The Adjoint of the Sesquilinear Two-Sided Operator
Introduction
The sesquilinear two-sided operator $\mathcal{T}_{a,b}(x)=ax^{*}b$ is the two-sided operator of a sesqualgebra, the smallest operator that uses the whole structure: one slot carries the involution, two carry the multiplication, and the result is $\varsigma$-semilinear of the same kind as the conjugate left multiplication. Its adjoint for the pairing $\varphi(x,y)=\tau(xy^{*})$ is taken by the twisted rule, and it is the two-sided operator with the two parameters interchanged,
$$ \mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a} . $$
So the adjoint acts on the pair $(a,b)$ by the transposition and not by the involution in each factor. That single identity carries the article: from it come the self-adjoint two-sided operators, which for invertibles are the pairs with $b=ap$ for a central unit $p$, and in particular the Hermitian pairs $\mathcal{T}_{a,a}$; the unitarity condition, under which the two-sided operators with unitary parameters are unitary operators, $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{b,a}$; and the comparison with the bilinear and the graded cases, where the adjoint keeps the order of the pair and conjugates its entries instead.
The subject belongs to the operator theory of the sesqualgebras and completes the adjoint theory of the operators: the one-sided case is The Adjoint of the Conjugate Left Multiplication, the ternary case is The Adjoint of the Ternary Product, and the present article is the two-sided case, from which the other two are read off.
Throughout, $R$ is a commutative ring with $1$ without zero divisors, $\varsigma$ is an involution of $R$, and $A$ is a unital associative $R$-algebra with a $\varsigma$-semilinear involution $*$, so that $A$ is a sesqualgebra in the sense of Sesqualgebras; the unit satisfies $1^{*}=1$. The function $\tau:A\to R$ is $R$-linear, central, $\tau(ab)=\tau(ba)$, and compatible with the involution, $\tau(x^{*})=\varsigma(\tau(x))$, and the pairing $\varphi(x,y)=\tau(xy^{*})$ is $R$-linear in its first variable, $\varsigma$-semilinear in its second and perfect. The derived product is $x\star y=xy^{*}$, and the operators are
$$ \mathcal{T}_{a,b}(x)=ax^{*}b,\qquad L_{p}R_{q}(x)=pxq,\qquad L_{a}=\mathcal{T}_{a,1},\qquad R_{b}=L_{1}R_{\,b^{*}} . $$
The adjoint of a $\varsigma$-semilinear operator $S$ is the operator $S^{\dagger}$ with $\varphi(Sx,y)=\varsigma\bigl(\varphi(x,S^{\dagger}y)\bigr)$, and the adjoint of an $R$-linear operator $T$ is the operator $T^{\dagger}$ with $\varphi(Tx,y)=\varphi(x,T^{\dagger}y)$; both exist and are unique, and every two-sided operator is $\varsigma$-semilinear, so its adjoint is taken by the twisted rule.
The Adjoint of the Two-Sided Operator
The Computation
Definition. For $a,b\in A$ the sesquilinear two-sided operator is the operator
$$ \boxed{\;\mathcal{T}_{a,b}:A\longrightarrow A,\qquad \mathcal{T}_{a,b}(x)=ax^{*}b\;} ; $$
it is additive and $\varsigma$-semilinear, $\mathcal{T}_{a,b}(\lambda x)=\varsigma(\lambda)\mathcal{T}_{a,b}(x)$, by The Sesquilinear Two-Sided Operator.
Theorem. For all $a,b\in A$ the adjoint of the two-sided operator is the two-sided operator with the parameters interchanged,
$$ \boxed{\;\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}\;} . $$
Proof. By the twisted rule it suffices to verify $\varsigma\bigl(\varphi(x,\mathcal{T}_{b,a}y)\bigr)=\varphi(\mathcal{T}_{a,b}x,y)$ for all $x,y$. The left-hand side is $\varsigma\bigl(\tau\bigl(x(by^{*}a)^{*}\bigr)\bigr)$, and $(by^{*}a)^{*}=a^{*}yb^{*}$, so it is $\varsigma\bigl(\tau(xa^{*}yb^{*})\bigr)=\tau\bigl((xa^{*}yb^{*})^{*}\bigr)=\tau(by^{*}ax^{*})$, using $\varsigma\circ\tau=\tau\circ{}^{*}$. By the centrality of $\tau$ this is $\tau(ax^{*}by^{*})$, which is $\varphi(\mathcal{T}_{a,b}x,y)$.
