(16th December 2001)

Linear Optical CNOT Gate in the Coincidence Basis

T. C. Ralph, N. K. Langford, T. B. Bell and A. G. White

Centre for Quantum Computer Technology,

Department of Physics,
University of Queensland,
arXiv:quant-ph/0112088v1 16 Dec 2001

QLD 4072, Australia

Fax: +61 7 3365 1242 Telephone: +61 7 3365 3412
email: ralph@physics.uq.edu.au

Abstract
We describe the operation and tolerances of a non-deterministic, coincidence basis, quantum
CNOT gate for photonic qubits. It is constructed solely from linear optical elements and requires
only a two-photon source for its demonstration.

1
I. INTRODUCTION

Qubits based on the polarisation state of individual photons have the advantage of low
decoherence rates and are easily manipulated at the single qubit level. Optical parametric
amplification experiments have been very successful in producing and analysing a large range
of two photon entangled states [1, 2, 3, 4]. A key “trick” in these types of experiments is to
work in the coincidence basis in which only events where two photons are detected in the
same, narrow time window are recorded. The entangled state postselected this way may be
a pure Bell state even though the total state is non-deterministic and may have experienced
considerable mixing from photon loss. Such systems are not scaleable in the quantum
computational sense in their present form but none-the-less provide an excellent testing
ground for quantum information concepts. Useful application of this type of technology
seems much closer in the realm of quantum communications.
A key two qubit gate is the Controlled Not (CNOT) gate. A deterministic CNOT gate
would require either very high non-linearities [5] or very complex linear networks [6]. Build-
ing on the latter ideas a linear, coincidence basis CNOT has been described [7] which could
be a useful test-bed. However it requires a 4-photon input, which is challenging. Two pho-
ton, coincidence basis gates, which perform some, but not all of the operations of a CNOT
gate have also been described [8, 9, 10].
In this paper we discuss a linear, coincidence basis gate which performs all the operations
of a CNOT gate and requires only a two photon input [11]. In section 2 we describe its
construction and ideal operation. In section 3 we consider the effect of imperfections in its
construction, particularly focusing on the effect of beamsplitter and mode-matching errors
on the gates efficacy as a Bell state analyser. In section 4 we conclude. Recently Hofmann
and Takeuchi have independently described a very similar gate [12]. Our analysis should
also apply to their construction.

II. THE GATE

The gate is shown in Fig.1. All beamsplitters, B1, B2, B3, B4, and B5, are assumed
asymmetric in phase. That is, it is assumed that the operator input/output relations (the
Heisenberg equations) between the two input mode operators (ain and bin ) and the corre-
sponding output operators (aout and bout ) for the beamsplitters have the general form
√ q
aout = ηain +
1 − ηbin
q √
bout = 1 − ηain − ηbin (1)

where η (1 − η) is the reflectivity (transmittivity) of the beamsplitter. Reflection off the

bottom produces the sign change except for B1 and B2 which have a sign change by reflection
off the top. This phase convention simplifies the algebra but other phase relationships will
work equally well in practice. Beamsplitters B3 and B4 are both 50:50 (η = 1/2). The
beamsplitters B1, B2 and B5 have equal reflectivities of one third (η = 1/3).
We employ dual rail logic such that the “control in” qubit is represented by the two
bosonic mode operators cH and cV . A single photon occupation of cH with cV in a vacuum
state will be our logical 0, which we will write |Hic (to avoid confusion with the vacuum
state). Whilst a single photon occupation of cV with cH in a vacuum state will be our

