A Parallel Repetition Theorem for Entangled Projection Games

1. A Parallel Repetition Theorem for Entangled Projection Games Thomas VidickSimons Institute, Berkeley Joint work with Irit Dinur (Weizmann) and David Steurer (Cornell)

2. Direct Product Theorems Answer! –resource solver Instance of ? Success • Consider task ; requires resources to achieve success • What is the optimal way to solve independent instances ? • DPT: solving instances of with resourcesmust have success • Question arises in diverse settings: • Circuits; power of restricted models of computation; • Information theory; communication complexity; • Multiplayer games: hardness amplification, cryptography –resource solver ! ? independent instances of ! ?

3. Multiplayer games • Referee interacts with playersMakes choice of inputs (questions); observes outputs (answers) • Referee decides to accept/reject • := max. prob. accept • optimize over all strategies = state + measurements • Study originates in foundations of QM: EPR’35, Bell’64 → “Bell inequalities”;CHSH’69, Mermin’90s → nonlocal games • Renewed interest over past years:randomness certification [Col’09,…,MS’14],device-independent key distribution [BHK’05,…,VV’13],testing quantum systems [MY’98,…,RUV’13] • Independent line of work in classical complexity theory: PCP theorem, interactive proofs and hardness of approximation := max. prob. accept optimize over allquantum strategies := max. prob. accept optimize over allclassical strategies Accept! Reject!!

4. Parallel repetition of multiplayer games • a two-player game. : copies of in parallel • Select independent pairs of questions • Send all questions at once, receive all answers • Accept if and only if all instances would accept • Can we relate to ? • (play each game independently) • [Feige,Watrous] Simple game such that • Parallel repetition question: does ? At what rate? • Introduced in classical setting as means to “amplify soundness”;long sequence of results [Ver’94,…,Raz’98,Hol’09] yields almost-optimal bounds • Quantum? XOR games [CSUU’07], unique games [KRT’08]: based on tight SDP[KV’12]: general games, but only polynomial decay Check! Check! Check!

5. Results: parallel repetition of projection games • Projection game: for every pair of questions, any answer from B determines unique validanswer from A • Your favorite two-player game is a projection game! • Exists universal transformation such that projection game andCould have Main result: for every projection game , • Matches optimal bound in classical case, up to value of constant ([Hol’09]: ) Projectiongames Magic square XOR games Unique games 3SAT game Hidden matching game

6. Proof outline • Introduce a relaxationsuch that: (i) is a tight relaxation: (ii) is perfectly multiplicative: • Conclude: if , Rounding argument based on quantum correlated sampling lemma Uses projection game : semidefinite relaxation + positivity constraint on vector coefficients→ shares multiplicativity properties of semidefinite relaxations

7. The relaxation Game coefficients • Three steps: • Relax condition on measurements (“”) • SDP-like relaxation to vector-valued matrices • Additional positivity constraint on coordinates (i)(ii) Variables are vector-valued matrices:

8. LO A quantum correlated sampling lemma • Consider the following distributed task: • A,B share of their choice • Receive classical descriptions of , in s. t. • Goal: using local operations alone, create s.t. • : quantum state embezzlement [vDH’09]→ Use embezzling state • We give robust procedure; achieves Main challenge is dealing with near-degeneracies in spectrum • Generalizes classical correlated sampling lemma from [Hol’09] Procedure in [vDH’09] is not robust:

9. Summary • First exponential parallel repetition theorem for entangled games • Applies to projection games • Proof introduces tight, multiplicative relaxation inspired from semidefinite relaxation • Rounding argument based on quantum correlated sampling lemma • Extend to general games? (See next talk for different approach) • Low soundness case? Direct product results? (Exist in classical case!) • General paradigm: relaxation optimizes over “partial strategies”→ Useful in other contexts? Parallel repetition for QMA(2)? Channels? Questions

10. Thank you!