Presentation is loading. Please wait.

Presentation is loading. Please wait.

The Power of Unentanglement Scott Aaronson (MIT) Salman Beigi (MIT) Andrew Drucker (MIT) | Bill Fefferman (Caltech) Peter Shor (MIT) | |

Similar presentations

Presentation on theme: "The Power of Unentanglement Scott Aaronson (MIT) Salman Beigi (MIT) Andrew Drucker (MIT) | Bill Fefferman (Caltech) Peter Shor (MIT) | |"— Presentation transcript:

1 The Power of Unentanglement Scott Aaronson (MIT) Salman Beigi (MIT) Andrew Drucker (MIT) | Bill Fefferman (Caltech) Peter Shor (MIT) | |

2 It is not yet entirely clear what advances in our understanding of quantum computation and quantum information can be expected as a result of the study of quantitative measures of entanglement. Nielsen & Chuang (2000) In this work, we connect quantum complexity theory to entanglement theoryironically, by studying the power of lack of entanglement! Previous 3 talks, 3 talks at upcoming FOCS: Quantum multi-prover interactive proof systems where provers share entanglement Not what Ill be talking about today

3 Main Results Proving 3SAT With Õ( n) Qubits Let be a 3SAT instance of size n. Someone can prove to you that is satisfiable by giving you only O( n polylog n) qubitsprovided you know certain subsets of the qubits are unentangled with the rest Proof is nonrelativizing, and requires a tight PCP theorem Additivity Amplification and Collapse Multi-prover quantum MA can be amplified to exponentially small error, and three or more Merlins can be simulated with two, assuming the famous Additivity Conjecture from quantum information theory

4 QMA: Quantum Merlin-Arthur [Kitaev and Watrous, 2000] Class of languages L such that for all inputs x: x L exists a witness | with poly(n) qubits, causing polytime quantum verifier Arthur to accept w.p. 2/3 x L for all witnesses |, Arthur accepts w.p. 1/3 We know a reasonable amount about QMA: its contained in PP, allows amplification, has natural complete promise problems…

5 QMA(k) [Kobayashi, Matsumoto, Yamakami 2003] Classically, this is completely uninteresting: MA(k)=MA But quantumly, a single Merlin could cheat by entangling the k proofs! Class of languages L such that for all inputs x: x L there exist witnesses | 1,…,| k causing Arthur to accept w.p. 2/3 x L for all | 1,…,| k, Arthur accepts w.p. 1/3

6 Do Multiple Quantum Proofs Ever Actually Help? Liu, Christandl, Verstraete: Natural problem from quantum chemistry, pure state N-representability, which is in QMA(2) but not known to be in QMA Blier and Tapp (independent of us) : 3-COLORING admits a 2-prover QMA protocol with witnesses of size log(n), and a (1/n 6 ) probability of catching cheating provers This work: A protocol for 3SAT with Õ( n) quantum witnesses of size log(n), and constant soundness

7 Our Protocol for 3SAT Well work not with 3SAT but with 2-out-of-4-SAT: Theorem: We can get all of this using known classical reductions from 3SAT (including Dinurs gap amplification procedure), incurring a polylog(n) blowup in the number of variables and clauses. We need our 2-out-of-4-SAT instance to be balanced (each variable occurs in O(1) clauses), as well as a PCP (either satisfiable or -far from satisfiable) x i + x j + x k + x l = 2 (mod 4)

8 So suppose Arthur has done all this, to obtain a 2-out- of-4-SAT instance. And suppose Merlin sends him a log(n)-qubit state of the form Then Arthur can measure | in a basis corresponding to the clauses of, obtaining the outcome where x 1,…,x n is a claimed satisfying assignment for. (I.e., a proper state.) for some clause C=(x i,x j,x k,x l ). A further measurement reveals whether C is satisfied with (1) probability.

