You are on page 1of 6

Quantum Information (WiSe 22/23) Last updated: 2022-12-22

Chapter 3.2: Quantum Communication: Complexity


Instructor: Dr. Christian Schilling Scriber: Ignacio Cirac, Sirui Lu, and Rahul Trivedi

1 Introduction
Communication complexity This field analyzes protocols for communication tasks and how many
bits have to be exchanged to achieve them. Quantum Communication Complexity analyzes the
same but using quantum physics and in terms of qubits.

Example: dense coding We have already seen one example where quantum physics enables a
more efficient solution of a certain communication task, and thus there is a quantum advantage.
With dense coding, Alice and Bob can send 2n bits of information by sending n qubits if they pre-
share entangled pairs. In this chapter, we will see other examples displaying a quantum advantage
in the context of quantum communication complexity.

1, σx , σy , σz

|Φ+ i

A B

Figure 1: A schematic of dense coding.

Example: agenda problem Alice and Bob want to know if there are time slots in their agendas
for which both of them are available. For that, they will have to send their whole agenda, and thus
a large number of bits. What if they have quantum channels to communicate? Can they solve the
task by transmitting fewer qubits? A related problem is that Alice and Bob want to propose a
date for a meeting, and they want to see if it is the same. This can be formulated in an abstract
form as follows: Alice and Bob are given a bit string x, y ∈ {0, 1}n , respectively, and want to see
if they are the same. We count the number of qubits that have to be communicated in order to
solve this problem, and compare it with the number of bits. These numbers will depend on n, in
general. The goal of quantum communication complexity is to compare them in the limit n  1.

Big O notation Since we are interested in the limit n  1, we are only concerned with the
asymptotic behavior of functions that count the number of bits or qubits. For that, we write
f (n) = O(g(n)) to indicate that ∃c > 0 √
such that the function f (n) 6 cg(n) as n → ∞. We will
often write expressions such as O(n) O( n) or O(log n) to denote different rates of growth.

Complexity In general, we write the classical and quantum communication complexity as fC (n)
and fQ (n), referring to the number of bits and qubits that have to be communicated in order to
solve a task. We will compare the classical and quantum communication complexity in asymptotic
limit n  1. If for some α < 1, the classical and quantum communication complexity for a certain
task are
fC = O(n), fQ = O(nα ), (1.1)
we say there is a polynomial advantage, whereas if
fC = O(n), fQ = O(log n) (1.2)
we say there is an exponential advantage.

Resources For some problems, Alice and Bob may have additional resources. For instance, they
may share a list of random bit strings, or a set of entangled pairs. The above complexity may
change in those cases.

1
Chapter 3.2 Quantum Information (WiSe 22/23)

2 Classical communication complexity

x ∈ {0, 1}n y ∈ {0, 1}n

A B
Classical communication
f (x, y)

Figure 2: Model of classical communication complexity.

Bit strings Typically, in a communication task, Alice and Bob are each given a bitstring and
they have to compute some function f together. A bit string is a sequence of bits, x = x1 . . . xn ,
where xi = 0, 1. The set of all possible bit strings of length n is denoted by {0, 1}n .

Formulation In communication complexity, the problem is usually formulated as follows. Given


a function f : X × Y → {0, 1} where Alice holds the first set of bits x ∈ X = {0, 1}n and Bob holds
the second set of bits y ∈ Y = {0, 1}n , we would like to know how much communication between
Alice and Bob is required to compute f (x, y). The complexity of the problem depends on the
type of communication (classical, quantum), the degree of accuracy that must be achieved (exact,
bounded-error, etc.), and other resources (shared randomness, entanglement), that we specify in
the coming subsections. First, let us see some examples.
Example 2.1 (The module 2 inner product
Pn problem). Alice and Bob want to compute the function
f (x, y) = x · y mod 2, where x · y = i=1 xi yi .
Example 2.2 (The equality problem). Alice and Bob want to compute the function f (x, y) =
(
1 if x = y
. This is related to the agenda problem mentioned in the introduction.
0 if x 6= y

