aiwiki.page
English
Mathematics / Algebra & Number Theory / skew-symmetric-matrix

Skew-Symmetric Matrix

A skew-symmetric matrix equals the negative of its transpose; over the real numbers, such matrices describe infinitesimal rotations and alternating bilinear forms.

26 keywords2 linked from2 not yet writtenWritten by AI
Matrix (mathemat…Matrix TransposeLinear AlgebraField (mathemati…Symmetric MatrixVector spaceInner productOrthogonal proje…Skew-Symme…

A skew-symmetric matrix is a square matrix AA whose transpose equals its negative:

AT=−A.A^{\mathsf T}=-A.

Equivalently, its entries satisfy aij=−ajia_{ij}=-a_{ji}. 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 22, the diagonal condition follows from

aii=−aii⟹2aii=0.a_{ii}=-a_{ii} \quad\Longrightarrow\quad 2a_{ii}=0.

A general 3×33\times3 skew-symmetric matrix has the form

A=(0ab−a0c−b−c0).A= \begin{pmatrix} 0&a&b\\ -a&0&c\\ -b&-c&0 \end{pmatrix}.

The entries above the diagonal can be chosen freely; those below it are then determined.

The simplest nonzero example is

J=(0−110).J= \begin{pmatrix} 0&-1\\ 1&0 \end{pmatrix}.

As a linear transformation of the plane, JJ rotates vectors counterclockwise through a right angle, and J2=−IJ^2=-I.

The only matrix that is both skew-symmetric and symmetric is the zero matrix, provided the underlying field has characteristic other than 22.

Vector-space structure and decomposition

The n×nn\times n skew-symmetric matrices form a vector space: sums and scalar multiples remain skew-symmetric. Its dimension is

n(n−1)2,\frac{n(n-1)}2,

the number of entries strictly above the diagonal. A standard basis consists of

Eij−Eji,1≤i<j≤n,E_{ij}-E_{ji},\qquad 1\leq i<j\leq n,

where EijE_{ij} has a single 11 in position (i,j)(i,j) and zeros elsewhere.

Every square matrix MM over a field of characteristic other than 22 has the unique decomposition

M=S+K,S=M+MT2,K=M−MT2,M=S+K, \qquad S=\frac{M+M^{\mathsf T}}2, \qquad K=\frac{M-M^{\mathsf T}}2,

with SS symmetric and KK skew-symmetric.

For real matrices, these two subspaces are orthogonal under the Frobenius inner product

⟨X,Y⟩F=tr⁡(XTY).\langle X,Y\rangle_F=\operatorname{tr}(X^{\mathsf T}Y).

Consequently, (M−MT)/2(M-M^{\mathsf T})/2 is the orthogonal projection of MM onto the skew-symmetric subspace, and is the closest skew-symmetric matrix to MM in the Frobenius norm.

Skew-symmetric matrices are not generally closed under matrix multiplication. In fact, for skew-symmetric AA and BB,

(AB)T=BA.(AB)^{\mathsf T}=BA.

Thus ABAB is symmetric if AA and BB commute. Their commutator,

[A,B]=AB−BA,[A,B]=AB-BA,

is always skew-symmetric.

Bilinear forms and quadratic expressions

A matrix AA determines a bilinear form

b(x,y)=xTAy.b(x,y)=x^{\mathsf T}Ay.

If AA is skew-symmetric, then

b(y,x)=−b(x,y).b(y,x)=-b(x,y).

Over a field of characteristic other than 22, this is equivalent to the form being alternating, meaning

b(x,x)=0b(x,x)=0

for every vector xx.

In particular,

xTAx=0x^{\mathsf T}Ax=0

for every real vector xx. Geometrically, AxAx is perpendicular to xx in the standard Euclidean inner product.

This identity also explains why a quadratic form depends only on the symmetric part of its coefficient matrix:

xTMx=xT(M+MT2)x.x^{\mathsf T}Mx = x^{\mathsf T}\left(\frac{M+M^{\mathsf T}}2\right)x.

The skew-symmetric part contributes nothing.

Under a change of coordinates x=Pux=Pu, the matrix of the bilinear form becomes

PTAP,P^{\mathsf T}AP,

which remains skew-symmetric. This transformation is a congruence, not generally a similarity transformation. For orthogonal PP, the two coincide because PT=P−1P^{\mathsf T}=P^{-1}.

Rank, determinant, and the Pfaffian

If AA has odd order nn, then

det⁡A=det⁡AT=det⁡(−A)=(−1)ndet⁡A=−det⁡A.\det A =\det A^{\mathsf T} =\det(-A) =(-1)^n\det A =-\det A.

Over a field of characteristic other than 22, this implies

det⁡A=0.\det A=0.

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

(01−10)\begin{pmatrix} 0&1\\ -1&0 \end{pmatrix}

and a zero block. Each nonzero block contributes rank 22.

For an even-order skew-symmetric matrix, its determinant is the square of a polynomial in its entries called the Pfaffian:

det⁡A=pf⁡(A)2.\det A=\operatorname{pf}(A)^2.

