On the relation between states and maps in infinite dimensions
Janusz Grabowski, Marek Kus, Giuseppe Marmo

TL;DR
This paper extends the fundamental relations between states and maps from finite to infinite-dimensional quantum systems using geometrical and tensor product frameworks, proving new theorems and exploring entanglement measures.
Contribution
It provides a rigorous geometric formulation of state-map relations in infinite dimensions, generalizes the Choi Theorem, and analyzes the limitations of isomorphisms for operators beyond trace-class.
Findings
Generalization of state-map relations to infinite-dimensional Hilbert spaces
Proof of a generalized Choi Theorem in infinite dimensions
Identification of limitations for extending isomorphisms to certain operator classes
Abstract
Relations between states and maps, which are known for quantum systems in finite-dimensional Hilbert spaces, are formulated rigorously in geometrical terms with no use of coordinate (matrix) interpretation. In a tensor product realization they are represented simply by a permutation of factors. This leads to natural generalizations for infinite-dimensional Hilbert spaces and a simple proof of a generalized Choi Theorem. The natural framework is based on spaces of Hilbert-Schmidt operators and the corresponding tensor products of Hilbert spaces. It is proved that the corresponding isomorphisms cannot be naturally extended to compact (or bounded) operators, nor reduced to the trace-class operators. On the other hand, it is proven that there is a natural continuous map…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsQuantum Mechanics and Applications · Quantum Mechanics and Non-Hermitian Physics · Spectral Theory in Mathematical Physics
On the relation between states and maps
in infinite dimensions
Janusz Grabowski111email: [email protected]
Polish Academy of Sciences, Institute of Mathematics,
- Śniadeckich 8, P.O. Box 21, 00-956 Warsaw, Poland
*Marek Kuś222email: [email protected]
Center for Theoretical Physics, Polish Academy of Sciences,
*Aleja Lotników 32/46, 02-668 Warszawa, Poland
*Giuseppe Marmo333email: [email protected]
Dipartimento di Scienze Fisiche, Università “Federico II” di Napoli
and Istituto Nazionale di Fisica Nucleare, Sezione di Napoli,
Complesso Universitario di Monte Sant Angelo,
Via Cintia, I-80126 Napoli, Italy
Abstract
Relations between states and maps, which are known for quantum systems in finite-dimensional Hilbert spaces, are formulated rigorously in geometrical terms with no use of coordinate (matrix) interpretation. In a tensor product realization they are represented simply by a permutation of factors. This leads to natural generalizations for infinite-dimensional Hilbert spaces and a simple proof of a generalized Choi Theorem. The natural framework is based on spaces of Hilbert-Schmidt operators and the corresponding tensor products of Hilbert spaces. It is proved that the corresponding isomorphisms cannot be naturally extended to compact (or bounded) operators, nor reduced to the trace-class operators. On the other hand, it is proven that there is a natural continuous map from trace-class operators on (with the nuclear norm) into compact operators mapping the space of all bounded operators on into trace class operators on (with the operator-norm). Also in the infinite-dimensional context, the Schmidt measure of entanglement and multipartite generalizations of state-maps relations are considered in the paper.
1 Introduction
Entanglement is one of the most counterintuitive features of quantum mechanics. Already in the early years of the birth of quantum theory, Erwin Schrödinger realized that this aspect is a consequence of the mathematical structure of the theory [1]. It is a characteristic property of quantum mechanics, not present in other physical theories described by linear equations like, for instance, classical electrodynamics. Entanglement was considered with embarrassment in connection with non locality, as pointed out by the Einstein-Podolsky-Rosen gedanken experiment [2]. Its role was clarified with the discovery of Bell’s inequalities [3, 4]. It was shown that these inequalities can be violated in quantum mechanics but have to be satisfied by all local realistic theories. The violation of the inequalities demonstrates the presence of entanglement. In the near past it was realized that entanglement could be a great resource for quantum information theory. The promising applications of this peculiar quantum property has induced intensive experimental efforts to build entangled quantum states and major theoretical efforts to understand the mathematical structure of entanglement [5]. To put the present paper into perspective, let us briefly consider how the problem arises.