Corollary. The adjoint is an involution of the two-sided family, $\bigl(\mathcal{T}_{a,b}^{\dagger}\bigr)^{\dagger}=\mathcal{T}_{a,b}$, so the family is carried to itself and the adjoint acts on it as the involution $\mathcal{T}_{a,b}\mapsto \mathcal{T}_{b,a}$. On the monoid of the two families of The Sesquilinear Two-Sided Operator, generated by the two-sided operators and the two-sided multiplications, the adjoint is an anti-automorphism of order two.
Proof. The theorem applied twice gives $\mathcal{T}_{b,a}^{\dagger}=\mathcal{T}_{a,b}$, which is the first statement, and the family is stable because $\mathcal{T}_{b,a}$ is again a two-sided operator. On the monoid the adjoint reverses products, $(XY)^{\dagger}=Y^{\dagger}X^{\dagger}$, it is of order two, and by the theorem and the linear rule it carries each family to itself, $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}$ and $L_{p}R_{q}^{\dagger}=L_{p^{*}}R_{q^{*}}$; so it is an anti-automorphism of the monoid of order two.
Remark (the dependence on the conjugate of $b$). Read through the two families of The Sesquilinear Two-Sided Operator, where ${}^{*}\circ L_{p}R_{q}=\mathcal{T}_{q^{*},p^{*}}$, the adjoint is
$$ \mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}={}^{*}\circ L_{a^{*}}R_{\,b^{*}} , $$
so the adjoint carries the pair $(a,b)$ to the ordinary two-sided multiplication by the conjugate parameters in the reverse order, composed with the involution. The dependence is on $a^{*}$ and $b^{*}$ together, and the transposition of the parameters is the record of the single involution that the two-sided operator already carries in its middle slot.
The Interpretation through the Factors
Remark (a second proof). The two-sided operator is the composite $\mathcal{T}_{a,b}=L_{a}R_{b}\circ{}^{*}$ of the two-sided multiplication with the involution, and the adjoint of a composite is the composite of the adjoints in the reverse order, $(XY)^{\dagger}=Y^{\dagger}X^{\dagger}$. The involution is self-adjoint, $\bigl({}^{*}\bigr)^{\dagger}={}^{*}$, and the two-sided multiplication has adjoint $L_{a}R_{b}^{\dagger}=L_{a^{*}}R_{b^{*}}$ by the linear rule, since it is $R$-linear. Hence
$$ \mathcal{T}_{a,b}^{\dagger}=\bigl(L_{a}R_{b}\circ{}^{*}\bigr)^{\dagger}=\bigl({}^{*}\bigr)^{\dagger}\circ L_{a}R_{b}^{\dagger}={}^{*}\circ L_{a^{*}}R_{b^{*}}=\mathcal{T}_{b,a} , $$
which is the theorem again, read factor by factor.
Remark (contrast with the linear family). The two-sided multiplication therefore obeys $L_{p}R_{q}^{\dagger}=L_{p^{*}}R_{q^{*}}$, the adjoint acting on the two parameters by the involution in each factor and preserving the order, while the two-sided operator obeys $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}$, the adjoint exchanging them. The two rules are the two halves of the table of The Sesquilinear Two-Sided Operator: on the linear family the adjoint acts on the pair without transposing it, and on the sesquilinear family it transposes the pair and puts the conjugates inside the linear factor.
The Self-Adjoint Two-Sided Operators
Theorem. Let $a,b$ be units. Then $\mathcal{T}_{a,b}$ is self-adjoint if and only if $b=ap$ for a central unit $p$; in particular $\mathcal{T}_{a,a}$ is self-adjoint for every $a$.
Proof. Self-adjointness is $\mathcal{T}_{b,a}=\mathcal{T}_{a,b}$ by the theorem. By the proposition on the kernel of the parametrisation of The Sesquilinear Two-Sided Operator, two such operators with units satisfy $\mathcal{T}_{c,d}=\mathcal{T}_{a',b'}$ exactly when $c=a'p$ and $d=p^{-1}b'$ for a central unit $p$. Read with $\mathcal{T}_{c,d}=\mathcal{T}_{b,a}$ and $\mathcal{T}_{a',b'}=\mathcal{T}_{a,b}$, this is $b=ap$ and $a=p^{-1}b$ for a central unit $p$. The second relation is $b=pa$, which is the same as the first because $p$ is central, so the condition is $b=ap$ with $p$ a central unit. The case $a=b$ is $p=1$.
Corollary. If $b=ap$ for a central unit $p$, then $\mathcal{T}_{a,b}$ is self-adjoint for every $a$, unit or not; and on a commutative algebra every invertible two-sided operator is self-adjoint.
