CS286.2 Lecture 14: Quantum de Finetti Theorems II

Similar documents
Notes on the stability of dynamic systems and the use of Eigen Values.

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!") i+1,q - [(!

On One Analytic Method of. Constructing Program Controls

Mechanics Physics 151

Mechanics Physics 151

V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS

. The geometric multiplicity is dim[ker( λi. number of linearly independent eigenvectors associated with this eigenvalue.

Mechanics Physics 151

Part II CONTINUOUS TIME STOCHASTIC PROCESSES

Epistemic Game Theory: Online Appendix

Lecture 18: The Laplace Transform (See Sections and 14.7 in Boas)

Graduate Macroeconomics 2 Problem set 5. - Solutions

FI 3103 Quantum Physics

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4

. The geometric multiplicity is dim[ker( λi. A )], i.e. the number of linearly independent eigenvectors associated with this eigenvalue.

Comparison of Differences between Power Means 1

Linear Response Theory: The connection between QFT and experiments

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany

NATIONAL UNIVERSITY OF SINGAPORE PC5202 ADVANCED STATISTICAL MECHANICS. (Semester II: AY ) Time Allowed: 2 Hours

GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS. Youngwoo Ahn and Kitae Kim

Solution in semi infinite diffusion couples (error function analysis)

( ) () we define the interaction representation by the unitary transformation () = ()

Comb Filters. Comb Filters

Department of Economics University of Toronto

Online Appendix for. Strategic safety stocks in supply chains with evolving forecasts

Let s treat the problem of the response of a system to an applied external force. Again,

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5

Online Supplement for Dynamic Multi-Technology. Production-Inventory Problem with Emissions Trading

Lecture 2 M/G/1 queues. M/G/1-queue

Appendix H: Rarefaction and extrapolation of Hill numbers for incidence data

Scattering at an Interface: Oblique Incidence

2/20/2013. EE 101 Midterm 2 Review

TSS = SST + SSE An orthogonal partition of the total SS

Ordinary Differential Equations in Neuroscience with Matlab examples. Aim 1- Gain understanding of how to set up and solve ODE s

Relative controllability of nonlinear systems with delays in control

SOME NOISELESS CODING THEOREMS OF INACCURACY MEASURE OF ORDER α AND TYPE β

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL

FTCS Solution to the Heat Equation

Density Matrix Description of NMR BCMB/CHEM 8190

Econ107 Applied Econometrics Topic 5: Specification: Choosing Independent Variables (Studenmund, Chapter 6)

Density Matrix Description of NMR BCMB/CHEM 8190

Math 128b Project. Jude Yuen

Normal Random Variable and its discriminant functions

PHYS 705: Classical Mechanics. Canonical Transformation

Existence and Uniqueness Results for Random Impulsive Integro-Differential Equation

Tight results for Next Fit and Worst Fit with resource augmentation

How about the more general "linear" scalar functions of scalars (i.e., a 1st degree polynomial of the following form with a constant term )?

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD

UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 2017 EXAMINATION

( t) Outline of program: BGC1: Survival and event history analysis Oslo, March-May Recapitulation. The additive regression model

2.1 Constitutive Theory

A Deza Frankl type theorem for set partitions

Motion of Wavepackets in Non-Hermitian. Quantum Mechanics

J i-1 i. J i i+1. Numerical integration of the diffusion equation (I) Finite difference method. Spatial Discretization. Internal nodes.

, t 1. Transitions - this one was easy, but in general the hardest part is choosing the which variables are state and control variables

THE PREDICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS

THEORETICAL AUTOCORRELATIONS. ) if often denoted by γ. Note that

Dynamic Team Decision Theory. EECS 558 Project Shrutivandana Sharma and David Shuman December 10, 2005

( ) [ ] MAP Decision Rule

Advanced Macroeconomics II: Exchange economy

arxiv: v1 [math.pr] 6 Mar 2019

CHAPTER 10: LINEAR DISCRIMINATION

Track Properities of Normal Chain

Example: MOSFET Amplifier Distortion

arxiv: v1 [cs.sy] 2 Sep 2014

Should Exact Index Numbers have Standard Errors? Theory and Application to Asian Growth

