Professional Documents
Culture Documents
Substituting Quantum Entanglement For Communication
Substituting Quantum Entanglement For Communication
Substituting Quantum Entanglement For Communication
If a set of entangled particles are individually measured, n bits of communication turn out to be necessary ~see Ref.
the resulting outcomes can exhibit ‘‘nonlocal’’ effects @1–5#. @6# for a proof of this!. Thus, even though the goal is for
These are effects that, from the perspective of ‘‘classical’’ Alice to acquire a single bit of information, this bit depends
physics, cannot occur unless ‘‘instantaneous communica- on the 2n bits distributed among Alice and Bob in such a
tions’’ occur among the particles, which convey information way that they must exchange n bits between them in order
about each particle’s measurement to the other particles. for Alice to determine this bit. For a function f : $ 0,1% n
On the other hand, no communication actually occurs 3 $ 0,1% n → $ 0,1% , the minimum number of bits that must be
among the entangled particles when they are measured. To communicated between Alice and Bob in order for Alice to
phrase this in operational terms, entangled particles cannot determine f (x,y) is called the communication complexity of
be used to simulate communication. For example, if two f . Several aspects of communication complexity are sur-
physically separated parties, Alice and Bob, initially possess veyed in Ref. @6#.
particles whose quantum states are entangled and then Bob The question that we consider is whether or not a prior
obtains a bit of information x, there is no operation that Bob quantum entanglement can reduce communication complex-
can apply to his particles that will have the effect of convey- ity. For example, if Alice and Bob initially possess entangled
ing x to Alice when she performs measurements on her par- particles, can they compute some functions using less com-
ticles. Moreover, entanglement cannot even be used to com- munication than would be required without the entangled
press information: for Bob to convey n bits ~with arbitrary particles? Although we do not presently know the answer for
values! to Alice, he must send n bits—sending n21bits will this two-party scenario, we exhibit an analogous three-party
not suffice. Also, similar results apply to communications scenario where entanglement does reduce communication
involving more than two parties. complexity. The function is based on Mermin’s version @5#
Consider the following related but different scenario. Al- of ‘‘Bell nonlocality without probabilities.’’
ice obtains an n-bit string x, and Bob obtains an n-bit string Consider the following three-party scenario. Alice, Bob,
y and the goal is for Alice to determine f (x,y), for some and Carol receive n-bit strings x, y, and z, respectively,
function f : $ 0,1% n 3 $ 0,1% n → $ 0,1% ,with as little communica- which are subject to the condition that
tion between Alice and Bob as possible. This can always be
accomplished by Bob sending his n bits to Alice, but fewer x1y1z51, ~3!
bits may suffice. For example, for the function
where 1 is applied bitwise ~modulo two! and
entanglement, two bits of communication are sufficient for First, consider the case where x Ai x Bi x Ci 5111. In this case,
Alice to determine f (x,y,z). Thus, even though entangle- no H transformation is applied to any of
ment cannot be used to simulate communication, it can nev- qAi ,qBi ,qCi .Therefore, qAi ,qBi ,qCi is measured in state ~5!,
ertheless act as a substitute for communication when the goal which implies that s Ai 1s Bi 1s Ci 515x Ai •x Bi •x Ci .
is to compute a function with distributed data. We also show Next, in the case where x Ai x Bi x Ci 5001, an H transforma-
that the lower bound of three in the case of no entanglement tion is applied to qAi and to qBi but not to qCi . Therefore,
cannot be improved. This is done by exhibiting a three-bit
qAi qBi qCi is measured in state
protocol.
Recently, Grover @7# has independently demonstrated that
H ^ H ^ I„ 21 ~ u 001& 1 u 010& 1 u 100& 2 u 111& )…
quantum entanglement can reduce communication complex-
ity in a different context.
5 21 ~ u 011& 1 u 101& 1 u 000& 2 u 110& ) ~8!