Proof. With $p$ central and $b=ap$ one has $\mathcal{T}_{a,b}(x)=a x^{*}a p$ and $\mathcal{T}_{b,a}(x)=a p x^{*}a=a(x^{*}p)a=a x^{*}p a=a x^{*}a p$, the middle steps moving $p$ past $x^{*}$ and past $a$ by its centrality; the two operators are equal for every $a$, unit or not. On a commutative $A$ the element $a^{-1}b$ is central, and it is invertible when $a$ and $b$ are.
Remark (the inner involution). The two-sided operator $\mathcal{T}_{u,u^{*}}=\sigma_{u}$ of a unitary $u$ is the inner involution of The Involutions of a Sesqualgebra, and the criterion reads $u^{*}=up$ for a central unit $p$, that is $u^{-2}=p$; so the inner involution is self-adjoint exactly when $u^{2}$ is central, which is exactly the condition under which it is an involution. The two conditions of the family therefore meet at the same elements, $u^{2}\in Z(A)$.
The Unitary Two-Sided Operators
Theorem (the unitary parameters). Let $a$ and $b$ be unitary elements of $A$. Then the two-sided operator is a unitary operator, with
$$ \boxed{\;\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}=\mathcal{T}_{a,b}^{-1}\;} . $$
Proof. The inverse of a two-sided operator is $\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{(b^{-1})^{*},\,(a^{-1})^{*}}$, by The Sesquilinear Two-Sided Operator. For a unitary $b$ one has $(b^{-1})^{*}=(b^{*})^{*}=b$, and likewise $(a^{-1})^{*}=a$, so the inverse is $\mathcal{T}_{b,a}$, which is the adjoint by the theorem.
Corollary. For all $a,b$ the composites of the two-sided operator with its adjoint are the ordinary two-sided multiplications
$$ \mathcal{T}_{a,b}^{\dagger}\mathcal{T}_{a,b}=L_{b b^{*}}R_{\,a^{*}a},\qquad \mathcal{T}_{a,b}\mathcal{T}_{a,b}^{\dagger}=L_{a a^{*}}R_{\,b^{*}b}, $$
so the two-sided operator is unitary exactly when both composites are the identity, that is when $b b^{*}$ and $a a^{*}$ are central units with $a^{*}a=(b b^{*})^{-1}$ and $b^{*}b=(a a^{*})^{-1}$. The two conditions are not redundant over a general ring; for invertibles $a,b$ the single condition $\mathcal{T}_{b,a}\mathcal{T}_{a,b}=\mathrm{id}$ suffices, because $\mathcal{T}_{a,b}$ is then invertible and a one-sided inverse is the inverse. The case of unitary parameters is $b b^{*}=a^{*}a=a a^{*}=b^{*}b=1$.
Proof. The composites are the multiplication table of The Sesquilinear Two-Sided Operator, $\mathcal{T}_{c,d}\mathcal{T}_{a,b}=L_{cb^{*}}R_{\,a^{*}d}$, read with $(c,d)=(b,a)$ for the first and $(c,d)=(a,b)$ for the second; and $L_{p}R_{q}=\mathrm{id}$ exactly when $p$ is a central unit with $q=p^{-1}$, since at $x=1$ it forces $pq=1$ and then $p$ must be central. If $a,b$ are invertible then $\mathcal{T}_{a,b}$ is invertible, so $\mathcal{T}_{b,a}\mathcal{T}_{a,b}=\mathrm{id}$ already makes $\mathcal{T}_{b,a}$ the two-sided inverse and gives the second identity.
Remark (the kernel of the parametrisation and a wider family). The invertible two-sided operators are the quotient of $A^{\times}\times A^{\times}$ by the action of the central units, $\mathcal{T}_{ap,p^{-1}b}=\mathcal{T}_{a,b}$, so a two-sided operator with unitary parameters depends on the pair only through the central unitary elements. The criterion of the corollary is wider than the unitary parameters: for $a=pV$ and $b=p^{-1}U$ with $U,V$ unitary and $p$ a central unit one has $a^{*}a=p^{*}p$ and $b b^{*}=(p^{*}p)^{-1}$, central units inverse to each other, and $a$ and $b$ are invertible, so the single condition suffices and $\mathcal{T}_{a,b}$ is unitary although neither parameter need be, the case $p=1$ being the unitary pair.
