Fact-checked by Grok 2 weeks ago

Involutory matrix

An involutory matrix is a square matrix A that is equal to its own inverse, meaning A^2 = I, where I is the identity matrix of the same order.^[1] This property implies that applying the matrix twice returns the original vector unchanged, making it a linear transformation of order at most 2.^[2] In linear algebra, involutory matrices exhibit several key properties that distinguish them from general invertible matrices. They are always diagonalizable over fields where the characteristic polynomial splits, such as the complex numbers, because their minimal polynomial divides x^2 - 1 = (x-1)(x+1), which has distinct linear factors.^[3] Consequently, the eigenvalues of an involutory matrix are restricted to +1 and -1, corresponding to eigenspaces that represent fixed points and reflections, respectively.^[3] Additionally, the identity matrix I is the trivial example of an involutory matrix, as I^2 = I, and the negative identity -I also satisfies the condition since (-I)^2 = I.^[1] Notable non-trivial examples include Householder reflection matrices, which are used in numerical algorithms like QR decomposition for their orthogonal and involutory nature, ensuring H^2 = I where H = I - 2 \frac{v v^T}{v^T v} for a nonzero vector v.^[4] Involutory matrices also appear in applications such as cryptography, where they simplify key management in systems like the Hill cipher by making encryption and decryption operations identical modulo a prime.^[5] Their role in representing involutions—symmetries that are their own inverses—extends to group theory and geometry, underscoring their importance in modeling reversible transformations.^[6]

Fundamentals

Definition

An involutory matrix is a square matrix A of order n \times n over a field, typically the real or complex numbers, that satisfies the equation A^2 = I, where I is the n \times n identity matrix.^[7] This property implies that A is invertible and A^{-1} = A, since A \cdot A = I.^[7] Only square matrices can be involutory, as the identity matrix I is defined exclusively for square dimensions, precluding non-square matrices from satisfying the condition.^[7] The term "involutory" derives from the Latin involvere, meaning "to roll up," which evokes the self-inverse nature of the matrix, whereby applying it twice "unrolls" back to the identity. The concept has been studied in linear algebra since the 19th century.^[8]^[9]

Basic Properties

An involutory matrix A satisfies A^2 = I, where I is the identity matrix of the same order, implying that A is invertible with A^{-1} = A.^[10] This relation leads to a simple pattern in the powers of A: for any positive integer k, A^k = A if k is odd, and A^k = I if k is even and greater than or equal to 2. To see this, note that A^3 = A^2 A = I A = A and A^4 = (A^2)^2 = I^2 = I; higher powers follow by induction, as odd powers reduce to multiplication by A and even powers to powers of I.^[10] The determinant of an n \times n involutory matrix A over the real or complex numbers is \pm 1. This follows from the multiplicative property of determinants: \det(A^2) = \det(A) \det(A) = [\det(A)]^2 = \det(I) = 1, so \det(A) = \pm 1.^[10]^[11] For an n \times n involutory matrix A, the trace \operatorname{tr}(A) is an integer satisfying -n \leq \operatorname{tr}(A) \leq n.^[12] The involutory property is invariant under similarity transformations. Specifically, if B = P^{-1} A P for some invertible matrix P, then B^2 = (P^{-1} A P)(P^{-1} A P) = P^{-1} A (P P^{-1}) A P = P^{-1} A^2 P = P^{-1} I P = I.^[10] The minimal polynomial m_A(x) of an involutory matrix A divides x^2 - 1 = (x-1)(x+1), and hence has degree at most 2. This is because m_A(A) = 0 and A^2 - I = 0, so m_A(x) annihilates A and must divide any such polynomial.^[10]

