---
title: Multiplicative Spectrum Preservers
url: https://www.emergentmind.com/topics/multiplicative-spectrum-preservers
type: topic
---

# Multiplicative Spectrum Preservers

Searching arXiv for the cited work on multiplicative spectrum preservers and closely related operator-algebraic classifications.
Multiplicative spectrum preservers are maps determined by identities that compare the spectrum, or the peripheral spectrum, of multiplicative expressions before and after transformation. In the operator-algebraic and matrix-theoretic settings represented here, the defining constraints take forms such as
\[
\sigma\bigl(\varphi(x)\varphi(y)\bigr)=\sigma(xy),\qquad
\sigma_\pi(\Phi(A_1)*\cdots *\Phi(A_k))=\sigma_\pi(A_1*\cdots *A_k),
\]
or analogous equalities for skew products, power-type products, and products of stochastic matrices. A recurrent conclusion is rigidity: despite the weakness of the spectral data being preserved, the map is forced into a narrow class of automorphic, anti-automorphic, Jordan, unitary, anti-unitary, transpose, or permutation-implemented forms [1305.7100] [1309.4032] [2404.04563] [2509.22743].

## 1. Definitions and ambient settings

The literature treats several related notions under the heading of multiplicative spectrum preservation. In standard operator algebras on complex Banach spaces, one fixes a generalized product
\[
T_1 * T_2 * \cdots * T_k := T_{i_1}T_{i_2}\cdots T_{i_m},
\]
where each \(i_j\in\{1,\ldots,k\}\), every index appears at least once, and at least one index appears exactly once; the integer \(m\) is the width of the product. The associated spectral invariant in Zhang–Hou is the peripheral spectrum
\[
\sigma_\pi(A):=\{\lambda\in \sigma(A):|\lambda|=r(A)\},
\]
where \(r(A)\) is the spectral radius. In unital \(C^*\)-algebras, Mori–Oi consider surjective maps between self-adjoint parts \(A_{sa}\to B_{sa}\) satisfying \(\sigma(\varphi(x)\varphi(y))=\sigma(xy)\). In Hilbert-space operator algebras, Taghavi–Hosseinzadeh study surjections preserving the spectrum of products such as \(A^rBA^s\), \(|A|^rB\), and \(|A|^rB|A|^s\). In stochastic-matrix theory, multiplicative spectrum preservers are lists of maps \(\phi_i:\mathcal S\to\mathcal S\) satisfying
\[
\operatorname{spec}\bigl(\phi_1(A_1)\cdots \phi_m(A_m)\bigr)=\operatorname{spec}(A_1\cdots A_m)
\]
for all inputs in \(\mathcal S\) [1305.7100] [1309.4032] [2404.04563] [2509.22743].

| Setting | Preserved quantity | Classified form |
|---|---|---|
| Standard operator algebras on Banach spaces | \(\sigma_\pi\) of generalized products | \(\Phi(A)=\omega TAT^{-1}\) or \(\omega TA^*T^{-1}\) |
| Hilbert spaces, skew generalized products | \(\sigma_\pi\) of skew products | \(\Phi(A)=cUAU^*\) or \(cUA^tU^*\) |
| Self-adjoint parts of unital \(C^*\)-algebras | \(\sigma(\varphi(x)\varphi(y))\) | \(\varphi(x)=s\Theta(x)\) |
| \(B(H)\) and \(S(H)\) | \(\sigma(A^rBA^s)\), \(\sigma(|A|^rB)\), \(\sigma(|A|^rB|A|^s)\) | \(\varphi(A)=TAT^*\) |
| \(DS_n,RS_n,CS_n\) and their spans | spectrum or trace of \(m\)-fold products | permutation or transpose-permutation forms |

These settings use different ambient categories, but they all formulate multiplicativity through a product and impose spectral equality on every admissible product. This suggests that the central problem is not merely linear preserver theory, but the interaction between multiplicative structure and spectral invariants.