The Comparison with the Bilinear and the Signed Cases
Remark. In the bilinear theory the two-sided operator is $L_{a}R_{b}(x)=axb$ and its adjoint is $L_{\hat a}R_{\hat b}$, where $\hat a=\alpha(a^{r})$ is the conjugation of The Adjoint of the Two-Sided Operator on a Hermitian Algebra: the adjoint acts on the pair by the involution in each factor and keeps the order. Here the pair is transposed and the conjugates are absorbed into the linear factor, $\mathcal{T}_{a,b}^{\dagger}={}^{*}\circ L_{a^{*}}R_{b^{*}}$. The difference is the involution that the sesquilinear two-sided operator already carries in its middle slot: it leaves the adjoint nothing to conjugate in the pair, and the adjoint trades the two slots instead. The graded companion, The Signed Adjoint of the Two-Sided Operator on a Hermitian Algebra, keeps the order as well: its adjoint is $R_{\hat b}\,\alpha\,L_{\hat a}$, so it carries the pair $(\hat a,\hat b)$ to $(\alpha(\hat a),\alpha(\hat b))$ through the involution $\alpha$. In the bilinear and the signed cases alike the adjoint therefore conjugates the two parameters and preserves their order, and it is the single involution carried by the middle slot of the sesquilinear two-sided operator that makes its adjoint exchange them instead; the transposition of the parameters is exactly the trace of the semilinearity of the two-sided operator.
Worked Cases
The Complex Matrices
Let $A=M_{n}(\mathbb{C})$ with the conjugate transpose, $\varsigma$ the complex conjugation, $\tau=\operatorname{tr}$ and $\varphi(X,Y)=\operatorname{tr}(XY^{*})$. The two-sided operator is $\mathcal{T}_{A,B}(X)=AX^{*}B$ and the theorem reads $\mathcal{T}_{A,B}^{\dagger}=\mathcal{T}_{B,A}$, so $\mathcal{T}_{A,A}$ is self-adjoint for every matrix $A$, and for a unitary $U$ the reflection $\mathcal{T}_{U,U}$ is self-adjoint with $\mathcal{T}_{U,U}^{2}=\mathrm{id}$. A two-sided operator with unitary parameters is unitary, $\mathcal{T}_{U,V}^{\dagger}=\mathcal{T}_{V,U}=\mathcal{T}_{U,V}^{-1}$, while the inner involution $\mathcal{T}_{U,U^{*}}=\sigma_{U}$ is self-adjoint exactly when $U^{2}$ is a scalar matrix, so the witness $U=\operatorname{diag}(i,1)$ of The Involutions of a Sesqualgebra has $U^{2}=\operatorname{diag}(-1,1)$ noncentral and $\mathcal{T}_{U,U^{*}}$ not self-adjoint.
The Field
Let $A=\mathbb{C}$ over $R=\mathbb{C}$ with $\varsigma$ the conjugation, $\tau$ the identity and $\varphi(x,y)=x\overline{y}$. The two-sided operator is $\mathcal{T}_{a,b}(x)=a\overline{x}b=ab\,\overline{x}$, and since the algebra is commutative $\mathcal{T}_{a,b}=\mathcal{T}_{b,a}$, so every two-sided operator is self-adjoint, in agreement with the corollary. The inverse is $\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{1/\overline{b},\,1/\overline{a}}$, the map $x\mapsto\overline{x}/(\overline{a}\,\overline{b})$, and the unitary condition is $|ab|=1$, where the adjoint and the inverse agree, $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}=\mathcal{T}_{a,b}^{-1}$; the parameters of unit length are the case $|a|=|b|=1$.
The Quaternions
Let $A=\mathbb{H}$ over $R=\mathbb{R}$ with $\varsigma=\mathrm{id}$, $*$ the quaternion conjugation, $\tau$ the real part and $\varphi(u,v)=\operatorname{Re}(u\overline{v})$. The two-sided operator $\mathcal{T}_{a,b}(q)=a\overline{q}b$ is $\mathbb{R}$-linear, so the twisted rule is the linear rule, and $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}$. The centre of $\mathbb{H}$ is $\mathbb{R}$, so the central units are the nonzero reals, the criterion of self-adjointness reads $a^{-1}b\in\mathbb{R}$, and the parameters of unit length give unitary two-sided operators. Since $\mathbb{H}$ is noncommutative the transposition is visible: for $a=i$ and $b=k$ one has $a^{-1}b=-ik=j$, whose imaginary part is not zero, so $\mathcal{T}_{i,k}$ is not self-adjoint.
Summary
The two-sided operator $\mathcal{T}_{a,b}(x)=ax^{*}b$ is the two-sided operator of the standard example, $\varsigma$-semilinear in its variable, and its adjoint for the pairing $\varphi(x,y)=\tau(xy^{*})$ is the two-sided operator with the parameters exchanged,
$$ \mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}={}^{*}\circ L_{a^{*}}R_{b^{*}} . $$
The adjoint is an involution of the family; it transposes the pair, against the two-sided multiplication, whose adjoint $L_{p}R_{q}^{\dagger}=L_{p^{*}}R_{q^{*}}$ conjugates the pair and keeps the order. The self-adjoint two-sided operators with units are the pairs $b=ap$, $p$ a central unit, and every diagonal two-sided operator $\mathcal{T}_{a,a}$ is self-adjoint; on a commutative algebra every invertible two-sided operator is self-adjoint.
