The Polar Decomposition from the Two Halves at $n=2$
Introduction
Every invertible $2\times2$ complex matrix is a unitary matrix times a positive definite Hermitian matrix, the two factors are unique, and the positive factor is the square root of $A^{*}A$, so the decomposition is built from the Hermitian half and the unitary group.
The article states the polar decomposition, constructs the positive factor from $A^{*}A$, proves the unitarity of the other factor, proves uniqueness, treats the singular case where the factors are not unique, and lists the real, symmetric and complex variants. It assumes The Exponential of $\operatorname{Sym}_2$: Stretches and the Positive Cone, The Spectral Theorem: the Hermitian Case and the Sphere of Idempotents and Normal $2\times2$ Matrices and the Common Spectral Theorem.
It follows the biquaternion article The Polar Decomposition of a Biquaternion for the same split written with the unit $e_t$ and the algebraic norm $N$.
The Statement
Theorem. Every invertible matrix $A$ is $A=UP$ with $U$ unitary and $P$ positive definite Hermitian, and the two factors are unique.
Proof. The construction and the uniqueness are proved in the two sections below.
Corollary. The polar decomposition is a bijection between the group of invertible matrices and the product of the unitary group with the positive definite cone.
Proof. The decomposition gives the map from the product to the group, and the uniqueness gives its injectivity.
Corollary. The determinant of $A$ is the determinant of $U$ times the positive number $\det P$, so the modulus of the determinant is $\det P$ and the phase is $\det U$.
Proof. The determinant is multiplicative, and a positive definite matrix has a positive determinant.
The Construction
Proposition. Let $P=(A^{*}A)^{1/2}$ be the positive definite square root. Then $U=AP^{-1}$ is unitary.
Proof. The matrix $A^{*}A$ is Hermitian, and it is positive definite because $v^{*}A^{*}Av=\|Av\|^{2}>0$ for a nonzero $v$ when $A$ is invertible; its square root $P$ is therefore positive definite and invertible. Then $U^{*}U=P^{-1}A^{*}AP^{-1}=P^{-1}P^{2}P^{-1}=I$, so $U$ is unitary.
Corollary. The matrix $A$ is the product $UP$, and the factor $P$ is Hermitian and positive definite.
Proof. The definition of $U$ gives $A=UP$, and $P$ was chosen Hermitian and positive definite.
Corollary. The factors are obtained from the two eigenvalues of $A^{*}A$, which are the squares of the singular values.
Proof. The square root is taken on the eigenvalues, and the singular values are their square roots, so the eigenvalues of $P$ are the singular values.
Uniqueness
Proposition. If $A=U_{1}P_{1}=U_{2}P_{2}$ with unitary $U_{j}$ and positive definite $P_{j}$, then $U_{1}=U_{2}$ and $P_{1}=P_{2}$.
Proof. Taking adjoints in $U_{1}P_{1}=U_{2}P_{2}$ and multiplying gives $P_{1}^{2}=A^{*}A=P_{2}^{2}$. Two positive definite matrices with equal squares are equal, because each is the unique positive definite square root of its square, so $P_{1}=P_{2}$. Then $U_{1}P_{1}=U_{2}P_{1}$ with $P_{1}$ invertible gives $U_{1}=U_{2}$.
Corollary. The decomposition is unique, so the notation $U(A)$ and $P(A)$ is well defined.
Proof. The two factors are determined by $A$.
Corollary. The positive factor depends only on $A$ through $A^{*}A$, so two matrices have the same positive factor exactly when they have the same $A^{*}A$.
Proof. The positive factor is the square root of $A^{*}A$, and the square root is unique.
The Two Halves
Proposition. The positive factor lies in the Hermitian half and the unitary factor lies in the exponential of the anti-Hermitian half, so both factors come from the two halves.
Proof. A positive definite matrix is Hermitian, and a unitary matrix is the exponential of an anti-Hermitian matrix by The Exponential of a $2\times2$ Matrix and Its Image.
Corollary. The polar decomposition writes the group as the product of the exponential of the anti-Hermitian half with the exponential of the Hermitian half.
Proof. The unitary factor is the exponential of an anti-Hermitian matrix, and the positive factor is the exponential of a Hermitian matrix, by the logarithm of The Exponential of $\operatorname{Sym}_2$: Stretches and the Positive Cone.
