Arxiv:1804.00918V3 [Math-Ph]

Home , Operator theory

arXiv:1804.00918v3 [math-ph] 12 Mar 2019 b) a) Sem Quantum-Dynamical Discrete-Time of Dilations Unitary Semigroups Quantum-Dynamical Discrete-Time of Dilations Unitary crepnigato) -al [email protected] E-mail: author); (corresponding -al [email protected] E-mail: nagdcoe unu ytm ..w osrc ntr dilation ( unitary semigroup a quantum-dynamical construct time we i.e. system, quantum closed enlarged eso httedsrt-ieeouino noe unu syste quantum channel open quantum an single of a evolution discrete-time the that show We 2019) March 14 (Dated: Würzburg, Germany 97074 unu system; quantum quantum-dynamical dilation; unitary channel; quantum Keywords: 03.67.-a 03.65.Fd, of 02.30.Tb, case 02.20.-a, the numbers: PACS in as result dilation unitary similar channel. a prove and channels man finitely tum by generated furth systems We control finite-dimensional. quantum discrete-time (partially) chosen be may space auxiliary 3) München, Germany 80799 2) Germany Garching, 85747 1) rdrkvmEnde vom Frederik nvri¨tWuzug ntttfu ahmtk Emil-F für Mathematik, Universität Würzburg, Institut uihCnr o unu cec n ehooy(CS) S (MCQST), Technology and Science Quantum for Centre Munich ehiceUiesta ¨nhn et hm,Lichtenbe Chem., Universität München, Dept. Technische ,2, 1, a) n ute Dirr Gunther and T a eebde ntedsrt-ieeouino an of evolution discrete-time the in embedded be can 1 T n 3, ) n b) ∈ N 0 ntecs facci channel cyclic a of case the In . shrSrß 40, ischer-Straße igroups gtae4, rgstraße omtn quan- commuting y eiru;open semigroup; eeae by generated m hligta 4, chellingstrae ftediscrete- the of rinvestigate er single a T the , Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

I. INTRODUCTION

Stimulated by the seminal work of Arveson1, Lindblad2, Gorini, Kossakowski and Sudarshan3 in the mid 1970s many eﬀorts have been made to obtain dilation results of various degrees of generality for semigroups of completely positive operators. 4 For instance, Davies (Ch. 9, Thm. 4.3) proved that for any continuous semigroup (Tt) R+ t∈ 0 of completely positive, unital operators acting on a ﬁnite-dimensional Hilbert spaces (or, H more precisely, on the corresponding W ∗-algebra ( ) of all bounded linear operators) there B H exists a Hilbert space , a pure state ω in ( ), and a strongly continuous one-parameter K B K group (U ) R of unitaries on such that t t∈ H⊗K

T (A)=tr (U †(A id )U ) (1) t ω t ⊗ K t

holds for all A ( ) and t R+. For infinite-dimensional , there is a whole zoo of ∈ B H ∈ 0 H similar results. While Davies5,6, Evans7, and Evans & Lewis8,9 focused primarily on one- 10 parameter semigroups (Tt) R+ of different continuity type, Kümmerer discussed at great t∈ 0 n length the discrete-time case (T )n∈N. However, to the best of our knowledge, for arbitrary Hilbert spaces a dilation result of the above form (1) is still not available. In the following, we give a short chronological overview on those contributions which are relevant and closely related to our work. Further results and a brief survey over the latest developments can be found in11–14 and15. For the readers’ convenience we collected some standard terminology and basic results on dilations and (completely) positive maps, which are well known to experts in this area, in the glossary of Appendix B.

7 In (Thm.1 and Thm.2), Evans shows that every family (Tg)g∈G of completely positive, unital operators acting on a unital C∗-algebra and indexed by an arbitrary group G admits A a unitary dilation, i.e.

† Tg(A)= E(Ug J(A)Ug)

for all A and g G, where (U ) is a unitary representation of G on some Hilbert space ∈A ∈ g g∈G and E a conditional expectation with corresponding injection J into ( ). Remarkably, K B K he need not assume that g T is a group homomorphism. His result can be regarded as 7→ g C∗-counterpart to Sz.-Nagy’s16,17 and Stroescu’s18 work on isometric dilations on Hilbert and Banach spaces, respectively. While possible generalizations to W ∗-algebras are addressed by

2 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Evans, continuity issues of the map g U are disregarded completely. His proof is based 7→ g on Stinespring’s representation T (x) = V ∗π (x)V which of course exists for all g G. g g g g ∈ However, he did not exploit the fact that one can choose a common Hilbert space for all πg which leads to a substantial simpliﬁcation in our approach. In8 (Thm.2) Evans & Lewis focus on norm-continuous semigroups (Tt) R+ of ultraweakly continuous, completely positive and t∈ 0 unital operators acting on a separable Hilbert space . They obtain a unitary dilation H † Tt(A)= E(Ut J(A)Ut) ,

+ for all A ( ) and t R , where (U ) R is a strongly continuous group of unitary ∈ B H ∈ 0 t t∈ operators acting on some extended Hilbert space (and E,J as above). Their proof exploits K the fact that the explicit form of the inﬁnitesimal generator of (Tt) R+ is well-known due to t∈ 0 the work of Lindblad2. In9 Evans & Lewis provide an overview on dilation results known at that time including some minor generalisations of their previous work8.

For locally compact groups G, Davies5 (Thm.2.1 and Thm.3.1) obtains the following

rather general result: Let (Tg)g∈G be a strongly continuous family of ultraweakly continuous, completely positive and unital operators on ( ). Then there exists a Hilbert space , a B H K strongly continuous unitary representation U of G on and conditional expectations H⊗K E : ( ) ( ) (for all n N) such that n B H⊗K → B H ∈ † Tg(A) = lim En(Ug (A idK)Ug) (2) n→∞ ⊗ holds for all A ( ) and all g G in the weak operator topology. Here, E is of the ∈ B H ∈ n form E (A) := V †AV where V : are isometric embeddings. This seems to n n n n H→H⊗K be the result which is closest to (1) in infinite dimensions, but it is not known whether the limit in (2) is necessary or not5 (cf. p. 335). For discrete-time systems T := T n, n N or, n ∈ more accurately, for an appropriate extension to G = Z Davies’ approach and ours are quite similar—in particular due to the fact that in this case, the limit in (2) can be avoided as G is discrete. More precisely, Davies first extends the state space from to L2(Z, ) = ℓ (Z) H H ∼ 2 ⊗H such that (T ) Z can be regarded as one completely positive, unital operator from ( ) to n n∈ B H (L2(Z, )). He then applies Stinespring’s representation theorem to obtain an dilation of B H 2 (T ) Z on a larger state space L (Z, ) . We, however, exploit Stinespring’s result first n n∈ H ⊗K to guarantee for all n N a dilation of the form ∈ T n(A)=tr ((U †)n(A id )U n) , ω ⊗ K

3 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

where ω and are independent of n N, to then enlarge the state space to ℓ (Z) = K ∈ 2 ⊗H⊗K ∼ L2(Z, ). Although both approaches differ only in the order of the construction steps the H⊗K resulting dilations behave quite differently: While Davies’ construction is more “flexible” as one can see, e.g., in Section III B, Remark 12.2, ours yields the desired partial trace structure of (1) which is in general not satisfied for (2) even if the limit can be avoided19 .

10 n Kümmerer discussed the discrete-time case (T )n∈N in detail. However, his setting significantly differs from ours. In his sense, a discrete-time quantum dynamical system consists of a triple ( ,ϕ,T ), where ( ) is a W ∗-algebra, T an ultraweakly continuous, A A ⊂ B H completely positive and unital operator which acts on and leaves a faithful normal state A ϕ ∗ invariant, i.e. ϕ T = ϕ. The latter condition can be thermodynamically motivated ∈A ◦ as ϕ can be interpreted as an equilibrium state which is preserved under composition with T and every power of it. This constraint on the quantum channel T obviously narrows down the possible choices of T . Even more restrictive is Kümmerer’s definition of a first order dilation of ( ,ϕ,T ). Here, he requires the existence of a reversible quantum dynamical A system (A, ϕ,ˆ Tˆ), i.e. Tˆ is a -automorphism on A and E is a conditional expectation with ∗ corresponding injection J such that

T (A)= E Tˆ(J(A)) and ϕ E =ϕ ˆ . ◦ for all A . In doing so, the condition ϕ E =ϕ ˆ is the delicate part. For instance, the ∈ A ◦ standard Kraus/Stinspring representation which constitutes a (first order) dilation does in general not satisfy this condition—note that, by definition,ϕ ˆ has to be a faithful normal state—and therefore even first order dilations in Kümmerer’s sense need not exist as the existence of a ϕ-adjoint is not guaranteed, cf.10 (Prop. 2.1.8 ff.). Within his setting, Kümmerer proved (cf. Thm 4.2.1, Cor 4.2.3) that a quantum dynamical system ( ( ),ϕ,T ) has a di- B H lation of first order if and only if it admits a Markovian one of first order which in turn implies that ( ( ),ϕ,T ) also allows a Markovian dilation of arbitrary order. His definition B H of Markovianity can be regarded as a W ∗-algebra counterpart of a well-known subspace condition20 which guarantees for contractions on Hilbert spaces that a first order unitary dilation T = P U is already a dilation (of arbitrary order), i.e. T n = P U n holds for H |H H |H all n N. To achieve a Markovian dilation he imbedded the given W ∗-algebra = ( ) ∈ A B H in an infinite product/sum of W ∗-algebras. Our approach considerably deviates from his construction since we use first order Stinespring/Kraus dilations for T n which of course exist