In the Dirac-Schrödinger picture of quantum mechanics, one identifies the carrier space of quantum evolution with a complex separable Hilbert space . The probabilistic interpretation of quantum mechanics requires that states be identified with rays, points of the complex projective space of . By using the Hermitian inner product one defines an action of the unitary group with an associated momentum map [6, 7, 8]. This map relates rays with rank-one projectors, i.e. operators, elements of the dual vector space of the Lie algebra of the unitary group. Thanks to this immersion, it becomes possible to consider convex combinations of elements in the image of this map and therefore to construct density states, also called density operators. In this way observables and states are represented both by means of operators, even though with qualifying different properties to take into account their corresponding physical interpretations. A similar situation results in the -algebraic approach to quantum mechanics, originated by Heisenberg and developed by Segal and Haag [9, 10]. Here one considers states as nonnegative normalized linear functionals on the space of observables, real elements of the -algebra and associates with them density states by means of Gleason’s theorem [11].In either approach states are identified with appropriate operators.
Composite systems are mathematically formed as tensor products of the Hilbert spaces associated with the system we are composing, called subsystems. Similarly within the -algebra approach, the consideration of states as maps has boosted a search for various procedures to characterize separability and entanglement of states by exploiting as much as possible what is available for the classification of maps [12, 13, 14]. The difficulties in a straightforward application of known classification procedures rely on the fact that the very definition of states as convex combination of rank-one projectors provides them with a positivity property which is not preserved under tensorial products, in general the product of positive maps does not result into a positive one. While the existing literature is concerned with the relation between maps and states restricted to finite dimensional Hilbert spaces or -algebras, the aim of this paper is to present a careful analysis of these various relations between states and maps for composite quantum systems in the more realistic situation of infinite dimensions.
Relations between states and maps are well known for systems in finite-dimensional Hilbert spaces. In the second section of the paper we reformulate them without invoking any particular matrix realization of the states. This allows us to generalize in the following sections the known results to infinite-dimensional Hilbert spaces and Hilbert-Schmidt operators acting as maps between them. As a result we can describe in the infinite-dimensional setting connections between positivity and complete positivity of maps and separability properties of the corresponding states on the composite spaces proven by Jamiołkowski and Choi for the finite-dimensional case. We discuss briefly generalization to multipartite systems and show that the infinite-dimensional Jamiołkowski isomorphism can be neither sensibly extended to the larger class of bounded operators nor reduced to a smaller set of the trace-class operators.
2 The Jamiołkowski isomorphism
Let and be two Hilbert spaces. In a finite dimensional case, , the Jamiołkowski isomorphism [15, 16] is a mapping:
[TABLE]
where by we denote the space of all complex linear maps from to , with an abbreviation for the space of all linear endomorphisms of .
We prefer to define objects in a basis-independent way, so we prefer to speak about and instead of and complex matrices. Therefore we start with the following observations. First, there is a natural anti-isomorphism between and its dual complex vector space,
[TABLE]
induced by the scalar product on , which in the Dirac notation reads
[TABLE]
The dual space is canonically a Hilbert space with the Hermitian product , where the latter is the scalar product on , which shows that is an anti-unitary. In the following we usually skip the subscripts specifying the Hermitian products in various spaces, if this does not lead to a confusion.
We clearly have up to an obvious identification . Moreover, the anti-isomorphism (2) induces an anti-isomorphism
[TABLE]
of the corresponding spaces of complex linear operators, where . Here, clearly, the adjoint operator is defined by and is the dual map. By definition, (2) intertwines with , i.e. . The notation is consistent, because (up to an obvious identification ) for the Hermitian product on .
The point here is that respects the composition, , while . This means that, restricting ourselves to the groups of invertible complex operators on the Hilbert spaces, we have a canonical group isomorphism , while is an anti-isomorphism. This group isomorphism restricts to an isomorphism of the unitary groups , as in this case
[TABLE]
Note that in the physics literature one usually identifies with by fixing an orthonormal basis in and putting
[TABLE]
It is a true isomorphism which, however, depends on the choice of the basis, and not the canonical anti-isomorphism we speak about.
