Entanglement Purification vs.
Quantum Error Correction
A major challenge in quantum communication is the accurate transmission
of quantum information between Alice and Bob when the channel connecting them
is noisy. The fidelity of a quantum
state communicated across a noisy quantum channel decreases (exponentially, in
general) with the length of the channel.
Thus, communication would be limited to extremely short distances if it
weren’t for the method of teleportation.
This requires shared entangled particles between Alice and Bob so that
local operations and classical communication allow “transmission” of quantum
states.
The problem remains, however, how can
The first protocol to be discussed was developed by Bennett (1996) and
involves projecting with some probability to a state of higher entanglement
fidelity. This method will first be
explored in its simplest one-particle ensemble form and will be generalized to
two-particle ensembles. For example,
pure entangled EPR pairs are required for quantum teleportation. Another two-particle purification protocol, quantum
privacy amplification, first developed by Deutsch (1996), designed for
cryptographic purposes will be discussed and its superior efficiency will be
shown.
Entanglement purification protocols, like quantum error correction
codes, can be considered as a form of protecting quantum states from their
interaction with the environment.
Quantum error correction codes can be used to purify EPR pairs between
Alice and Bob. For example,
Conversely,
Consider a collection of qubits with available states |0ñ and |1ñ, in a mixture of states. This could be an ensemble of spin ½ particles
with polarization spin up and spin down, along the spin axis, partially
polarized along that axis. The corresponding
density matrix of the ensemble is
. (1)
The fraction f is called the entanglement
fidelity of the ensemble.
The qubits can be measured in the {|0ñ, |1ñ} basis to obtain a sub-ensemble of particles in the |0ñ state by discarding those projected into the |1ñ state. For the case of spin ½ particles, an inhomogeneous magnetic field (such as one created by two magnets north facing south) spatially separates spin up from spin down particles. By selecting only one path, say the spin down direction, |0ñ, the particles can be thought of as purified, with the projection distilling the ensemble to the desired pure state.
Assume now that the measurement destroys the qubit
(for example, the mechanism could absorb the particle) while still conveying
the result of measurement. For example,
the apparatus would click if the particle followed the spin up path before
being absorbed. Instead of sending the
actual particle through the measurement apparatus, a copy could instead be
used. Of course, the no-cloning theorem
proves it is impossible to create a copy of a general quantum state; however,
it is possible to copy the selected basis states using the CNOT gate. For example, a superposition
is transformed to
, so that the no-cloning theorem is not violated. For a target qubit with the density matrix |0ñá0|,
performing a CNOT with (1) as the control qubit gives the resulting entangled
state
, (2)
see Figure 1a.
Measuring the target qubit in the {|0ñ, |1ñ}
basis, if measurement yields a click (i.e. the target qubit is measured to be
in the |0ñ)
state then we have with certainty the control qubit in the state |0ñá0|;
otherwise, the control qubit is in the state |1ñá1|. To purify the ensemble to states |0ñá0|,
the measurements yielding a |1ñ are discarded, distilling the mixture to a pure
subset. In this setup, the fraction of
qubits discarded is
, leaving the remaining fraction in the desired pure state.
Above, we’ve assumed a source of existing pure states |0ñá0| for the target qubit. If these were available, however, there would be no need to distil our mixed state (1). Can the distillation be performed without these?

