Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Dan Popescu asked me the following question, and since I'm not an expert I'm throwing his question on MO.

Suppose that $A$ is a finite-dimensional vector space over an ordered field $k$ with $\operatorname{char}(k) = 0$, equipped with:

  • a commutative associative $k$-linear multiplication $\circ\\,$;
  • a positive-definite inner product $\langle \cdot , \cdot \rangle \\,$.

For each $a \in A$, let $L_a \colon A \to A \colon x \mapsto a \circ x$. This is a linear operator on $A$; consider the adjoint operator w.r.t. the given inner product:

$$\langle L_a x, y \rangle = \langle x, L_a^* y \rangle$$ for all $x,y \in A$. Now write $y \star a := L_a^* y$.

Popescu's question is the following:

Is there a name for this operation $\star$? Has it been studied in the literature? Is there anything known about algebraic properties or identities involving $\star$ (possibly also involving the other data $\circ$ and $\langle \cdot , \cdot \rangle$)?

share|improve this question

3 Answers 3

up vote 8 down vote accepted

In case $k=\mathbb{C}$, what you're describing is a finite-dimensional H*-algebra. More generally, these are Banach algebras, whose carrier space is a Hilbert space, satisfying the adjoint property you mention.

It is natural to make a nondegeneracy assumption: $A$ is called proper when $\forall a \in A\\,.\\, a \circ A = 0 \Rightarrow a = 0$. This turns out to be equivalent to the adjoint $L_a^\ast$ being unique, or, in other words, $\ast$ being an involution. Every H*-algebra is a direct sum of a proper one and an algebra in which $a \circ b=0$ for all $a$ and $b$.

There is a neat structure theorem by Warren Ambrose (see http://www.jstor.org/stable/1990182), showing that proper H*-algebras are always direct sums of full matrix algebras. In particular, commutative H*-algebras are direct sums of 1-dimensional algebras, and hence correspond precisely to orthogonal basis of their carrier space!

I don't know about other fields $k$, but Ambrose's proof basically comes down to carefully analyzing idempotents, which is feasible to repeat for other fields $k$.

share|improve this answer
In $H^*$-algebras, the adjoint of a multiplication operator $L_a$ is always itself a multiplication operator. Is this automatic in the setting that I described, or is this an additional assumption? –  Tom De Medts Aug 25 '11 at 14:47
(As pointed out by Dima Shlyakhtenko in his answer ---which is in fact a response to this comment--- this is indeed an additional assumption.) –  Tom De Medts Aug 26 '11 at 9:02
I'll accept this answer; it's not exactly what I had in mind, but it's definitely very useful --- thanks! –  Tom De Medts Aug 29 '11 at 7:00

(This is too long for a comment, my apologies for posting this as an answer).

I don't think it is in general true that $L_a^*$ is a multiplication operator. For example, let $S$ be an invertible $n\times n$ matrix, and consider $$A=\{ SDS^{-1} : D \textrm{ diagonal}\}$$ as an abelian subalgebra of $n\times n$ matrices $M_{n\times n}$. Put an inner product on $M_{n\times n}$ by $\langle X,Y\rangle = Tr(XY^*)$, and endow $A$ with its restriction. Finally, let $P : M_{n\times n}\to A$ be the orthogonal projection onto $A$.

For a matrix $X\in M_{n\times n}$ denote by $\mathscr{L}_X$ the operator $$\mathscr{L}_X : M_{n\times n} \ni Y \mapsto XY \in M_{n\times n}.$$ It is not hard to see that $\mathscr{L}_X^* = \mathscr{L}_{X^*}$.

Then for $a = S D S^{-1} \in A$ your $L_a$ is given by $L_a = P \mathscr{L}_{SDS^{-1}} P $ and so its adjoint (as a linear map from $A$ to $A$) is given by $$P \mathscr{L}_{ (S^{-1})^* D S^*} P$$ and its value on some $b = SD' S^{-1}$ is given by $P( (S^{-1})^*DS^* S D'S^{-1})$ so that $L_a^*$ is no longer a multiplication operator.

It looks like, at least for abelian subalgebras of $M_{n\times n}(\mathbb{C})$ with the given inner product,the necessary and sufficient condition for $L_a^*$ to belong to $L_A$ is that $A$ be closed under the adjoint operation, i.e. be a von Neumann algebra.

share|improve this answer
Thanks for this explicit counterexample; it was indeed my intuition that this condition would not be automatically fulfilled. That brings me back to the original question... –  Tom De Medts Aug 26 '11 at 7:58
Nice example! The last paragraph confuses me, though. All finite-dimensional von Neumann algebras are direct sums of full matrix algebras, and hence must be H*-algebras. So that cannot be the characterizing condition for each $L_a^*$ to be a multiplication operator again. But the norm induced by the trace inner product does not satisfy the C*-condition $\|X\|^2=\|X^*X\|$, so at least there is no contradiction. –  Chris Heunen Aug 26 '11 at 10:14
@Chris: what I meant was that you are embedding $A$ into linear operators $End(A)$ on the vector space $A$ and that you have (by the choice of your inner product on $A$) chosen a $*$-operation on $End(A)$; together with the operator norm $\Vert T\Vert = \sup_{\langle xi,xi\rangle=1}\langle A\xi ,A\xi\rangle$ and this $*$-operation, $End(V)$ is a von Neumann algebra. To require that $L_a^* \in L_A$ you end up requiring that the image of $A$ (under $a\mapsto L_a$) is closed under the $*$-operation on $End(V)$, i.e. be a von Neumann subalgebra of $End(V)$. –  Dima Shlyakhtenko Aug 27 '11 at 2:43

This operation has been extensively studied for the needs of (weighted) automata theory.

In the case of the free algebra (this is a non-commutative version of the question of the MO) Here
pp 8 (end) and 9 (18 and 19 of the .pdf) and in the case of the shuffle algebra (which, now, is commutative).

Note that, if the adjoints $L_a^{\star}$ are computed within the algebraic dual, the linear forms which have a finite dimensional orbit by the collection $(L_a^{\star})_{a\in A}$ is exactly the Sweedler's dual of the algebra $A$ (it is therefore a coalgebra) Sweedler's duals and Schützenberger's calculus

If you are interested do not hesitate to interact.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.