4 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

for all n N but in general do not satisfy K¨ummerer’s faithful state condition. ∈

Probably one of the strongest semigroup dilation results so far was presented by Gaebler14 (Thm. 5.10). Using Sauvageot’s theory he showed that for a norm-continuous semigroup

(Tt)t∈R+ of ultraweakly continuous, completely positive and unital operators acting on a ∗ W -algebra with separable pre-dual, there exists a unital dilation (A, (σt) R+ , J, E) of A t∈ 0 (Tt) R+ (cf. Def. 18) where A has separable pre-dual and ((σt) R+ , J, E) satisﬁes the strong t∈ 0 t∈ 0 dilation property, i.e. T E = E σ for all t R+. The strength of this result, however, t ◦ ◦ t ∈ 0 comes at the cost of lacking any partial trace structure of the form (1).

The paper is organized as follows: After some preliminaries on trace-class operators and quantum channels, we present our main results in Section III: (i) For discrete-time quantum-dynamical semigroups on separable Hilbert spaces, a unitary dilation of the form (1) is proved. (ii) If the semigroup in question is generated by a cyclic quantum channel, then the auxiliary Hilbert space can be chosen partially ﬁnite-dimensional. (iii) Finally, for discrete-time quantum control systems, the control of which can be switched between a ﬁnite number of commuting channels, a unitary dilation of the form (1) is derived.

II. PRELIMINARIES AND NOTATION

In this section, we ﬁx our notation and recall some basic material on Schr¨odinger and Heisenberg quantum channels. These results should be known to experts in this area.

Henceforth, let , be infinite-dimensional separable complex Hilbert spaces and , G H X Y real or complex Banach spaces. By convention, all scalar products on complex Hilbert spaces are assumed to be conjugate linear in the first argument and linear in the second. Moreover, let ( ), ( ) denote the set of all bounded operators acting on , and let ( ), ( ) B G B H G H B X B Y be defined respectively.

Recall that an operator A ( ) on a complex Hilbert space is said to be positive ∈ B H semi-definite, denoted by A 0, iff x, Ax 0 for all x . Because we consider complex ≥ h i ≥ ∈ H Hilbert spaces, A 0 directly implies that A is self-adjoint via the polarization identity, cf.21 ≥ (Prop. 2.4.6)—else, self-adjointness would have to be required in the definition of A 0. ≥ 5 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

A. Quantum Channels

Let 1( ) ( ) be the subset of all trace-class operators, i.e. 1( ) is the largest B H ⊂ B H B H subspace of ( ) which allows to deﬁne the trace of an operator A via B H

tr(A) := ei, Aei (3) Xi∈I h i such that the right-hand side of (3) is ﬁnite and independent of the choice of the orthonormal basis (e ) . More precisely, 1( ) can be deﬁned either as the set of all compact operators i i∈I B H A ( ) whose singular values σ (A) are summable, i.e. ∈ B H n

ν1(A) := σn(A) < (4) Xn∈N ∞ or, equivalently22 (Thm.VI.21), as the set of all A ( ) such that ∈ B H

† ei, √A Aei < (5) Xi∈I h i ∞ is summable for some orthonormal basis (e ) of . i i∈I H Because of √A†A 0, all summands in (5) are non-negative and therefore22 (Thm.VI.18), ≥ the value of the left-hand side of (5) is independent of the chosen orthonormal basis. More- over, one has tr(√A†A) = ν (A) for all A 1( ) which readily implies tr(A) = ν (A) if 1 ∈ B H 1 A 0. Finally, we note that for ﬁnite-dimensional Hilbert spaces the sets ( ) and 1( ) ≥ B H B H coincide with the set of all linear operators acting on and that for arbitrary Hilbert spaces, H 1( ) constitutes a Banach space with respect to the trace norm ν given by (4). For more B H 1 on these topics we refer to22 (Ch.VI.6) and23 (Ch.16).

An operator ρ 1( ) which is positive semi-deﬁnite and fulﬁlls tr(ρ) = 1 is called a ∈ B H state and the set of all states is denoted by

D( ) := ρ 1( ) ρ is state . H { ∈ B H | } A state ρ is said to be pure if it has rank one. Certainly, the corresponding definitions apply to 1( ) and D( ). After these preliminaries, we can introduce the key concepts. B G G Definition 1. (a) A linear map T : 1( ) 1( ) is said to be positive if T (A) 0 for B H → B G ≥ all positive semi-definite A 1( ). ∈ B H (b) A linear map T : 1( ) 1( ) is said to be completely positive if for all n N the B H → B G ∈ maps T id : 1( Cn) 1( Cn) are positive. ⊗ n B H ⊗ → B G ⊗ 6 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

(c) A Schr¨odinger quantum channel is a linear, completely positive and trace-preserving map T : 1( ) 1( ). Furthermore, we deﬁne B H → B G

Q ( , ) := T : 1( ) 1( ) T is Schr¨odinger quantum channel S H G { B H → B G | }

and Q ( ) := Q ( , ). S H S H H

Note that Deﬁnition 1 (a) and (b) also make sense for maps from ( ) ( ) instead of B H → B G 1( ) 1( ). B H → B G Clearly, every positive, trace-preserving map and thus every Schr¨odinger quantum channel maps states to states. Further algebraic and topological properties of Q ( ) which are S H crucial in the following are summarized in the following theorem, the proof of which can be found in Appendix A.

Theorem 1. The set Q ( ) is a convex subsemigroup of ( 1( )) with unity element S H B B H 1 id 1 . Moreover, Q ( ) is closed in ( ( )) with respect to the weak operator, strong B (H) S H B B H operator and uniform operator topology.

Remark 2. 1. Here one should emphasize that it is not necessary to require boundedness of Schr¨odinger quantum channels. In fact, one can easily prove that any positive linear map is bounded automatically4 (Ch. 2, Lemma 2.1).

2. In Proposition 2 we will see that Q ( ) is actually a closed convex subset of the unit S H sphere of ( 1( )). Note that the existence of non-trivial convex subsets on the unit B B H sphere of ( 1( )) is a consequence of the non-strict convexity of the operator norm. B B H

The following beautiful and well-known representation result for Schr¨odinger quantum channels which can be traced back to Kraus is the starting of our work.

Theorem 2. For every T Q ( ) there exists a separable Hilbert space , a pure ω D( ) ∈ S H K ∈ K and a unitary U ( ) such that ∈ B H⊗K

T (A)=tr (U(A ω)U †) (6) K ⊗ for all A 1( ). ∈ B H 7 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Here tr : 1( ) 1( ) is the partial trace with respect to the Hilbert space which K B H⊗K → B H K is deﬁned via

tr(B tr (A)) = tr((B id )A) (7) K ⊗ K for all B ( ) and all A 1( ). ∈ B H ∈ B H⊗K For a complete proof of Theorem 2, see24 (second part of Thm.2). Here, we only emphasize that the separable auxiliary space can be chosen independently of T ; for instance, := K K ℓ (N) constitutes such a universal auxiliary space. Moreover, once is ﬁxed, ω D( ) can 2 K ∈ K be chosen as any orthogonal rank-1 projection. Thus ω is pure and independent of T , too.

Corollary 1 (General Stinespring Dilation). For every T Q ( , ) there exists a separable ∈ S H G Hilbert space , pure ω D( ), ω D( ) and a unitary U ( ) such that K G ∈ G K ∈ K ∈ B H⊗G⊗K T (A) = (tr tr )(U(A ω ω )U †) H ◦ K ⊗ G ⊗ K for all A 1( ). ∈ B H Proof. Consider arbitrary ω D( ) and ω D( ) of rank one. Applying Theorem 2 G ∈ G H ∈ H to X( ) := ω T (tr ( )) Q ( ) which is obviously a composition of Schrödinger · H ⊗ G · ∈ S H ⊗ G quantum channels yields a separable Hilbert space , a pure ω D( ) and a unitary K K ∈ K U ( ) such that X is of form (6). For any A 1( ) one gets ∈ B H⊗G⊗K ∈ B H T (A)=tr (X(A ω )) = (tr tr )(U(A ω ω )U †) H ⊗ G H ◦ K ⊗ G ⊗ K with ω ω D( ) rank one. G ⊗ K ∈ G⊗K The last result of this subsection provides a characterization of invertible quantum channels which leads to a nice simplification later on (cf. Remark 9). For finite dimensions, this was essentially shown in25 (Coro. 3).

Proposition 1. Let T Q ( ) be bijective. Then the following statements are equivalent. ∈ S H (a) T −1 is positive.

