universal recoverability in quantum information · if ˙is a gaussian state and nis a gaussian...

Universal recoverability in quantum information

Mark M. Wilde

Hearne Institute for Theoretical Physics,Department of Physics and Astronomy,Center for Computation and Technology,

Louisiana State University,Baton Rouge, Louisiana, USA

[email protected]

Based on arXiv:1505.04661, 1506.00981, 1509.07127,1511.00267, 1601.01207, 1608.07569, 1610.01262

with Berta, Buscemi, Das, Dupuis, Junge, Lami, Lemm, Renner, Sutter, Winter

QMATH 2016, Atlanta, Georgia, USA

Mark M. Wilde (LSU) 1 / 28

Main message

Entropy inequalities established in the 1970s are a mathematicalconsequence of the postulates of quantum physics

They have a number of applications: for determining the ultimatelimits on many physical processes (communication, thermodynamics,uncertainty relations, cloning)

Many of these entropy inequalities are equivalent to each other, so wecan say that together they constitute a fundamental law of quantuminformation theory

There has been recent interest in refining these inequalities, trying tounderstand how well one can attempt to reverse an irreversiblephysical process

We discuss progress in this direction


Background — entropies

Umegaki relative entropy [Ume62]

The quantum relative entropy is a measure of dissimilarity between twoquantum states. Defined for state ρ and positive semi-definite σ as

D(ρ‖σ) ≡ Tr{ρ[log ρ− log σ]}

whenever supp(ρ) ⊆ supp(σ) and +∞ otherwise

Operational interpretation (quantum Stein’s lemma) [HP91, NO00]

Given are n quantum systems, all of which are prepared in either the stateρ or σ. With a constraint of ε ∈ (0, 1) on the Type I error ofmisidentifying ρ, then the optimal error exponent for the Type II error ofmisidentifying σ is D(ρ‖σ).


Fundamental law of quantum information theory

Monotonicity of quantum relative entropy [Lin75, Uhl77]

Let ρ be a state, let σ be positive semi-definite, and let N be a quantumchannel. Then

D(ρ‖σ) ≥ D(N (ρ)‖N (σ))

“Distinguishability does not increase under a physical process”Characterizes a fundamental irreversibility in any physical process

Proof approaches (among many)

Lieb concavity theorem [L73]

relative modular operator method (see, e.g., [NP04])

quantum Stein’s lemma [BS03]


Equality conditions [Pet86, Pet88]

When does equality in monotonicity of relative entropy hold?

D(ρ‖σ) = D(N (ρ)‖N (σ)) iff ∃ a recovery map Pσ,N such that

ρ = (Pσ,N ◦ N )(ρ), σ = (Pσ,N ◦ N )(σ)

This “Petz” recovery map has the following explicit form [HJPW04]:

Pσ,N (ω) ≡ σ1/2N †(

(N (σ))−1/2ω(N (σ))−1/2)σ1/2

Classical case: Distributions pX and qX and a channel N (y |x). Thenthe Petz recovery map P(x |y) is given by the Bayes theorem:

P(x |y)qY (y) = N (y |x)qX (x)

where qY (y) ≡∑

x N (y |x)qX (x)


Approximate case

Approximate case would be useful for applications

Approximate case for monotonicity of relative entropy

What can we say when D(ρ‖σ)− D(N (ρ)‖N (σ)) = ε ?

Does there exist a CPTP map R that recovers σ perfectly from N (σ)while recovering ρ from N (ρ) approximately? [WL12]


One-shot measure of similarity for quantum states

Fidelity [Uhl76]

Fidelity between ρ and σ is F (ρ, σ) ≡ ‖√ρ√σ‖21. Has a one-shot

operational interpretation as the probability with which a purification of ρcould pass a test for being a purification of σ.


New result of [Wil15, JSRWW15]

Recoverability Theorem

Let ρ and σ satisfy supp(ρ) ⊆ supp(σ) and let N be a channel. Then

D(ρ‖σ)− D(N (ρ)‖N (σ)) ≥ −∫ ∞−∞

dt p(t) log[F(ρ,Pt/2

σ,N (N (ρ)))],

where p(t) is a distribution and Ptσ,N is a rotated Petz recovery map:

Ptσ,N (·) ≡

(Uσ,t ◦ Pσ,N ◦ UN (σ),−t

)(·) ,

Pσ,N is the Petz recovery map, and Uσ,t and UN (σ),−t are defined from

Uω,t(·) ≡ ωit (·)ω−it , with ω a positive semi-definite operator.