Characterization and Structure

Spectral Properties

An involutory matrix A \in \mathbb{C}^{n \times n} satisfies A^2 = I_n, where I_n is the n \times n identity matrix. The eigenvalues \lambda of A must satisfy \lambda^2 = 1, implying that every eigenvalue is either +1 or -1. The algebraic multiplicities of these eigenvalues sum to n, the dimension of the space. The trace of A, denoted \operatorname{tr}(A), equals the sum of its eigenvalues (counted with algebraic multiplicity). If k is the algebraic multiplicity of the eigenvalue +1, then the multiplicity of -1 is n - k, yielding \operatorname{tr}(A) = k \cdot 1 + (n - k) \cdot (-1) = 2k - n. The eigenspace corresponding to \lambda = 1 is \ker(A - I_n), and for \lambda = -1 it is \ker(A + I_n). Since the minimal polynomial of A divides x^2 - 1 = (x - 1)(x + 1) and has distinct roots, the space \mathbb{C}^n decomposes as the direct sum \ker(A - I_n) \oplus \ker(A + I_n), with the dimensions of these eigenspaces adding to n. Over the complex numbers, every involutory matrix is diagonalizable because its minimal polynomial splits into distinct linear factors. Thus, there exists an invertible matrix P such that P^{-1} A P = \operatorname{diag}(I_k, -I_{n-k}) for some k. For real involutory matrices, the eigenvalues remain \pm 1 (real numbers), and the minimal polynomial again splits into distinct linear factors over \mathbb{R}, ensuring diagonalizability over the reals as well.

Jordan Canonical Form

An involutory matrix A satisfies A^2 = I, implying that its minimal polynomial m_A(x) divides x^2 - 1 = (x-1)(x+1), a polynomial with distinct linear factors over both \mathbb{R} and \mathbb{C}. Consequently, A is diagonalizable over these fields, and its Jordan canonical form is a diagonal matrix consisting of entries \pm 1.^[13] The absence of repeated roots in the minimal polynomial ensures that there are no non-trivial Jordan blocks; all blocks are 1×1 with eigenvalues \pm 1. This structure holds over \mathbb{C}, where the characteristic polynomial splits completely into linear factors.^[14] Over \mathbb{R}, the eigenvalues \pm 1 are real, so A is diagonalizable over \mathbb{R} as well, yielding the same diagonal Jordan canonical form with no larger blocks.^[13] There exists an invertible matrix P such that P^{-1} A P = D, where D = \operatorname{diag}(\lambda_1, \dots, \lambda_n) and each \lambda_i = \pm 1; the columns of P form a basis of eigenvectors for A. If A is symmetric, the spectral theorem guarantees that P can be chosen orthogonal, yielding an orthogonal diagonalization.^[15]

Examples

Algebraic Examples

The identity matrix I_n of order n provides the simplest algebraic example of an involutory matrix, satisfying I_n^2 = I_n by definition.^[16] Likewise, the scalar multiple -I_n is involutory, as (-I_n)^2 = I_n.^[16] Permutation matrices associated with involutions in the symmetric group S_n form another class of involutory matrices. An involution \sigma \in S_n satisfies \sigma^2 = \mathrm{id}, and the corresponding permutation matrix P_\sigma, with entries (P_\sigma)_{i,j} = \delta_{i, \sigma(j)}, then obeys P_\sigma^2 = I_n. For instance, the transposition \sigma = (1\ 2) yields the matrix

P = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix},

and direct computation confirms P^2 = I_2. The exchange matrix J_n, defined with 1's along the anti-diagonal and 0's elsewhere, is a specific involutory permutation matrix known as the reversal matrix. It satisfies J_n^2 = I_n for any n, as repeated reversal restores the original order.^[17] An example for n=3 is

J_3 = \begin{pmatrix} 0 & 0 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 0 \end{pmatrix},

with J_3^2 = I_3.^[17] Block diagonal matrices constructed as direct sums of involutory blocks are also involutory. If A \in M_m(\mathbb{C}) and B \in M_k(\mathbb{C}) are involutory, then the direct sum A \oplus B = \begin{pmatrix} A & 0 \\ 0 & B \end{pmatrix} satisfies (A \oplus B)^2 = I_{m+k}.^[16] This follows from the block-diagonal multiplication rule, preserving the property across independent subspaces. For concrete 2×2 involutory matrices over \mathbb{R}, solving A^2 = I_2 for A = \begin{pmatrix} a & b \\ c & d \end{pmatrix} yields two cases: diagonal matrices with entries \pm 1 on the diagonal, or non-diagonal forms where d = -a and bc = 1 - a^2.^[18] A representative non-diagonal example is