Figure 1. Selection of spin up or spin down using a Stern-Gerlag apparatus (SGA) (polarized particles are deflected with a double magnet). (a) The state of an impure qubit is copied onto an auxiliary pure qubit for projection onto a pure state; (b) the impure state is copied onto another similar impure state for projection into a purified state (taken from Bouwmeester, 2000).
Consider replacing the target qubit with another of our ensemble of mixed state particles. That is divide the collection of qubits to be distilled in half: the first half destined as control qubits to be distilled, the second half used as target bits to be sacrificed. After the CNOT is performed on each pair (Figure 1b), we obtain the ensemble of entangled state with density matrix
. (3)
If the target qubit is then measured and found to be in the |0ñ state, the (renormalized) density matrix of the control qubit becomes
(4)
where
. The function
is plotted in Figure
2. Clearly, for
the remaining control
qubits (those for which the target qubit was measured in state |0ñ) are a purified ensemble since
and there is an
increased fraction of particles in the desired state |0ñ. If the process is iterated,
the ensemble can be distilled arbitrarily closely to the pure state |0ñ; however, to distil to a pure state requires infinitely many qubits
to be wasted.
The number of qubits discarded per
iteration is
where the first term
represents the sacrificed target qubits and the second term arises from the
discarded control qubits projected into the equal superposition
.
Consider the nth iteration. The purity of the sample is
while the fraction of
qubits discarded in this iteration is
, where
. The total fraction
discarded in the n iterations is
. (5)

Figure 2. The fidelity after an iteration of the purification algorithm. Note that f' > f for f > ½.
Figure 3 shows the number of iterations required to reach fidelity 99% as well as the fraction of discarded qubits versus the initial entanglement fidelity.

Figure 3. The efficiency of the purification protocol for the two-particle EPR pair.
Now consider the case
of mixed entangled states, such as the two-particle entangled EPR pairs shared
by Alice and Bob, where the particles are spatially separated. For simplicity consider the superposition of
the
and
(6)
with
the density matrix
. (7)
Note
that unless f = ½, the state is inseparable, so that it is indeed a
mixed entangled state with fidelity f.
Clearly, if Alice and Bob send their particles through a measurement
(such as measurement of spin in the case of spin ½ particles) they could
distinguish the two
both measure the same thing (up-up or
down-down); for pairs in the state
each measures the
opposite (up-down or down-up). This
isn’t of much use since this destroys any existing entanglement and the
particles leave the apparatus in a product state.
Applying the same
trick as employed in the case of one-particle purification, consider the
bilateral CNOT with
as the target bits
(one of each of the entangled particles is used for each of Alice’s and Bob’s
CNOT gate), depicted in Figure 4
and summarized in Table 1. The CNOT maps
as follows to the
(8)
so
that
plays the analogue to
the |1ñ state of the single particle system,
while
is analogous to |0ñ. Thus measurement of the
target bits of the CNOT operations would signal whether the control bits had
been projected into the
or
Bell state (as before,
Alice and Bob measure the same thing in the case of
and opposites for
).

Figure 4. Purification of a mixed entangled pair by local unitary CNOT gates, measurement and classical communication. Note the two CNOT's compose a bilateral CNOT (B-CNOT) (from Bouwmeester, 2000).
But were these pure
states available, no
purification scheme would be necessary, as before. Splitting the ensemble of EPR pairs in two,
one for purification, the other sacrificial half as the indicator qubit, we
attempt the same scheme as was employed in the one-particle case. After the bilateral CNOT transformation the
qubits are in the entangled state
(9)
and each of Alice
and Bob measure their target particle and transmit their result classically to
determine whether or not to discard the first particle. If they both measure the same thing, they
keep the first entangled particle, whose fidelity has increased to
, as in the single particle analogue. If their measurements differ, their control
particles are in the state