(b) There exists unitary U ( ) such that T (A)= UAU † for all A 1( ). ∈ B H ∈ B H In particular if one (and thus both) conditions are fulﬁlled, then T Q ( ) is invertible as ∈ S H a channel, i.e. T is bijective and T −1 Q ( ). ∈ S H 8 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Proof. (b) (a): X (a) (b): The proof idea is the same as in26 (Prop. 4.31). Consider the ⇒ ⇒ 1 restricted channel T D : D( ) ( ) which by assumption is convex-linear and injective. | H → B H As T and T −1 are linear, trace-preserving and, by assumption, positive, the restricted channel satisﬁes T (D( )) D( ) T −1(D( )) D( ) , H ⊆ H H ⊆ H so T D : D( ) D( ) is surjective and thus a state automorphism, i.e. convex-linear | H → H and bijective. Then Corollary 3.2 in4 or, more explicitely, Theorem 2.63 in26 imply the † existence of unitary or anti-unitary U such that T D( ) = U( )U . If U were anti-unitary, | · · then T would not be completely positive26 (Prop. 4.14) hence U has to be unitary. Due to 1 1 spanC(D( )) = ( ), this representation extends linearily to all of ( ) which concludes H B H B H the proof.

B. Dual Channels

It is well known23 (Prop.16.26) that the dual space of 1( ) is isometrically isomorphic B H to ( ) by means of the map ψ : ( ) ( 1( ))′, B ψ (B) with B H H B H → B H 7→ H

(ψH(B))(A) := tr(BA)

for all A 1( ). Note that the weak- -topology and the ultraweak topology on ( ) ∈ B H ∗ B H coincide under the above identiﬁcation ( 1( ))′ = ( ), cf.4 (Section 1.6). B H ∼ B H Now, since every positive linear map T : 1( ) 1( ) is bounded (cf. Remark 2.1) B H → B G the dual map

T ′ :( 1( ))′ ( 1( ))′ X T ′(X) := X T B G → B H 7→ ◦ is well deﬁned and this allows us to construct the so called dual channel of T

T ∗ : ( ) ( ) B T ∗(B) := (ψ−1 T ′ ψ )(B) B G → B H 7→ H ◦ ◦ G which then satisﬁes

tr(BT (A)) = tr(T ∗(B)A) (8) for all B ( ) and A 1( ). Alternatively, onc can use (8) as deﬁning equation for ∈ B G ∈ B H T ∗. Furthermore, one has T = T ∗ by deﬁnition of T ∗, because T and T ′ have the same k k k k ∗ operator norm and ψG and ψH are isometric isomorphisms. Some basic properties of T are:

9 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

(a) T ∗ is positive and ultraweakly continuous.

(b) T ∗ is completely positive if and only if T is completely positive.

∗ ∗ (c) T is unital (i.e. T (idG) = idH) if and only if T is trace-preserving.

For more details and proofs we refer to24 (p. 35) or26 (Ch.4.1.2).

Definition 3. A Heisenberg quantum channel is a linear, ultraweakly continuous27, completely positive and unital map S : ( ) ( ). Furthermore, we define B G → B H Q ( , ) := S : ( ) ( ) S is Heisenberg quantum channel H G H { B G → B H | } and Q ( ) := Q ( , ). H H H H H By the properties listed above, it is evident that the map : Q ( , ) Q ( , ) which ∗ S H G → H G H to any quantum channel assigns its dual channel is well-defined. Furthermore, it is—as we will see next—bijective.

Theorem 3. (a) For every S : ( ) ( ) linear, ultraweakly continuous and positive B G → B H there exists unique T : 1( ) 1( ) linear and positive such that T ∗ = S. B H → B G (b) For every S Q ( , ) there exists unique T Q ( , ) such that T ∗ = S. ∈ H G H ∈ S H G Proof. (a) By the above construction of the dual channel it is obvious that the map is ∗ one-to-one. Therefore, it suffices to show its surjectivity. First one shows, similar to4 (Ch. 1, Lemma 6.1), that for every positive, linear and ultraweakly continuous functional λ : ( ) C there exists a unique positive semi-definite B G → ρ 1( ) such that λ( ) = tr(ρ( )). Next choose arbitrary positive semi-definite A 1( ) ∈ B G · · ∈ B H and consider the linear functional

B tr(S(B)A) 7→ which by assumption on S is ultraweakly continuous. Our preliminary consideration yields a unique positive semi-definite ρ 1( ) such that tr(S(B)A) = tr(Bρ ) for all B ( ). A ∈ B G A ∈ B G This allows to define an R+-linear map Tˆ on the positive semi-definite elements of 1( ) via B H Tˆ(A) := ρ . Finally, Tˆ(A) can be uniquely extended to a positive, linear map T : 1( ) A B H → 1( ) satisfying T ∗ = S. B G Now (b) follows from (a) together with the above connections between properties of a positive, linear map and its dual channel.

10 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Remark 4. 1. Note that, again, boundedness is not required in the deﬁnition of a Heisen- berg quantum channel because, similar to Schr¨odunger quantum channels, they are automatically bounded, see Proposition 2 below.

2. In ﬁnite dimensions, ultraweak continuity is of course always satisﬁed and the -map ∗ is an involution as the sets of trace-class operators and bounded operators coincide.

Proposition 2. Let T Q ( , ) and S Q ( , ). Then T =1 and S =1. ∈ S H G ∈ H G H k k k k Proof. As each S Q ( , ) in particular is linear, positive and unital it has operator ∈ H G H norm S = 1 as a consequence of the Russo-Dye Theorem, cf.28 (Cor. 1) or Rem. 19.1. k k This directly implies T = T ∗ = 1. k k k k Alternatively, one can prove Proposition 2 via the general Stinespring dilation (Corollary 1) because all maps involved in the Stinespring representation have operator norm one. Either way, with this one readily veriﬁes that Q ( ) forms a convex subsemigroup of the Banach H H space ( ( )) with unity element id . B B H B(H) The partial trace tr : ( ) ( ) with respect to a state ω D( ) is deﬁned via ω B H⊗K → B H ∈ K

tr(tr (B)A) = tr(B(A ω)) (9) ω ⊗

for all B ( ), A 1( ), cf.4 (Ch. 9, Lemma 1.1). Be aware that the map tr ∈ B H⊗K ∈ B H K from (7) and the extension

i : 1( ) 1( ) A A ω ω B H → B H⊗K 7→ ⊗

with some state ω D( ) are Schr¨odinger quantum channels so we immediatly get their ∈ K ∗ ∗ dual channels iω = trω and trK = iK with

i : ( ) ( ) B B id . K B H → B H⊗K 7→ ⊗ K

This leads to the following result.

Corollary 2. For every S Q ( , ) there exists a separable Hilbert space , pure states ∈ H G H K ω D( ) and ω D( ) and a unitary U ( ) such that G ∈ G K ∈ K ∈ B H⊗G⊗K

S(B) = (tr tr )(U †(id B id )U) ωG ◦ ωK H ⊗ ⊗ K

11 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

for all B ( ). For = this reduces to ∈ B G G H

S(B)=tr (U †(B id )U) (10) ωK ⊗ K

for all B ( ) where the unitary operator U now acts on . ∈ B H H⊗K Proof. Note that (8) implies (T T )∗ = T ∗ T ∗ for arbitrary positive, linear maps T and 1 ◦ 2 2 ◦ 1 1 T2. Hence this is a simple consequence of Theorem 2, Theorem 3 (b) and Corollary 1.

Remark 5. The result in Corollary 2 is a more structured version of Stinespring’s theorem29 for Heisenberg quantum channels due to the following: Let S Q ( ) (the same argument ∈ H H works for S Q ( , )) and ω D( ) be the state from (10) of rank one, i.e. ω = y, y ∈ H G H K ∈ K K h ·i for some y with y = 1. Deﬁning the isometric embedding V : , x x y, ∈K k k y H→H⊗K 7→ ⊗ one readily veriﬁes via (9) that tr (B)= V †BV for all B ( ). Now (10) becomes ωK y y ∈ B H⊗K

S( )= V †π( )V · ·

with the auxiliary Hilbert space being of tensor form, the Stinespring isometry H⊗K V = UV and the unital -homomorphism π : ( ) ( ) being π(B) := B id . y ∗ B H → B H⊗K ⊗ K To the best of our knowledge, the above representation (10) so far only appeared in an unpublished (as of now) book by S. Attal30 (Thm. 6.15).

The above concept of dual channels will be useful to transfer dilation results from the Schr¨odinger to the Heisenberg picture and vice versa so one is independent of the used quantum-mechanical framework.