A = \begin{pmatrix} 2 & 1 \\ -3 & -2 \end{pmatrix},

verifying A^2 = I_2 via direct multiplication.^[18]

Geometric Examples

In Euclidean space, involutory matrices often arise as linear transformations representing reflections, which geometrically flip points across a hyperplane while leaving the hyperplane fixed. A prominent example is the Householder reflection matrix, defined as H = I - 2 \mathbf{u} \mathbf{u}^T, where \mathbf{u} is a unit vector normal to the hyperplane of reflection and I is the identity matrix. This transformation satisfies H^2 = I, confirming its involutory nature, and it is both symmetric (H^T = H) and orthogonal (H^T H = I), preserving lengths and angles in the space.^[19] Orthogonal reflections in real Euclidean space are inherently involutory, as applying the reflection twice returns every point to its original position. When the reflection matrix is real and symmetric, it remains orthogonal, ensuring the transformation is an isometry. More generally, symmetric involutory matrices over the reals are precisely the orthogonal matrices whose eigenvalues are exclusively \pm 1; the eigenspace for eigenvalue 1 consists of fixed points on the hyperplane, while the eigenspace for -1 comprises directions that are negated, corresponding to reflections or improper rotations that reverse orientation.^[1] A simple geometric example in two dimensions is the reflection across the x-axis, represented by the matrix

\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}.

This matrix has eigenvalues 1 (with eigenvector along the x-axis, fixed) and -1 (with eigenvector along the y-axis, flipped), illustrating how the transformation preserves the reflection line while inverting the perpendicular direction; squaring the matrix yields the identity, verifying involutoriness.^[1] In higher dimensions, Householder reflections generalize this concept to reflections over hyperplanes orthogonal to a unit vector \mathbf{u}, with the same algebraic properties holding: symmetry, orthogonality, and involution. The eigenvalue 1 has multiplicity n-1 (spanning the fixed hyperplane), while -1 has multiplicity 1 (along \mathbf{u}). For a symmetric involutory matrix A, the expression (I + A)/2 defines the orthogonal projection onto the +1 eigenspace (the subspace of fixed points), and (I - A)/2 projects onto the -1 eigenspace (the reflected directions); these projections are idempotent and sum to the identity.^[19]^[20] As noted in the spectral properties, the eigenvalues \pm 1 directly underpin this decomposition into fixed and reflected subspaces.

Applications

In Computing and Graphics

In computer graphics, reflection transformations rely on involutory matrices to mirror objects across axes or planes efficiently. For example, the 2D reflection over the x-axis is represented by the diagonal matrix \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}, which is involutory since its square equals the identity matrix, allowing repeated applications—such as toggling mirror effects—without computing an inverse.^[21]^[22] This property simplifies rendering pipelines, as the self-inverse nature reduces computational overhead in real-time animations or symmetric scene manipulations.^[23] Householder matrices, a class of symmetric involutory orthogonal matrices defined as H = I - 2 \frac{v v^T}{v^T v} where v is a nonzero vector, are fundamental in numerical linear algebra for QR decomposition.^[24] In computing applications, including graphics, this decomposition solves least-squares problems for tasks like curve fitting or lighting calculations, with the involutory property enabling efficient iterative orthogonalization steps that revert to the original vector upon reapplication.^[25] The method's stability and low communication cost make it suitable for parallel implementations on GPUs, accelerating matrix factorizations in graphics software.^[26] In image processing, involutory matrices facilitate symmetry operations such as horizontal or vertical flips, which transform pixel coordinates via self-inverse linear mappings to generate augmented data. For instance, a horizontal flip matrix \begin{pmatrix} -1 & 0 \\ 0 & 1 \end{pmatrix} mirrors the image and, being involutory, allows instant reversal without additional matrix inversion.^[27] These transformations are integral to data augmentation in deep learning pipelines, enhancing model robustness to symmetries while minimizing storage and computation due to the self-inverse trait.^[28] In optimization algorithms, involutory matrices appear as preconditioners or mirror maps in iterative methods like alternating mirror descent, where their self-inverse property simplifies updates by eliminating explicit inversion steps. For example, in symplectic analyses of such descent methods, an involutory matrix A with A^2 = I preserves geometric structures, accelerating convergence in constrained optimization problems common in graphics rendering.^[29] Similarly, in distributed gradient tracking, involutory transformations coordinate updates across nodes, reducing synchronization overhead in large-scale computations.^[30] This efficiency is particularly beneficial for real-time applications, where iterative solvers handle high-dimensional parameter spaces.