A TWO-BIT QUANTUM PROTOCOL so s Ai 1s Bi 1s Ci 505x Ai •x Bi •x Ci . The cases where x Ai x Bi x Ci
We now show that if A~lice!, B~ob!, and C~arol! initially 5010 and 100 are similar by the symmetry of state ~5!. h
share a certain entanglement of qubits then there is a proto- Now, it follows that
S( D S( D S( D
col in which B and C each send a single bit to A, which n n n
enables A to determine f (x,y,z) @as defined by Eqs. ~3! and s A 1s B s C 5 s Ai 1 s Bi 1 s Ci
~4!#. i51 i51 i51
The entanglement involves 3n qubits, with each party n
having n of them. Call the n qubits that party pP $ A,B,C %
starts with q1p ,...,qnp . For each iP $ 1,...,n % , let the triple
5 ( ~ s Ai 1s Bi 1s Ci !
i51
qAi qBi qCi be in state n
1
2 ~ u 001& 1 u 010& 1 u 100& 2 u 111& ). ~5! 5 ( x Ai •x Bi •x Ci
i51
The bit that Bob broadcasts is some function respective values of f (x,y,z) are 0, 1, 0. No binary partition-
f : $ 0,1% 3 → $ 0,1% of his input data y alone. The function f ing of z will work for both possibilities.
partitions $ 0,1% 3 into two classes f 21 (0) and f 21 (1). Call Case 2.1.3 (000, 001, 110PS 0 ): Consider Alice’s per-
these two classes S 0 and S 1 ,and assume ~without loss of spective. If x5010 then, the possible values for (x,y,z) in-
generality! that 000PS 0 . After Bob broadcasts his bit, what clude ~010,000,101!, ~010,001,100!, ~010,110,011! for which
Alice and Carol each learn is whether yPS 0 or yPS 1 . For a the respective values of f (x,y,z) are 0, 0, 1; whereas, if x
two-bit protocol to be correct, it must always be possible at 5011 then the possible values for (x,y,z) include
this stage for Carol to send one bit to Alice that will enable ~011,000,100!, ~011,001,101!, ~011,110,010! for which the
Alice to completely determine the value of f (x,y,z). We respective values of f (x,y,z) are 0, 1, 1. No binary partition-
shall show that, whatever the partitioning S 0 ,S 1 is, there is ing of z will work for both possibilities.
an instance where Alice cannot determine the value of Case 2.1.4 (000, 001, 111PS 0 ): Consider Alice’s per-
f (x,y,z). There are 128 different possible partitionings, and spective. If x5010 then, the possible values for (x,y,z) in-
each is one of the seven types that are examined below. clude ~010,000,101!, ~010,001,100!, ~010,111,010! for which
Case 1 ( u S 0 u <2): Recall our convention that 000PS 0 . If the respective values of f (x,y,z) are 0, 0, 1; whereas, if x
S 0 has a second element then, by symmetry, no generality is 5011 then the possible values for (x,y,z) include
lost if we assume that it is either 100, 110, or 111. ~011,000,100!, ~011,001,101!, ~011,111,011! for which the
Thus, without loss of generality, 001,010,011PS 1 . Now, respective values of f (x,y,z) are 0, 1, 0. No binary partition-
should the bit that Bob broadcasts specify to Alice and Carol ing of z will work for both possibilities.
that yPS 1 , Carol must send one bit to Alice from which Case 2.2 ( u S 0 u contains no string of weight 1): We con-
Alice can completely determine the value of f (x,y,z).The sider the following three subcases.
bit that Carol sends induces a partition of the possible values Case 2.2.1 (111P ” S 0 ): In this case, 011,010,100,111
of z into two classes. If x5001 then, from Alice’s perspec- PS 1 . Suppose that Bob’s bit specifies that yPS 1 . Consider
tive, after receiving Bob’s bit but before receiving Carol’s Alice’s perspective. If x5001 then, the possible values for
bit, the possible values of (x,y,z) include ~001,001,111!, (x,y,z) include ~001,001,111!, ~001,010,100!, ~001,100,
~001,010,100!, ~001,011,101!, and the respective values of 010!, ~001,111,001! for which the respective values of
f (x,y,z) on these points are 1,0,1. Therefore, for the proto- f (x,y,z) are 1, 0, 0, 1; whereas, if x5010 then the possible
col to be successful in this case, the partition that Carol’s bit values for (x,y,z) include ~010,001,100!, ~010,010,111!,
induces in z must place 111 and 101 together in one class ~010,100,001!, ~010,111,010! for which respective values of
and 100 in the other class @otherwise Alice would not be f (x,y,z) are 0, 1, 0, 1. No binary partitioning of z will work
able to determine f (x,y,z) when x5001]. On the other for both possibilities.
hand, if x5011 then, from Alice’s perspective, the possible Case 2.2.2 (111PS 0 ): In this case, S 0 must contain an
values of (x,y,z) include ~011,001,101!, ~011,010,110!, element of weight 2. Without loss of generality, 011PS 0 .