Example 2.3 (The intersection problem). Alice and Bob want to compute the function f (x, y) =
(
1 if ∃i, xi = yi = 1
.
0 otherwise

2.1 Problems
The function f (x, y) : D ⊆ X × Y → {0, 1}, to be computed in the communication task with the
domain D being a subset of all possible strings, is a
• Total function if the domain D = X × Y . The functions used in Examples 2.1, 2.2, 2.3 are
all total functions.
• Promise function: if the domain of f D ⊂ X × Y (i.e. it is promised that some arguments
x, y are not encountered).

Example 2.4 (An example of promise function). The even-length 2n-bit inputs x and y are
promised such that either all the bit strings are the same or just differ in half of the bits.
Alice and Bob would like to find out which case is true.

2.2 Protocols
To solve the communication task, Alice and Bob will use a protocol consisting of several rounds
of communication. In each round, they can communicate several bits of information. In the end,
Bob announces f (x, y). The communication cost N (x, y) is the total number of bits communicated
while computing f (x, y).

2
Chapter 3.2 Quantum Information (WiSe 22/23)

Complexity The value of N , in general, depends on Alice’s and Bob’s inputs. The communication
complexity is the worst case communication cost,

fC (n) = max N (x, y) (2.1)


x,y

where the maximization is done over all strings x, y with length n.

Deterministic and non-deterministic protocol There are two kinds of protocols, deterministic
and non-deterministic (called Bounded-error protocol).

(1) Deterministic protocol: To compute f (x, y) with probability p = 1.


(2) Bounded-error protocol: To compute f (x, y) with probability p > 2/3 ∀x, y.

Note: that for practical purposes, a bounded error protocol is enough for solving a task. If we
repeat L times and use the majority vote to decide if f (x, y) = 0, 1, the error probability will
decrease exponentially. By majority vote we mean that we choose f (x, y) = 0 if more than half of
the times we obtained 0, and f (x, y) = 1 otherwise. The error probability will be
L/2    i  L−i
X L 2 1
Perror = . (2.2)
i=0
i 3 3

Thus, taking e.g., L = 100, the probability of failing is about 10−4 . Note that the choice of 2/3 is
arbitrary. We could have taken other numbers > 0.5 and the asymptotic scaling will not change.

2.3 Randomness
In some protocols, Alice and Bob are allowed to share an arbitrarily large number of random bits.
This may change the complexity of the problem. Let us take a problem in which we see randomness
and bounded error in action.
Example 2.5 (A bounded-error
( protocol for the equality problem). Alice and Bob compute the
1 if x = y
function f (x, y) = . Since Alice and Bob can only compare their bits one by one,
0 if x 6= y
in the worst-case they will only realize that they are different in the last bit. Thus, the exact
communication complexity is n. However, they are able to do it more efficiently if they share a set
of random bitstring r1 , . . . , rm = {0, 1}n and they work within the bounded error model. In this
setup, Alice and Bob can proceed in the following way. In round t, Alice computes rt · x mod 2
and sends it to Bob, who compares it with rt · y mod 2. They output f (x, y) = 1 if in every round
they obtain the same results and zero otherwise. The error probability scales exponentially with the
number of rounds L
1
P (x 6= y ∧ all the results are the same) = L . (2.3)
2
After two rounds, the probability of guessing correctly is 3/4, which is already larger than the
required 2/3. The bounded error communication complexity is f (n) = O(1), which is an exponential
gain compared to the case that Alice and Bob do not share randomness. Note that in this case the
randomness allows them to get around the worst case.

3 Quantum communication complexity


Now Alice and Bob can send qubits to each other. There are different ways of defining quantum
protocols.

Yao’s model In this model, Alice and Bob send qubits to each other, which they can act on and
send back. Then the quantum communication complexity fQ (n) is equivalent to the number of
qubits sent through the quantum channel.