2
logical 1, which we will write |V ic . Superposition states can also be formed via beamsplitter
interactions. Similarly the “target in” is represented by the bosonic mode operators tH and
tV and the states |Hit and |V it , with the same interpretations as for the control. The use
of H and V to describe the states of the qubits of course alludes to the usual encoding in
polarisation [13]. To go from polarisation encoding to dual rail spatial encoding and vice
versa in the lab requires a polarising beamsplitter and half-wave plate.
The Heisenberg equations relating the control (cH , cV ) and target (tH , tV ) input modes
to the their corresponding outputs are
1 √
cHO = √ ( 2vc + cH )
3
1
cVO = √ (−cV + tH + tV )
3
1
tHO = √ (cV + tH + vt )
3
1
tVO = √ (cV + tV − vt )
3
1 √
vcO = √ (−vc + 2cH )
3
1
vtO = √ (tH + tV − vt ) (2)
3
Ancillary, vacuum input modes, vc and vt , complete the network. The gate operates by
causing a sign shift in the interferometer formed by the splitting and remixing of the target
modes, conditional on the presence of a photon in the cV mode. Thus the target modes swap
if the control is in the state |V ic but do not if the control is in state |Hic. This is always
true when a coincidence is measured between the control and target outputs (photons are
detected at the same time). However such coincidences only occur one ninth of the time, on
average. The other eight times out of nine either the target or the control or both do not
contain a photon. This can be seen explicitly by calculating the output state of the system
in the Schrödinger picture. Consider the general input state
|φi = (α|HHi + β|HV i + γ|V Hi + δ|V V i)|00i
= (αc†H t†H + βc†H t†V + γc†V t†H + δc†V t†V )|0000i|00i (3)

where the ordering in the kets is |ncH ncV ntH ntV i|nvc nvt i with ncH = c†H cH etc and we use
the short hand |1010i = |HHi etc where appropriate. For a time symmetric linear network
such as that in Fig.1, the output state can be directly obtained from the input state, Eq.3,
by substituting input operators for the output operators given by Eq.2. Thus we obtain

|φiout = (α c†HO t†HO + β c†HO t†VO + γ c†VO t†HO + δ c†VO t†VO )|0000i|00i
1
= {α|HHi + β|HV i + γ|V V i + δ|V Hi
3√ √
+ 2(α + β)|0100i|10i + 2(α − β)|0000i|11i + (α + β)|1100i|00i
+(α − β)|1000i|01i + α|0010i|10i + β|0001i|10i
−(γ + δ)|0200i|00i − (γ − δ)|0100i|01i + γ|0020i|00i
+(γ − δ)|0010i|01i + (γ + δ)|0011i|00i + (γ − δ)|0001i|01i + δ|0002i|00i} (4)

3
The state postselected in the coincidence basis is then just

|φicb = α|HHi + β|HV i + γ|V V i + δ|V Hi (5)

occuring with probability one ninth. The relationship between Eq.3 and Eq.5 is a CNOT
transformation.
It is also useful to look at the coincidence number expectation values, obtained directly
from the Heisenberg equations (Eq.2). These can be interpreted as the predicted output
coincident count rates normalized to the input pair rate. An example is given in Table 1
which shows the count rates for logical basis inputs. A more interesting case is to use the
four Bell-states,
1
|ψ ± i = √ (|Hic |Hit ± |V ic |V it )
2
1
|φ± i = √ (|Hic |V it ± |V ic |Hit ) (6)
2
as inputs and to detect the control in the superposition basis by mixing the control outputs
on a 50:50 beamsplitter before detection:
1
cS1 = √ (cHO + cVO )
2
1
cS2 = √ (cHO − cVO ) (7)
2
In Table 2 the count rates for this arrangement are presented showing the ability to distin-
guish all four Bell states (albeit with non unit efficiency). Such a Bell state analyser could
have significant applications in quantum communications. In the next section we will use
this application as an example in order to investigate the effect of non-optimal parameters
on the gate.