Lecture 6: Learning for Control (Generalised Linear Regression)

Clustering (Bishop ch 9)

Testing a new idea to solve the P = NP problem with mathematical induction

Robust and Accurate Cancer Classification with Gene Expression Profiling

Notes for Lecture 17-18

ON THE WEAK LIMITS OF SMOOTH MAPS FOR THE DIRICHLET ENERGY BETWEEN MANIFOLDS

A Reinforcement Procedure Leading to Correlated Equilibrium

Chapter 6: AC Circuits

Bundling with Customer Self-Selection: A Simple Approach to Bundling Low Marginal Cost Goods On-Line Appendix

ECON 8105 FALL 2017 ANSWERS TO MIDTERM EXAMINATION

Advanced time-series analysis (University of Lund, Economic History Department)

2 Aggregate demand in partial equilibrium static framework

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model

Survival Analysis and Reliability. A Note on the Mean Residual Life Function of a Parallel System

CH.3. COMPATIBILITY EQUATIONS. Continuum Mechanics Course (MMC) - ETSECCPB - UPC

Variants of Pegasos. December 11, 2009

2 Aggregate demand in partial equilibrium static framework

Time-interval analysis of β decay. V. Horvat and J. C. Hardy

F-Tests and Analysis of Variance (ANOVA) in the Simple Linear Regression Model. 1. Introduction

e-journal Reliability: Theory& Applications No 2 (Vol.2) Vyacheslav Abramov

GORDON AND NEWELL QUEUEING NETWORKS AND COPULAS

Lecture 11 SVM cont

SELFSIMILAR PROCESSES WITH STATIONARY INCREMENTS IN THE SECOND WIENER CHAOS

H = d d q 1 d d q N d d p 1 d d p N exp

Robustness Experiments with Two Variance Components

Volatility Interpolation

Motion in Two Dimensions

Panel Data Regression Models

EP2200 Queuing theory and teletraffic systems. 3rd lecture Markov chains Birth-death process - Poisson process. Viktoria Fodor KTH EES

THERMODYNAMICS 1. The First Law and Other Basic Concepts (part 2)

January Examinations 2012

Performance Analysis for a Network having Standby Redundant Unit with Waiting in Repair

Lecture VI Regression

Dual Approximate Dynamic Programming for Large Scale Hydro Valleys

Transcription:

CS286.2 Lecure 14: Quanum de Fne Theorems II Scrbe: Mara Okounkova 1 Saemen of he heorem Recall he las saemen of he quanum de Fne heorem from he prevous lecure. Theorem 1 Quanum de Fne). Le ρ Dens C 2 ) n), k n and n k. Then here exss an m such ha f for any [n] k, j [n] m and x [d 8 ] m, ρ j,x s he pos-measuremen sae on quds... k obaned afer measurng j... j m usng Λ and obanng oucomes x 1... x m, we have E E j,x ρ j,x ρ j,x... ρ j,x k 1 4 lnd)18d)k k 2. 1) Ths saemen s dfferen from he prevous saemen of he quanum de Fne heorem as ρ s no assumed o be permuaon nvaran hs s replaced by he fac ha he ndces j ha are measured, and he ndces ha are kep, are dsrbued unformly n [n] subjec o beng dsnc ndces), he expecaon over x s aken ousde f we move nsde, we oban a convex combnaon of producs as n he prevous de Fne, bu akng he expecaon value ousde makes he saemen sronger), and because he bound has a larger dependence on he dmenson d k nsead of he prevous d). Ths las pon s he man drawback of he heorem and s drecly due o he use of he nformaonally complee measuremen Λ s an open problem wheher hs can be avoded). Theorem 1 has a classcal analogue, wh he same defnons of n, m, k, and : Theorem 2 Classcal Analogue). Le P be a dsrbuon on S n for some se S of sze d. Le = 1... k, j = j1... j m, and x = x 1... x m be sampled accordng o he margnal P j. Le P j,x be he margnal on condoned on he varables ndexed by j beng equal o he values specfed by x. Then E E j,x P j,x 2 Classcal = Quanum P j,x... P j,x k 1 2k2 ln d. 2) Our pah now s o prove ha he quanum saemen follows from he classcal saemen, and hen prove he classcal saemen, and fnally o apply hs o QPCP. Proof. Consder he dsrbuon P obaned from measurng all quds n ρ usng Λ we assume ha Λ s nformaonally complee wh dsoron 18d) k/2, as before, and s measured as a ensor produc across all qus). By he classcal saemen of he heorem, we have, 1