In Physics and Cryptography

In quantum mechanics, the Pauli matrices \sigma_x, \sigma_y, and \sigma_z serve as fundamental involutory operators, satisfying \sigma_i^2 = I for i = x, y, z, where I is the 2×2 identity matrix.^[31] These matrices represent spin operators for spin-1/2 particles and are employed as Pauli gates in quantum circuits, enabling operations like state flipping. For instance, the Pauli X gate (\sigma_x) interchanges the computational basis states |0\rangle and |1\rangle, and its self-inverse property—applying it twice returns the original state—facilitates reversible computations without additional inverse gates.^[32] In broader physical contexts, involutory matrices model symmetry operators in particle physics, such as the parity operator P, which inverts spatial coordinates and satisfies P^2 = I, preserving or reflecting wave functions based on intrinsic parity eigenvalues \pm 1.^[33] Similarly, the time-reversal operator T in systems without spin often acts as an involutory transformation, with T^2 = I, reversing momenta while leaving positions unchanged to enforce time-reversal invariance in scattering processes and Hamiltonian dynamics.^[34] These representations underpin analyses of conservation laws and symmetry breaking in high-energy physics. In cryptography, involutory maximum distance separable (MDS) matrices provide efficient diffusion in block cipher designs, achieving optimal branch numbers while being self-inverse, thus allowing shared hardware for encryption and decryption rounds. The Skinny family of lightweight block ciphers utilizes such an involutory MDS matrix in its linear layer to minimize circuit overhead in resource-constrained environments like IoT devices.^[35] This self-inverse structure simplifies round functions, as the same matrix-vector multiplication applies bidirectionally, reducing implementation complexity without compromising security margins.^[36] Recent developments emphasize orthogonal involutory MDS matrices, which combine self-inversivity with orthogonality to further optimize gate counts and power consumption in lightweight ciphers, as explored in structural analyses over finite fields.^[37] These constructions support variants of AES-inspired designs and emerging standards, enhancing resistance to differential and linear cryptanalysis through provable diffusion properties.