Two tools for proof

Renyi generalization of a relative entropy difference and theStein–Hirschman operator interpolation theorem


Universal Recovery

Universal Recoverability Corollary

Let ρ and σ satisfy supp(ρ) ⊆ supp(σ) and let N be a channel. Then

D(ρ‖σ)− D(N (ρ)‖N (σ)) ≥ − log F (ρ,Rσ,N (N (ρ))),

where

Rσ,N ≡∫ ∞−∞

dt p(t)Pt/2σ,N

(follows from concavity of logarithm and fidelity)


Universal Distribution

−3 −2 −1 0 1 2 3

0.8

0.6

0.4

0.2

0

t

p(t)

Figure: This plot depicts the probability density p(t) := π2

(cosh(πt) + 1

)−1as a

function of t ∈ R. We see that it is peaked around t = 0 which corresponds tothe Petz recovery map.


Renyi generalizations of a relative entropy difference

Definition from [BSW14, SBW14]

∆α(ρ, σ,N ) ≡ 2

α′log∥∥∥(N (ρ)−α

′/2N (σ)α′/2 ⊗ IE

)Uσ−α

′/2ρ1/2∥∥∥2α,

where α ∈ (0, 1) ∪ (1,∞), α′ ≡ (α− 1)/α, and US→BE is an isometricextension of N .

Important properties

limα→1

∆α(ρ, σ,N ) = D(ρ‖σ)− D(N (ρ)‖N (σ)).

∆1/2(ρ, σ,N ) = − log F (ρ,Pσ,N (N (ρ))).


Stein–Hirschman operator interpolation theorem (setup)

Let S ≡ {z ∈ C : 0 < Re {z} < 1} , and let L (H) be the space of boundedlinear operators acting on H. Let G : S → L(H) be an operator-valuedfunction bounded on S , holomorphic on S , and continuous on theboundary ∂S . Let θ ∈ (0, 1) and define pθ by

1

pθ=

1− θp0

+θ

p1,

where p0, p1 ∈ [1,∞].


Stein–Hirschman operator interp. theorem (statement)

Then the following bound holds

log ‖G (θ)‖pθ ≤∫ ∞−∞

dt(αθ(t) log

[‖G (it)‖1−θp0

]+ βθ(t) log

[‖G (1 + it)‖θp1

]),

where αθ(t) ≡ sin(πθ)

2(1− θ) [cosh(πt)− cos(πθ)],

βθ(t) ≡ sin(πθ)

2θ [cosh(πt) + cos(πθ)],

limθ↘0

βθ(t) = p(t).


Proof of Recoverability Theorem

Tune parameters

Pick G (z) ≡(

[N (ρ)]z/2 [N (σ)]−z/2 ⊗ IE

)Uσz/2ρ1/2,

p0 = 2, p1 = 1, θ ∈ (0, 1) ⇒ pθ =2

1 + θ

Evaluate norms

‖G (it)‖2 =∥∥∥(N (ρ)it/2N (σ)−it/2 ⊗ IE

)Uσit/2ρ1/2

∥∥∥2≤∥∥∥ρ1/2∥∥∥

2= 1,

‖G (1 + it)‖1 =[F(ρ,Pt/2

σ,N (N (ρ)))]1/2

.


Proof of Recoverability Theorem (ctd.)

Apply the Stein–Hirschman theorem

log∥∥∥([N (ρ)]θ/2 [N (σ)]−θ/2 ⊗ IE

)Uσθ/2ρ1/2

∥∥∥2/(1+θ)

≤∫ ∞−∞

dt βθ(t) log

[F(ρ, (Pt/2

σ,N ◦ N )(ρ))θ/2]

.

Final step

Apply a minus sign, multiply both sides by 2/θ, and take the limit asθ ↘ 0 to conclude.


Specializing to the Holevo Bound

Specializing to the Holevo bound leads to a refinement. Given

ρXB ≡∑x

pX (x)|x〉〈x |X⊗ρxB , ωXY ≡∑y

〈ϕy |BρXB |ϕy 〉B |y〉〈y |Y .

Then the following inequality holds

I (X ;B)ρ − I (X ;Y )ω ≥ −2 log∑x

pX (x)√F (ρxB , EB(ρxB)),

where EB is an entanglement-breaking map of the form

EB(·) ≡∫ ∞−∞

dt β0(t)∑y

〈ϕy |B(·)|ϕy 〉Bρ(1+it)/2B |ϕy 〉〈ϕy |Bρ

(1−it)/2B

〈ϕy |BρB |ϕy 〉B.


