A limit on nonlocality in any
world in which communication
complexity is not trivial
IFT6195
Alain Tapp
In collaboration with…
Gilles Brassard
Harry Buhrman
Naoh Linden
André Allan Methot
Falk Unger
Quant-ph/0508042
Motivation
What would be the consequences if the
non local collerations in our world were
stronger than the one given by quantum
mechanics?
Theoretical computer science?
Foundation of physics?
Philosophy?
Perfect Non Local Boxes
Bob
Alice
a
x
NLB
ab x y
b
y
NLB and communication
One bit of communication is enough to
implement a NLB.
Alice sends a to Bob and output x=0
Bob outputs y a b
NLB and communication
NLBs does not allow for communication.
We can have a perfect box for which x and y
are uniformly distributed and independent of
(respectively) a and b.
NLB, classical deterministic strategies
0 0 a 0 0 b a b
00
01
10
11
yes
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
yes
NLB classical implementation
There is a probabilistic strategy with
succes probability ¾ on all input.
There is no classical déterministic strategy
with success proportion greater than ¾.
There is no probabilistic strategy with
success probability greater than ¾.
¾
NLB quantum strategy
1
00 11
2
cos( ) sin( )
R( )
sin( ) cos( )
Alice and Bob have the same strategy.
If input=0 applies R ( / 16) otherwise R (3 / 16)
Measure and output the result.
This strategy works on all inputs with probability:
cos ( / 8) 85%
2
NLB quantum strategy
cos ( / 8) 85%
2
Tsirelson proved in 1980 that this is
optimal whatever the entanglement
shared by the players.
Bell theorem
85% cos ( / 8) 3 / 4
2
The classical upper bound and the
quantum lower bound do not match.
We can derive an inequality from this that
provides a Bell theorem proof.
This is known as the CHSH inequality.
Classical Communication Complexity
Alice
Bob
z R {0,1}k
x
f ( x, y )
y
Quantum Communication Complexity
Alice
x
f ( x, y )
1
1
00
11
2
2
k
Bob
y
Communication Complexity
The classical/quantum probabilistic
communication complexity of f, C(f)/Q(f)
is the amount of classical communication
required by the best protocol that
succeeds on all input with probability at
least 1 / 2 when the players have
unlimited prior classical/quantum
correlation.
Inner product (IP)
IP ( x, y ) x y
x x1 x2 xn
y y1 y2 yn
x y ( x1 y1 ) ( x2 y3 ) ( xn yn )
n
x y xi yi (mod 2)
i 1
Inner product (IP)
C ( IP ) n O(1)
Q( IP ) (n)
Most functions are difficult
For most functions f
C ( f ) n O(1)
Q( f ) n O(1)
Equality
x y EQ( x, y ) 1
x y EQ( x, y ) 0
Alice and Bob each have a very large file and
they want to know if it is exactly the same.
How much do they need to communicate?
Equality
z R 0,1
n
Alice x
ma x z
Bob y
mb y z
mb
Output ma mb
Equality
x y Pz x z y 1
1
x y Pz x z y
2
By repeating the protocol twice we have
success probability of at least ¾.
C (EQ ) 2
Scheduling
Alice and Bob want to find a time where
they are both available for a meeting.
x x1 x2 xn
y y1 y2 yn
S ( x, y ) ( x1 y1 ) ( x2 y3 ) ( xn yn )
n
S ( x, y ) xi yi 0
i 1
Scheduling
C ( S ) ( n )
Q ( S ) ( n )
Raz separation
There exists a problem such that:
n
C ( S )
log( n)
1/ 4
Q ( S ) O (log( n))
IP using NLB
( Ai , Bi ) NLB( xi , yi )
Ai Bi xi yi
x y ( x1 y1 ) ( x2 y3 ) ( xn yn )
x y ( A1 B1 ) ( A2 B2 ) ( An Bn )
x y ( A1 A2 An ) ( B1 B2 Bn )
Perfect NLB implies trivial CC
Any function can be computed with a serie of
AND gates and negations.