3
Chapter 3.2 Quantum Information (WiSe 22/23)

|Φ+ i
A B
A B
Quantum communication

Figure 3: Yao’s model of quantum com- Classical communication


munication complexity.
Figure 4: Cleve and Buhrman’s model
of quantum communication
complexity.

Cleve and Buhrman model In this model, Alice and Bob are assumed to share an arbitrary
amount of maximally entangled states and communicate classically. Then the quantum communi-
cation complexity fQ (n) is equivalent to the number of bits sent through the classical channel.

Remark Since we can use teleportation to trade ebits for bits, Cleve and Buhrman’s model
includes Yao’s model up to a constant factor of 2. Additionally, Cleve and Buhrman’s model
includes the shared randomness case because they can simply measure a pair in the Z basis to
obtain the same random numbers. Here, we will use Yao’s model.
In the following, we will give several examples of problems where there is a quantum advantage.
Note that the field of quantum communication complexity is a very active area of research, and
relatively few results are known.

3.1 Example: Promise function


Now we consider an example of promise functions, D 6= X × Y . Let us take a function f defined
through (
1 if x = y
f= (3.1)
0 if x and y differ in exactly n/2 bits
The inputs are constrained such that only these two cases will be asked.
Example 3.1. x = (10100101) and y = (11000011) differ in 4 (i.e., half of the) bits.

Quantum protocol Consider for simplicity n = 2m .


1. Let us define
n
1 X
|ri = √ (−1)xi |ii (3.2)
n i=1

where |1i = |00 · · · 0i, · · · , |ni = |11 · · · 1i are states of m qubits.

2. Given x = (x1 , · · · , xn ), Alice sends m qubits in a state


n
1 X
|ψi = √ (−1)xi |ii . (3.3)
n i=1

3. Bob applies the unitary transformation U : |ii → (−1)yi |ii


1 X
|ψi → |ψ 0 i = U |ψi = √ (−1)xi +yi |ii . (3.4)
n i

If x = y then |ψ 0 i = |ri. Otherwise, |ψ 0 i is orthogonal to |ri.


4. Bob measures in the basis {|ri hr| , 1 − |ri hr|}.
The number of qubits need to be sent from Alice to Bob in this quantum protocol is m = log n.
Therefore, the quantum communication complexity is fQ (n) = O(log n).

4
Chapter 3.2 Quantum Information (WiSe 22/23)

Analysis The classical complexity depends on whether we consider deterministic or bounded error
protocols.
(a) Deterministic protocol. fC (n) = 0.007n ≥ fQ (n). Thus, there is an exponential quantum
advantage.
(b) Bounded-error protocol with randomness: if Alice and Bob have access to a public coin the
complexity is shown to be fC (n) = O(1). If they only have access to private coins, the
complexity is fC (n) = O(log n). Thus, in practice, the quantum protocol is not better than
a classical one.

3.2 Example: Intersection problem


There are problems for which there is a quantum advantage even under the bounded error setup.
An example is the intersection
√ problem defined in the example 2.3. In that case, fC (n) = O(n)
whereas fQ (n) = O( n log n). Thus, there is a polynomial advantage.

3.3 Example: The vector in the subspace problem


This is a problem for which there is an exponential advantage, even under the bounded error setup.
The problem is very academic because it is just constructed to demonstrate that an exponential
quantum advantage is possible. Additionally, it is a promise problem because D 6= X × Y .

Formulation Alice is given a vector ~v ∈ Rm with kvk = 1. Bob is given two orthogonal projectors
acting on Rm , P0 and P1 , fulfilling P0 +P1 = 1. That is, they project onto two orthogonal subspaces
of Rm . The promise is that ~v lies in one or the other subspace, i.e., either P0~v = ~v or P1~v = ~v .
Note that Bob just has to know P0 because P1 = 1 − P0 . Thus, the promise function is
(
0 if P0~v = ~v
f ({~v , P0 , P1 }) = . (3.5)
1 if P0~v = 0

The fact that it is a promised function relies on the fact that not every pair ~v , P0 are permitted.

