The Adjoints and the Polar Decomposition of an Operator at $n=2$

Introduction

The endomorphism algebra of the matrix space is a Hilbert space of real dimension thirty-two, each operator has an adjoint, and every invertible operator is the product of a unitary operator with a positive operator, by the same theorem that gives the polar decomposition of a matrix.

The article computes adjoints of operators, identifies the self-adjoint, the unitary and the positive operators, proves the polar decomposition inside the endomorphism algebra, applies it to the one-sided and the two-sided families, and computes the positive part of a two-sided operator in terms of its two parameters. It assumes The Length and the Two Ranks at $n=2$, The Polar Decomposition from the Two Halves at $n=2$ and The Hilbert–Schmidt Form at $n=2$.

It follows the biquaternion article The Endomorphisms of the Biquaternion Algebra for the same adjoints on the eight real coordinates.

The Adjoint of an Operator

The Hilbert–Schmidt form gives every operator an adjoint, and the adjoint is the conjugate transpose in the basis of matrix units.

Definition. The adjoint $T^{*}$ of an operator $T$ is the unique operator with $\langle TZ,W\rangle=\langle Z,T^{*}W\rangle$ for all $Z,W$.

Proposition. The adjoint is conjugate-linear, involutive and reversing: $(T^{*})^{*}=T$ and $(TS)^{*}=S^{*}T^{*}$.

Proof. The form is linear in the second argument and conjugate-linear in the first, and the two identities are checked by moving the form from one side to the other twice.

Corollary. In the basis of matrix units the adjoint of an operator is the conjugate transpose of its matrix.

Proof. The Hilbert–Schmidt form is the standard Hermitian form on the four complex coordinates, whose adjoint is the conjugate transpose.

Corollary. The trace of an operator is the trace of its adjoint up to conjugation, and the length of an operator equals the length of its adjoint.

Proof. The trace of $T^{*}$ is the conjugate of the trace of $T$, and the squared lengths of the two agree.

The Adjoints of the Two Families

Proposition. The adjoint of $L_{X}$ is $L_{X^{*}}$ and the adjoint of $T_{X,Y}$ is $T_{X^{*},Y^{*}}$.

Proof. The first was computed in The One-Sided Operators at $n=2$; for the second, $\langle XZY,W\rangle=\operatorname{tr}(Y^{*}Z^{*}X^{*}W)=\langle Z,X^{*}WY^{*}\rangle$.

Corollary. A two-sided operator is self-adjoint exactly when both parameters are Hermitian.

Proof. The adjoint is the two-sided operator with the conjugate transposed parameters, and the two operators agree exactly when both parameters are Hermitian.

Corollary. A two-sided operator is unitary exactly when both of its parameters are unitary.

Proof. The composition $T_{X,Y}^{*}T_{X,Y}=T_{X^{*}X,YY^{*}}$ is the identity exactly when both of its parameters are the identity, since a two-sided operator is the identity exactly when both parameters are.

The Special Operators

Proposition. The self-adjoint operators of the matrix space form a real subspace of dimension sixteen, the unitary operators form a group of real dimension sixteen, and the positive operators form a cone inside the self-adjoint space.

Proof. The self-adjoint operators are the Hermitian matrices of the four-dimensional complex space, a real space of dimension sixteen; the unitary operators are the unitary matrices of that space, a group of real dimension sixteen; and the positive operators are the positive definite Hermitian matrices of the same space, an open cone.

Corollary. The self-adjoint operators of the one-sided family are exactly the left multiplications by Hermitian matrices, a subspace of real dimension four.

Proof. The left multiplication is self-adjoint exactly when its parameter is Hermitian, and the Hermitian half has dimension four.

Corollary. The unitary operators of the one-sided family are exactly the left multiplications by unitary matrices, and they form a group of real dimension four.

Proof. The left multiplication is unitary exactly when its parameter is unitary, and the unitary group at order two has real dimension four.

The Polar Decomposition of an Operator

Theorem. Every invertible operator $T$ of the matrix space is uniquely $T=UP$ with $U$ unitary and $P$ positive definite self-adjoint.

Proof. This is The Polar Decomposition from the Two Halves at $n=2$ applied to the four-dimensional complex space of matrices, with the Hilbert–Schmidt form as the Hermitian form.

Corollary. The endomorphism algebra is the product of its unitary group with its positive cone, and the product map is a diffeomorphism.

Proof. The decomposition is unique and continuous in both directions.

Corollary. The positive part is the square root of $T^{*}T$ and its eigenvalues are the singular values of $T$.

Proof. The construction of the matrix case applies verbatim, and the singular values of an operator are the eigenvalues of its positive part.

The Positive Part of a Two-Sided Operator

Proposition. For a two-sided operator with invertible parameters the operator $T^{*}T$ is the two-sided operator $T_{X^{*}X,Y Y^{*}}$.

Proof. The adjoint of $T_{X,Y}$ is $T_{X^{*},Y^{*}}$, and the composition gives the two-sided operator with the two products on the two sides.

Corollary. The positive part of an invertible two-sided operator is $T_{|X|,|Y^{*}|}$, with $|X|=(X^{*}X)^{1/2}$ and $|Y^{*}|=(YY^{*})^{1/2}$.

Proof. The square root of a two-sided operator with one parameter on each side is computed on each side, and the square roots of $X^{*}X$ and of $YY^{*}$ are the two absolute values.

Corollary. The unitary part of an invertible two-sided operator is $T_{X|X|^{-1},\,|Y^{*}|^{-1}Y}$, whose two parameters are the unitary factors of the polar decompositions of $X$ and of $Y$.

Proof. Multiplying the operator by the inverse of its positive part gives the stated two-sided operator, and each parameter of that operator is unitary because it is the unitary factor of a polar decomposition.

The Geometry of the Decomposition

Proposition. The polar decomposition of an operator separates the rotation of the matrix space from its stretching, and the stretching is measured on the four singular values.

Proof. The unitary factor preserves the length of every matrix, and the positive factor multiplies the length along four orthogonal directions by the four singular values.

Corollary. The volume distortion of an operator is the product of its four singular values, which is the absolute value of its determinant.

Proof. The determinant is the product of the eigenvalues, and the absolute value of the determinant is the product of the singular values.

Corollary. An operator preserves the Hilbert–Schmidt length exactly when its positive part is the identity, which is the case of the unitary operators.

Proof. The positive part is the identity exactly when all the singular values are one.

Corollary. The condition number of an operator is the ratio of its largest to its smallest singular value, and it is one exactly for the unitary operators up to a scalar.

Proof. The ratio is computed from the eigenvalues of the positive part, and it equals one exactly when they are all equal; the positive part is then a scalar multiple of the identity, and the operator is a scalar multiple of a unitary one.

Summary

Every operator of the matrix space has an adjoint, which is the conjugate transpose in the basis of matrix units and is computed on the two parameters of a two-sided operator by conjugating and transposing each of them. The self-adjoint operators form a real space of dimension sixteen, the unitary group has real dimension sixteen, and the positive operators form a cone inside the self-adjoint space.

Every invertible operator is uniquely the product of a unitary operator with a positive operator, and the positive part is the square root of $T^{*}T$ with the singular values on its diagonal. For a two-sided operator the positive part is read from the two products $X^{*}X$ and $YY^{*}$.

Summary of Notation

Symbol Meaning
$T^{*}$ the adjoint for the Hilbert–Schmidt form
$T=UP$ the polar decomposition of an operator
$P=(T^{*}T)^{1/2}$ the positive part
$U$ the unitary part
$|X|$ the positive square root of $X^{*}X$

Further Reading