## 2. Peripheral-spectrum preservation on standard operator algebras

For standard operator algebras \(\mathcal A_1\subseteq B(X_1)\) and \(\mathcal A_2\subseteq B(X_2)\) on complex Banach spaces, Zhang–Hou prove a classification theorem for maps \(\Phi:\mathcal A_1\to\mathcal A_2\) whose range contains all operators of rank at most two and which satisfy
\[
\sigma_\pi(\Phi(A_1)*\cdots *\Phi(A_k))=\sigma_\pi(A_1*\cdots *A_k)
\]
for every \(k\)-tuple. The theorem states that this condition is equivalent to one of two possibilities. Either there exists an invertible \(T\in B(X_1,X_2)\) and a root of unity \(\omega\in\mathbb C\) with \(\omega^m=1\) such that
\[
\Phi(A)=\omega\,TAT^{-1}\qquad \forall A\in\mathcal A_1,
\]
or \(X_1,X_2\) are reflexive, the index-sequence satisfies the quasi-semi-Jordan symmetry, and
\[
\Phi(A)=\omega\,TA^*T^{-1}\qquad \forall A\in\mathcal A_1.
\]
If the product is not quasi-semi-Jordan, only the first case occurs [1305.7100].

The quasi-semi-Jordan condition is imposed on the fixed index-sequence \((i_1,\ldots,i_m)\). Writing the unique index as \(i_p\), the sequence is generalized quasi-semi-Jordan when deleting \(i_p\) leaves a block that reverses about its center:
\[
(i_{p+1},\ldots,i_m,i_1,\ldots,i_{p-1})
=
(i_{p-1},\ldots,i_1,i_m,\ldots,i_{p+1}).
\]
When this holds, the generalized product is called quasi-semi-Jordan. The role of this symmetry is decisive: it is precisely the condition under which the adjoint-type alternative survives in the classification.

The proof strategy is organized around rank-one operators. A spectral-radius argument yields a rank-one characterization: \(A\) has rank \(1\) if and only if, for every \(B\), the peripheral spectrum \(\sigma_\pi(B^rAB^s)\) is a singleton. From the product-preservation identity, \(\Phi\) sends rank-one operators bijectively onto rank-one operators. A standard rank-one-injectivity argument then forces linearity and gives, on rank-one tensors \(x\otimes f\),
\[
\Phi(x\otimes f)=\omega\,T(x\otimes f)T^{-1}
\quad\text{or}\quad
\Phi(x\otimes f)=\omega\,T(x\otimes f)^*T^{-1}.
\]
Linearity and the rank-one-generated ideals extend this to the full algebra, and comparison on reversed monomials detects the quasi-semi-Jordan symmetry in the adjoint case.

The Hilbert-space refinement treats skew generalized products
\[
T_1\star T_2\star\cdots\star T_k
=
T_{i_1}\cdots T_{i,p-1}T_{i_p}^*T_{i,p+1}\cdots T_{i_m}.
\]
If \(\Phi\) preserves the peripheral spectrum of every such skew product and its range contains all rank-\(\le 2\) operators, then there exist a unitary or conjugate-unitary \(U:H\to K\) and a scalar \(c\in\{\pm 1\}\), with \(c=+1\) when \(m\) is odd, such that either
\[
\Phi(A)=c\,UAU^*
\]
for all \(A\), or, in the skew quasi-semi-Jordan case,
\[
\Phi(A)=c\,UA^tU^*.
\]
Here \(A^t\) is the transpose in some fixed orthonormal basis. The proof passes through rank-one projections and then invokes the Uhlhorn–Wigner theorem after constructing a bijection on unit spheres that preserves \(|\langle x,y\rangle|\).

## 3. Spectrum-preserving maps for power-type operator products

On an infinite-dimensional complex Hilbert space \(H\), Taghavi–Hosseinzadeh study surjective maps on \(S(H)\) and \(B(H)\) preserving the spectrum of specific products. The products include \(A^rBA^s\) with \(r,s>0\) rational and \(r+s>1\), \(|A|^rB\) with \(r>1\) rational, and \(|A|^rB|A|^s\) with \(r,s>0\) rational and \(r+s>1\). The corresponding theorems show that, in each case, a surjective map \(\varphi\) satisfying the prescribed spectral identity must have the form
\[
\varphi(A)=TAT^*
\]
for every \(A\) in the domain, where \(T:H\to H\) is a bounded bijection that is either complex-linear or conjugate-linear and satisfies \(T^*=T^{-1}\) [1309.4032].