III. MAIN RESULTS

A. Unitary Dilation of Discrete-Time Quantum-Dynamical Systems

Consider a discrete-time quantum-dynamical system, the evolution of which is described by

ρ = T (ρ ), ρ D( ) (11) n+1 n 0 ∈ H

for arbitrary but ﬁx T Q ( ). Obviously, the explicit solution of (11) is given by ∈ S H

n ρn = T (ρ0)

12 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

for all n N By Theorem 1, one has T n Q ( ) and thus Theorem 2 yields separable ∈ 0 ∈ S H Hilbert spaces , pure states ω D( ) and unitaries U ( ) such that Kn n ∈ Kn n ∈ B H⊗Kn

T n(A)=tr U (A ω )U † (12) Kn n ⊗ n n

for all A 1( ) and all n N . Now our goal is to simplify the right-hand side of (12) ∈ B H ∈ 0 in the following sense: We want to embed the evolution of ρ0 into an evolution of a closed discrete-time quantum-dynamical system, i.e. we want to replace the r.h.s. of (12) by

n † n tr ˜ V (A ω˜)(V ) K ⊗ where V is an appropriate unitary operator and the separable Hilbert space ˜ as well as the K pure stateω ˜ does no longer depend on n N . Our established result reads as follows ∈ 0

Theorem 4. For every T Q ( ) there exists a separable Hilbert space , a pure state ∈ S H K n ω D( ) and a unitary V ( ) such that ( , (V ) Z, i , tr ) is a unitary ∈ K ∈ B H⊗K H⊗K n∈ ω K n 1 dilation of (T ) N (in the sense of Deﬁnition 21.2). In particular, for all A ( ) and n∈ 0 ∈ B H n N , one has ∈ 0

T n(A)=tr V n(A ω)(V †)n . (13) K ⊗

Proof. First we consider the n-dependence of and ω . By construction, cf. Theorem 2, Kn n does not depend on T n anymore, thus we can choose ˜ with a countably inﬁnite basis, Kn K for example ˜ = ℓ (N), and replace every with ˜. Moreover, also by construction, the K 2 Kn K pure state ω is determined via and thus can be chosen independently of n, too. Hence n Kn we obtain a joint Hilbert space ˜ and a pure stateω ˜ such that K

n † T (A)=tr ˜ U (A ω˜)U . K n ⊗ n

for all n N . Finally, in order to remove the n-dependence of the unitary operators U we ∈ 0 n deﬁne := ˜ ℓ (Z) and U = id ˜ for all n 0. Furthermore, let (e ) Z denote the K K ⊗ 2 n H⊗K ≤ n n∈ † standard basis of ℓ2(Z) so σ : ℓ2(Z) ℓ2(Z) given by σ = Z eie yields the right shift → Pi∈ i−1 on ℓ (Z). With this U, W : ( ) ( ) are deﬁned by 2 B H⊗K → B H⊗K

† † U := UnUn− enen and W := id ˜ σ . Xn∈Z 1 ⊗ H⊗K ⊗

13 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Thus U can be visualised as follows: . ..    idH⊗K˜   ←  U   1    ,  U U †   2 1   †   U U   3 2   .   ..

↑ where the arrows indicate the zero-zero entry of this both-sided “infinite matrix”. A simple calculation shows that U, W and therefore also V := UW are unitary. Next, using the results from Section II B, one readily verifies that the maps E := trK and J := iω (where ω :=ω ˜ e e† D( ) is obviously pure) satisfy the conditions from Definition 21.1. Then, ⊗ 0 0 ∈ K by induction, one shows

V n(A ω)(V †)n = U (A ω˜)U † e e† ⊗ n ⊗ n ⊗ n n

1 for all A ( ) and n N . Finally, id ˜ = id ˜ id implies tr ˜ = tr ˜ tr so ∈ B H ∈ 0 K⊗ℓ2 K ⊗ ℓ2 K⊗ℓ2 K ◦ ℓ2

n † n † † † n tr V (A ω)(V ) = tr ˜ (tr (U (A ω˜)U e e )) = tr ˜ (U (A ω˜)U )= T (A) K ⊗ K ℓ2 n ⊗ n ⊗ n n K n ⊗ n

1 n for all A ( ) and n N . Hence we constructed a unitary dilation of (T ) N of the ∈ B H ∈ 0 n∈ 0 form (13) which concludes the proof.

Remark 6. Note that i is trace-preserving because ω D( ), so the above (tensor type) ω ∈ K dilation is trace-preserving.

Now we can easily extend this result to Heisenberg quantum channels.

Corollary 3. For every S Q ( ) there exists a separable Hilbert space , a pure state ∈ H H K † n ω D( ) and a unitary V ( ) such that ( , ((V ) ) Z, i , tr ) is a unitary ∈ K ∈ B H⊗K H⊗K n∈ K ω n dilation of (S ) N (in the sense of Deﬁnition 20.2). In particular, for all B ( ) and n∈ 0 ∈ B H n N , one has ∈ 0

Sn(B)=tr (V †)n(B id )V n . ω ⊗ K 14 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Proof. By Theorem 3 (b) there exists a unique T Q ( ) such that S = T ∗ and therefore ∈ S H Sn = (T ∗)n = (T n)∗. Now Theorem 4 yields a separable Hilbert space , a pure state K ω D( ) and a unitary V such that (13) holds. By duality we obtain ∈ K ∗ n n † n † n n S (B)= trK V iω( )(V ) (B)=trω (V ) (B idK)V · ⊗ for all B ( ) and for all n N . ∈ B H ∈ 0 Remark 7. 1. Due to id id = id we even constructed a unital dilation (of tensor H ⊗ K H⊗K type).

2. Recall that a “classical” unitary dilation T n = P U n inc of some Hilbert space H ◦ ◦ H contraction T : (cf. Rem. 17.3), where P denotes the orthogonal projection H → H H onto and inc the inclusion map, is called minimal if the domain of U ( ) is H H ∈ B K minimal in the sense of = U n . (14) K _ H n∈Z Here the right-hand side of (14) denotes the smallest closed subspace of which K contains all images U n , n Z, cf.31,32. K¨ummerer10 captures this idea and deﬁnes a H ∈ dilation T n(A)= E Tˆn(J(A)) of an ultraweakly continuous, completely positive and unital map T : on a W ∗-algebra to be minimal if A→A A = Tˆn(i( )) (15) A _ A n∈Z holds, where the right-hand side of (15) now denotes the smallest closed W ∗-algebra which contains all images Tˆn(i( )), n Z, cf.10 (Def. 2.1.5). It is easy to see that our A ∈ constructions in Theorem 4 / Corollary 3 do in general not lead to a minimal dilation in the above sense. However, one can always restrict a given dilation to the right-hand side of (15) to obtain a minimal one.

∞ −n 3. As seen above in (14) the space := Z U and its forward and backward H−∞ Wn∈ H invariant counterparts

∞ := U n , and := U −n , H _ H H−∞ _ H n∈N0 n∈N0 play an essential role in the theory of “classical” unitary dilations. In particular, they admit orthogonal decompositions

∞ = ˆ∞ , = ˆ and ∞ = ˆ∞ ˆ (16) H H ⊕ H H∞ H ⊕ H−∞ H−∞ H ⊕ H ⊕ H−∞ 15 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

such that ˆ∞ and ˆ are invariant under U and U −1, respectively, cf.32 (Lemma H H−∞ VI.3.1) and33. Eventually, (16) establishes the relation to K¨ummerer’s notion of Markovianity, cf.10 (Prop. 2.2.3 (b)).

Next we want to improve Theorem 4 for cyclic T , i.e. in the case of T m = T for some m N 1 . ∈ \{ } Definition 8. In doing so, we define a modified modulo function

ν : N 1 N N \{ } × → (m, n) (n 1) mod(m 1)+1 7→ − − as well as

µ : N 1 N N \{ } × → 0 n ν(m, n) (m, n) − . 7→ m 1 − To connect ν(m, n) to the above cyclicity condition of T we represent n 1 as − n 1= j(m 1) + r (17) − − with unique j N and r 0,...,m 2 . This yields ν(m, n) = r + 1 as well as ∈ 0 ∈ { − } µ(m, n)= j N and we obtain the following result. ∈ 0 Lemma 1. Let T Q ( ) be cyclic so T m = T for m N 1 . Then ∈ S H ∈ \{ } T n = T ν(m,n) for all n N. ∈ Proof. Via (17) we get T n = T j(m−1)+r+1 = T r+1−j(T m)j = T r+1−jT j = T r+1 = T ν(m,n).

Thus µ(m, n) indicates how often the cyclicity condition of T can be applied to reduce the exponent n to its remaining non-cyclic portion ν(m, n). With this we obtain the following simpliﬁcation of Theorem 4.

Theorem 5. Let T Q ( ) be cyclic, i.e. T m = T for some m N 1 . Then for ∈ S H ∈ \{ } n n the unitary dilation ( , (V ) Z, i , tr ) of (T ) N from Theorem 4, one can choose H⊗K n∈ ω K n∈ 0 = ˜ Cm such that (after modifying V and ω accordingly) K K ⊗ T n(A)=tr V n+µ(m,n)(A ω)(V †)n+µ(m,n) K ⊗ for all A 1( ) and all n N . Note that ω D( ) still is a pure state. ∈ B H ∈ 0 ∈ K 16 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Proof. Choose ˜ andω ˜ D( ˜) as in the proof of Theorem 4. For every T,...,T m−1 there K ∈ K again exist unitary U ,...,U ( ˜) satisfying Theorem 2. This allows to deﬁne 1 m−1 ∈ B H ⊗ K

m † † m † U := UiUi−1 eiei and W := idH⊗K˜ ei+1ei Xi=1 ⊗ ⊗ Xi=1 Cm where em+1 := e1 and U0 := idH⊗K˜ =: Um. Then W represents a cyclic shift acting on and U is of the following form.