Corollary. The unitary group is the compact factor and the positive cone is the noncompact factor, and the multiplication map from their product to the group of invertible matrices is a bijection.
Proof. The decomposition supplies the map, the uniqueness makes it injective, and every invertible matrix is the image of its own pair of factors.
The Singular Case
Proposition. A matrix $A$ of rank one is $A=UP$ with $U$ unitary and $P$ positive semidefinite of rank one, and the factors are not unique.
Proof. The matrix $A^{*}A$ is positive semidefinite of rank one and its square root $P$ is defined. On the range of $P$, which is a line, the factor $U$ is forced by $U(Pv)=Av$, and that rule is an isometry there; it extends to the orthogonal complement, which is the kernel of $P$, in a circle of ways, so the factors are not unique.
Corollary. The polar decomposition exists for every matrix with $P$ positive semidefinite, and it is unique exactly when the matrix is invertible.
Proof. The kernel of $A$ is nonzero exactly when the matrix is singular, and the freedom in $U$ is precisely the freedom on that kernel.
Corollary. At rank one the freedom is a circle, the choice of a unitary on the kernel line.
Proof. The kernel is a line, and the unitary group of a line is a circle.
The Real and Symmetric Cases
Proposition. A real invertible matrix has a real polar decomposition with an orthogonal factor and a positive definite symmetric factor.
Proof. The matrix $A^{\mathsf T}A$ is real symmetric positive definite, its square root is real symmetric, and the factor $U=A P^{-1}$ satisfies $U^{\mathsf T}U=I$.
Corollary. A real matrix with positive determinant has the decomposition with a rotation factor, and one with negative determinant has a reflection factor.
Proof. The determinant of $U$ is the sign of the determinant of $A$, and the orthogonal group splits into rotations and reflections accordingly.
Corollary. The decomposition of a self-adjoint matrix is trivial on the positive part and gives the sign on the negative part.
Proof. For a Hermitian $A$ the matrix $A^{*}A=A^{2}$ has the square root $|A|$, and $U$ is the unitary that is the identity on the positive eigenspace and minus the identity on the negative one.
The Biquaternion Comparison
Proposition. In the biquaternion language the polar decomposition is the split of an element into its norm and its direction, and the algebraic norm $N$ is the determinant of the positive factor.
Proof. The algebraic norm $N$ of inertia $(2,2)$ restricts to the determinant on the matrix model, and the positive factor has the determinant equal to $N$ of the element.
Corollary. The biquaternion polar decomposition of The Polar Decomposition of a Biquaternion is the same statement under the identification of the two algebras.
Proof. The two decompositions agree under the matrix model, and both are unique on the invertible elements.
Corollary. The unitary factor carries the direction and the positive factor carries the length, so the four-dimensional unitary group and the three-dimensional cone account for the seven real dimensions of the invertible set.
Proof. The unitary group has real dimension four, the positive cone has dimension three, and the product covers the seven-dimensional complement of the singular set inside the eight-dimensional algebra.
Summary
Every invertible $2\times2$ complex matrix is $A=UP$ with $U$ unitary and $P$ positive definite Hermitian, and the factors are unique. The positive factor is the square root of $A^{*}A$ and its eigenvalues are the singular values, while the unitary factor is $AP^{-1}$.
Both factors come from the two halves: the positive factor is the exponential of a Hermitian matrix and the unitary factor is the exponential of an anti-Hermitian matrix. For a singular matrix the decomposition survives with a positive semidefinite factor but is not unique, the freedom being the choice of a unitary on the kernel.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $A=UP$ | the polar decomposition |
| $U$ | the unitary factor |
| $P$ | the positive definite (or semidefinite) factor |
| $A^{*}A$ | the Hermitian matrix whose square root is $P^{2}$ |
| $N$ | the algebraic norm, the determinant on the model |
Further Reading
- Rajendra Bhatia, Matrix Analysis (Springer, 1997), for the polar decomposition and the singular values.
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge University Press, 2013), for the existence and uniqueness statements.
- The Exponential of $\operatorname{Sym}_2$: Stretches and the Positive Cone, for the positive factor as an exponential.
- Normal $2\times2$ Matrices and the Common Spectral Theorem, for the unitary diagonalisation used here.
- The Polar Decomposition of a Biquaternion, for the same split in the biquaternion language.