For self-adjoint operators, the first theorem assumes
\[
\sigma\bigl(A^rBA^s\bigr)=\sigma\bigl(\varphi(A)^r\varphi(B)\varphi(A)^s\bigr)
\]
for all \(A\in P_1(H)\cup\{I\}\) and all \(B\in S(H)\), and concludes that \(\varphi(A)=TAT^*\) on \(S(H)\). The second theorem replaces \(A^r\) by \(|A|^r\) and reaches the same conclusion on \(S(H)\). For the full algebra \(B(H)\), the third and fourth theorems impose the spectrum-preservation identities for all \(A,B\in B(H)\) and again conclude unitary or anti-unitary implementation.

The paper rephrases the conclusion as the statement that there is a unitary or anti-unitary \(U\) on \(H\) such that
\[
\varphi(A)=UAU^*
\]
for all operators in the domain. In that sense, these power-type multiplicative spectrum preservers are exactly the Jordan-\(*\)-automorphisms of \(B(H)\) or \(S(H)\).

The proofs follow a common pattern. First, one shows \(\varphi(0)=0\), \(\varphi(I)=I\), and preservation of rank-one projections in both directions. Next, functional-calculus arguments and results on rational powers imply preservation of rank-one operators and the trace pairings
\[
\operatorname{tr}(PB)=\operatorname{tr}(\varphi(P)\varphi(B))
\]
for rank-one projections \(P\). These trace relations yield additivity and \(\mathbb R\)-homogeneity on \(S(H)\), and then full complex-linearity on \(B(H)\). A rank-one-preserver theorem then produces the conjugation formula. The operator-algebraic significance is that no a priori linearity or continuity is required beyond the stated spectral condition and surjectivity.

## 4. Self-adjoint parts of unital \(C^*\)-algebras

Mori–Oi consider surjective maps
\[
\varphi:A_{sa}\longrightarrow B_{sa}
\]
between the self-adjoint parts of unital \(C^*\)-algebras satisfying
\[
\sigma\bigl(\varphi(x)\varphi(y)\bigr)=\sigma(xy)\qquad \text{for all }x,y\in A_{sa}.
\]
Their main theorem states that such a map is characterized by a central symmetry and a Jordan \(*\)-isomorphism: there exist a central symmetry \(s\in B_{sa}\), meaning a self-adjoint unitary in the center of \(B\), and a Jordan \(*\)-isomorphism \(\Theta:A\to B\) such that
\[
\varphi(x)=s\,\Theta(x),\qquad x\in A_{sa}.
\]
Equivalently, \(\varphi\) factors as \(s\circ \Theta|_{A_{sa}}\) [2404.04563].

Two intermediate lemmas organize the proof. The real-spectrum test says that for \(x\in A_{sa}\), the condition \(\sigma(xy)\subset\mathbb R\) for every \(y\in A_{sa}\) is equivalent to
\[
x_+\,y\,x_-=0\qquad \text{for every }y\in A.
\]
The uniqueness lemma says that if \(a,b\in A_{sa}\) satisfy
\[
\sigma(ay)=\sigma(by)\qquad\text{for every invertible positive }y\in A_+,
\]
then \(a=b\). The proof also uses Jacobson’s lemma, square-root identities for positive elements, and an inductive spectral-inclusion argument.

The normalization step sets \(s=\varphi(1)\). Since \(\sigma(s^2)=\sigma(1)=\{1\}\), one gets \(\sigma(s)\subset\{\pm 1\}\), so \(s\) is a self-adjoint unitary. Surjectivity and the real-spectrum lemma then force \(s\) to be central. Defining
\[
T(x)=s\,\varphi(x)
\]
produces a unital surjection preserving the multiplicative spectrum on \(A_{sa}\). On the cone \(A_+^{-1}\) of invertible positive elements, the spectral condition shows that \(T\) is an isometry for the Thompson metric
\[
d_T(u,v)=\|\log(u^{-1/2}vu^{-1/2})\|.
\]
A further application of the uniqueness lemma yields positive homogeneity: \(T(tx)=tT(x)\) for all \(t>0\). Hatori–Molnár’s theorem on Thompson-isometric positive-homogeneous surjections then gives a Jordan \(*\)-isomorphism \(\Theta\) extending \(T\).