9 Problem: How can Arthur force Merlin to send him a proper state? (E.g., what if Merlin cheats by putting all amplitude on a few computational basis states?) Solution: More Merlins! | | | n log(n)

10 The Protocol With 1/3 prob. Pick a random | k and do the Satisfiability Test described earlier With 1/3 prob. Pick two random | k s and do a Swap-Test (Ensures most | k s are pretty much identical) With 1/3 prob. Do a Uniformity Test to make sure the | ks are close to proper states

11 The Uniformity Test Pick a random matching M on [n] for each (i,j) M. Measure each witness | in a basis containing Since there are n witnesses, by the Birthday Paradox, with constant probability well see a collision: two outcomes involving the same edge (i,j). |3 -|4 |8 +|10 |2 +|9 |8 -|10

12 If both outcomes are |i +|j or both |i -|j, accept. If one outcome is |i +|j and the other is |i -|j, reject. Accepts with certainty if the witnesses are identical and proper Theorem: Rejects with (1) probability if witnesses are close to each other but far from proper Proof: So intuitively obvious, it takes 14 pages to prove Why doesnt our protocol work with entangled witnesses? Because the Merlins could send a state that passes all Swap-Tests, yet doesnt produce collisions

13 Amplification For QMA, its easy to amplify success probability, even if Merlin cheats by entangling the witnesses Entanglement Swapping So then whats the problem with amplifying QMA(2)? Witness 1 Witness 2 Witness 3 Witness 1 Witness 2 Witness 3 Witness 1 Witness 2 Witness 3 Merlin 1 : Merlin 2 : Uh-oh! Accept

14 Yet it seems possible to defend against this bizarre behavior… n 100 pairs of witnesses, of which we only measure a random n Does any tiny amount of entanglement thats created during this protocol spread itself thinly across the registers in a reasonable way? To answer this question, we need a way to measure entanglement… W1W1 W2W2 W3W3 W4W4 W5W5 W6W6 W7W7 W1W1 W2W2 W3W3 W4W4 W5W5 W6W6 W7W7

15 Entanglement of Formation E F ( AB ) Intuitively, minimum # of EPR pairs needed to prepare AB Fun Facts: E F can only increase by 2K when we act on a K-qubit register If E F ( AB ) 0, then AB is close to a separable state in trace distance Is E F superadditive? Shor 2003: Equivalent to proving the additivity of quantum channel capacity, a famous open problem Good for us

16 Assuming the Additivity Conjecture, we show that… QMA(2) protocols can be amplified to exponentially small error QMA(2)=QMA(k) for all 2 k poly(n) (building on [KMY]) SymQMA(k)=QMA(k) (SymQMA(k): All k Merlins send the same state) For every fixed polynomial p, p(n) entanglement gives the Merlins no extra power to cheat

17 Upper Bounds for QMA(2) Its obvious that QMA(2) NEXP. Embarrassingly, we still dont have a better upper bound! On the other hand: If amplification that reuses both witnesses is possible, then PSPACE=NEXP! Our Result: QMA(2) PSPACE, assuming Strong Amplification of QMA(2) protocols (Amplification that reuses one of the Merlins witnesses over and over) PSPACE EXP NEXP PP QMA(2) ?

18 Does QMA(2)=QMA? Right now, even proving an oracle separation between them seems way beyond reach! We show: If you want a perfect disentangler, then the input Hilbert space needs to be infinite-dimensional. Conjecture (Watrous): Theres no way to simulate QMA(2) in QMA by taking an arbitrary polynomial-size witness, and disentangling it to produce an arbitrary roughly-separable witness All separable states All states

19 More Open Problems In our 3SAT protocol, can the assumption of unentanglement be removed? If so, then we get a 2 Õ( n) quantum algorithm for 3SAT! Conjecture: Our protocol can be modified to require only two provers sending Õ( n) qubits each In defining QMA(2), does it matter whether amplitudes are real or complex? Are there natural group-theoretic problems in QMA(2)? Does QMA(2) have natural complete promise problems? Can we improve on Õ( n), or get evidence against this? Remove additivity/amplification assumptions!

Download ppt "The Power of Unentanglement Scott Aaronson (MIT) Salman Beigi (MIT) Andrew Drucker (MIT) | Bill Fefferman (Caltech) Peter Shor (MIT) | |"

Similar presentations

Ads by Google