In , generalized inverses address ill-posed inverse problems like , where truncated approximations of A^+ regularize the solution by damping small singular values, thus reducing noise amplification in recovering original signals from blurred observations. Advancements, such as modified truncated randomized -based pseudoinverses, have been integrated with compressive sensing to reconstruct sparse signals from undersampled measurements, improving efficiency in applications like and . Contemporary algebraic research has advanced characterizations of generalized inverses through polar decompositions and extensions to abstract structures. Between 2023 and 2025, new representations of the Moore-Penrose inverse leverage the canonical of square matrices, employing invertible factors to express the inverse in terms of unitary and components, aiding in operator computations. Recent work has introduced and characterized the weighted generalized core-EP inverse in Banach *-algebras for approximations in .

Properties

Existence and Uniqueness

In the context of matrices, a reflexive generalized inverse always exists for any matrix A \in \mathbb{R}^{m \times n} or \mathbb{C}^{m \times n}. This follows from a rank factorization of A: if \operatorname{rank}(A) = r, then A = BC where B \in \mathbb{R}^{m \times r} has full column rank and C \in \mathbb{R}^{r \times n} has full row rank. A reflexive generalized inverse G can then be constructed as G = C^\dagger B^\dagger, where ^\dagger denotes the Moore-Penrose pseudoinverse, satisfying both AGA = A and GAG = G. Among specific types, the Moore-Penrose pseudoinverse exists for every A and is unique, as it is the only matrix satisfying the four Penrose conditions: AGA = A, GAG = G, (AG)^* = AG, and (GA)^* = GA. This uniqueness stems directly from the symmetry and idempotence requirements imposed by the final two conditions. In contrast, the Drazin inverse of a square A \in \mathbb{C}^{n \times n} with finite (always the case for finite-dimensional matrices, where the is at most n-1) also exists and is unique, defined as the unique A^D satisfying A^{k+1} A^D = A^k, A^D A A^D = A^D, and A A^D = A^D A for k = \operatorname{ind}(A). One-sided generalized inverses, such as right inverses G where AG = I, exist if A is surjective (full row rank) but are not unique; if \ker(A) is nontrivial (\dim \ker(A) > 0), there are infinitely many such G, forming an affine space of dimension \dim \ker(A). In more general algebraic structures like s, a generalized inverse for an element a \in S exists a is , meaning there exists y \in S such that a = a y a. In this case, y serves as a generalized inverse satisfying a y a = a. This condition defines regularity in the S. A fundamental states that a generalized inverse G for A exists A = A^2 Y for some Y. To see this, suppose G is a generalized inverse, so A G A = A; setting Y = G A yields A = A^2 Y. Conversely, if A = A^2 Y, define E = A Y; then E^2 = A Y A Y = A (Y A Y) = A Y = E, so E is an idempotent. Moreover, A E = A (A Y) = A^2 Y = A, and E A = (A Y) A = A (Y A). Setting G = Y E = Y (A Y) gives A G A = A (Y A Y) A = (A Y A) Y A = A Y A = A, confirming G is a generalized inverse. This proof leverages the idempotent E to construct the inverse explicitly.

Consistency and Transformations

A generalized inverse A^g of a matrix A is said to be consistent if, for any b such that the Ax = b is solvable (i.e., b lies in the column space of A), the equation A A^g b = b holds. This property ensures that the generalized inverse preserves solvability by projecting b back onto the range of A without introducing inconsistencies. The Moore-Penrose pseudoinverse A^+ satisfies this condition for all matrices, as it yields an exact to consistent systems while minimizing the Euclidean norm of the vector. A key manifestation of this consistency is the orthogonal projection property of the Moore-Penrose pseudoinverse: A A^+ is the orthogonal projection onto the column space (range) of A, denoted \operatorname{proj}_{\operatorname{range}(A)}. Similarly, A^+ A projects orthogonally onto the row space of A. This projection ensures that A A^+ b = b precisely when b \in \operatorname{range}(A), reinforcing the in linear systems. Reflexivity of the Moore-Penrose inverse, where A^+ A A^+ = A^+, further supports this by maintaining in the projected solutions. Under linear transformations, the Moore-Penrose pseudoinverse exhibits invariance properties that preserve its structure. Specifically, if P and Q are invertible matrices, then (P A Q)^+ = Q^{-1} A^+ P^{-1}. For unitary matrices U and V (where U^{-1} = U^* and V^{-1} = V^*), the pseudoinverse transforms as (U A V^*)^+ = V A^+ U^*, reflecting the unitary invariance inherent in the underlying the pseudoinverse. These properties ensure that the pseudoinverse behaves analogously to a true under change-of-basis operations. For the Drazin inverse, transformation properties include additive results under commutativity conditions. If A and B are Drazin invertible and commute (i.e., AB = BA), then A + B is Drazin invertible, with explicit expressions for (A + B)^D in terms of A^D and B^D. Recent extensions in the generalize this to weaker commutativity, such as A-weak commutativity where there exists C such that AB = CA and BA = AC, providing conditions for the Drazin invertibility of A + B and AB in Banach algebras.