U  1  U U †  2 1   .  U =  ..       U U †   m−1 m−2   †   Um−1

Obviously, U, W and thus V := UW are unitary. Again choosing E := trK and J := iω with pure state ω :=ω ˜ e e† D( ), one readily veriﬁes via indiction ⊗ m m ∈ K V n+µ(m,n)(A ω)(V †)n+µ(m,n) = U (A ω)U † e e† . ⊗ ν(m,n) ⊗ ν(m,n) ⊗ ν(m,n) ν(m,n) for all A 1( ) and n N. Together with Lemma 1 one gets ∈ B H ∈

n+µ(m,n) † n+µ(m,n) † † trK V (A ω)(V ) = tr ˜ trCm U (A ω˜)U e e ⊗ K ◦ ν(m,n) ⊗ ν(m,n) ⊗ ν(m,n) ν(m,n) † ν(m,n) n = tr ˜ U (A ω˜)U = T (A)= T (A) K ν(m,n) ⊗ ν(m,n) for all A 1( ) and n N. ∈ B H ∈ Remark 9. Note that quantum channels which have an inverse channel (or are “just” bijective with positive inverse) can be written as a unitary conjugation AdU , cf. Prop. 1.

For such channels, Theorem 4 is trivially fulﬁlled by choosing = C, E = J = id 1 and K B (H) V = U. The same holds for cyclic quantum channels which are bijective because cyclicity implies T −1 = T m−2 Q ( ). ∈ S H

B. Unitary Dilation of Discrete-Time Quantum-Control Systems

Here, we investigate discrete-time quantum-mechanical control systems of the form

ρ = T (ρ ), ρ D( ) (18) n+1 n n 0 ∈ H

17 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

where T , n N is regarded as control input which can be chosen freely from some subset n ∈ 0 Q ( ). We deﬁne ρ( , (T ) N , ρ ) to be the unique solution of (18) generated by the C ⊂ S H · n n∈ 0 0 control sequence (Tn)n∈N0 and the initial value ρ0. In the sequel, we are interested in whether the dynamics of (18) can be embedded in the dynamics of a unitary discrete-time quantum control system of the same form.

Deﬁnition 10. Let RN (ρ0) denote the set of all states which can be reached from ρ0 in N time steps via (18), i.e.

R (ρ ) := ρ(N, (T ) N , ρ ) (T ) N arbitrary control sequence . N 0 { n n∈ 0 0 | n n∈ 0 }

Moreover, the overall reachable set of ρ0 is deﬁned by

R(ρ0) := RN (ρ0). [N∈N0 For the remaining section, we assume := T,S where T and S are commuting but other- C { } wise arbitrary quantum channels over . Then the following result is a direct consequence H of the fact that T and S commute.

Lemma 2. For all N N one has R (ρ ) := T kSN−kρ k =0,...,N . ∈ 0 N 0 { 0 | } Based on this we are interested in dilations of quantum channels of the form T kSN−k.

Theorem 6. Let T,S Q ( ) be commuting. Then there exists a separable Hilbert space ∈ S H , a pure state ω D( ) and unitary U, V ( ) such that K ∈ K ∈ B H⊗K T kSN−k(A)=tr U kV N−k(A ω)(V †)N−k(U †)k . K ⊗ for all A 1( ), N N and k =0,...,N. ∈ B H ∈ 0 Proof. For ﬁxed N N and k =0,...,N, one has T kSN−k Q ( ) by Theorem 1 and thus ∈ ∈ S H Theorem 2 yields a separable Hilbert space , a pure state ω D( ) and unitary KN,k N,k ∈ KN,k U ( ) such that N,k ∈ B H⊗K T kSN−k(A)=tr U (A ω )U † . KN,k N,k ⊗ N,k N,k for all A 1( ). The same line of arguments as in the proof of Theorem 4 show that ∈ B H KN,k and ωN,k can be chosen independently of N and k, so there exists some mutual auxiliary space ˜ as well as a mutual pure stateω ˜ D( ˜) such that K ∈ K k N−k † T S (A)=tr ˜ U (A ω˜)U (19) K N,k ⊗ N,k

18 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

for all A 1( ), N N and k =0,...,N. In particular, to every T kSN−k we can assign ∈ B H ∈ some unitary U ( ˜) such that (19) holds. Now, choose := ˜ ℓ (Z) ℓ (Z) N,k ∈ B H ⊗ K K K ⊗ 2 ⊗ 2 and again E := tr and J := i with pure state ω :=ω ˜ e e† e e† D( ). Moreover, by K ω ⊗ 0 0 ⊗ 0 0 ∈ K means of the right shift σ from the proof of Theorem 4 one deﬁnes

† † † W := id ˜ σ σ , U := U U e e e e , 1 H⊗K ⊗ ⊗ 1 X m,n m−1,n−1 ⊗ m m ⊗ n n m,n∈Z

† † W := id ˜ σ id , U := U U e e id , 2 H⊗K ⊗ ⊗ ℓ2 2 X n,0 n−1,0 ⊗ n n ⊗ ℓ2 n∈Z where U := id ˜ if m < 1 or n / 0,...,m . Obviously, W and W are unitary. m,n H⊗K ∈ { } 1 2 The unitarity of U1, and U2 is readily veriﬁed via the unitarity of UN,k so U := U1W1 and

V := U2W2 are unitary, too. As before, by induction one shows

V j(A ω)(V †)j = U (A ω˜)U † e e† e e† (20) ⊗ j,0 ⊗ j,0 ⊗ j j ⊗ 0 0 for all A 1( ) and j N and based on this ∈ B H ∈ 0

U kV N−k(A ω)(V †)N−k(U †)k = U (A ω˜)U † e e† e e† (21) ⊗ N,k ⊗ N,k ⊗ N N ⊗ k k for all A 1( ), N N and k =0,...,N. Note that the case k = 0 reproduces (20) and ∈ B H ∈ 0 thus can be omitted. Finally, (19) and (21) imply

k N−k † N−k † k † † † tr U V (A ω)(V ) (U ) = tr ˜ (tr (U (A ω˜)U e e e e )) K ⊗ K ℓ2⊗ℓ2 N,k ⊗ N,k ⊗ N N ⊗ k k † k N−k = tr ˜ (U (A ω˜)U )= T S (A) K N,k ⊗ N,k for all A 1( ), N N and k =0,...,N which concludes this proof. ∈ B H ∈

Remark 11. The statement of Theorem 6 can be extended to ﬁnitely many commuting channels T ,...,T Q ( ). Obviously, it is natural to choose 1 m ∈ S H

= ˜ ℓ (Z) ... ℓ (Z) K K ⊗ 2 ⊗ ⊗ 2 m-times | {z } as common auxiliary space. The rest of the proof is completely analogous.

We can now transfer the above result to obtain a characterization of the reachable set of the control system (18).

19 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Corollary 4. There exists a separable Hilbert space , a pure state ω D( ) and unitary K ∈ K U, V ( ) such that ∈ B H⊗K

k N−k † N−k † k ρ(N, (T ) N , ρ )=tr U V (ρ ω)(V ) (U ) n n∈ 0 0 K 0 ⊗ for all controls (T ) N , initial states ρ D( ) and N N , where k = k(N, (T ) N ) n n∈ 0 0 ∈ H ∈ 0 n n∈ 0 ∈ 0,...,N counts how often T occurs in the control sequence (T ) N during the ﬁrst N { } n n∈ 0 time steps.

Proof. By Deﬁnition 10, ρ(N, (T ) N , ρ ) R (ρ ) and hence by Lemma 2 there exists k n n∈ 0 0 ∈ N 0 ∈ k N−k 0,...,N such that ρ(N, (T ) N , ρ ) = T S (ρ ). Thus the result follows immeditely { } n n∈ 0 0 0 from Theorem 6.

Corollary 5. Let , ω D( ) and U, V ( ) be as in Corollary 4. Then, for all K ∈ K ∈ B H⊗K N N and ρ D( ) one has ∈ 0 0 ∈ H

R (ρ ) tr (R˜ (ρ ω)) (22) N 0 ⊆ K N 0 ⊗ and thus R(ρ ) tr (R˜(ρ ω)). Here, R˜(˜ρ ) and R˜ (˜ρ ) denote the reachable sets of the 0 ⊆ K 0 ⊗ 0 N 0 discrete-time closed quantum control system

ρ˜ = U ρ˜ U † , ρ˜ D( ) n+1 n n n 0 ∈ H⊗K with U U, V for all n N . n ∈{ } ∈ 0 Proof. By Lemma 2 and Theorem 6, one has

R (ρ )= T kSN−k(ρ ) k =0,...,N N 0 { 0 | } = tr U kV N−k(ρ ω)(V †)N−k(U †)k k =0,...,N { K 0 ⊗ | } = tr ( U kV N−k(ρ ω)(V †)N−k(U †)k k =0,...,N ) tr (R˜ (ρ ω)) . K { 0 ⊗ | } ⊆ K N 0 ⊗

Remark 12. 1. Note that the unitary channels U and V of Corollary 5 do in general not commute, so (22) states a proper inclusion rather than an equality for N > 1.