We will use the following canonical identification of Hilbert spaces,
[TABLE]
Under this identification for and . Moreover, the Hilbert-Schmidt scalar product on coincides with the standard scalar product on . Here, the adjoint operator
[TABLE]
is defined in an obvious way by (or, in the tensor product realization, ). Indeed, for being an arbitrary orthonormal basis in ,
[TABLE]
We have a canonical -action on :
[TABLE]
which in the tensor product realization takes the form
[TABLE]
This action can be reduced to an -action which is unitary, as in this case
[TABLE]
If , then one can reduce the above action to a diagonal action of (or ): .
The canonical isomorphism gives rise to an identification
[TABLE]
Moreover,
[TABLE]
with the obvious pairing
[TABLE]
In particular,
[TABLE]
Note that we have further natural identifications
[TABLE]
and
[TABLE]
Definition 1**.**
The Jamiołkowski isomorphism (1), up to above natural identifications, is defined as a natural transposition in the tensor products
[TABLE]
consisting of interchanging of the second and fourth factors, i.e.
[TABLE]
The twisted Jamiołkowski isomorphism
[TABLE]
comes in a similar way from the permutation
[TABLE]
As Jamiołkowski isomorphisms are simply permutations in the tensor product, they are automatically unitary. Moreover, it is completely clear that Jamiołkowski isomorphisms intertwine the canonical actions of the group on the tensor product :
[TABLE]
and its corresponding permutations, so the following is immediate.
Theorem 1**.**
The Jamiołkowski isomorphisms are unitary and intertwine the canonical -actions.
3 Infinite dimensions
The above definitions can be extended to infinite-dimensional Hilbert spaces and as follows. In this case we define the Hilbert-Schmidt tensor product as the closure of the algebraic tensor product with respect to the scalar product which on simple tensors reads . In this way, elements of represent Hilbert-Schmidt operators from into and can be viewed as infinite combinations , where and are orthonormal bases in and , respectively, and is the squared Hilbert-Schmidt norm of (it does not depend on the choice of bases). It is well-known [17] that every compact operator admits the so called Schmidt decomposition, , with and being (not necessarily complete) orthonormal sets, and as . The Hilbert-Schmidt norm can be equivalently defined as . In fact, the coefficients can be chosen positive. In the following we will denote simply as . Since the Hermitian conjugation is also a transposition of the tensor product, is Hilbert-Schmidt if is.
Now, the (Hillbert-Schmidt) tensor product represents the space of the Hilbert-Schmidt operators, i.e. the Hilbert space of those complex linear maps from to such that , for some (thus all) orthonormal basis in , and with the Hermitian form
[TABLE]
Note that the trace is well-defined, since any composition of Hilbert-Schmidt operators is known to be a trace-class operator [17].
We will abbreviate to for an arbitrary Hilbert space . The symbol is now reserved for all bounded complex linear maps, so that is the space of all bounded operators from to with the operator-norm topology. The latter makes sense also in the Banach category.
In complete analogy with the finite-dimensional case we have also natural canonical identifications
[TABLE]
and
[TABLE]
Thus, the Jamiołkowski isomorphism, defined on the level of tensor products by the same transposition (13), is now interpreted as
[TABLE]
and the twisted Jamiołkowski isomorphism, as
[TABLE]
Both isomorphisms are clearly unitary. Moreover, since the Hilbert-Schmidt operators form an operator ideal, the Hilbert-Schmidt tensor products are invariant with respect to the canonical -actions and the Jamiołkowski isomorphisms intertwines these actions.
Note also that the original definitions are obviously equivalent to the following properties of the Jamiołkowski isomorphisms.
Proposition 1**.**
The Jamiołkowski unitary isomorphisms (18) and (19) can be uniquely characterized, respectively, by the identities
[TABLE]
and
[TABLE]
which must be fulfilled for all , . Equivalent formulations of the above identities are, respectively,
[TABLE]
and
[TABLE]
Example 1**.**
For , orthonormal bases and in and , and for
[TABLE]
let us consider the map
[TABLE]
i.e.