and are no longer entangled. This is similar to the one particle case where the rejected control qubits were in the equal superposition of |0ñ and |1ñ. In that case, the purification scheme is useless (recall we needed f > ½), just as here all entanglement is lost.
The fraction of wasted entangled pairs is the same as in the one particle protocol and the number of iterations for appropriate purification is also the same (Figure 3). Note that in the two-particle purification protocol the exchange of classical information was used. This is in fact an integral part of any purification protocol between parties.
Consider next the more general mixed two-particle state. Bennett first developed the following method (1996). As before, we define the fidelity of the sample as
![]()
where we are now interested
in distilling to the state
.
The first step has Alice and Bob perform a random
bilateral rotation, that is each performs one of {I, Rx Ry
Rz} at a random angle on their half of the pair chosen at random,
independently. A normal bilateral
rotation is simply when both Alice and Bob perform a rotation about the
same axis, but in different directions (that is, Bob performs the inverse
transformation of
. (10)
Note that the state
is invariant under
each of the bilateral rotations, while the remaining
entry).
Table 1. The unilateral and bilateral rotation
transformations on the four
|
Source qubit |
|
|
|
|
|
||
|
Bilateral p/2 rotations |
Bx |
|
|
|
|
|
|
|
By |
|
|
|
|
|
||
|
Bz |
|
|
|
|
|
||
|
Unilateral p rotations |
sx |
|
|
|
|
|
|
|
sy |
|
|
|
|
|
||
|
sz |
|
|
|
|
|
||
|
Bilateral CNOT |
|
|
|
|
|
(source) |
|
|
|
|
|
|
|
(target) |
||
|
|
|
|
|
|
(source) |
||
|
|
|
|
|
|
(target) |
||
|
|
|
|
|
|
(source) |
||
|
|
|
|
|
|
(target) |
||
|
|
|
|
|
|
(source) |
||
|
|
|
|
|
|
(target) |
||
Operations
on the qubits required for Bennett’s algorithm (summarised in Table 1) are:
Unilateral Pauli rotations: one of the
Pauli transformations {I, sx, sy, sz} is applied to one of the
particles in the entangled pair. The
various
(i) The bilateral p/2 rotations that were already discussed.
(ii) The bilateral CNOT (B-CNOT) gate depicted in Figure 4.
(iii) Measurement by Alice and Bob of each of their qubits. Recall that this destroys any entanglement between the particles.
The algorithm itself requires two mixed
entangled pairs with fidelity f > ½ shared by
(A1)
A unilateral sy rotation is performed on each of the
two pairs, converting them from mostly
to mostly
states:
![]()
(A2)
A B-CNOT is performed as in
Figure 4 after which the target pair is measured along the z-axis. If Alice and Bob measure the same result
(i.e. the input was a true
state) they keep the
control qubit; otherwise, they discard it.
After measurement, the density matrix of the control qubit becomes
(11)
which is easily worked out using the transformation rules of Table 1.
(A3)
If the control qubit is not
discarded, then it is converted back to mostly
states via a
unilateral sy
rotation, and is made rotationally symmetric as before by a random bilateral
rotation.
The fidelity of the output state, after an iteration of the algorithm, has improved to
(12)
which is plotted in Figure 5. The fraction of discarded qubits is the same as in the last protocol and is given by (5). Since f¢(f) > f for f > ½, with an initial fidelity exceeding ½, repeated iteration of the algorithm yields arbitrarily high fidelity (of course sacrificing many qubits in the process). Figure 6 shows the number of iterations required to reach fidelity 99% as well as the fraction of discarded qubits versus the initial entanglement fidelity, just as was done in the simplified protocol in the previous section.

Figure 5. The fidelity after one iteration of
Bennett's algorithm. Note that f'
> f for f > ½.

Figure 6. The efficiency of the purification of a general mixed entangled state using Bennett's algorithm.
The quantum privacy amplification scheme (QPA) was developed to
account for noise in the quantum communication channel over which information
must be sent as part of a quantum cryptographic scheme. Prior to the inclusion of entanglement
purification protocols in the quantum cryptography schemes, security of such
schemes was only proven over idealized (noiseless) channels. An eavesdropper (Eve) in such protocols is
detected by measuring whether the generated EPR pair is entangled with any
third party. With the possibility of
noise, however, whether the entanglement of their particles is with a potential
eavesdropper or with the environment becomes indistinguishable.
In the idealized cryptography schemes, Alice and Bob have a supply of
maximally entangled states
(6), where each has
one of the entangled particles. In the
presence of noise, each pair would become entangled with each other and with
the environment and would be described by a superposition of the four
is the entanglement
fidelity of the ensemble.
As in the other
purification protocol, divide the ensemble of mixed entangled pairs into two,
one for purification, the other for measurement.
(13)
on each of her two qubits (from each of the two divisions) while Bob performs the inverse operation
(14)
on
each of his. The effect of such a
transformation in the
.
Table 2. The
transformation in the
|
Input |
|
|
|
|
|
Transformed |
|
|
|
|
Thus after this first step, the diagonal entries of the density matrix become {a, d, c, b}.
Next Alice and Bob perform a CNOT with the control bit from the first division of the entangled pairs (Figure 4) with the entangled state