Distributed bit
x xa xb
Input bit
xa x1 , xb 0
Negation:
AND
Outcome
x xa xb
Two NLBs
y ya yb
Bob sends
yb to Alice
AND
( A1 , B1 ) NLB( xa , yb )
( A2 , B2 ) NLB( ya , xb )
A1 B1 xa yb
A2 B2 ya xb
x y ( xa xb ) ( ya yb )
x y ( xa ya ) ( xa yb ) ( xb ya ) ( xb yb )
x y (( xa ya ) A1 A2 ) ( B1 B2 ( xb yb ))
Main result
In any world where non local boxes can be
implemented with accuracy larger than 0.91
communication complexity is trivial.
1 1
NLB
2
6
f , C* ( f ) 1
CC with a bias
We say that a function f can be computed
with a bias if Alice and Bob can produce a
distributed bit z such that
z z a zb
1
P[ f ( x, y ) z ]
2
CC with a bias
Every function can be computed with a bias.
Alice’s input: x
Bob’s input: y
Alice and Bob share z
Alice outputs a=f(x,z)
Bob outputs b=0 if y=z and a random bit otherwise.
1
1 1
1
P[ f ( x, y ) a b] n 1 n
2 2 2
2
Idea
We want a bounded bias.
Let’s amplify the bias.
Repetition and majority?
Idea
Maj
Maj
Maj
~
~
~
f ( x) f ( x) f ( x)
Maj
~
~
~
f ( x) f ( x) f ( x)
Maj
Maj
~
~
~
f ( x) f ( x) f ( x)
Maj
~
~
~
f ( x) f ( x) f ( x)
~
f ( x)
Maj
~
~
~
f ( x) f ( x) f ( x)
Maj
Maj
~
~
~
f ( x) f ( x) f ( x)
Maj
Maj
Maj
~
~
~
f ( x) f ( x) f ( x)
~
~
~
f ( x) f ( x) f ( x)
~
~
~
f ( x) f ( x) f ( x)
Non local majority
y Maj ( x1 , x2 , x3 ) 1 iff
x1 x2 x3 2
x1 x x
1
a
1
b
x2 x x
2
a
2
b
x3 x x
3
a
3
b
y y a yb
ya yb Maj ( x x , x x , x x )
1
a
1
b
2
a
2
b
3
a
3
b
NLM > 5/6
If NLM can be computed with probability
stricly greather than 5/6 than every
fonction can be computed with a bounded
bias.
Below that treshold NLM makes things
worst.
NLM > 5/6
p 1/ 2
( 0)
q 5/ 6
( 0)
h( p) q( p 3 3 p 2 (1 p)) (1 q)(3 p(1 p) 2 (1 p)3 )
1
3
1
s
2 2 1 3 2
1 / 2 p s h( p ) p
Non local equality
y NLE ( x1 , x2 , x3 ) 0 iff
x1 x2 x3
x1 x x
1
a
1
b
x2 x x
2
a
2
b
x3 x x
3
a
3
b
y y a yb
ya yb NLE ( x x , x x , x x )
1
a
1
b
2
a
2
b
3
a
3
b
NLE implies NLM
ya yb NLE ( x x , x x , x x )
1
a
1
b
z a ya x x x
1
a
2
a
2
a
2
b
3
a
3
b
3
a
z b yb x x x
1
b
2
b
3
b
z a zb Maj ( x x , x x , x x )
1
a
1
b
2
a
2
b
3
a
3
b
2 NLB implies NLE
x1 x1a xb1
x 2 xa2 xb2
x 3 xa3 xb3
z1a zb1 NLB( x1a xa2 , xb2 xb3 )
z a2 zb2 NLB(1 xa2 xa3 , xb1 xb2 )
NLE ( x1 , x 2 , x 3 )
( x1 x 2 ) ( x 2 x 3 )
( x1a xb1 xa2 xb2 ) ( xa2 xb2 3a xb3 )
(( x1a xa2 ) ( xa2 xa3 )) z1a z a2
(( xb1 xb2 ) ( xb2 xb3 )) zb1 zb2
To conclude the proof
1 1
5
5
NLB
NLE Maj
2
6
6
6
•Compute f several times with a bias
•Use a tree of majority to improve the bias.
•Bob sends his share of the outcome to Alice.
Open question
Show some unacceptable
consequences of correlations
epsilon-stronger than the one
predicted by quantum mechanics.
© Copyright 2026 Paperzz