References

  1. [1]
    [PDF] Lecture 6: Generalized inverse and pseudoinverse
    Remark. If A is square and invertible, then it has one and only one generalized inverse which must coincide with the ordinary inverse A−1.
  2. [2]
    [PDF] Generalized Inverses: How to Invert a Non-Invertible Matrix
    Definition. If A is an m × n matrix, then G is a generalized inverse of A. if G is an n × m matrix with.
  3. [3]
    [PDF] A generalized inverse for matrices - KFUPM
    A generalized inverse for matrices. 407. Proof. I first show that equations (4) and (5) are equivalent to the single equation. XX*A* = X. (7). Equation (7) ...
  4. [4]
    A generalized inverse for matrices | Mathematical Proceedings of ...
    Oct 24, 2008 · A generalized inverse for matrices - Volume 51 Issue 3. ... However, as you have access to this content, a full PDF is available via the 'Save PDF ...
  5. [5]
    Contributions to the Theory of Generalized Inverses - jstor
    [27] C. C. MACDUFFEE, The Theory of Matrices, Springer Verlag, Berlin, 1933. [28] N. H. McCoy, Rings and Ideals, The Carus Mathematical Monographs, No. 8 ...
  6. [6]
    [PDF] A Short History of Inverse Semigroups - University of York
    Feb 16, 2011 · A partial bijection on a set X is a bijection A → B, where A,B ⊆ X. We compose partial transformations α, β (left → right) on the domain.
  7. [7]
    Inverse semigroups: The theory of partial symmetries - ResearchGate
    Aug 7, 2025 · The algebraic theory of semigroups emerged in the mid-twentieth century in a form that is sometimes referred to as the 'classical' theory ...
  8. [8]
    Core inverse of matrices - Taylor & Francis Online
    The Core inverse is introduced as an alternative to the group inverse, and its properties are derived for possible applications.
  9. [9]
    Geometric approach to the Moore-Penrose inverse and the polar ...
    We study the Moore-Penrose inverse of perturbations by a symmetrically-normed ideal of a closed range operator on a Hilbert space.
  10. [10]
    [2403.14196] The $W$-weighted $m$-weak core inverse - arXiv
    Mar 21, 2024 · The main aim of this paper is to extend the concept of m-weak core inverse for complex rectangular matrices. This extension is called the W-weighted m-weak ...
  11. [11]
    [PDF] onesided inverses for semigroups - D-MATH
    Abstract. In any semigroup S, we say that elements a and b are left inverses of each other if a = aba, b = bab and a L b, in which case we write a γ b.
  12. [12]
    [PDF] 1 The Generalized Inverse - Jeff Gill
    Penrose, R. A. 1955. “A Generalized Inverse for Matrices.” Proceedings of the Cambridge Philo- sophical Society 51, 406-13. Searle, S. R. 1971. Linear ...
  13. [13]
    On best approximate solutions of linear matrix equations
    Oct 24, 2008 · The purpose of the present note is to give a further application which has relevance to the statistical problem of finding 'best' approximate solutions.
  14. [14]
    Drazin Inverse -- from Wolfram MathWorld
    The Drazin inverse is a matrix inverse-like object derived from a given square matrix. In particular, let the index k of a square matrix be defined as the ...Missing: properties | Show results with:properties
  15. [15]
    THE DRAZIN INVERSE FOR CLOSED LINEAR OPERATORS ... - jstor
    The purpose of this paper is to introduce the Drazin inverse AD of a closed. linear operator Л on a Banach space X which is defined if 0 is merely an isolated. ...
  16. [16]
    Drazin inverse - Planet Math
    Mar 22, 2013 · The following are some other useful properties of the Drazin inverse: ... 5. If the matrix is represented explicitly by its Jordan canonical form ...
  17. [17]
    On a generalized core inverse - ScienceDirect.com
    The paper introduces the concept of a generalized core inverse of a matrix, which extends the notion of the core inverse defined by Baksalary and Trenkler [1].
  18. [18]
    Condition number of Bott–Duffin inverse and their condition numbers