[TABLE]
Hence, the Jamiołkowski isomorphism applied to is a map represented by
[TABLE]
i.e. is just the one-dimensional operator . In particular, if , the operator is just the standard Kraus map and its Jamiołkowski image is the Hermitian ”projection” on the vector . This is a true projection if the length of is 1. The fact that we deal with a unitary isomorphism implies easily that there is an orthonormal basis in consisting of operators of the form for a basis in .
Since is unitary, the Hilbert-Schmidt norm of equals the Hilbert-Schmidt norm of this projection, i.e. . Recall that the space of Hilbert-Schmidt operators is an operator ideal in the space of all bounded operators as , where is the operator-norm. Since , the operator-norm of is not bigger than the square of the operator-norm of , i.e. . But the operator-norm of the projection is still . Since we can easily find a sequence with such that , this shows that the Jamiołkowski isomorphism is not continuous in the operator-norm topology. In other words, does not admit a natural extension to a map
[TABLE]
where denotes the space of compact operators – the operator-norm closure of the space of all Hilbert-Schmidt operators. One can think that the above suggests that the inverse Jamiołkowski isomorphism is continuous in the norm topology, as
[TABLE]
But this is also not true, since is 1 for , , and the operator-norm of the corresponding Kraus map
[TABLE]
is at lest . Indeed, the projection on the subspace spanned by has the Hilbert-Schmidt norm , while its image
[TABLE]
has the Hilbert-Schmidt norm .**
Any composition of Hilbert-Schmidt operators is well known to be a trace-class operator, called also nuclear operator (see e.g. [18, Chapter VII]). The space of nuclear operators consists of operators admitting a decomposition into one-dimensional operators: (in they are represented as tensors that can be written in the form ) with . They can be equivalently described as these operators for which is Hilbert-Schmidt on . The nuclear norm can be defined as \|T\|_{1}=\sum_{\mathchar 28939\relax}{\|Tf_{\mathchar 28939\relax}\|}=\mbox{\text{Tr}}\left(\sqrt{TT^{\dagger}}\right), as for any Schmidt decomposition with and being (not necessarily complete) orthonormal sets, or as the infimum of for all possible realizations. The latter has the advantage that it applies also in the Banach space context.
Example 2**.**
For an orthonormal base in , a vector of length 1, and for a sequence of complex numbers , the operator ,
[TABLE]
is nuclear with the nuclear norm . Its Jamiołkowski image is Hermitian
[TABLE]
with eigenvalues , so . Since there are sequences with infinite -norm, the Jamiołkowski isomorphism does not map nuclear operators into nuclear ones. A similar fact can be proved for the inverse Jamiołkowski isomorphism.**
Let us summarize the conclusions of the above examples in the following proposition.
Proposition 2**.**
The Jamiołkowski isomorphism (18) and its inverse cannot be extended to all compact operators nor restricted to nuclear (trace-class) operators.
Despite of the above negative result, it is obvious that the map , associated with , can be viewed as a map .
Theorem 2**.**
There is a unique continuous map
[TABLE]
such that for all .
Proof.- Let us start with computing the operator-norm of . First of all, for an orthonormal basis in , we have , so
[TABLE]
Hence,
[TABLE]
But, as easily seen, , so and . Moreover, if is an orthonormal basis in , then
[TABLE]
which shows that is bounded (continuous) with the operator-norm . One can easily see that this norm is actually 1. Let us see that the operators are compact. Indeed,
[TABLE]
But the operator
[TABLE]
is finite-dimensional and
[TABLE]
has the operator-norm
[TABLE]
which is arbitrary small for large , so is an operator-norm limit of finite-dimensional operators. Further, is clearly compact for of length 1 if .
The operator we will call the Choi map.
The trace-class operators appear in Quantum Mechanics as quantum states. The convex set of quantum states consists of trace-class non-negative Hermitian operators with trace 1. It follows from the spectral theorem that each quantum state can be written in a form , where is a sequence of one-dimensional orthogonal projectors, with for , and , . In other words, is the smallest convex set in closed in the nuclear topology which contains all pure states – one-dimensional orthogonal projectors. The Choi map associates with quantum states , with , on a Kraus maps from with the operator sum representation .
