1 Definition and basic properties

An orthogonal matrix is a square matrix whose columns form an orthonormal set of vectors. The same is true of its rows. In practical terms, such a matrix acts on Euclidean space without changing lengths or angles. This makes orthogonal matrices fundamental in geometry and many computational methods.

1.1 Orthonormal rows and columns

If the columns of a matrix are orthonormal, each column has unit length and any two distinct columns are perpendicular. The same condition on the rows is equivalent for square matrices. This symmetry is one of the defining features of orthogonal matrices and distinguishes them from more general invertible matrices.

1.2 Equivalent matrix equations

Orthogonality can be expressed compactly using matrix multiplication. A square matrix \(Q\) is orthogonal when multiplying it by its transpose yields the identity matrix. This condition encodes the orthonormality of the rows and columns in algebraic form.

1.2.1 Transpose and inverse relationship

For an orthogonal matrix, the transpose is also its inverse. In symbols, \(Q^T = Q^{-1}\). This property is especially useful because it makes inversion computationally simple: the inverse is obtained by swapping rows and columns.

1.2.2 Identity matrix condition

The defining equation is \(Q^TQ = QQ^T = I\), where \(I\) is the identity matrix. This means that applying \(Q\) and then undoing it with \(Q^T\) leaves every vector unchanged. It also shows that orthogonal matrices preserve the standard inner product structure of Euclidean space.

1.3 Determinant and invertibility

Every orthogonal matrix is invertible, since its transpose serves as an inverse. Its determinant must be either \(1\) or \(-1\). A determinant of \(1\) corresponds to a transformation that preserves orientation, while \(-1\) indicates a reversal of orientation, such as a reflection.

1.4 Geometric interpretation

Geometrically, orthogonal matrices represent rigid motions about the origin. They preserve distances, angles, and areas or volumes in the appropriate dimension, up to possible sign changes in orientation. In two and three dimensions, they describe rotations and reflections, and in higher dimensions they generalize these familiar transformations.

2 Examples

Orthogonal matrices appear in many standard forms. Some are simple and immediately recognizable, while others arise from combining basic geometric operations.

2.1 Identity matrix

The identity matrix is orthogonal because its columns and rows are the standard basis vectors, which are orthonormal. It leaves every vector unchanged and serves as the simplest example of an orthogonal matrix.

2.2 Reflection matrices

A reflection matrix flips vectors across a line or plane through the origin. Such matrices are orthogonal because reflection preserves lengths and angles, even though it reverses orientation. In two dimensions, a reflection across a line makes a familiar example of an orthogonal transformation with determinant \(-1\).

2.3 Rotation matrices

Rotation matrices are orthogonal matrices with determinant \(1\). In two dimensions, a rotation matrix is determined by an angle and rotates every vector around the origin. In three dimensions, rotations can occur about an axis, and the corresponding matrices remain orthogonal.

2.4 Permutation matrices

A permutation matrix is obtained by reordering the rows or columns of the identity matrix. Its rows and columns are orthonormal because each row and column contains a single \(1\) and the rest zeros. These matrices are orthogonal and often appear in algorithms that rearrange data.

3 Characterizations

Orthogonal matrices can be identified in several equivalent ways. Besides the standard transpose condition, they may be recognized by their action on dot products, norms, and eigenvalues.

3.1 Preservation of dot products

A matrix is orthogonal if it preserves dot products between vectors. For any vectors \(x\) and \(y\), an orthogonal matrix \(Q\) satisfies \((Qx)\cdot(Qy)=x\cdot y\). This property captures the idea that the transformation does not distort angles.

3.2 Preservation of norms and distances

Because dot products are preserved, vector lengths are preserved as well. Hence \(\|Qx\|=\|x\|\) for every vector \(x\). Distances between points are also unchanged, since the difference of two vectors has the same norm before and after applying \(Q\).

3.3 Eigenvalue properties

The eigenvalues of an orthogonal matrix are constrained by its length-preserving nature. They lie on or within the complex unit circle, with real eigenvalues restricted to special values. These facts are often used in theoretical analysis.

3.3.1 Complex eigenvalues

If an orthogonal matrix has complex eigenvalues, they occur in conjugate pairs. Each complex eigenvalue has absolute value \(1\). This reflects the fact that repeated application of the matrix cannot produce unbounded growth in any direction.

3.3.2 Real eigenvalues

The real eigenvalues of an orthogonal matrix can only be \(1\) or \(-1\). An eigenvalue of \(1\) corresponds to a fixed direction, while \(-1\) indicates reversal along that direction. These cases are closely connected to rotations and reflections.

4 Special classes

Orthogonal matrices are often divided into subclasses according to determinant and group structure. These categories help describe their geometric behavior and algebraic organization.

4.1 Proper orthogonal matrices

Proper orthogonal matrices are those with determinant \(1\). They preserve orientation and are commonly associated with rotations. In many applications, these are the orthogonal transformations regarded as “pure” rotational motions.

4.2 Improper orthogonal matrices

Improper orthogonal matrices have determinant \(-1\). They reverse orientation and include reflections and transformations formed by combining a rotation with a reflection. Such matrices still preserve lengths and angles but do not preserve handedness.