The invertible two-sided operators have $\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{(b^{-1})^{*},(a^{-1})^{*}}$, and the two-sided operator is unitary exactly when $\mathcal{T}_{b,a}=\mathcal{T}_{(b^{-1})^{*},(a^{-1})^{*}}$; the composites with the adjoint are $\mathcal{T}_{a,b}^{\dagger}\mathcal{T}_{a,b}=L_{bb^{*}}R_{a^{*}a}$ and $\mathcal{T}_{a,b}\mathcal{T}_{a,b}^{\dagger}=L_{aa^{*}}R_{b^{*}b}$, so the unitary two-sided operators are those with $bb^{*}$ and $aa^{*}$ central units and $a^{*}a=(bb^{*})^{-1}$, $b^{*}b=(aa^{*})^{-1}$, the first condition sufficing when $a,b$ are invertible. In particular a two-sided operator with unitary parameters is a unitary operator, $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{b,a}$, and the inner involution $\mathcal{T}_{u,u^{*}}$ is self-adjoint exactly when $u^{2}$ is central. The family of the two-sided operators is thus closed under the adjoint, with the adjoint acting inside it as the transposition of the parameters.
Summary of Notation
| symbol | meaning |
|---|---|
| $\mathcal{T}_{a,b}(x)=ax^{*}b$ | the sesquilinear two-sided operator, $\varsigma$-semilinear |
| $L_{p}R_{q}(x)=pxq$ | the ordinary two-sided multiplication, $R$-linear |
| $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{b,a}$ | the adjoint of the two-sided operator, the parameters transposed |
| $\mathcal{T}_{a,b}^{\dagger}={}^{*}\circ L_{a^{*}}R_{b^{*}}$ | the adjoint read through the two families |
| $L_{p}R_{q}^{\dagger}=L_{p^{*}}R_{q^{*}}$ | the adjoint of the two-sided multiplication, for contrast |
| $\mathcal{T}_{a,a}^{\dagger}=\mathcal{T}_{a,a}$ | the Hermitian pairs are self-adjoint |
| $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{a,b}\iff b=ap$, $p\in Z(A)^{\times}$ | the criterion of self-adjointness, units $a,b$ |
| $\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{(b^{-1})^{*},(a^{-1})^{*}}$ | the inverse of the two-sided operator |
| $\mathcal{T}_{a,b}^{\dagger}\mathcal{T}_{a,b}=L_{bb^{*}}R_{a^{*}a}$ | the right square, a two-sided multiplication |
| $\mathcal{T}_{a,b}\mathcal{T}_{a,b}^{\dagger}=L_{aa^{*}}R_{b^{*}b}$ | the left square, a two-sided multiplication |
| $a,b$ unitary $\Rightarrow$ $\mathcal{T}_{a,b}^{\dagger}=\mathcal{T}_{a,b}^{-1}=\mathcal{T}_{b,a}$ | the unitary case |
| $\mathcal{T}_{u,u^{*}}=\sigma_{u}$ self-adjoint $\iff$ $u^{2}\in Z(A)$ | the inner involution |
Further Reading
- I. N. Herstein, Rings with Involution (University of Chicago Press, 1976), for the two-sided operators of a ring with an involution, the conjugate-linear maps attached to them, and the adjoint of a two-sided operator.
- Nathan Jacobson, Structure of Rings (American Mathematical Society Colloquium Publications 37, 1956), for the triple products of a ring with an involution and the operators they induce, among which the two-sided operator is the two-parameter family.
- Richard S. Pierce, Associative Algebras, Graduate Texts in Mathematics 88 (Springer, 1982), for the inner automorphisms, the central units and the indeterminacy of the two-sided representation that the criterion of self-adjointness reads.
- Sterling K. Berberian, Baer $\ast$-Rings (Springer, 1972), for the adjoint operation in a ring with an involution and the trace pairing against which it is taken, the setting of the whole article.
- The companion articles of this series: The Sesquilinear Two-Sided Operator, The Sesquilinear Adjoint Operator, The Adjoint of the Conjugate Left Multiplication, The Adjoint of the Ternary Product, The Adjoint of the Two-Sided Multiplication Operator, The Involutions of a Sesqualgebra, Invertibles and the Unitary Elements and The Unitary Operators of a Sesqualgebra.