E E j,x P j,x P j,x... P j,x k 1 2k2 lnd 8 ), snce P s a dsrbuon on he se [d 8 ] n of all possble oucomes of Λ. Now observe ha by defnon P j,x = Λ k ρ j,x ) and P j,x l = Λρ j,x l ) for l = 1,..., k. Hence from he above we ge E E j,x Λ k ρ j,x ) Λρ j,x )... Λρ j,x k ) 1 2k2 lnd 8 ) Usng he fac ha Λ k s nformaonally complee wh dsorson 18d) k/2 yelds he quanum heorem. 3 Proof of Classcal Saemen Now our am s o prove he classcal analogue of he heorem, whch we wll do usng an nformaon heorec proof. Frs, we wll defne varous nformaon heorec quanes boh classcal and quanum) o help us n he effor. Le ρ AB DensC d C d ) be a densy marx, wh reduced saes ρ A and ρ B. Defnon 3. The enropy of a sae ρ A s defned as Sρ A ) = Trρ A ln ρ A ) = Sλρ A )), 3) where λρ A ) are he egenvalues of ρ A hese form a dsrbuon snce ρ A s a densy marx). Noe ha f ρ AB = ψ ψ s pure hen he egenvalues of ρ A are he squares of Schmd coeffcens n he decomposon of ψ across he A : B paron. Defnon 4. The muual nformaon IA : B) ρ s defned as IA : B) ρ = Sρ A ) + Sρ B ) Sρ AB ). 4) We wll also use condonal muual nformaon. If ρ ABX s a qqc sae,.e. ρ ABX = x p x ρ x AB x x for some dsrbuon p x and densy marces ρ x AB, hen IA : B X) ρ ABX = x p x IA : B) ρ x AB = SA X) + SB X) SAB X) where SA X) = SAX) SX). Example 5. If ρ = ρ A ρ B, hen Sρ A ρ B ) = Sρ A ) + Sρ B ) and so IA : B) ρ = 0. In oher words, n a produc sae, neher subsysem carres nformaon abou he oher. Example 6. On he oher hand, f ρ = ψ ψ s pure, hen Sρ) = 0 and IA : B) = 2Sρ A ). In parcular, s ψ = 1 d s he maxmally enangled sae, hen IA : B) = 2 ln d. Clam 7. In fac hese wo exreme examples saurae he followng nequales, whch always hold: 0 IA : B) 2 mnln d, ln d ) 5) One key propery of muual nformaon s gven by Pnsker s nequaly. 2

Clam 8. For any bpare sae ρ AB, IA : B) ρ 1 2 ρ AB ρ A ρ B 2 1. 6) Noe ha, f IA : B) = 0, hen Pnsker s nequales mples ha he sae mus be a ensor produc. Muual nformaon has several useful properes. Proposon 9. Muual nformaon sasfes he chan rule and monooncy under measuremens M A on A, as well as racng ou: IA : BX) = IA : X) + IA : B X) 7) IA : B) MA Id B ρ AB ) IA : B) ρab 8) IA : B) TrC ρ ABC ) IAC : B) ρ ABC. 9) Monooncy s nuve: gven a sysem A, we canno ncrease he nformaon ha A conans abou B by performng an operaon on A alone. In he classcal case s easy o prove, bu n he quanum case requres much more work! We wll ake hese properes for graned here. We are ready o prove he classcal saemen of he de Fne heorem. For smplcy we wll do he proof for k = 2, bu he proof for general k s vrually he same. We show he followng: Clam 10. Under he same noaon as Theorem 2, E 0 m< E =1, 2 ), j=j1...j m ) IX 1 : X 2 X j1... X jm ) 1 2 E IX : X ) 10) where X... X n ) are random varables wh dsrbuon P and s shorhand for [n]\{}. Frs le s show ha he clam leads o he classcal heorem. Proof of Theorem 2. Snce X ranges over a se of sze d we have IX : X ) 2 ln d. Applyng Pnsker s nequaly, So we oban as desred. P jx, 2 P jx P jx 2 2 1 2IX : X 1 X j1... X jm ). E m E j E x P jx, 2 P jx P jx 2 1 2 ln d, Now le s prove he clam. Proof of Clam 10. By monooncy, and hen by he chan rule, we have 1 E IX : X ) 1 E,j 1,...j IX : X j1... X j ) where he las lne s a smple relabelng of coordnaes. = 1 1 E,j 1,...j IX : X jm+1 X j1... X jm ) m=0 = E 0 m< E =1, 2 ), j=j1...j m ) IX 1 : X 2 X j1... X jm ), 3