References

[1]
Linear Algebra Glossary - UC Davis Math
Feb 15, 2012 · Involutory Matrix. An involutory matrix A has the property that. A * A = I. Simple facts about an involutory matrix A: The identity matrix is ...
[2]
Inverse functions and the identity function - UTSA
Nov 14, 2021 · A matrix that is its own inverse (i.e., a matrix A such that A = A−1 and A2 = I), is called an involutory matrix.
[3]
[PDF] T E T EE AND ONE A O S a "$$4 Instructions: Do < of the 8 ...
32 linear operator T on4/ is said to be involutory if65 -¼ =*5 . (a) Prove that a linear operator on)/ is involutory if and only if it is diagonalizable.
[4]
Sage Tutorial, part 2.1 (Positive matrices)
\) Involutory matrices are all square roots of the identity matrix. Another example of the involutory matrix gives the Householder matrix (or Householder ...
[5]
[PDF] Affine Ciphers - Koc Lab
An involutory matrix is the inverse of itself: A2 = I. This way, the encryption and decryption keys are the same: Encryption function: v = Au (mod 26).
[6]
[PDF] Numerical Matrix Analysis - Ilse Ipsen
(ii) What can you say about an involutory matrix that is also unitary (orthogonal)? ... involutory matrix . . . . 11, 11, 18, 20 inverse of ...
[7]
Involutory Matrix -- from Wolfram MathWorld
A square matrix A such that A^2=I , where I is the identity matrix. An involutory matrix is its own matrix inverse. Explore with Wolfram|Alpha.
[8]
Involution - Etymology, Origin & Meaning
Late 14c. origin from Latin involutio meaning "a rolling up," describes a twisted, coiled condition or fold, originally used in anatomy to denote ...
[9]
None
Below is a merged summary of all mentions of "involutory" in *Matrix Analysis* (2nd Edition) by Roger A. Horn and Johnson, consolidating information from the provided segments into a comprehensive response. To retain as much detail as possible, I will use a structured format with tables where appropriate, followed by a narrative summary for additional context. The response avoids redundancy while ensuring all unique information is included.
[10]
[PDF] Products of involutions in the stable general linear group - HAL
Dec 20, 2021 · Every involutory matrix has determinant ±1, while every U2-matrix has determinant. 1. Note also that In is both an involution and a U2-matrix ...
[11]
[PDF] Each n-by-n matrix with n > 1 is a sum of 5 coninvolutory matrices
Aug 8, 2016 · trace(A) = trace(J) is an integer. ... (a) Each square matrix over an infinite field is a sum of an involutory matrix and a diagonalizable matrix.
[12]
[PDF] Diagonalizable matrix
Apr 19, 2013 · Put in another way, a matrix is diagonalizable if each block in its Jordan form has no nilpotent part; i.e., one-by-one matrix. Diagonalization.
[13]
[PDF] the minimal polynomial and some applications - Penn Math
Theorem 4.11. Let A: V → V be a linear operator. Then A is diagonalizable if and only if its minimal polynomial in F[T] splits in F[T] and has distinct roots.Missing: involutory | Show results with:involutory
[14]
[PDF] Orthogonally Diagonalizable Matrices
The Spectral Theorem says that the symmetry of is also. E. E sufficient: a real symmetric matrix must be orthogonally diagonalizable. This is the part of the.Missing: involutory | Show results with:involutory
[15]
[PDF] The Sum of Involutory Matrices 1 Introduction
Definition 1. Let A ∈ Mn(K). Then A is an involution or an involutory matrix if A2 = I. We denote by Γ+ n the set of all involutions in Mn (K). Let Ω+ n be ...
[16]
[PDF] Characterization and properties of matrices with k-involutory ...
Abstract. We say that a matrix R ∈ Cn×n is k-involutory if its minimal poly- nomial is xk − 1 for some k ≥ 2, so Rk−1 = R−1 and the eigenvalues.
[17]
[PDF] 587 A NOTE ON SPECIAL MATRICES Roselin Antony1
version of the identity matrix. The exchange matrix is also called the reversal matrix, backward identity, or standard involutory permutation. Page 7. a note ...<|control11|><|separator|>
[18]
Involutory Matrix - GeeksforGeeks
Aug 21, 2025 · Properties of Involutory Matrix · If A and B are two involutory matrices of the same order and AB = BA, then AB is also an involutory matrix.Missing: linear | Show results with:linear
[19]
[PDF] Types of square matrices