    The definition of the Bott–Duffin inverse is given below. Definition 1.1. Given A∈Rn×n, a subspace L⊂Rn, if APL+PL⊥ is nonsingular then we denote the Bott– ...
  19. [19]
    New Characterizations and Representations of the Bott–Duffin Inverse
    Aug 1, 2023 · The paper focuses on the class of the Bott–Duffin inverses. Several original features of the class are identified and new properties are ...
  20. [20]
  21. [21]
    [PDF] Existence of Generalized Inverse: Ten Proofs and Some Remarks
    We now turn to the existence of g-inverses of an arbitrary matrix. 1. A Proof using Rank Factorization. Any matrix admits a rank factorization. Thus if A is an ...
  22. [22]
    [PDF] Generalized Inverse and Pseudoinverse - Iowa State University
    holds. ♢ 𝐴. # is reflexive if and only if rank 𝐴#. = rank 𝐴 . (See [4], Theorem 20.3.4, page 501). If 𝐴 has full rank then it is reflexive. Example 3 See [1], p ...Missing: factorization | Show results with:factorization
  23. [23]
    [PDF] PROPERTIES OF A GENERALIZED INVERSE WITH ... - DTIC
    This paper presents some properties and applications of a Rao general- ized inverse A" of an arbitrary matrix A. After both V and the better-known. Penrose- ...<|control11|><|separator|>
  24. [24]
    [0903.1785] On generalized inverses and Green's relations - arXiv
    Mar 10, 2009 · We study generalized inverses on semigroups by means of Green's relations. We first define the notion of inverse along an element and study its properties.
  25. [25]
    Full article: Projections for generalized inverses
    For an idempotent p in a ring R, every can be written as (2.6). or in matrix form. The decomposition (Equation2.6 (2.6) ) is known as the Pierce decomposition.
  26. [26]
    DrazinInverse: Drazin generalized inverse of a square matrix ...
    The Drazin inverse satisfies the relations and . » The nilpotency index of a matrix is defined as the size of the largest Jordan block corresponding to the ...
  27. [27]
    (PDF) On the regular elements of Z_n - ResearchGate
    Aug 5, 2025 · An element a \in R is called a regular element if there exists b \in R such that a=a2b, the element b is called a von Neumann inverse for a. A ...<|control11|><|separator|>
  28. [28]
    None
    ### Summary of Generalized Inverses in Semigroups Using Green's Relations (arXiv:0903.1785)
  29. [29]
    Generalized Inverses: Theory and Applications - SpringerLink
    In stockAvailable as PDF; Read on any device; Instant download; Own it forever ... Adi Ben-Israel, Thomas N. E. Greville. Series Title: CMS Books in Mathematics.
  30. [30]
    [PDF] Generalized inverses and solutions of linear systems
    The connection is exposed between the properties of the generalized inverses in (a) and (b) and the fact that among all matrices. X satisfying AXA = A, that ...Missing: algebra | Show results with:algebra
  31. [31]
    [PDF] Restricted Linear Constrained Minimization of quadratic functionals
    Sep 17, 2010 · 3 The Generalized inverse in linear constrained minimization. The Generalized inverse of an operator or matrix plays a crucial role in many.
  32. [32]
    Theory and Application of Constrained Inverse of Matrices - SIAM.org
    The concept of a constrained inverse of a matrix, introduced by Bott and Duffin [1], is extended so as to bring together the various generalized inverses ...
  33. [33]
    [PDF] a new tikhonov regularization method
    Let A† denote the Moore-Penrose pseudoinverse of A. We are interested in com- puting an approximation of the solution x = A†b of minimal Euclidean norm of the.
  34. [34]
    Projectors, Generalized Inverses and the Blue'S - Oxford Academic
    Summary. It is well known that in the Gauss–Markov model (Y, Xβ, σ2 V) with | V | ≠ 0, the BLUE (best linear unbiased estimator) of Xβ is Y1, the orthogona.
  35. [35]
    [PDF] The Use of the Generalized Inverse in the General Linear Statistical ...