Alice and Bob then measure the target qubits of the CNOT and communicate classically their result. If their outcomes coincide they keep their control qubits, otherwise they discard them, yielding the density matrix
.
After an iteration of this procedure, the average (taken over different coincident outcomes) of the new diagonal elements of the density matrix are {an+1, b n+1, c n+1, d n+1} (Deutsch, 1996)
(15)
where
is the probability
that Alice and Bob measure the same outcome on their target qubits and {an,
b n, c n, d n} are the diagonal entries of the
density matrix prior to this iteration.
Note that {a, b, c, d} = {1, 0, 0, 0} and
{0, 0, 1, 0} are fixed points of the mapping (15); however, these fixed points
aren’t necessarily approached monotonically.
For ao > ½, the first is the only fixed point; whereas for
co > ½, the second is the only fixed point. Each is a local attractor. Computer simulations show that in their
separate regions they are global attractors (Deutsch, 1996) and an analytical
proof was recently presented. The latter
also proved that for an initial state bo > ½ (and of course for ao
> ½) the iteration converges to
, and similarly for co > ½ or do >
½ the iteration converges to the pure state
. Thus if initially the mixture has fidelity exceeding ½, but
which is otherwise in an arbitrary state, then the states of pairs surviving
each iteration always converge to the unit-fidelity pure state
.
Figure 7 shows the convergence of the special initial states with bo = co = do for a range of initial fidelities. The number of qubits discarded in the process is about 1000 times more efficient for initial fidelity close to 0.5 than using Bennett’s algorithm (Deutsch, 1996).

Figure 7. Average fidelity as a
function of the initial fidelity and the number of iterations of the quantum
privacy amplification (from Bouwmeester, 2000).
The fundamentals of
entanglement purification have been explored.
The main idea behind such protocols is the distilling of a large
ensemble of not-so-pure entangled states (the ones examined required an initial
fidelity exceeding ½) down to a smaller (generally much smaller) sub-ensemble
of more pure entangled states.
Generally, the algorithms are repeated to attain the required
fidelity. The yield of such iterations
was examined in the simpler methods (Figure 3); the efficiency was discussed
for the various protocols.
In addition to the methods discussed above, other purification protocols are being researched, such as purification via entanglement swapping (Bose, 1998). There also exist protocols whose aim is to create maximally entangled EPR pairs between spatially distant particles, such as quantum repeaters (Briegel, 1998).
C. H. Bennett, G. Brassard, S. Popescu, B. Schumacher, J. Smolin, and W. K. Wootters, Phys. Rev. Lett. 76, 722 (1996).
S. Bose, V. Vedral, and P. L. Knight, Phys. Rev. A 57, 822 (1998).
H.-J. Briegel, W. Dür, J. I. Cirac, and P. Zoller, Phys. Rev. Lett. 81, 5932 (1998).
H.-J. Briegel, Principles of Entanglement Purification,
from D. Bouwmeester, A. Ekert, and A. Zeilinger, The Physics of Quantum
Information.
D. Deutsch, A. Ekert, R. Jozsa, C. Macchiavello, S. Popescu, and A. Sanpera, Phys. Rev. Lett. 77, 2818 (1996).
C. Macchiavello, Quantum Privacy Amplification, from
D. Bouwmeester, A. Ekert, and A. Zeilinger, The Physics of Quantum
Information.