Feb 7, 2025 · Gram matrix of a lin. indep. set of vectors. Principal submatrices ... Symmetric and involutory. Eigenvalues are +1. -1 is a simple ...
[20]
[PDF] Zero-Sum Triangles for Involutory, Idempotent, Nilpotent and ... - arXiv
Such unipotent matrices can be used in matrix construction or factorization for Jordan canonical form blocks with prescribed eigenvalues or Frobenius normal ...
[21]
Computer Graphics - Reflection Transformation in 3D - GeeksforGeeks
Jun 22, 2022 · Reflection is nothing but a mirror image of an object. Three kinds of Reflections are possible in 3D space: Reflection along the X-Y plane.
[22]
Transformation of Graphs Using Matrices - Reflection - Varsity Tutors
Reflections using matrices work by applying the matrix to each point's coordinates, effectively flipping points across the desired axis or line using matrix ...
[23]
Reflection (geometry) - Citizendium
Oct 10, 2024 · It follows that the eigenvalues of σ are ±1. The product of the eigenvalues being its determinant, −1, the sets of eigenvalues of σ are either { ...
[24]
Householder matrix - StatLect
A matrix is said to be involutory if it is equal to its inverse. ... Householder reduction and QR decomposition. The Householder matrix analyzed ...Definition · An incredibly useful reflector · Householder reduction
[25]
Householder Reflections and the QR Decomposition » Cleve's Corner
Oct 3, 2016 · A Householder reflection is characterized by a vector u, which, following Pete's convention, is normalized to have ||u||= √ 2.
[26]
[PDF] Householder transformations - Cornell: Computer Science
Householder transformations are orthogonal reflections, expressed as H = I − 2vvT, used to introduce subdiagonal zeros in Gaussian elimination.
[27]
[PDF] Encoding Involutory Invariances in Neural Networks - arXiv
This paper introduces architectures that embed invariance to involutory linear/affine transformations up to parity p = ±1, where NN(X) = pNN(AX).
[28]
[PDF] Encoding Involutory Invariance in Neural Networks
Jun 7, 2021 · P is the diagonalizing matrix that converts the involutory matrix A to its standard diagonal form. (refer Appendix B). The algorithm assumes ...
[29]
[PDF] A Symplectic Analysis of Alternating Mirror Descent - arXiv
May 17, 2025 · Furthermore, when A is an involutory matrix, i.e., A2 = Id, then ΩA ... The Jordan canonical form of a product of a Hermi- tian and a ...
[30]
[PDF] Triggered Gradient Tracking for Asynchronous Distributed ... - arXiv
Since T1 is an involutory matrix (i.e., it coincides with its inverse), the change of coordinates (A.21) transforms (A.20) in. ˙˜y. ˙˜η. = −2L I. −L2. 0. ˜y. ˜η.
[31]
Pauli Matrix - an overview | ScienceDirect Topics
The Pauli matrices (traditionally used in QM for particle spin) are prominent, and have the property of squaring to the identity operator I. There is a special ...Missing: involutory | Show results with:involutory
[32]
XGate (latest version) | IBM Quantum Documentation
This is done by recursively reversing all sub-instructions. It does not invert any gate. Returns. a new instruction with. sub-instructions reversed. Return type.
[33]
[PDF] Parity Operator and Eigenvalue
☞ P2 = I. ☞ Parity operator is unitary. □ If the interaction Hamiltonian (H) conserves parity. ☞ [H,P] = 0. ☞ P.
[34]
[PDF] 13 Symmetries in Particle Physics
There are three important discrete symmetries: parity (P), charge conjugation (C) and time reversal (T). ... The parity operator changes the direction of motion ...Missing: involutory | Show results with:involutory
[35]
[PDF] Lightweight Involutory MDS Matrices - Cryptology ePrint Archive
The SKINNY family of block ciphers and its low-latency variant MANTIS. In Advances in. Cryptology - CRYPTO 2016 - 36th Annual International Cryptology ...
[36]
[PDF] MDS Matrices with Lightweight Circuits - Inria
In particular, a number of lightweight ciphers use an involutory matrix, so that the same implementation can be used for encryption and ... The. SKINNY family of ...
[37]
New Insights into Involutory and Orthogonal MDS Matrices - arXiv
Involutory and orthogonal MDS matrices offer additional benefits by allowing identical or nearly identical circuitry for both encryption and ...