1. Faithful Squashed Entanglement Fernando G.S.L. Brandão 1 Matthias Christandl 2 Jon Yard 3 1. Universidade Federal de Minas Gerais, Brazil 2. ETH Zürich, Switzerland 3. Los Alamos Laboratory, USA with applications to separability testing and quantum Merlin-Arthur games

2. Mutual Information vs Conditional Mutual Information Mutual Information: Measures the correlations of A and B in ρ AB I(A:B) ρ := S(A) ρ + S(B) ρ – S(AB) ρ

3. Mutual Information vs Conditional Mutual Information Mutual Information: Measures the correlations of A and B in ρ AB I(A:B) ρ := S(A) ρ + S(B) ρ – S(AB) ρ Always positive: I(A:B) ρ ≥ 0 (subadditivity of entropy)

4. Mutual Information vs Conditional Mutual Information Mutual Information: Measures the correlations of A and B in ρ AB I(A:B) ρ := S(A) ρ + S(B) ρ – S(AB) ρ Always positive: I(A:B) ρ ≥ 0 (subadditivity of entropy) When does it vanish? I(A:B) ρ = 0 iff ρ AB = ρ A ρ B

5. Mutual Information vs Conditional Mutual Information Mutual Information: Measures the correlations of A and B in ρ AB I(A:B) ρ := S(A) ρ + S(B) ρ – S(AB) ρ Always positive: I(A:B) ρ ≥ 0 (subadditivity of entropy) When does it vanish? I(A:B) ρ = 0 iff ρ AB = ρ A ρ B Approximate version? Pinsker’s inequality: Remark: dimension-independent! Useful in many application in QIT (e.g. decoupling, QKD, …)

6. Mutual Information vs Conditional Mutual Information Conditional Mutual Information: Measures the correlations of A and B relative to E in ρ ABE I(A:B|E) ρ := S(AE) ρ + S(BE) ρ – S(ABE) ρ – S(E) ρ

7. Mutual Information vs Conditional Mutual Information Conditional Mutual Information: Measures the correlations of A and B relative to E in ρ ABE I(A:B|E) ρ := S(AE) ρ + S(BE) ρ – S(ABE) ρ – S(E) ρ Always positive: I(A:B|E) ρ ≥ 0 (strong-subadditivity of entropy) (Lieb, Ruskai ‘73)

8. Mutual Information vs Conditional Mutual Information Conditional Mutual Information: Measures the correlations of A and B relative to E in ρ ABE I(A:B|E) ρ := S(AE) ρ + S(BE) ρ – S(ABE) ρ – S(E) ρ Always positive: I(A:B|E) ρ ≥ 0 (strong-subadditivity of entropy) When does it vanish? I(A:B|E) ρ = 0 iff ρ ABE is a “Quantum Markov Chain State” E.g. (Hayden, Jozsa, Petz, Winter ‘04) (Lieb, Ruskai ‘73)

9. Mutual Information vs Conditional Mutual Information Conditional Mutual Information: Measures the correlations of A and B relative to E in ρ ABE I(A:B|E) ρ := S(AE) ρ + S(BE) ρ – S(ABE) ρ – S(E) ρ Always positive: I(A:B|E) ρ ≥ 0 (strong-subadditivity of entropy) When does it vanish? I(A:B|E) ρ = 0 iff ρ ABE is a “Quantum Markov Chain State” E.g. Approximate version??? … (Hayden, Jozsa, Petz, Winter ‘04) (Lieb, Ruskai ‘73)

10. Outline I(A:B|E)≈0 characterization Applications: Squashed Entanglement de Finetti-type bounds Algorithm for Separability A new characterization of QMA Proof

11. No-Go For Approximate Version A naïve guess for approximate version (à la Pinsker):

12. No-Go For Approximate Version A naïve guess for approximate version (à la Pinsker): It fails badly! (Christandl, Schuch, Winter ’08) = O(|A| -1 ) = Ω(1) E.g. Antisymmetric Werner state

