A skew-symmetric matrix is a square matrix whose transpose equals its negative:
Equivalently, its entries satisfy . Over the real or complex numbers, every diagonal entry is therefore zero. Skew-symmetric matrices are fundamental in linear algebra, where they represent alternating bilinear forms and, over the real numbers, infinitesimal rotations. Their structure imposes strong restrictions on their rank, determinant, and eigenvalues.
Definition and examples
For a matrix over a field of characteristic other than , the diagonal condition follows from
A general skew-symmetric matrix has the form
The entries above the diagonal can be chosen freely; those below it are then determined.
The simplest nonzero example is
As a linear transformation of the plane, rotates vectors counterclockwise through a right angle, and .
The only matrix that is both skew-symmetric and symmetric is the zero matrix, provided the underlying field has characteristic other than .
Vector-space structure and decomposition
The skew-symmetric matrices form a vector space: sums and scalar multiples remain skew-symmetric. Its dimension is
the number of entries strictly above the diagonal. A standard basis consists of
where has a single in position and zeros elsewhere.
Every square matrix over a field of characteristic other than has the unique decomposition
with symmetric and skew-symmetric.
For real matrices, these two subspaces are orthogonal under the Frobenius inner product
Consequently, is the orthogonal projection of onto the skew-symmetric subspace, and is the closest skew-symmetric matrix to in the Frobenius norm.
Skew-symmetric matrices are not generally closed under matrix multiplication. In fact, for skew-symmetric and ,
Thus is symmetric if and commute. Their commutator,
is always skew-symmetric.
Bilinear forms and quadratic expressions
A matrix determines a bilinear form
If is skew-symmetric, then
Over a field of characteristic other than , this is equivalent to the form being alternating, meaning
for every vector .
In particular,
for every real vector . Geometrically, is perpendicular to in the standard Euclidean inner product.
This identity also explains why a quadratic form depends only on the symmetric part of its coefficient matrix:
The skew-symmetric part contributes nothing.
Under a change of coordinates , the matrix of the bilinear form becomes
which remains skew-symmetric. This transformation is a congruence, not generally a similarity transformation. For orthogonal , the two coincide because .
Rank, determinant, and the Pfaffian
If has odd order , then
Over a field of characteristic other than , this implies
Every odd-order skew-symmetric matrix is therefore singular.
More generally, the rank of a skew-symmetric matrix over such a field is always even. Under an invertible congruence transformation, it can be reduced to a block diagonal matrix consisting of blocks
and a zero block. Each nonzero block contributes rank .
For an even-order skew-symmetric matrix, its determinant is the square of a polynomial in its entries called the Pfaffian:
For example, if
then
In particular, every real skew-symmetric matrix has nonnegative determinant, including zero in the singular case.
Eigenvalues and real canonical form
A real skew-symmetric matrix satisfies
where denotes the conjugate transpose. It is therefore a special case of a skew-Hermitian matrix.
Every eigenvalue of a real skew-symmetric matrix is purely imaginary or zero. Nonzero eigenvalues occur in conjugate pairs
with .
Such a matrix is normal, so the complex spectral theorem gives a unitary diagonalization. Over the real numbers, the corresponding result is an orthogonal block decomposition: there exists an orthogonal matrix such that
where
Each block acts in an invariant two-dimensional plane, while the zero block acts on the null space. This decomposition makes the even-rank property explicit. It also shows that each positive singular value occurs twice.
Furthermore,
Thus is symmetric positive semidefinite, since
Rotations and matrix exponentials
For a real skew-symmetric matrix , its matrix exponential
is orthogonal for every real :
Its determinant is , so it is a rotation in the higher-dimensional sense.
For a single canonical block,
A general real skew-symmetric matrix therefore generates simultaneous rotations in mutually orthogonal planes, leaving its null space fixed.
Conversely, if is a differentiable curve of orthogonal matrices with , differentiating
at shows that is skew-symmetric.
The real skew-symmetric matrices, equipped with the commutator, form the Lie algebra of the rotation Lie group . This is the precise sense in which they describe infinitesimal rotations.
A related construction is the Cayley transform
For real skew-symmetric , is invertible and is orthogonal with determinant . Unlike the exponential map, this parametrization cannot produce orthogonal matrices having eigenvalue .
Three-dimensional cross products
For , define
Then
where is the cross product.
Every real skew-symmetric matrix has this form for a unique vector . When , its null space is the line spanned by , and its eigenvalues are
The exponential is rotation about the axis through angle , with orientation determined by the right-hand rule.
This correspondence is widely used in rigid-body kinematics, robotics, and classical mechanics to express angular velocity as a linear operation on position vectors.
Complex matrices and characteristic two
Two qualifications are essential when extending the real theory.
Complex entries. A complex skew-symmetric matrix satisfies , whereas a skew-Hermitian matrix satisfies . These are different conditions. The purely imaginary eigenvalue property belongs to skew-Hermitian matrices and need not hold for complex skew-symmetric matrices. For example,
is complex skew-symmetric but has eigenvalues and .
Characteristic two. In a field of characteristic , , so the equation merely says that is symmetric. It does not force the diagonal to vanish. To represent an alternating bilinear form in this setting, zero diagonal must be required separately. Alternating matrices still have even rank, but matrices satisfying the transpose equation alone need not.
Applications to differential equations
Consider the linear ordinary differential equation
with real skew-symmetric . Its solution is , and
The motion preserves Euclidean length rather than producing growth or decay. The same norm-preservation calculation holds for a time-dependent real skew-symmetric coefficient matrix .
This structure is important in numerical analysis because a time-stepping method may fail to preserve the geometry of the exact flow. For example, forward Euler gives
and hence
so it generally increases the norm. The implicit midpoint method instead uses the update matrix
a Cayley transform that is orthogonal in exact arithmetic.