III. NON-OPTIMAL OPERATION

The accuracy with which the gate operates will be determined by how closely the pa-
rameters of the constructed gate correspond to those of the idealized gate of the previous
section. We can identify three potential sources of error: incorrect beamsplitter ratios; non-
unit mode matching and; timing errors. One advantage of working in the coincidence basis
is that losses and detector inefficiency can be ignored because they take the system out of
the coincidence basis and thus their only effect is to reduce the count rate.
Timing Errors. Correct gate operation depends on indistinguishability of the paths taken
by the two photons through the network. This means that they must arrive simultaneously
at the central beamsplitter to an accuracy of a fraction of their coherence length. Photon
coherence length in down conversion experiments is generally determined by pre-detection
frequency filtering and can be of order one hundred wave-lengths. Locking path lengths on
this scale should not be a major problem.
Beamsplitter ratios. The effect of non-optimal beamsplitter ratios can be investigated by
deriving the operator equations (Eq.2) more generally, with arbitrary beamsplitter ratios.
For simplicity we assume that the beamsplitters all came from the same “production-run”

4
such that any deviation from the optimal value is common. That is, we might suppose
that both the 50:50 beamsplitters actually have a reflectivity of η ′ whilst the three 33:67
beamsplitters all actually have reflectivities η. The Heisenberg equations are then
√ q
cH O = ηcH + 1 − ηvc
√ q q
cVO = − ηcV + (1 − η)η ′ tH + (1 − η)(1 − η ′ )tV
√ q q q
tHO = η(1 − 2η ′)tV + 2 η(1 − η ′ )η ′ tH + ηη ′ cV + (1 − η)(1 − η ′ )vt
q √ ′
q q
tVO = 2 η(1 − η )η tV + η(1 − 2η )tH + (1 − η)(1 − η )cVO − (1 − η)η ′ vt
′ ′ ′

√ q
vcO = − ηvc + 1 − ηcHO
q q √
vtO = (1 − η)(1 − η ′ )tH + (1 − η)η ′ tV − ηvt (8)

In general the effect of varying the beamsplitter ratios is input state dependent. However
for small deviations from the optimum values Bell state analysis is approximately state
independent and serves as a useful diagnostic [14]. In Fig.2 we plot the error probability in
distinguishing the Bell states as a function of η and η ′ in the region close to their optimum
values. The dependence of the error probability on η ′ is mirror imaged between the |ψ ± i and
the |φ± i Bell states. However this dependence is negligible in the region close to η ′ = 1/2.
The dependence on η is more pronounced. For an η of 1/3 ± 0.01 (and η ′ of 1/2 ± 0.05) error
rates of about 0.7% are predicted. Such uncertainties are standard with current beamsplitter
technology, and we conclude that errors below 1.0 % are realistic.
Mode matching errors. Mode matching in non-classical interference experiments is gener-
ally quite difficult and may be identified as a major contributor to non-unit visibility. Given
the key role of non-classical interference in the CNOT gate we may expect mode matching
errors to be of some significance.
In order to model the mismatch of input modes at the central beamsplitter, ancillary
modes v1 , v2 and v3 (originally in the vacuum state) are introduced to interact with the
propagating mismatch mode. The additional output modes are labelled cVm , cHm and tVm
(see Fig.3). The mode cv is assumed to be the source of the mismatch, after having passed
through some kind of optical element that has misaligned it.
q q
cv1 = ξ cV + 1 − ξ v1
q q
cv2 = − 1 − ξ cV + ξ v1

The parameter ξ quantifies the degree of mode matching between the control and target
modes at the central beamsplitter. So long as the modes are matched reasonably well, cv1
can be considered a sort of “primary” mode. It interacts with the output from beamsplitter
B3 in the same way as for the case neglecting mode matching. The mismatch component
cv2 interacts only with the newly introduced vacuum modes.
The equations for the output modes of the quantum CNOT gate, including the effects of
a mode mismatch, are
1 √
vcO = √ (−vc + 2cH )
3
1 √
cHO = √ ( 2vc + cH )
3