13. Main Result Thm: (B., Christandl, Yard ’10)

14. Main Result Thm: (B., Christandl, Yard ’10) (Euclidean norm or (one-way) LOCC norm)

15. Main Result Thm: (B., Christandl, Yard ’10) (Euclidean norm or (one-way) LOCC norm) Pointed out yesterday by David Reeb

16. Main Result Thm: (B., Christandl, Yard ’10) The Euclidean (Frobenius) norm: ||X|| 2 = tr(X T X) 1/2 (Euclidean norm or (one-way) LOCC norm)

17. Main Result Thm: (B., Christandl, Yard ’10) The trace norm: ||X|| 1 = ½ + ½ max 0≤A≤I |tr(AX)| ||ρ-σ|| 1 : optimal bias (Euclidean norm or (one-way) LOCC norm)

18. Main Result Thm: (B., Christandl, Yard ’10) The LOCC norm: ||X|| LOCC = ½ + ½ max 0≤A≤I |tr(AX)| : {A, I-A} in LOCC ||ρ-σ|| LOCC : optimal bias by LOCC (Euclidean norm or (one-way) LOCC norm)

19. Main Result Thm: (B., Christandl, Yard ’10) The one-way LOCC norm: ||X|| LOCC(1) = ½ + ½ max 0≤A≤I |tr(AX)| : {A, I-A} in one-way LOCC ||ρ-σ|| LOCC : optimal bias by one-way LOCC (Euclidean norm or (one-way) LOCC norm)

20. The Power of LOCC Thm: (B., Christandl, Yard ’10) (Euclidean norm or (one-way) LOCC norm) (Matthews, Wehner, Winter ‘09) For X in A B Interesting one, uses a covariant random local measurement

21. Squashed Entanglement (Christandl, Winter ‘04) Squashed entanglement: E sq (ρ AB ) = inf π { ½ I(A:B|E) π : tr E (π ABE ) = ρ AB } Open question: Is it faithful? i.e. Is E sq (ρ AB ) > 0 for every entangled ρ AB ?

22. Squashed Entanglement (Christandl, Winter ‘04) Squashed entanglement: E sq (ρ AB ) = inf π { ½ I(A:B|E) π : tr E (π ABE ) = ρ AB } Open question: Is it faithful? i.e. Is E sq (ρ AB ) > 0 for every entangled ρ AB ? Corollary:

23. Squashed Entanglement (Christandl, Winter ‘04) Squashed entanglement: E sq (ρ AB ) = inf π { ½ I(A:B|E) π : tr E (π ABE ) = ρ AB } Corollary Proof: From Follows: General two-way LOCC: monotonicity of squashed entanglement under LOCC

24. Entanglement Zoo

26. Entanglement Monogamy Classical correlations are shareable: Def. ρ AB is k-extendible if there is ρ AB1…Bk s.t for all j in [k] tr \ Bj (ρ AB1…Bk ) = ρ AB Separable states are k-extendible for every k. A B1B1 B2B2 B3B3 B4B4 BkBk …

27. Entanglement Monogamy Quantum correlations are non-shareable: ρ AB entangled iff ρ AB not k-extendible for some k - Follows from: Quantum de Finetti Theorem (Stormer ’69, Hudson & Moody ’76, Raggio & Werner ’89) E.g. - Any pure entangled state is not 2-extendible - The d x d antisymmetric state is not d-extendible

28. Entanglement Monogamy Quantitative version: For any k-extendible ρ AB, - Follows from: finite quantum de Finetti Theorem (Christandl, König, Mitchson, Renner ‘05)

29. Entanglement Monogamy Quantitative version: For any k-extendible ρ AB, - Follows from: finite quantum de Finetti Theorem (Christandl, König, Mitchson, Renner ‘05) Close to optimal: there is a state ρ AB s.t. (guess which? ) For other norms ( ||*|| 2, ||*|| LOCC, … ) no better bound known.