4 Applcaon o QPCP Now we apply hs saemen of he quanum de Fne heorem o he quanum PCP conjecure o show he followng. Theorem 11. Le H be a 2-local Hamlonan on quds such ha he neracon graph has degree exacly D. Then here exss a produc sae ψ = ψ 1... ψ n such ha d ψ H ψ λ 0 H) + 2 ) 1/3 ln d O m, 11) D where m = nd 2 s he number of local erms n H. Ths heorem can be seen as a srong no-go for QPCP. Indeed, shows ha QPCP s false on D- regular graphs for any value of he gap parameer γ = Ωd 2 ln d/d) 1/3 assumng QMA =NP). Indeed, he heorem shows ha, for hese values of γ here s always a produc sae ha has energy less han b = λ 0 H) + γ and can serve as a classcal wness for hs fac. Snce QPCP assers hardness for consan γ, we see ha he conjecure can only be rue on graphs of consan degree D unless he local dmenson d s allowed o grow faser han D). In an earler lecure we also saw as an exercse) ha QPCP requred graphs of hgh dmenson n order o be rue. Togeher hese wo saemens narrow down he knd of local Hamlonans for whch he conjecure may be rue: her neracon degree should have hgh dmenson canno be embedded on any lace of dmenson Olog n)), bu consan degree. Proof. Le ψ be he ground sae of H, and he Λ be he IC measuremen, defned as before. Le ρ = ψ ψ. j,x, defne σ j,x = j ρ j,x 1 d m I j 1...j m 12) In oher words, hs sae s ρ j,x f j and Id /d oherwse. By he quanum de Fne heorem appled o ρ wh k = 2 and o be deermned laer, we have E 1, 2 E j,x ρ j,x ρ j,x ρ j,x d 2 1 = 2 ) O ln d 13) For he sae we defned, we have E, j,x ρ j,x σ j,x σ j,x d 2 1 2 ) O ln d + 2 n 14) where he las quany comes from he rare cases when j. Resrcng he expecaon on = 1, 2 ) o he Dn/2 pars ha correspond o edges of he consran graph of he local Hamlonan and applyng Jensen s nequaly we oban E =1, 2 ) E E j,x ρ j,x σ j,x σ j,x 1 2 = O d 2 ln d + ) n 15) n D 4

By defnon of he race norm and usng ha for any j he reduced densy of Ex ρ j,x on he n quds ha were no measured s he same as ha of ρ, we oban E j ψ H ψ ψ E x σ j,x ψ O d 2 ln d + ) n nd + D, 16) n D where here he mulplcave nd comes from swchng he expecaon over edges o a sum over edges snce H s gven as a sum of local erms), and he addonal D corresponds o mposng a maxmum energy penaly of 1 for each edge nvolvng one of he measured qubs. Choosng he ha mnmzes he rgh-hand sde, nd 2/3 ln d 1/3 /D 1/3, we oban he desred resul: here wll exs a leas one choce of j such ha he separable sae Ex σ j,x has low energy; usng lneary of he Hamlonan here s a leas one pure produc sae n he suppor of E x σ j,x ha has a mos he same energy. 5