4.3 Orthogonal groups

The orthogonal matrices of a given size form a mathematical group under matrix multiplication. This structure makes it possible to study them systematically using group theory. The set splits naturally into two related families.

4.3.1 The group O(n)

The group \(O(n)\) consists of all \(n \times n\) orthogonal matrices. It includes both proper and improper orthogonal transformations. This group is important in geometry because it describes all linear isometries of \(\mathbb{R}^n\) that fix the origin.

4.3.2 The special orthogonal group SO(n)

The special orthogonal group \(SO(n)\) consists of orthogonal matrices with determinant \(1\). It is the subgroup of \(O(n)\) made up of orientation-preserving transformations. In low dimensions, it corresponds to ordinary rotations.

5 Algebraic properties

Orthogonal matrices satisfy several useful algebraic rules. These properties make them stable under many common operations and convenient in theoretical and computational settings.

5.1 Closure under multiplication

The product of two orthogonal matrices is again orthogonal. This follows from the transpose condition and shows that composing two distance-preserving linear transformations produces another one. Closure under multiplication is one reason the orthogonal matrices form a group.

5.2 Inverse and transpose

For an orthogonal matrix, the inverse equals the transpose. This relationship simplifies calculations and often improves numerical efficiency. Instead of performing a general matrix inversion, one can simply transpose the matrix.

5.3 Powers and repeated application

Any integer power of an orthogonal matrix is also orthogonal. Positive powers correspond to repeated application of the same transformation, while negative powers use the inverse, which is again the transpose. This behavior is especially relevant for rotations, where repeated application can produce periodic motion.

5.4 Block diagonal orthogonal matrices

A block diagonal matrix is orthogonal if each diagonal block is orthogonal and the off-diagonal blocks are zero. This construction allows larger orthogonal matrices to be built from smaller ones. It is often useful when a problem naturally separates into independent subspaces.

6 Construction and decomposition

Orthogonal matrices can be constructed from arbitrary sets of vectors or used as building blocks in matrix factorization. Several standard procedures in linear algebra generate or decompose them efficiently.

6.1 Gram-Schmidt process

The Gram-Schmidt process converts a linearly independent set of vectors into an orthonormal set. When these vectors are arranged as columns of a matrix, the result is an orthogonal factor. This procedure is foundational in many numerical algorithms, though care is needed to maintain stability in computation.

6.2 QR decomposition

In QR decomposition, a matrix is factored into an orthogonal matrix \(Q\) and an upper triangular matrix \(R\). This decomposition is widely used in solving linear systems and least squares problems. The orthogonal factor helps simplify calculations by preserving norms and reducing numerical error.

6.3 Householder transformations

Householder transformations are orthogonal matrices that reflect vectors across hyperplanes. They are commonly used to zero out selected components of a vector or matrix column. Because they are numerically stable, they play a major role in matrix factorization algorithms.

6.4 Givens rotations

Givens rotations are orthogonal transformations that act nontrivially only on a two-dimensional subspace. They rotate coordinates in a chosen plane while leaving the remaining coordinates unchanged. This localized behavior makes them useful for introducing zeros in sparse or structured matrices.

7 Applications

Orthogonal matrices are widely used because they combine geometric clarity with numerical reliability. Their ability to preserve structure makes them valuable in many areas of applied mathematics and engineering.

7.1 Geometry and computer graphics

In geometry, orthogonal matrices represent rigid motions of coordinate systems. In computer graphics, they are used to rotate models, camera views, and coordinate frames without distortion. Their preservation of lengths and angles helps maintain the visual integrity of scenes.

7.2 Least squares and numerical stability

Orthogonal matrices are central in least squares computation because they help transform a problem into a simpler form without changing its essential content. They are also valued for numerical stability, since operations involving orthogonal matrices tend to avoid large amplification of rounding errors. For this reason, many high-quality algorithms are designed around orthogonal factorizations.

7.3 Signal processing

In signal processing, orthogonal transforms are used to separate, compress, or analyze data. Their energy-preserving property is particularly important, since it allows signal content to be redistributed without changing total power. Many transform methods rely on orthogonal or nearly orthogonal matrices.

7.4 Physics and rigid body motion

Orthogonal matrices describe rigid rotations of bodies and coordinate systems in physics. They are used in mechanics, robotics, and related fields to represent orientation changes. Since they preserve distance, they are well suited to modeling motion without deformation.

Several other matrix classes are closely connected to orthogonal matrices. Some are complex analogues, while others impose extra symmetry or structured constraints.

8.1 Unitary matrices

Unitary matrices are the complex analogues of orthogonal matrices. They satisfy \(U^*U=I\), where \(U^*\) is the conjugate transpose. Like orthogonal matrices, they preserve inner products, norms, and angles, but they act on complex vector spaces.

8.2 Symmetric orthogonal matrices

A symmetric orthogonal matrix is both symmetric and orthogonal. Such matrices satisfy \(Q^T=Q\) as well as \(Q^TQ=I\). They are closely related to reflections, since symmetry together with orthogonality strongly restricts their possible form.

8.3 Stiefel matrices

Stiefel matrices are matrices with orthonormal columns but not necessarily square. They generalize the column structure of orthogonal matrices to rectangular settings. These matrices are important in geometry and optimization, where orthonormal frames of lower dimension are studied.