Applying to Entropy

Special case: Entropy gain (also called Entropy Production)

Specializing to entropy gives the following bound for a unitalquantum channel N :

H(N (ρ))− H(ρ) ≥ − log F (ρ,N †(N (ρ)))

A different approach [BDW16] gives a stronger bound and applies tomore general maps. For N a positive, subunital, trace-preserving map:

H(N (ρ))− H(ρ) ≥ D(ρ‖N †(N (ρ))) ≥ 0


Application to entropy uncertainty relations [BWW15]

Let ρABE be a state for Alice, Bob, and Eve, and let X ≡ {PxA} and

Z = {QzA} be projection-valued measures for Alice’s system

Define the post-measurement states:

σXBE ≡∑x

|x〉〈x |X ⊗ σxBE where

σxBE ≡ TrA{(PxA ⊗ IBE )ρABE}

ωZBE ≡∑z

|z〉〈z |Z ⊗ ωzBE where

ωzBE ≡ TrA{(Qz

A ⊗ IBE )ρABE}

Then

H(Z |E )ω + H(X |B)σ

≥ − log maxx ,z‖Px

AQzA‖2∞ − log F (ρAB ,RXB→AB(σXB))


Case of quantum Gaussian channels

If σ is a Gaussian state and N is a Gaussian channel, then the Petzrecovery map Pσ,N is a Gaussian channel (result with Lami and Das).

We have an explicit form for the Petz recovery map in terms of itsaction on the mean vector and covariance matrix of a quantumGaussian state.

We have the same for rotated Petz recovery maps.


Quantum cloning, partial trace, and recovery [LW16]

Let ω(n) be a state with support in the symmetric subspace of(Cd)⊗n, let πd ,nsym denote the maximally mixed state on this symmetricsubspace, let Ck→n denote a universal quantum cloning machine, andPn→k the symmetrize partial trace. Then

D(ω(n)‖πd ,nsym) ≥ D(Pn→k(ω(n))‖Pn→k(πd ,nsym))

+ D(ω(n)‖(Ck→n ◦ Pn→k)(ω(n))).

With the same notation, the following inequality holds

D(ω(k)‖πd ,ksym) ≥ D(Ck→n(ω(k))‖Ck→n(πd ,ksym))

+ D(ω(k)‖(Pn→k ◦ Ck→n)(ω(k))).

So cloning machines and partial trace are dual to each other in theabove sense.


Generality of approach [DW15]

Technique is very general and can be used to prove inequalities fornorms of multiple operators chained together (called “Swiveled RenyiEntropies” in [DW15], due to presence of “unitary swivels”)

Example: The following quantity

L′α (ρA1···Al) ≡ 2

α′max{VρS}S

log

∥∥∥∥∥[ ∏S∈P ′

ρ−aSα′/2S VρS

]ρ1/2A1···Al

∥∥∥∥∥2α

,

where α′ = (α− 1) /α is monotone increasing in α for α ∈ [1/2,∞].

Another example: for positive semi-definite operators C1, . . . , CL, aunitary VCi

commuting with Ci , and p ≥ 1, the quantity

maxVC1

,...,VCL

∥∥∥C 1/p1 VC1 · · ·C

1/pL VCL

∥∥∥pp

is monotone decreasing in p for p ≥ 1. (See also [Wil16])


Generality of approach (ctd.) [DW15]

Another example: Let C1, . . . ,CL be positive semi-definite operators,and let p > q ≥ 1. Then the following holds [DW15, Wil16]:

log∥∥∥C 1/p

1 C1/p2 · · ·C 1/p

L

∥∥∥pp

≤∫ ∞−∞

dt βq/p(t) log∥∥∥C (1+it)/q

1 C(1+it)/q2 · · ·C (1+it)/q

L

∥∥∥qq.

By taking a limit: Let C1, . . . ,CL be positive definite operators, andlet q ≥ 1. Then the following inequality holds [DW15, Wil16]:

log Tr {exp {logC1 + · · ·+ logCL}}

≤∫ ∞−∞

dt β0(t) log∥∥∥C (1+it)/q

1 C(1+it)/q2 · · ·C (1+it)/q

L

∥∥∥qq.


Conclusions

The result in [Wil15, JSRWW15] applies to relative entropydifferences, has a brief proof, and yields a universal recovery map(depending only on σ and N ).

Applications in a variety of areas, including entropy gain [BDW16],entropic uncertainty [BWW15], quantum cloning [LW16], quantumGaussian channels, etc.

Later results of [DW15] clarify how the approach is very general andleads to many other inequalities

It has been conjectured that the recovery map can be the Petzrecovery map alone (not a rotated Petz map), but it is unclearwhether this will be true.


References I

[BLW14] Mario Berta, Marius Lemm, and Mark M. Wilde. Monotonicity of quantumrelative entropy and recoverability. December 2014. arXiv:1412.4067.

[BWW15] Mario Berta, Stephanie Wehner, and Mark M. Wilde. Measurementreversibility and entropic uncertainty. November 2015. arXiv:1511.00267.