In matrix algebras \(A=B=M_n(\mathbb C)\), the theorem specializes to four possibilities because every Jordan \(*\)-automorphism of \(M_n(\mathbb C)\) is either \(a\mapsto uau^*\) or \(a\mapsto ua^Tu^*\), while every central symmetry is \(\pm I_n\). Hence every surjective multiplicatively spectrum-preserving map on \(M_n(\mathbb C)_{sa}\) is of one of the four forms
\[
\varphi(a)=\pm uau^*,\qquad \varphi(a)=\pm ua^Tu^*.
\]

## 5. Stochastic matrices and their linear spans

For stochastic matrices, the multiplicative spectrum-preserving problem is formulated on \(\mathcal S\) equal to the set of \(n\times n\) doubly stochastic matrices \(DS_n\), row stochastic matrices \(RS_n\), column stochastic matrices \(CS_n\), or the real linear span of one of these sets. A list of maps
\[
\phi_i:\mathcal S\to\mathcal S,\qquad i=1,\ldots,m,
\]
is a multiplicative spectrum preserver if
\[
\operatorname{spec}\bigl(\phi_1(A_1)\phi_2(A_2)\cdots \phi_m(A_m)\bigr)
=
\operatorname{spec}(A_1A_2\cdots A_m)
\]
for all \(A_1,\ldots,A_m\in\mathcal S\). When \(m=1\), linearity is assumed. The same paper also treats multiplicative trace preservers and proves that when \(m\ge 3\), multiplicative trace preservers always coincide with multiplicative spectrum preservers [2509.22743].

A structural decomposition underlies the classification. With \(e\) the all-ones column and
\[
R_n=\{ev^T:v^Te=0\},\qquad C_n=\{ve^T:v^Te=0\},
\]
one has
\[
M_n(F)=\langle DS_n\rangle\oplus R_n\oplus C_n,\qquad
\langle RS_n\rangle=\langle DS_n\rangle\oplus R_n,\qquad
\langle CS_n\rangle=\langle DS_n\rangle\oplus C_n.
\]
Thus every \(A\in M_n(F)\) decomposes uniquely as
\[
A=A_{DS}+A_R+A_C.
\]
Moreover, a suitable orthogonal similarity shows \(\operatorname{spec}(A)=\operatorname{spec}(A_{DS})\), and if \(A_i\in \langle RS_n\rangle\) or \(\langle CS_n\rangle\), then
\[
(A_1\cdots A_m)_{DS}=(A_1)_{DS}\cdots (A_m)_{DS}.
\]
Hence both spectrum and trace of products depend only on the \(DS\)-parts.

On \(DS_n\), the one-map case is classical in form: a linear \(\phi:DS_n\to DS_n\) satisfies \(\operatorname{spec}(\phi(A))=\operatorname{spec}(A)\) for all \(A\) if and only if there exists \(P\in P_n\) such that
\[
\phi(A)=PAP^T
\quad\text{or}\quad
\phi(A)=PA^TP^T.
\]
For two maps \(\phi_1,\phi_2:DS_n\to DS_n\), preservation of \(\operatorname{tr}(\phi_1(A)\phi_2(B))=\operatorname{tr}(AB)\), equivalently \(\operatorname{spec}(\phi_1(A)\phi_2(B))=\operatorname{spec}(AB)\), holds if and only if there are \(P,Q\in P_n\) with
\[
\phi_1(A)=PAQ^T,\qquad \phi_2(A)=QAP^T,
\]
or the transpose version
\[
\phi_1(A)=PA^TQ^T,\qquad \phi_2(A)=QA^TP^T.
\]
For \(m\ge 3\), the classification becomes cyclic: there exist \(P_1,\ldots,P_m\in P_n\) with \(P_{m+1}=P_1\) such that
\[
\phi_i(A)=P_iAP_{i+1}^T,\qquad i=1,\ldots,m.
\]