30. Exponentially Improved de Finetti type bound Corollary For any k-extendible ρ AB, with||*|| equals ||*|| 2 or ||*|| LOCC Bound proportional to the (square root) of the number of qubits: exponential improvement over previous bound

31. Exponentially Improved de Finetti type bound Proof: E sq satisfies monogamy relation (Koashi, Winter ’05) For ρ AB k-extendible: Corollary For any k-extendible ρ AB, with||*|| equals ||*|| 2 or ||*|| LOCC

32. Exponentially Improved de Finetti type bound (Close-to-Optimal) There is k-extendible state ρ AB s.t. Corollary For any k-extendible ρ AB, with||*|| equals ||*|| 2 or ||*|| LOCC

33. Exponentially Improved de Finetti type bound

34. The separability problem When is ρ AB entangled? - Decide if ρ AB is separable or ε-away from separable Beautiful theory behind it (PPT, entanglement witnesses, symmetric extensions, etc) Horribly expensive algorithms State-of-the-art: 2 O(|A|log (1/ε)) time complexity (Doherty, Parrilo, Spedalieri ‘04)

35. The separability problem When is ρ AB entangled? - Decide if ρ AB is separable or ε-away from separable (Gurvits ‘02) NP-hard with ε=1/exp(d) (d:= (|A||B|) 1/2 ) (Gharibian ‘08, Beigi ‘08) NP-hard with ε=1/poly(d) (Beigi&Shor ‘08) Favorite separability tests fail (Harrow&Montanaro ‘10) No exp(O(d 2-δ) ) time algorithm for membership in any convex set within ε=Ω(1) trace distance to SEP, unless ETH fails ETH (Exponential Time Hypothesis): SAT cannot be solved in 2 o(n) time Hardness results: (Impagliazzo&Paruti ’99)

36. Quasi-polynomial Algorithm Corollary There is a exp(O(ε -2 log|A|log|B|)) time algorithm for deciding separability (in ||*|| 2 or ||*|| LOCC )

37. Quasi-polynomial Algorithm Corollary There is a exp(O(ε -2 log|A|log|B|)) time algorithm for deciding separability (in ||*|| 2 or ||*|| LOCC ) The idea (Doherty, Parrilo, Spedalieri ‘04) Search for a k=O(log|A|/ε 2 ) extension of ρ AB by SDP Complexity SDP of size |A| 2 |B| 2k = exp(O(ε -2 log|A|log|B|))

38. Quasi-polynomial Algorithm Corollary There is a exp(O(ε -2 log|A|log|B|)) time algorithm for deciding separability (in ||*|| 2 or ||*|| LOCC ) NP-hardness for ε = 1/poly(d) is shown using ||*|| 2 From corollary: the problem in ||*|| 2 cannot be NP-hard for ε = 1/polylog(d), unless ETH fails

39. Best Separable State Problem BSS(ε) Problem: Given X, approximate to additive error ε Corollary There is a exp(O(ε -2 log|A| log|B| (||X|| 2 ) 2 )) time algorithm for BSS(ε)

40. Best Separable State Problem BSS(ε) Problem: Given X, approximate to additive error ε Corollary There is a exp(O(ε -2 log|A| log|B| (||X|| 2 ) 2 )) time algorithm for BSS(ε) The idea Optimize over k=O(log|A|ε -2 (||X|| 2 ) 2 ) extension of ρ AB by SDP

41. Best Separable State Problem BSS(ε) Problem: Given X, approximate to additive error ε Corollary There is a exp(O(ε -2 log|A| log|B| (||X|| 2 ) 2 )) time algorithm for BSS(ε) (Harrow and Montanaro ‘10): BSS(ε) for ε=Ω(1) and ||X|| ∞ ≤ 1 cannot be solved in exp(O(log 1-ν |A| log 1-μ |B|)) time for any ν + μ > 0 unless ETH fails

42. QMA

44. - Quantum analogue of NP (or MA) - Local Hamiltonian Problem, … Is QMA a robust complexity class? (Aharonov, Regev ‘03) superverifiers doesn’t help (Marriott, Watrous ‘05) Exponential amplification with fixed proof size (Beigi, Shor, Watrous ‘09) logarithmic size interaction doesn’t help

45. New Characterization QMA Corollary QMA doesn’t change allowing k = O(1) different proofs if the verifier can only apply LOCC measurements in the k proofs

46. New Characterization QMA Corollary QMA doesn’t change allowing k = O(1) different proofs if the verifier can only apply LOCC measurements in the k proofs Def QMA m,s,c (k): analogue of QMA with k proofs, proof size m, soundness s and completeness c.