Detail Since we deal with real numbers, in principle one needs an infinite number of bits to encode
the components of ~v and the matrix elements of P0 . This is why we just take an approximation
and encode each real entry in log m bits. Under this approximation, the total number of bits, n,
required to encode ~v and P0 are
n = m log m + m2 log m = O(m2 log m). (3.6)
Pm
Quantum protocol Alice prepares |ψi = i=1 vi |ii using log m qubits and sends them to Bob,
who measures in the basis {P0 , P1 }. We thus have fQ (n) = log m = O(log n).

Classical protocol One can show that fC (n) ≥ O(n1/3 ). Thus, as announced, there is an expo-
nential quantum advantage.

3.4 Example: Quantum fingerprint


We finish this part with an example that does not fit the setup defined above but also displays
a quantum advantage. It is based on a simplified model called the simultaneous message passing
model (see Fig. 5 for an illustration). In this model, Alice and Bob are again given two bit strings
x and y, but there is a trusted third referee who takes on the responsibility of computing the
function f (x, y). The constraints are that Alice and Bob cannot communicate with each other,
and only are allows to send messages to the referee, who cannot send any information back.

Equality problem To illustrate this simultaneous message passing model, we will use the equality
function (
1 if x = y
f (x, y) = . (3.7)
0 if x 6= y

5
Chapter 3.2 Quantum Information (WiSe 22/23)

Referee: decide if x = y using SWAP test.


i |φ
x yi

Alice: x ∈ {0, 1}n Bob: y ∈ {0, 1}n

Figure 5: Quantum fingerprint protocol (the equality problem).

Quantum protocol
(a) Alice and Bob each holds bit-string x ∈ {0, 1}n and y ∈ {0, 1}n . They encode their bit
strings into C(x) ∈ {0, 1}N and C(y) ∈ {0, 1}N . The encoding function C satisfies the
following property: (i) N = O(n); (ii) has distance D[C(x), C(y)] ∈ [0.49N, 0.51N ] if x 6= y;
(iii) D[C(x), C(x)] = 0. Here D is the number of bits in which they differ (the Hamming
distance).
(b) Alice and Bob prepare log(n) qubit states |φx i and |φy i and send them to the referee.
N N
1 X 1 X
|φx i = √ (−1)C(x)j |ji, |φy i = √ (−1)C(y)j |ji. (3.8)
N j=1 N j=1

(c) The referee perform the SWAP test on the received states |φx i and |φy i. The SWAP test
consists of the following steps.
(1) Take an auxiliary qubit in the state |+i, so that the initial state is |+i |φx i |φy i. Apply
the controlled swap operation U = |0i h0| 1 + |1i h1| SWAP, using the auxiliary qubit as
the control qubit. The resulting state is
1 1
|ηi = √ |0i |φx i |φy i + √ |1i |φy i |φx i . (3.9)
2 2

(2) Then we measure the ancilla qubit in the basis {|+i , |−i}. The probabilities of obtaining
+1 and −1 are
1
P± = |h±|ηi|2 = |(1 ± |hφx |φy i|2 ). (3.10)
2
Therefore, if x = y, P+ = 1 and P− = 0. Otherwise, if x 6= y, |hφx |φy i|2 ∈ [−0.02, 0.02],
we have P+ = 1/2 + , with  ∈ [−0.02, 0.02]. To get a guess probability larger than
2/3, we can repeat the process for a constant number of times. Hence, the quantum
communication complexity for this problem is fQ (n) = O(log n).

Classical protocol Classically, the communication complexity is shown to be fC (n) = O( n).
Therefore in this problem with quantum computers and quantum channels one can achieve an
exponential saving in communication complexity.

♣♣♣

Acknowledgement.— We thank Adrian O. Paulus for providing their handwritten notes.

These lecture notes were kindly provided by Ignacio Cirac, Sirui Lu and Rahul Trivedi.

You might also like