2. Consider the dual problem of (18), i.e. let T,S Q ( ) be two commuting Heisenberg ∈ H H channels. Of course, one can translate the above results—which we will omit here—into the Heisenberg picture via Corollary 3. However, we want to comment on the result of

20 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Davies5 which was already mentioned in the introduction and yields a unitary dilation with commuting unitary channels, at the cost of our desired partical trace structure.

Let G = (Z Z, +) with subgroup S := (N,k) G N N and 0 k N and × { ∈ | ∈ 0 ≤ ≤ } deﬁne the family (T ) of Heisenberg channels via T := T kSN−k for g =(N,k) S g g∈G g ∈ 5 and Tg := idB(H) otherwise. Adjusting the proof of (Thm. 3.1) to discrete groups and using Corollary 2, one gets a Hilbert space , a unitary representation U of G K on and a conditional expectation E such that T (B)= E(U (B id )U †) for H⊗K g g ⊗ K g all B ( ). As (U ) is a representation of G we may consider the commuting ∈ B H g g∈G unitary operators U(1,1) =: U, U(1,0) =: V resulting in

T kSN−k(B)= T (B)= E(U kV N−k(B id )(V †)N−k(U †)k) g ⊗ K

for all B ( ), N N and k = 0,...,N. Observe that we did not use the fact ∈ B H ∈ 0 that T,S commute so this result even holds for arbitrary channels T and S with the drawback of E lacking any partial trace structure, see also19.

ACKNOWLEDGMENTS

This article is based on a master thesis34 which was written at the Institute of Mathemat- ics of the University of W¨urzburg. The authors are grateful to Michael M. Wolf for drawing their attention to several more recent publications on dilations of completely positive maps.

Appendix A: Topological Properties of Q ( ) S H

For the following deﬁnition, we refer to35 (Ch.VI.1).

Deﬁnition 13. Let and be arbitrary Banach spaces. X Y

(a) The strong operator topology (s.o.t.) on ( , ) is the locally convex topology induced B X Y by the family of seminorms of the form T T x with x . →k k ∈ X

(b) The weak operator topology (w.o.t.) on ( , ) is the locally convex topology induced B X Y by the family of seminorms of the form T y(T x) with (x, y) ′. →| | ∈X×Y 21 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Note that both topologies, the s.o.t. as well as the w.o.t., are Hausdorﬀ so limits are unique.

By the natural isomorphism ( 1( ))′ = ( ), see Section II B, one has the following B H ∼ B H equivalence: A net (T ) in ( 1( )) converges to T ( 1( )) in w.o.t. if and only if α α∈I B B H ∈ B B H

lim tr(BTα(A)) tr(BT (A)) =0 (A1) α∈I | − |

for all A 1( ) and B ( ). ∈ B H ∈ B H Remark 14 (Metrizability of s.o.t. and w.o.t. on bounded subsets). At this point one might ask whether the strong or weak operator topology is metrizable. If this is the case, closed and sequentially closed sets do coincide which, of course, is of interest for further investigations. The following is well known in the literature, cf.36 (Thm. 1.2 and 1.13): If is separable, X then the s.o.t. is metrizable on bounded subsets of ( ). If ′ is also separable, then the B X X w.o.t. is metrizable on bounded subsets of ( ). B X Now, recall that is assumed to be separable. Therefore it is evident that the subspace H of ﬁnite-rank operators ( ) and hence 1( ) itself, which is the ν -closure of ( ) (cf.35 F H B H 1 F H (Lemma XI.9.11)), is separable. Moreover, we already know from Proposition 2 that Q ( ) S H is a subset of the unit ball in ( 1( )). This implies that the s.o.t. on Q ( ) is metrizable B B H S H and thus convergence, closedness, continuity, etc. can be fully characterized by sequences. On the other hand, it is also well known that ( ) is not separable with respect to the B H operator norm topology as the non-separable space ℓ∞ can be isometrically embedded into ( ). Hence ( 1( ))′ is not separable and the above metrizability result does not apply to B H B H the w.o.t. on Q ( ). S H However, one could make use of the result that for convex sets in ( 1( )), the closures B B H with respect to the w.o.t. and the s.o.t coincide, cf.35 (Coro. VI.1.6). Therefore, in the proof of Theorem 1 one could focus on the s.o.t. On the other hand, Lemma 4 ﬀ. show that a direct approach via the w.o.t. is just as simple.

For clarity of the proof of Theorem 1, we ﬁrst state some auxiliary results.

Lemma 3. For every linear map S : 1( ) 1( ) the following statements are equivalent. B H → B G (a) S is positive.

(b) For all A 1( ) and B ( ) with A, B 0, one has tr(BS(A)) 0. ∈ B H ∈ B G ≥ ≥ 22 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Proof. (a) (b): For A, B 0 and S positive, we obtain S(A) 0 and thus ⇒ ≥ ≥ tr(BS(A)) = tr(√BS(A)√B) 0 , ≥ where √B 0 denotes the unique square root of B. ≥ (b) (a): Choosing B := x, x for arbitrary x yields B 0 and ⇒ h ·i ∈ G ≥ x, S(A)x = tr(BS(A)) 0 , h i ≥ for all A 0. Hence it follows S(A) 0 so S is positive. ≥ ≥ Lemma 4. Let (T ) be a net in ( 1( )) which converges to T ( 1( )) in w.o.t. α α∈I B B H ∈ B B H Then the following statements hold.

(a) If T is trace-preserving for all α I then T is trace-preserving. α ∈ (b) If T is positive for all α I then T is positive. α ∈

Proof. Both statements follow from (A1): (a) by choosing B = idH and (b) by applying Lemma 3 and taking into account that [0, ) is a closed subset of R. ∞ For the proof of our next result we recall that 1( Cm) and 1( ) Cm×m can be B H ⊗ B H ⊗ m m identiﬁed as follows. Any A ( C ) can be represented as A = Aij Eij with ∈ B H ⊗ Pi,j=1 ⊗ the standard basis (E )m of Cm×m and appropriate A ( ). Then, the following ij i,j=1 ij ∈ B H statements are equivalent24 (p. 33-34).

(a) A 1( Cm) ∈ B H ⊗ (b) A 1( ) for all i, j 1,...,m ij ∈ B H ∈{ } Lemma 5. Let (T ) be a net in ( 1( )) converging to T ( 1( )) in w.o.t. Then, α α∈I B B H ∈ B B H for all m N, the net (T id ) converges to T id ( 1( Cm)) in w.o.t. ∈ α ⊗ m α∈I ⊗ m ∈ B B H ⊗ Proof. According to (A1) we have to show

lim tr(B(Tα idm T idm)A) =0 (A2) α∈I | ⊗ − ⊗ | for all A 1( Cm) and B ( Cm). As seen above every A 1( Cm) and ∈ B H ⊗ ∈ B H ⊗ ∈ B H ⊗ B ( Cm) can be represented as ﬁnite linear combinations of elements A E ∈ B H ⊗ ij ⊗ ij ∈ 1( ) Cm×m and B E ( ) Cm×m, respectively, with i, j =1,...,m. Hence B H ⊗ ij ⊗ ij ∈ B H ⊗ m tr(B(Tα idm T idm)A) = tr(B((Tα T ) idm)A)= tr(Bij(Tα T )(Aji)) ⊗ − ⊗ − ⊗ Xi,j=1 − so convergence of (T id ) can easily be related to the convergence of (T ) . α ⊗ m α∈I α α∈I 23 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

Now for the main proof of this section.

Proof of Theorem 1. Since every linear and positive operator on 1( ) is naturally norm B H bounded as a simple consequence of 4 (Ch. 2, Lemma 2.1), the set Q ( ) of all Schrödinger S H channels is a bounded subset of ( 1( )). Now it is readily verified that Q ( ) is a convex B B H S H subsemigroup of ( 1( )), cf.26 (Ch.4.3). Next consider a net (T ) in Q ( ) converging B B H α α∈I S H to some T ( 1( )) in w.o.t. By Lemma 4 (a), the map T is trace-preserving and by ∈ B B H Lemma 5 (T id ) converges to T id with respect to the w.o.t. Then applying α ⊗ m α∈I ⊗ m Lemma 4 (b) to the net (T id ) yields that T is also m-positive for all m N. Hence α ⊗ m α∈I ∈ T is a Schrödinger quantum channel and Q ( ) is closed in ( 1( )) with respect to the S H B B H w.o.t. The well-known fact that the w.o.t. is weaker than the s.o.t. and the uniform operator topology concludes the proof.