    An experimental design application of the general linear sta- tistical model has been chosen to illustrate the use of the method in the less-than-full rank ...
  36. [36]
    On Best Linear Estimation and General Gauss-Markov Theorem in ...
    In this note we extend the Gauss–Markov theorem to the case of singular covariances and give an explicit formula for the BLUE of an estimable parameter. The ...
  37. [37]
    [PDF] s. R. Searle - Cornell eCommons
    This paper summarizes properties of generalized inverse matrices and some of their uses in statistics, particularly those relating to linear models. Neither ...
  38. [38]
    [PDF] Randomized Kernel Methods for Least-Squares Support Vector ...
    Mar 22, 2017 · Here we discuss several approximation algorithms for the least-squares support vector machine classifier. The proposed methods are based on ...
  39. [39]
    [PDF] Chapter 13 Applications of SVD and Pseudo-inverses - CIS UPenn
    The following properties, due to Penrose, characterize the pseudo-inverse of a matrix. We have already proved that the pseudo-inverse satisfies these ...<|control11|><|separator|>
  40. [40]
    [PDF] Solving Inverse Problems with Deep Linear Neural Networks - arXiv
    Feb 21, 2025 · Moreover, we let A† denote the Moore-Penrose pseudoinverse of A. We write σmin(A) for the smallest nonzero singular value of A. The ...<|control11|><|separator|>
  41. [41]
    A Fast Iterative Shrinkage-Thresholding Algorithm for Linear Inverse ...
    Abstract. We consider the class of iterative shrinkage-thresholding algorithms (ISTA) for solving linear inverse problems arising in signal/image processing.<|separator|>
  42. [42]
    Modified Truncated Randomized Singular Value Decomposition ...
    In this paper, we propose new randomization based algorithms for large scale linear discrete ill-posed problems with general-form regularization.Missing: signal 2020s
  43. [43]
    Some New Characterizations of Generalized Inverses
    Dec 2, 2023 · In this paper, we characterize the Moore–Penrose inverse of a square matrix based on the canonical polar decomposition by using an invertible matrix and a ...
  44. [44]
    New properties of weighted generalized core-EP inverse in Banach ...
    Jul 10, 2025 · We characterize the generalized weighted core-EP inverse via the canonical decomposition, utilizing a weighted core-EP invertible element and a quasinilpotent.Missing: extended operator theory quantum computing
  45. [45]
    [PDF] A.12 Generalized Inverse
    A = A holds (see Rao (1973a, p. 24). Theorem A.63 A generalized inverse always exists although it is not unique in general. Proof: Assume rank(A) = r.
  46. [46]
    [PDF] 1. Introduction. The Moore-Penrose generalized inverse of a matrix ...
    Soc., I, (1935). [2] R. PENROSE, A generalized inverse for matrices, Proc. Cambridge Philos. Soc.,. 51 (1955)' pp. 406-413. [3] R. RADO, Note on generalized ...
  47. [47]
  48. [48]
    [PDF] Generalized Inverses, Semigroups and Rings - HAL
    Sep 30, 2023 · First come generalized inverses, that inspired me in the first place and enlighten the path to more general semigroup theory and ring theory.
  49. [49]
    regular semigroup - PlanetMath
    Mar 22, 2013 · A regular semigroup is where all elements are regular, meaning x=xyx. A regular element has at least one inverse.
  50. [50]
    Generalized Inverses of Linear Transformations | SIAM Publications Library
    ### Summary of Generalized Inverses from https://epubs.siam.org/doi/book/10.1137/1.9780898719048
  51. [51]
    On the Drazin inverses involving power commutativity - ScienceDirect
    We explore the Drazin inverses of bounded linear operators with power commutativity ( P Q = Q m P ) in a Hilbert space. Conditions on Drazin invertibility ...
  52. [52]
    Additive and multiplicative properties of Drazin inverse under new ...
    Dec 8, 2023 · The objective of this paper is to study the Drazin invertibility of A+B and AB, when A, B\in\mathcal{B}(X)^D are A\text{ and }B-weakly commutative.