47. New Characterization QMA Corollary QMA doesn’t change allowing k = O(1) different proofs if the verifier can only apply LOCC measurements in the k proofs Def QMA m,s,c (k): analogue of QMA with k proofs, proof size m, soundness s and completeness c. Def LOCCQMA m,s,c (k): analogue of QMA with k proofs, proof size m, soundness s, completeness c and LOCC verification procedure along the k proofs.

48. New Characterization QMA Corollary QMA = LOCCQMA(k), k = O(1) LOCCQMA m,s,c (2) = QMA O(m 2 ε -2 ),s+ε,c Contrast: QMA m,s,c (2) not in QMA O(m 2-δ ε -2 ),s+ε,c for ε = O(1) and δ>0 unless Quantum ETH * fails * Quantum ETH: SAT cannot be solved in 2 o(n) quantum time Follows from Harrow and Montanaro ‘10 (based on Aaroson et al ‘08)

49. New Characterization QMA Corollary QMA = LOCCQMA(k), k = O(1) LOCCQMA m,s,c (2) = QMA O(m 2 ε -2 ),s+ε,c Idea to simulate LOCCQMA m,s,c (2) in QMA: Arthur asks for proof ρ on AB 1 B 2… B k with k = mε -2 He symmetrizes the B systems and apply the original verification prodedure to AB 1 Correcteness de Finetti bound implies:

50. Proof

51. Relative Entropy of Entanglement The proof is largely based on the properties of a different entanglement measure: Def Relative Entropy of Entanglement (Vedral, Plenio ‘99)

52. Entanglement Hypothesis Testing Given (many copies) of ρ AB, what’s the optimal probability of distinguishing it from a separable state?

53. Entanglement Hypothesis Testing Given (many copies) of ρ AB, what’s the optimal probability of distinguishing it from a separable state? Def Rate Function: D(ρ AB ) is maximum number s.t. there exists {M n, I-M n }, 0 < M n < I, D LOCC (ρ AB ) : defined analogously, but now {M, I-M} must be LOCC

54. Entanglement Hypothesis Testing Given (many copies) of ρ AB, what’s the optimal probability of distinguishing it from a separable state? Def Rate Function: D(ρ AB ) is maximum number s.t there exists {M n, I-M n }, 0 < M n < I, D LOCC (ρ AB ) : defined analogously, but now {M, I-M} must be LOCC (B., Plenio ‘08) D(ρ AB ) = E R ∞ (ρ AB ) Obs: Equivalent to reversibility of entanglement under non-entangling operations

55. Proof in 1 Line

56. (i) Quantum Shannon Theory: State redistribution Protocol (ii) Large Deviation Theory: Entanglement Hypothesis Testing (iii) Entanglement Theory: Faithfulness bounds Relative entropy of Entanglement plays a double role: (Devetak and Yard ‘07) (B. and Plenio ‘08)

57. First Inequality Non-lockability: (Horodecki 3 and Oppenheim ‘04) State Redistribution: How much does it cost to redistribute a quantum system? ABE F AE BF ½ I(A:B|E) (i) Apply non-lockability to and use state redistribution to trace out B at a rate of ½ I(A:B|E) qubits per copy

58. Second Inequality Equivalent to: Monogamy relation for entanglement hypothesis testing Idea Use optimal measurements for ρ AE and ρ AB achieving D(ρ AE ) and D LOCC(1) (ρ AB ), resp., to construct a measurement for ρ A:BE achieving D(ρ A:BE )

59. Third Inequality Pinsker type inequality for entanglement hypothesis testing Idea minimax theorem + martingale like property of the set of separable states

60. Open Question Can we prove a lower bound on I(A:B|E) in terms of distance to “markov quantum chain states”? Can we close the LOCC norm vs. trace norm gap in the results (hardness vs. algorithm, LOCCQMA(k) vs QMA(k))? Are there more applications of the bound on the convergence of the SDP relaxation? Are there more application of the main inequality?

61. Thanks!