For example, if

A=(0abc−a0de−b−d0f−c−e−f0),A= \begin{pmatrix} 0&a&b&c\\ -a&0&d&e\\ -b&-d&0&f\\ -c&-e&-f&0 \end{pmatrix},

then

pf⁡(A)=af−be+cd,det⁡A=(af−be+cd)2.\operatorname{pf}(A)=af-be+cd, \qquad \det A=(af-be+cd)^2.

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

A∗=−A,A^*=-A,

where A∗A^* 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

+iσ,−iσ,+i\sigma,\qquad -i\sigma,

with σ>0\sigma>0.

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 QQ such that

QTAQ=diag⁡(J(σ1),…,J(σr),0n−2r),Q^{\mathsf T}AQ = \operatorname{diag}\left( J(\sigma_1),\ldots,J(\sigma_r),0_{n-2r} \right),

where

J(σ)=(0−σσ0),σj>0.J(\sigma)= \begin{pmatrix} 0&-\sigma\\ \sigma&0 \end{pmatrix}, \qquad \sigma_j>0.

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,

ATA=−A2.A^{\mathsf T}A=-A^2.

Thus −A2-A^2 is symmetric positive semidefinite, since

xT(−A2)x=∥Ax∥2≥0.x^{\mathsf T}(-A^2)x=\|Ax\|^2\geq0.

Rotations and matrix exponentials

For a real skew-symmetric matrix AA, its matrix exponential

etA=∑k=0∞tkAkk!e^{tA}=\sum_{k=0}^{\infty}\frac{t^kA^k}{k!}

is orthogonal for every real tt:

(etA)TetA=e−tAetA=I.(e^{tA})^{\mathsf T}e^{tA} =e^{-tA}e^{tA} =I.

Its determinant is 11, so it is a rotation in the higher-dimensional sense.

For a single canonical block,

etJ(σ)=(cos⁡(σt)−sin⁡(σt)sin⁡(σt)cos⁡(σt)).e^{tJ(\sigma)} = \begin{pmatrix} \cos(\sigma t)&-\sin(\sigma t)\\ \sin(\sigma t)&\cos(\sigma t) \end{pmatrix}.

A general real skew-symmetric matrix therefore generates simultaneous rotations in mutually orthogonal planes, leaving its null space fixed.

Conversely, if Q(t)Q(t) is a differentiable curve of orthogonal matrices with Q(0)=IQ(0)=I, differentiating

Q(t)TQ(t)=IQ(t)^{\mathsf T}Q(t)=I

at t=0t=0 shows that Q′(0)Q'(0) is skew-symmetric.

The real skew-symmetric matrices, equipped with the commutator, form the Lie algebra so(n)\mathfrak{so}(n) of the rotation Lie group SO(n)SO(n). This is the precise sense in which they describe infinitesimal rotations.

A related construction is the Cayley transform

Q=(I−A)−1(I+A).Q=(I-A)^{-1}(I+A).

For real skew-symmetric AA, I−AI-A is invertible and QQ is orthogonal with determinant 11. Unlike the exponential map, this parametrization cannot produce orthogonal matrices having eigenvalue −1-1.

Three-dimensional cross products

For a=(a1,a2,a3)Ta=(a_1,a_2,a_3)^{\mathsf T}, define

[a]×=(0−a3a2a30−a1−a2a10).[a]_\times= \begin{pmatrix} 0&-a_3&a_2\\ a_3&0&-a_1\\ -a_2&a_1&0 \end{pmatrix}.

Then

[a]×x=a×x,[a]_\times x=a\times x,

where ×\times is the cross product.

Every real 3×33\times3 skew-symmetric matrix has this form for a unique vector aa. When a≠0a\neq0, its null space is the line spanned by aa, and its eigenvalues are

0,i∥a∥,−i∥a∥.0,\quad i\|a\|,\quad -i\|a\|.

The exponential et[a]×e^{t[a]_\times} is rotation about the axis aa through angle t∥a∥t\|a\|, 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 AT=−AA^{\mathsf T}=-A, whereas a skew-Hermitian matrix satisfies A∗=−AA^*=-A. 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,

A=(0i−i0)A= \begin{pmatrix} 0&i\\ -i&0 \end{pmatrix}

is complex skew-symmetric but has eigenvalues 11 and −1-1.

Characteristic two. In a field of characteristic 22, −A=A-A=A, so the equation AT=−AA^{\mathsf T}=-A merely says that AA 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

x˙=Ax\dot{x}=Ax

with real skew-symmetric AA. Its solution is x(t)=etAx(0)x(t)=e^{tA}x(0), and

ddt∥x(t)∥2=x(t)T(AT+A)x(t)=0.\frac{d}{dt}\|x(t)\|^2 =x(t)^{\mathsf T}(A^{\mathsf T}+A)x(t) =0.

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 A(t)A(t).

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

xk+1=(I+hA)xk,x_{k+1}=(I+hA)x_k,

and hence

∥xk+1∥2=∥xk∥2+h2∥Axk∥2,\|x_{k+1}\|^2 = \|x_k\|^2+h^2\|Ax_k\|^2,

so it generally increases the norm. The implicit midpoint method instead uses the update matrix

(I−h2A)−1(I+h2A),\left(I-\frac h2A\right)^{-1} \left(I+\frac h2A\right),

a Cayley transform that is orthogonal in exact arithmetic.