If , a quantum state we call separable if one can find a decomposition as above but with being simple tensors, , where and are one-dimensional orthogonal projectors in, respectively, and . Composite quantum states we call entangled if they are not separable. Replacing with we can speak, in an obvious sense, about separable and entangled quantum states on .
4 Basic features of
To proceed we shall need some further observations. First, let us see that maps from a linear subspace of , closed with respect to the Hermitian conjugation, into , which preserve hermiticity, commute with the operation of taking the adjoint. Indeed, assume that is a linear map that maps Hermitian operators into Hermitian ones, and define, for an arbitrary ,
[TABLE]
Clearly, is additive, , and antilinear, . Now,
[TABLE]
where we use the hermiticity of and the assumed property that maps Hermitian operators into Hermitian ones. Since was arbitrary, we can take instead of , hence
[TABLE]
which, upon additivity of and together with (30) gives . Thus we get the following proposition.
Proposition 3**.**
If is a linear subspace, closed with respect to Hermitian conjugation, and is a linear map that maps Hermitian operators into Hermitian ones, then commutes with Hermitian conjugation, .
Theorem 3**.**
A Hilbert-Schmidt operator preserves hermiticity if and only if is Hermitian.
Proof.- According to the above proposition, preserving hermiticity means commuting with the Hermitian conjugation. Since, fixing orthonormal bases and in and , respectively, , etc., commutes with the Hermitian conjugation if and only if . On the other hand, is Hermitian if and only if
[TABLE]
i.e., as above, if and only if .
We say that as above preserves positivity (this property is usually called also positivity that might be confused with positivity of a Hermitian operator), if it maps non-negatively defined Hermitian operators on (we will call them simply positive) into positive ones on . Using (20) we can prove now the following.
Theorem 4**.**
A Hilbert-Schmidt operator preserves positivity if and only if is a Hermitian operator on which is non-negatively defined on separable states, i.e. \mbox{\text{Tr}}({\cal J}(\Phi)\mathchar 28954\relax)\geq 0 for separable states on .
Proof.- We have to prove that preserves the positivity if and only if, for arbitrary , ,
[TABLE]
Indeed, assume that (31) holds. Then from (20)
[TABLE]
Hence, for each projection its Hilbert-Schmidt scalar product with evaluated on arbitrary is positive, so is positive for all and then positivity of for arbitrary positive-definite follows from the spectral decomposition of .
On the other hand, if preserves positivity, then evaluated on a positive operator it gives a positive operator for which the Hilbert-Schmidt scalar product with an arbitrary projection is non-negative, hence (32) and a fortiori (31) hold.
A natural question now is: what Hilbert-Schmidt operators correspond, via the Jamiołkowski isomorphism, to Hermitian operators which are positive on the whole .
Definition 2**.**
A Hilbert-Schmidt operator we call completely positive, if is Hermitian positive on .
We will show now that the above natural definition is equivalent to the standard concepts of complete positivity. Note however that we cannot consider tensor products with the identity on an infinite-dimensional Hilbert space, as the latter is not a Hilbert-Schmidt operator. Therefore, for an auxiliary Hilbert space with an orthonormal basis , consider the Hilbert-Schmidt operator on associated with a diagonal Hilbert-Schmidt matrix , . In other words,
[TABLE]
We know that is Hermitian positive,
[TABLE]
For , we can consider with , , with its Jamiołkowski image .
Take now arbitrarily chosen , . We have
[TABLE]
Indeed, for
[TABLE]
one can write
[TABLE]
Note that any vector in , thus any map in , can be approximated by vectors of the form . Similarly, vectors from and can be approximated by vectors of the form and as above. If , , are finite-dimensional, then we can actually get all these vectors taking the number of not exceeding the maximum of these dimensions. Since, according to the formula (35), if and only if , we can derive the following characterization of complete positivity that can be viewed as an infinite-dimensional version of Choi Theorem [19], cf. also [20].