Remark 15. Note that in the above proof we did not explicitely use the fact that domain and range of the operator T coincides. Therefore, the convexity and closedness results trivially extend to Q ( , ). S H G

Appendix B: Glossary on Dilations

For the sake of self-containedness, we recall some basic terminology concerning diﬀerent types of dilations of linear contractions. Let us start with the Banach space case.

Deﬁnition 16. Let be an arbitrary Banach space. X

1. Let T : be a linear contraction, i.e. T 1. A dilation ( , T,ˆ J, E) of T X → X k k ≤ Y consists of a Banach space and a triple of maps (Tˆ , J, E) with Y

T = E Tˆ J and E J = id , (B1) ◦ ◦ ◦ X

where the linear maps Tˆ, J and E satisfy:

(a) Tˆ : is a bi-isometry (i.e. Tˆ is bijective and T,ˆ Tˆ−1 are isometries) Y→Y (b) J : is an isometric embedding of in . X →Y X Y (c) E : has operator norm E =1. Y → X k k 24 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

2. Let S G be a semigroup of a group G and (T ) be a representation of S with ⊂ g g∈S values in the contraction semigroup of . A dilation ( , (Tˆ ) , J, E) of (T ) X Y g g∈G0 g g∈S consists of a Banach space , a subgroup G G and a triple ((Tˆ ) , J, E) with Y 0 ⊂ g g∈G0

T = E Tˆ J and E J = id g ◦ g ◦ ◦ X

for all g S G , where (Tˆ ) is a linear representation of G with values in the ∈ ⊂ 0 g g∈G0 0 isometry group of and J, E as before. Y Remark 17. 1. Note that E J = id implies that E is onto and J is injective. Fur- ◦ X thermore, J E : is a projection of norm 1 from onto the range of J. ◦ Y→Y Y

2. If S is assumed to be abelian and there exists a “dilation” of (Tg)g∈S such that Tˆg is well-deﬁned for all g S then (Tˆ ) obviously extends to a proper dilation in ∈ g g∈S the above sense, where G0 can be chosen to be the subgroup generated by S. For non-abelian S, however, this extension property is not obvious.

3. Choosing S := N and T := T n in Def. 16.2, where T : is a linear contraction, 0 n H → H we recover the “classical” concept of a linear diliation (see also Rem. 7.2). To distin- n guish such a dilation of T —which in our sense is actually a dilation of (T )n∈N0 —from a dilation of T in the sense of Def. 16.1 one sometimes calls the latter a “dilation of ﬁrst order”, cf.10.

4. Once continuity comes into play, things become more sublte as one can either require that the continuity properties of g T are preserved by g Tˆ (which in some 7→ g 7→ g cases is unfeasible) or allow that the continuity is relaxed, cf., e.g.,9 (Rem. 17.5). For our applications, however, this is not an issue as we are only concerned with the case

S = N0.

In the context of quantum channels (or, more generally, completely positive maps) various specializations of the above deﬁnitions to

abstract C∗- or W ∗-algebras • Heisenberg quantum channels • and Schr¨odinger quantum channels • 25 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups are available in the literature. For more details we refer to9,14.

Deﬁnition 18. Let be a unital C∗-algebra. A 1. Let T : be linear, completely positive and unital (i.e. identity preserving). A→A A dilation (A, Tˆ , J, E) of T consists of a unital C∗-algebra A and a triple of maps (Tˆ , J, E) with T = E Tˆ J and E J = id , (B2) ◦ ◦ ◦ A where Tˆ, J and E satisfy:

(a) Tˆ : A A is a -automorphism. → ∗ (b) J : A is a -homomorphism of into A. A→ ∗ A (c) E : A is linear and completely positive with operator norm E =1. →A k k 2. Let S G be a semigroup of a group G and let (T ) be a semigroup representa- ⊂ g g∈S tion of S with values in the set of completely positive, unital maps on . A dilation A (A, (Tˆ ) , J, E) of (T ) consists of a unital C∗-algebra A, a subgroup G G g g∈G0 g g∈S 0 ⊂ ˆ and a triple ((Tg)g∈G0 , J, E) with

T = E Tˆ J and E J = id , g ◦ g ◦ ◦ A for all g S G , where (Tˆ ) is a representation of G with values in the - ∈ ⊂ 0 g g∈G0 0 ∗ automorphism group of A and J, E as before.

If, in addition, J(idA) = idA then the dilation is said to be unital. On the other hand, if is even a W ∗-algebra, then all involved maps are in general assumed to be ultraweakly A continuous.

Remark 19. 1. Let , be unital C∗-algebras and let T : be unital. Then A B A → B positivity of T is equivalent to the norm condition T = 1, cf.28,37. In particular, one k k has T = T (id ) for every positive map T : so unitality of T implies that k k k A k A → B T is a contraction.

2. As every -homomorphism is trivially completely positive and every injective - ∗ ∗ homomorphism is always isometric, (B2) yields a dilation in the sense of (B1). Moreover, if a dilation is unital, then E is unital as well because (B2) implies

id =(E J)(id )= E(idA). A ◦ A 26 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

3. Every -automorphism Tˆ : A A on a unital C∗-algebra A is unital itself because of ∗ → −1 −1 Tˆ(idA)= Tˆ(idA) idA = Tˆ(idA)Tˆ(Tˆ (idA)) = Tˆ(idA T (idA)) = idA .

∗ ∗ 4. Let A be a C -subalgebra of a unital C -algebra A, i.e. idA A. Then a linear ∈ A ⊂ map E : A is said to be a conditional expectation (of A onto ) if it is completely →A A positive with norm E = 1 and satisﬁes k k E(AB)= AE(B) for all A and B A . (B3) ∈A ∈ Obviously, (B3) implies that E is a unital (cf. Rem. 19.1) projection onto , that is A E(A) = A for all A . The converse is also true, i.e. every projection E : A ∈ A → A of norm E = 1 is a conditional expectation, cf.37 (Thm. II.6.10.2) and38. Moreover, k k exploiting that E(B∗) = E(B)∗ for all B A, which results from the (complete) ∈ positivity of E, one can easily show that (B3) is equivalent to

E(BA)= E(B)A for all A and B A ∈A ∈ and, since is unital, also to A E(A BA )= A E(B)A for all A , A and B A . (B4) 1 2 1 2 1 2 ∈A ∈ In the literature, (B3) is often replaced by the “more symmetric” condition (B4). Now if A, but can be embedded into A via some unital, injective -homomorphism A 6⊂ A ∗ J : A, then E : A is said to be a conditional expectation with corresponding A→ →A injection J, if E is completely positive and E J = id . Note that in this case E ◦ A is also unital (because J is unital) and thus of norm one. Hence the composed map J E : A J( ) A is a projection of norm one and thus a conditional expectation ◦ → A ⊂ in the above sense. Thus every unital dilation gives rise to a conditional expectation E with corresponding injection J.

Now Deﬁnition 18 directly applies to Heisenberg channels. Taking into account that the only invertible channels are the unitary ones (cf. Prop. 1) we obtain the following concept.

Deﬁnition 20. 1. Let T Q ( ) be a Heisenberg quantum channel, i.e. T : ( ) ∈ H H B H → ( ) is linear, ultraweakly continuous, completely positive and unital. A unitary di- B H lation ( , U, J, E) of T consists of a Hilbert space and a triple of maps (U, J, E) K K with T = E Ad J and E J = id , ◦ U ◦ ◦ B(H) 27 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

where U, J and E satisfy

(a) U ( ) is unitary. ∈ B K (b) J : ( ) ( ) is an ultraweakly continuous -homomorphism of ( ) into B H → B K ∗ B H ( ). B K (c) E : ( ) ( ) is linear, ultraweakly continuous and completely positive with B K → B H operator norm E =1. k k 2. Let S G be a semigroup of a group G and let (T ) be a semigroup representation ⊂ g g∈S of S with values in the set of Heisenberg quantum channels Q ( ). A unitary dilation H H ( , (U ) , J, E) of (T ) consists of a Hilbert space , a subgroup G G and a K g g∈G0 g g∈S K 0 ⊂

triple ((Ug)g∈G0 , J, E) with

T = E Ad J and E J = id , g ◦ Ug ◦ ◦ B(H) for all g S G , where (U ) is a representation of G with values in the unitary ∈ ⊂ 0 g g∈G0 0 group on and J, E as before. K

If, in addition, J(idH) = idK then the dilation is said to be unital.

If the dilation is unital, then J Q ( , ) and E Q ( , ) are Heisenberg channels ∈ H H K ∈ H K H (cf. Rem. 19.2). Finally, this concept can be transferred to the Schr¨odinger quantum channels via duality (cf. Section II B).

Deﬁnition 21. 1. Let T Q ( ) be a Schr¨odinger quantum channel, i.e. T : 1( ) ∈ S H B H → 1( ) is linear, completely positive and trace-preserving. A unitary dilation ( , U, J, E) B H K of T consists of a Hilbert space and a triple of maps (U, J, E) with K

T = E Ad J and E J = id 1 , ◦ U ◦ ◦ B (H) where U, J and E satisfy

(a) U ( ) is unitary. ∈ B K (b) J : 1( ) 1( ) is linear and completely positive with operator norm J =1. B H → B K k k (c) E : 1( ) 1( ) is linear, completely positive and satisﬁes B K → B H E(E∗(B)A)= BE(A) for all B ( ) and A 1( ), (B5) ∈ B H ∈ B K where E∗ is the dual channel of E.