~011,011,111!, and the respective values of f (x,y,z) on these Therefore, 000,011,111PS 0 . Consider Alice’s perspective.
points are 1, 1, 0. Since we have established that Carol’s bit If x5010 then, the possible values for (x,y,z) include
does not distinguish between z5111 and z5101, Carol’s bit ~010,000,101!, ~010,011,110!, ~010,111,010! for which the
is not sufficient information for Alice to determine respective values of f (x,y,z) are 0, 1, 1; whereas, if x
f (x,y,z) in this case. 5110 then the possible values for (x,y,z) include
Case 2 ( u S 0 u >3): For this case, we consider the subcases ~110,000,001!, ~110,011,010!, ~110,111,110! for which the
where either S 0 contains a string of weight 1 ~i.e., that has respective values of f (x,y,z) are 0, 1, 0. No binary partition-
exactly one 1! or does not. ing of z will work for both possibilities.
Case 2.1 ( u S 0 u contains a string of weight 1!: Without loss This concludes the proof that there is no classical protocol
of generality, assume 001PS 0 . By our convention, 000 for computing f (x,y,z) in which only two bits are commu-
PS 0 , and, after disregarding the obvious symmetries, there nicated among Alice, Bob, and Carol.
are four distinct possibilities for a third element of S 0 : 010,
011, 110, 111 and these are considered separately.
Case 2.1.1 (000, 001, 010PS 0 ): The argument is similar A THREE-BIT CLASSICAL PROTOCOL
to that in Case 1 using S 0 instead of S 1 . Consider Alice’s
perspective. If x5001 then, the possible values for (x,y,z) Although one might suspect that, without the use of en-
include ~001,000,110!, ~001,001,111!, ~001,010,100! for tangled particles, n bits of communication are necessary for
which the respective values of f (x,y,z) are 0, 1, 0; whereas, Alice to determine f (x,y,z) in general, it turns out that three
if x5011 then the possible values for (x,y,z) include bits always suffice.
~011,000,100!, ~011,001,101!, ~011,010,110! for which the The idea behind the method is to count the total number
respective values of f (x,y,z) are 0, 1, 1. No binary partition- of 0’s among all the 3n inputs of Alice, Bob, and Carol.
ing of z will work for both possibilities. Note that, for each iP $ 1,...,n % , if x i •y i •y i 51 then there are
Case 2.1.2 (000, 001, 011PS 0 ): Consider Alice’s per- zero 0’s among x i ,y i ,z i ,and if x i •y i •y i 50 then there are
spective. If x5001 then, the possible values for (x,y,z) in- two 0’s among x i ,y i ,z i .Let the number of 0’s among
clude ~001,000,110!, ~001,001,111!, ~001,011,101! for which x 1 ,...,x n be r A , the number of 0’s among y 1 ,...,y n be r B ,
the respective values of f (x,y,z) are 0, 1, 1; whereas, if x and the number of 0’s among z 1 ,...,z n be r C . Let k be the
5011 then the possible values for (x,y,z) include total number of terms among x 1 •y 1 •z 1 ,...,x n •y n •z n that
~011,000,100!, ~011,001,101!, ~011,011,111! for which the have value 0. Then, from the above, r A 1r B 1r C 52k.
1204 RICHARD CLEVE AND HARRY BUHRMAN 56
@1# A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 @6# E. Kushilevitz and N. Nisan, Communication Complexity
~1935!. ~Cambridge University Press, Cambridge, 1996!; the proof that
@2# J. S. Bell, Physics ~Long Island City, NY! 1, 195 ~1964!. the communication complexity of the inner product function is
@3# J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt, Phys. n is in Chapter 1: Exercise 1.29 and the remark immediately
Rev. Lett. 23, 1881 ~1969!. preceding it on p. 14 imply a lower bound of n11 in the
@4# D. M. Greenberger, M. Horne, and A. Zeilinger, in Bell’s model where both parties must acquire the answer; this, in
Theorem, Quantum Theory, and Conceptions of the Universe, combination with a remark on p. 5, implies a lower bound of
edited by M. Kafatos ~Kluwer Academic, Dordrecht, 1989!, p. n in the model where only Alice must acquire the answer.
69. @7# L. K. Grover, e-print Quantum Telecomputation, Report No.
@5# N. D. Mermin, Phys. Today 43 ~6!, 9 ~1990!; Am. J. Phys. 58, quant-ph/9704012, 1997 ~unpublished!.
731 ~1990!. @8# L. Fortnow ~private communication!.