Theorem 5**.**
Let \Phi\in\mathcal{L}_{2}\big{(}\mathcal{L}_{2}(\mathcal{H}_{2}),\mathcal{L}_{2}(\mathcal{H}_{1})\big{)}. The following are equivalent:
(a) is Hermitian positive;
(b) For any finite-dimensional Hilbert space the operator preserves positivity;
(c) For an infinite-dimensional Hilbert space and for a Hermitian positive with trivial kernel, the operator preserves positivity.
If , , are finite-dimensional, then the dimensions of above auxiliary Hilbert spaces can be restricted to the maximum of the dimensions of and .
Of course, all the above has the corresponding counterpart for the other Jamiołkowski isomorphism . This version fits sometimes better to the language of bi-partite systems.
5 Schmidt rank and Schmidt measure
We know already that any element admits a Schmidt decomposition with and being (not necessarily complete) orthonormal sets, and being positive. The Hilbert-Schmidt norm can be equivalently defined as . The number of summands in this decomposition (which can be infinite if both Hilbert spaces are infinite-dimensional) we call the *Schmidt rank * of . Directly by definition, a pure state on is separable if and only if the Schmidt rank of is 1.
Proposition 4**.**
The Schmidt rank of is if and only if for an operator of rank .
Proof.- Since and for , the image of is spanned by , , thus its rank is .
This suggests the following extension of the concept of Schmidt rank.
Definition 3**.**
The Schmidt rank of is the operator-rank of .
In these terms we can state the following corollary, where we admit infinite-dimensional Hilbert spaces.
Corollary 1**.**
A pure state on is separable if and only if the Schmidt rank of is 1.
This easy characterization of separable pure states has been used by Terhal and Horodecki [21] to develop the concept of Schmidt number of an arbitrary density matrix (quantum state in finite dimensions). This number characterizes the minimum Schmidt rank of the pure states that are needed to construct such density matrix. The Schmidt number is non-increasing under local operations and classical communications, i.e. it provides a legitimate entanglement measure. We can construct an entanglement measure – Schmidt measure – which is additionally convex using the convex roof construction (see e.g. [22]). This construction, proposed as a general tool for entanglement measures (see e.g. [23, 6, 7]), can be repeated in infinite dimensions as
[TABLE]
where the infimum is taken over all possible realizations of as infinite-convex combinations with , and . Every quantum state admits such a realization and a reasoning analogous to the one in [6] shows that is infinite-convex, non-negative and vanishes exactly on separable states.
6 Multipartite generalizations
The diagram of the Jamiołkowski isomorphisms
[TABLE]
interpreted also as
[TABLE]
can be easily generalized to multipartite cases, where we replace with . Of course, the number of possible permutations grows quickly with . Part of them can be obtained by a trivial induction. For instance, we can replace with (or ) in (38), but we can also get
[TABLE]
or
[TABLE]
etc. We will not study here these isomorphisms in details, as the choice of a particular one depends on our interests in possible applications.
7 Acknowledgements
This work was supported by the Polish Ministry of Scientific and Higher Education under the (solicited) grant No PBZ-Min-008/P03/03, EU IP “SCALA” and partially supported by PRIN SINTESI.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] E. Schrödinger, Naturwissenschaften 23 , 807 (1935).
- 2[2] A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47 , 777 (1935).
- 3[3] J. Bell, Physics 1 , 195 (1964).
- 4[4] J. Bell, Rev. Mod. Phys. 38 , 447 (1966).
- 5[5] D. Bouwmeester, A. Ekert, and A. Zeilinger, The Physics of Quantum Information (Springer, Berlin, 2000).
- 6[6] J. Grabowski, M. Kuś, and G. Marmo, J. Phys. A: Math. Gen. 38 , 10217 (2005).
- 7[7] J. Grabowski, M. Kuś, and G. Marmo, Open Sys. Information Dyn. 13 , 343 (2006).
- 8[8] J. F. Carinena, J. Clemente-Gallardo, and G. Marmo, ar Xiv:math-ph/0701053 .