5
 1 q q
cVO = √ (− ξcV − 1 − ξv1 + tH + tV )
3
1 q q √
cVm = √ ( 1 − ξcV − ξv1 + 2v2 )
3
1 q q
tHO = √ ( ξcV + tH + 1 − ξv1 + vt )
3
s
1 q q 1 3
tHm = √ (− 1 − ξcV + ξv1 + √ v2 + v3 )
3 2 2
1 q q
tVO = √ ( ξcV + tV − 1 − ξv1 − vt )
3
s
1 q q 1 3
tVm = √ (− 1 − ξcV + ξv1 + √ v2 − v3 )
3 2 2
1
vtO = √ (tH − tV − vt ) (9)
3
Now, when measuring the coincidences, the detectors see a combination of the counts from
both the primary modes and the mismatch modes (see Fig.3). For example, when detecting
coincidences of horizontally polarised photons, the count rate becomes
hncHD ntHD i = hncHO (ntHO + ntHm )i
= hncHO ntHO i + hncHO ntHm i and similarly,
hncHD ntVD i = hncHO ntVO i + hncHO ntHm i
hncVD ntHD i = hncVO ntHO i + hncVO ntHm i + hncVm ntHO i + hncVm ntHm i
hncVD ntVD i = hncVO ntVO i + hncVO ntVm i + hncVm ntVO i + hncVm ntVm i (10)
These moments are summarised for logical inputs in Table 3. As expected, the mode mis-
match has not affected the CNOT operation when the control is “off” (i.e. when cH is
occupied). In this case, there is no interaction at beamsplitter B2 (Fig.3) and thus no non-
classical interference. However, when the control is “on”, the effects of the mismatch are
noticeable.
Interestingly, the mismatch adds extra terms rather than redistributing the probabilities
of the counts measured in the ideal case. Coincidence events which previously were disal-
lowed due to the non-classical interference can now appear as error events because of the
mismatch. Thus the probabilities that are being redistributed are those for the states that
were not detected in the ideal case (the states which had been postselected out).
We now consider the performance of the gate as a Bell state analyser in the presence
of mode mismatch. As in the ideal case, another beamsplitter is added to the outputs of
the control qubit. Another ancillary mode v4 must be added to interact with the mismatch
mode cVm .
The beamsplitter outputs are given in the Heisenberg picture by
1
cS1O = √ (cHO + cVO )
2
1
cS1M = √ (v4 + cVm )
2
1
cS2O = √ (cHO − cVO )
2

6
 1
cS2M = √ (v4 − cVm ) (11)
2
Each detector receives counts from both of the modes incident on it, so the expectation
values must be combined in a similar way to Eq. 10. The coincidence count rates are
given in Table 4. Using ξ = 1 yields the perfectly matched case calculated previously (see
Table 2). The error probability for Bell state discrimination is plotted in Fig.4. For small
mismatch the error is approximately equal to the percentage mismatch. Clearly good Bell
state discrimination will require accurate mode matching to the central beamsplitter.

IV. CONCLUSION

We have described a non-deterministic quantum CNOT gate, that operates with one
ninth efficiency, constructed solely from linear optical elements. We have investigated the
behaviour of the gate with variation in both the beamsplitter and mode match values and
conclude that a demonstration is feasible with current optical technology. Aside from its
value as a testbed system, such a gate could be made scaleable if photon-number QND
detectors were added to each output. This latter system would also act as efficient Bell
state analyser, which is an important component in some quantum algorithms, notably
quantum teleportation.