Exactly the same pattern holds on the span \(\langle DS_n\rangle\), except that the conjugating matrices \(P_i\) need only be invertible in the algebra \(\langle DS_n\rangle\). For \(m=1\) and \(n\ge 3\),
\[
\phi(A)=PAP^{-1}\quad\text{or}\quad \phi(A)=PA^TP^{-1}.
\]
For \(m=2\),
\[
\phi_1(A)=PAQ^{-1},\quad \phi_2(A)=QAP^{-1},
\]
or the transpose version; in the special \(n=2\) case one must further insert a central invertible \(C\) on one side. For \(m\ge 3\),
\[
\phi_i(A)=P_iAP_{i+1}^{-1},\qquad P_{m+1}=P_1.
\]

The \(RS_n\) and \(CS_n\) classifications are obtained by passing down from the \(DS_n\)-part. On \(\langle RS_n\rangle\), a linear one-map spectrum preserver agrees with a \(\langle DS_n\rangle\)-spectral automorphism on the \(DS\)-summand, while it is arbitrary on the \(R_n\)-summand. For two maps, the \(DS\)-parts satisfy the two-map classification and the \(R\)-parts are arbitrary maps into \(R_n\). On \(RS_n\) itself, a linear spectrum preserver is, for \(n\ge 3\), inner permutation-conjugation \(A\mapsto PAP^{-1}\); for \(n=2\) it has the form
\[
\phi(A)=A_{DS}+cA_R,\qquad c\in[-1,1].
\]
For \(m\ge 2\), the classification on \(RS_n\) is
\[
\phi_i(A)=P_iA_{DS}P_{i+1}^T+\gamma_i(A),
\]
where \(P_i\in P_n\), \(P_{m+1}=P_1\), and \(\gamma_i:RS_n\to R_n\) is any map such that the sum stays in \(RS_n\). If one \(\phi_\ell\) is linear, then \(\gamma_\ell\) is forced to be scalar on \(R_n\), and for \(n\ge 3\) one obtains the rigid form \(\phi_\ell(A)=P_\ell AP_{\ell+1}^T\).

## 6. Structural themes, proof mechanisms, and limitations

Across these settings, the classification theorems depend on a small number of recurring mechanisms. In the standard operator-algebra and Hilbert-space product problems, rank-one operators and rank-one projections are the decisive test objects; spectral preservation on carefully chosen products forces preservation of rank-one structure, after which linearity and inner implementation follow [1305.7100] [1309.4032]. In the \(C^*\)-algebraic setting, the key tools are spectral characterization on the self-adjoint part, normalization at the unit, the Thompson metric on invertible positives, and a passage from metric isometries to Jordan \(*\)-isomorphisms [2404.04563]. In the stochastic-matrix setting, the essential device is the decomposition into \(DS\)-, \(R\)-, and \(C\)-parts, combined with trace extremality of permutation matrices, the nondegenerate bilinear form \(\operatorname{tr}(AB)\), and Birkhoff’s theorem [2509.22743].

Several limitations are explicit in the classifications. Adjoint-type or transpose-type alternatives occur only under symmetry hypotheses: the Banach-space anti-isomorphism case requires quasi-semi-Jordan symmetry, and the skew generalized product transpose case occurs only in the skew quasi-semi-Jordan case [1305.7100]. In stochastic matrices, the transpose map \(A\mapsto A^T\) on \(DS_n\) cannot extend to a two-map preserver on \(RS_n\), which illustrates that the direct-sum freedom in \(R_n\) is essential but constrained by the \(RS_n\)-nonnegativity requirement [2509.22743]. In the operator-product theorems on \(S(H)\) and \(B(H)\), surjectivity is part of every theorem statement, and the test sets may be restricted to \(P_1(H)\cup\{I\}\) in the self-adjoint cases [1309.4032]. In the peripheral-spectrum theorem, the range assumption that \(\Phi\) contain all rank-\(\le 2\) operators is likewise built into the result [1305.7100].

Taken together, these results show that multiplicative spectrum preservation is not merely a weak spectral compatibility condition. Preserving only the peripheral spectrum of generalized products, preserving the full spectrum of self-adjoint products, or preserving product spectra on stochastic matrix sets all lead to forms dictated by the ambient algebraic symmetry. A plausible implication is that multiplicative spectral data often encode enough rank, order, or permutation structure to recover the underlying automorphism class, even when linearity is not assumed a priori.

Source: https://www.emergentmind.com/topics/multiplicative-spectrum-preservers