28 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

2. Let S G be a semigroup of a group G and let (T ) be a semigroup representation ⊂ g g∈S of S with values in the set of Schr¨odinger quantum channels Q ( ). A unitary dilation S H ( , (U ) , J, E) of (T ) consists of a Hilbert space , a subgroup G G and a K g g∈G g g∈S K 0 ⊂

triple ((Ug)g∈G0 , J, E) with

T = E Ad J and E J = id 1 , g ◦ Ug ◦ ◦ B (H)

for all g S G , where (U ) is a representation of G with values in the unitary ∈ ⊂ 0 g g∈G0 0 group on and J, E as before. K If, in addition, E is trace-preserving then the dilation is said to be trace-preserving.

Remark 22. 1. Property (B5) which looks quite similar to (B3) implies (by direct com- putation) that the dual channel E∗ is a -homomorphism. Moreover, E∗ is ultraweakly ∗ continuous as this holds for every dual channel. Conversely, for any ultraweakly continuous -homomorphism J from Definition 20 one can show that together with its ∗ pre-dual channel, it satisfies (B5). In this sense, the dilation definitions 20 and 21 are dual to each other. Similar as for (B3), one can conclude that (B5) is equivalent to

E(AE∗(B)) = E(A)B for all B ( ) and A 1( ). ∈ B H ∈ B K

2. If a dilation is trace-preserving, then J is trace-preserving as well (cf. Remark 19.2.) so in particular, J Q ( , ) and E Q ( , ) are Schr¨odinger channels. ∈ S H K ∈ S K H

3. Corollary 2 shows that for every Heisenberg channel T Q ( ) there exists a unitary ∈ H H (and even unital) dilation of T of the following type ( , Ad , i , tr ), where is a H⊗K U K ω K separable Hilbert space, ω D( ) a pure state and U ( ) a unitary operator. ∈ K ∈ B H⊗K Such a dilation is also said to be of tensor type. This result holds analogously for every T Q ( ) by Theorem 2. ∈ S H

REFERENCES

1W. Arveson, “Subalgebras of C∗-algebras II,” Acta Math. 128, 271–308 (1972). 2G. Lindblad, “On the Generators of Quantum Dynamical Semigroups,” Commun. Math. Phys. 48, 119–130 (1976).

29 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

3V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, “Completely Positive Dynamical Semigroups of N-Level Systems,” J. Math. Phys. 17, 821–825 (1976). 4E. Davies, Quantum Theory of Open Systems (Academic Press, St. John’s College, Oxford, 1976). 5E. B. Davies, “Dilations of Completely Positive Maps,” J. London Math. Soc. s2-17, 330–338 (1978). 6E. B. Davies, “Generators of Dynamical Semigroups,” J. Funct. Anal. 34, 421 – 432 (1979). 7D. E. Evans, “Positive Linear Maps on Operator Algebras,” Commun. Math. Phys. 48, 15–22 (1976). 8D. E. Evans and J. T. Lewis, “Dilations of Dynamical Semi-Groups,” Commun. Math. Phys. 50, 219–227 (1976). 9D. Evans and J. Lewis, Dilations of Irreversible Evolutions in Algebraic Quantum Theory, Series A: Theoretical Physics, Vol. 24 (Dublin Institute for Advanced Studies, Dublin, 1977). 10B. Kümmerer, “Markov Dilations on W ∗-Algebras,” J. Funct. Anal. 63, 139 – 177 (1985). 11B. V. R. Bhat, “An Index Theory For Quantum Dynamical Semigroups,” Trans. Amer. Math. Soc. 348, 561–583 (1996). 12B. V. R. Bhat and M. Skeide, “Tensor Product Systems of Hilbert Modules and Dilations of Completely Positive Semigroups,” Infin. Dimens. Anal. Quantum Probab. Relat. Top. 03, 519–575 (2000). 13P. Muhly and B. Solel, “Quantum Markov Processes,” Int. J. Math. 13, 863–906 (2002). 14D. J. Gaebler, “Continuous Unital Dilations of Completely Positive Semigroups,” J. Funct. Anal. 269, 998–1027 (2015). 15Y. Sawada, “A remark on the minimal dilation of the semigroup generated by a normal UCP-map,” (2018), to appear in Bull. Belg. Math. Soc. Simon Stevin; preprint at https://arxiv.org/abs/1705.09104. 16B. Sz.-Nagy, “Sur les contractions de l’espace de Hilbert,” Acta Sci. Math. (Szeged) 15, 87–92 (1953). 17B. Sz.-Nagy, “Modle de Jordan pour une classe d’oprateurs de l’espace de Hilbert,” Acta Sci. Math. (Szeged) 31, 91–115 (1970). 18E. Stroescu, “Isometric Dilations of Contractions on Banach Spaces,” Pacific J. Math. 47, 257–262 (1973).

30 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

19For G = Z one can show, e.g., via Corollary 2, that Davies’ construction leads to a dilation n † † of the form T (A)=tr ′ (U U (A id 2 Z )U U). On the one hand, (U U) Z does ω⊗ω n ⊗ L ( ,H)⊗K n n n∈ obviously not yield a unitary representation of G = Z although (Un)n∈Z does. On the other † † hand, the map X tr tr ′ (U XU) = tr ′ (U XU) can in general not be replaced by 7→ ω ω ω⊗ω X tr (X) so (2) does not have the desired structure. 7→ ρ 20For unitary dilations on Hilbert spaces the condition is necessary and sufficient (cf. Remark 7.2) and for W ∗-algebras Kümmerer proved sufficiency10 (Prop. 2.2.7). 21R. Kadison and J. Ringrose, Fundamentals of the Theory of Operator Algebras: Volume 1 - Elementary theory, Fundamentals of the Theory of Operator Algebras (Academic Press, New York, 1983). 22M. Reed and B. Simon, Functional Analysis, Methods of Modern Mathematical Physics, Vol. 1 (Academic Press, San Diego, 1980). 23R. Meise and D. Vogt, Introduction to Functional Analysis, Oxford Graduate Texts in Mathematics (Oxford University Press, New York, 1997). 24K. Kraus, States, Effects, and Operations, Lecture Notes in Physics (Springer-Verlag Berlin Heidelberg, 1983). 25M. M. Wolf and J. I. Cirac, “Dividing Quantum Channels,” Commun. Math. Phys. 279, 147–168 (2008). 26T. Heinosaari and M. Ziman, The Mathematical Language of Quantum Theory: From Uncertainty to Entanglement (Cambridge University Press, Cambridge, 2012). 27Due to the fact that S is linear and 1( ) is separable (as was assumed to be separable, B H H cf. Remark 14), ultraweak continuity of S and ultraweak sequential continuity coincide, cf.35 (Thm. V.5.1). 28B. Russo and H. A. Dye, “A Note on Unitary Operators in C∗-algebras,” Duke Math. J. 33, 413–416 (1966). 29W. Stinespring, “Positive Functions on C∗-Algebras,” Proc. Amer. Math. Soc. 6, 211–216 (1955). 30S. Attal, “Quantum Channels,” (2014), online course: http://math.univ-lyon1.fr/~attal/Quantum_Channels.pdf. Institut Camille Jordan, University of Lyon. 31B. Sz.-Nagy and C. Foia¸s, Harmonic Analysis of Operators on Hilbert Space (North- Holland Pub. Co., Amsterdam-Budapest, 1970).

31 Unitary Dilations of Discrete-Time Quantum-Dynamical Semigroups

32C. Foias and A. Frazho, The Commutant Lifting Approach to Interpolation Problems, Operator Theory: Advances and Applications, Vol. 44 (Birkhuser, Basel, 1990). 33More precisely, said Lemma shows the existence of the decompositions ∞ = ˆ∞ H H ⊕ H and = ˆ . Then it is easy to see that ˆ is orthogonal to ˆ because H∞ H ⊕ H−∞ H∞ H−∞ v,ˆ U −nh = U nv,ˆ h =0 forv ˆ ˆ , h and n N . h i h i ∈ H∞ ∈ H ∈ 0 34F. vom Ende, “Unitary Dilations of Discrete Quantum-Dynamical Semi-Groups,” (2016), master thesis. 35N. Dunford and J. Schwartz, Linear Operators: Spectral Theory, Pure and applied mathematics (Interscience Publishers, New York, 1963). 36J. Kim, “Compactness in B(X),” J. Math. Anal. Appl. 320, 619 – 631 (2006). 37B. Blackadar, Operator Algebras, Encyclopaedia of Mathematical Sciences, Vol. 122 (Springer-Verlag Berlin Heidelberg, 2006). 38J. Tomiyama, “On the Projection of Norm One in W ∗-Algebras,” Proc. Japan Acad. 33, 608–612 (1957).