[BS03] Igor Bjelakovic and Rainer Siegmund-Schultze. Quantum Stein’s lemmarevisited, inequalities for quantum entropies, and a concavity theorem of Lieb. July2003. arXiv:quant-ph/0307170.

[BSW14] Mario Berta, Kaushik Seshadreesan, and Mark M. Wilde. Renyi generalizationsof the conditional quantum mutual information. March 2014. arXiv:1403.6102.

[BDW16] Francesco Buscemi, Siddhartha Das, and Mark M. Wilde. Approximatereversibility in the context of entropy gain, information gain, and completepositivity . January 2016. arXiv:1601.01207.

[DW15] Frederic Dupuis and Mark M. Wilde. Swiveled Renyi entropies. June 2015.arXiv:1506.00981.


References II

[FR14] Omar Fawzi and Renato Renner. Quantum conditional mutual information andapproximate Markov chains. October 2014. arXiv:1410.0664.

[HJPW04] Patrick Hayden, Richard Jozsa, Denes Petz, and Andreas Winter. Structureof states which satisfy strong subadditivity of quantum entropy with equality.Communications in Mathematical Physics, 246(2):359–374, April 2004.arXiv:quant-ph/0304007.

[HP91] Fumio Hiai and Denes Petz. The proper formula for relative entropy and itsasymptotics in quantum probability. Communications in Mathematical Physics,143(1):99–114, December 1991.

[JSRWW15] Marius Junge, David Sutter, Renato Renner, Mark M. Wilde, and AndreasWinter. Universal recovery from a decrease of quantum relative entropy. September2015. arXiv:1509.07127.

[Lin75] Goran Lindblad. Completely positive maps and entropy inequalities.Communications in Mathematical Physics, 40(2):147–151, June 1975.

[L73] Elliott H. Lieb. Convex Trace Functions and the Wigner-Yanase-DysonConjecture. Advances in Mathematics, 11(3), 267–288, December 1973.


References III

[LR73] Elliott H. Lieb and Mary Beth Ruskai. Proof of the strong subadditivity ofquantum-mechanical entropy. Journal of Mathematical Physics, 14(12):1938–1941,December 1973.

[LW14] Ke Li and Andreas Winter. Squashed entanglement, k-extendibility, quantumMarkov chains, and recovery maps. October 2014. arXiv:1410.4184.

[LW16] Marius Lemm and Mark M. Wilde. Information-theoretic limitations onapproximate quantum cloning and broadcasting. August 2016. arXiv:1608.07569.

[NO00] Hirsohi Nagaoka and Tomohiro Ogawa. Strong converse and Stein’s lemma inquantum hypothesis testing. IEEE Transactions on Information Theory,46(7):2428–2433, November 2000. arXiv:quant-ph/9906090.

[NP04] Michael A. Nielsen and Denes Petz. A simple proof of the strong subadditivityinequality. arXiv:quant-ph/0408130.

[Pet86] Denes Petz. Sufficient subalgebras and the relative entropy of states of a vonNeumann algebra. Communications in Mathematical Physics, 105(1):123–131,March 1986.


References IV

[Pet88] Denes Petz. Sufficiency of channels over von Neumann algebras. QuarterlyJournal of Mathematics, 39(1):97–108, 1988.

[SBW14] Kaushik P. Seshadreesan, Mario Berta, and Mark M. Wilde. Renyi squashedentanglement, discord, and relative entropy differences. October 2014.arXiv:1410.1443.

[SFR15] David Sutter, Omar Fawzi, and Renato Renner. Universal recovery map forapproximate Markov chains. April 2015. arXiv:1504.07251.

[Uhl76] Armin Uhlmann. The “transition probability” in the state space of a *-algebra.Reports on Mathematical Physics, 9(2):273–279, 1976.

[Uhl77] Armin Uhlmann. Relative entropy and the Wigner-Yanase-Dyson-Lieb concavityin an interpolation theory. Communications in Mathematical Physics, 54(1):21–32,1977.

[Ume62] Hisaharu Umegaki. Conditional expectations in an operator algebra IV (entropyand information). Kodai Mathematical Seminar Reports, 14(2):59–85, 1962.


References V

[Wil15] Mark M. Wilde. Recoverability in quantum information theory. May 2015.arXiv:1505.04661.

[Wil16] Mark M. Wilde. Monotonicity of p-norms of multiple operators via unitaryswivels. October 2016. arXiv:1610.01262.

[WL12] Andreas Winter and Ke Li. A stronger subadditivity relation?http://www.maths.bris.ac.uk/∼csajw/stronger subadditivity.pdf, 2012.


universal recoverability in quantum information · if ˙is a gaussian state and nis a gaussian...

Documents