[1] P. G. Kwiat, K. Mattle, H. Weinfurter, A. Zeilinger, A. V. Sergienko, and Y. Shih Phys. Rev.
Lett. 75, 4337 (1995).
[2] J. Brendel, N. Gisin, W. Tittel, Phys. Rev. Lett. 82, 2594 (1999).
[3] A. G. White, D. F. V. James, P. H. Eberhard and P. G. Kwiat, Phys. Rev. Lett. 83, 3103
(1999).
[4] A. G. White, D. F. V. James, W J. Munro and P. G. Kwiat, Phys. Rev. A 65, 012301 (2001).
[5] G. J. Milburn, Phys. Rev. Lett. 62, 2124 (1988).
[6] E. Knill, R. Laflamme and G. Milburn, Nature 409, 46, (2001).
[7] T. C. Ralph, A. G. White, W. J. Munro and G. J. Milburn, Phys. Rev. A 65, 012314 (2001).
[8] D. Bouwmeester, J. W. Pan, K. Mattle, M. Eibl, H. Weinfurter and A. Zeilinger, Nature 390,
575 (1997).
[9] J. W. Pan, C. Simon, C. Brukner and A. Zeilinger, Nature 410, 1067 (2001).
[10] T. B. Pittman, B. C. Jacobs, J. D. Franson, quant-ph/0109128 (2001).
[11] T. C. Ralph, provisional patent, lodged August (2001).
[12] H. F. Hofmann, S. Takeuchi, quant-ph/0111092 (2001).
[13] Of course, other degrees of freedom, such as orbital angular momentum, could be used in
place of polarisation.
[14] Full characterization of the gate via quantum process tomography is currently under investi-
gation.

7
 Input hncHO ntHO i hncHO ntVO i hncVO ntHO i hncVO ntVO i
1
|Hic |Hit 9 0 0 0
1
|Hic |V it 0 9 0 0
1
|V ic |Hit 0 0 0 9
1
|V ic |V it 0 0 9 0

TABLE I: Coincident expectation values calculated for the four logical basis inputs.

Input hncS1 ntHO i hncS2 ntVO i hncS1 ntHO i hncS2 ntVO i

1
|ψ + i 9 0 0 0
1
|ψ − i 0 9 0 0
1
|φ+ i 0 0 9 0
1
|φ− i 0 0 0 9

TABLE II: Coincident expectation values calculated in the superposition basis for the four Bell
states.

Input hncHD ntHD i hncHD ntVD i hncVD ntHD i hncVD ntVD i

1
|Hic |Hit 9 0 0 0
1
|Hic |V it 0 9 0 0
2 1
|V ic |Hit 0 0 9 (1 − ξ) 9
1 2
|V ic |V it 0 0 9 9 (1 − ξ)

TABLE III: As for Table I, now allowing for mode matching ξ. (For perfect mode match, ξ = 1;
for complete mode mismatch ξ = 0).

Input hncS1 ntHD i hncS2 ntHD i hncS1 ntVD i hncS2 ntVD i

1 √ 1 √ 1 1
|ψ + i 18 (1 + ξ) 18 (1 − ξ) 18 (1 − ξ) 18 (1 − ξ)
− 1 √ 1 √ 1 1
|ψ i 18 (1 − ξ) 18 (1 + ξ) 18 (1 − ξ) 18 (1 − ξ)
1 1 1 √ 1 √
|φ+ i 18 (1 − ξ) 18 (1 − ξ) 18 (1 + ξ) 18 (1 − ξ)
1 1 1 √ 1 √
|φ− i 18 (1 − ξ) 18 (1 − ξ) 18 (1 − ξ) 18 (1 + ξ)

TABLE IV: As for Table II, now allowing for mode match ξ.

8
 ^
vc
^
vc O

1
B1 h1 =
c
^ 3 ^
cH
H O

^
cV
^
cV O

1
B2 h2 = 3
t
^ ^
tH
1 1
H

h3 = h4 =
O

B3 B4
^ 2 2 ^
t
V tV
1
O

B5 h5 = 3
^
v
^
vt
t O

FIG. 1: Schematic of the coincidence CNOT gate. Dashing indicates the surface from which a sign
change occurs upon reflection. The control modes are cH and cV . the target modes are tH and tV .
The modes vc and vt are unoccupied ancillary modes.

Relative Bit Error Rate (BER) for | f+/- > input states
Relative BER

0.04

0.02

0
0.55
0.38
0.5 0.36
0.34
0.32
0.3
0.45

h¢ h
Relative Bit Error Rate (BER) for | y+/- > input states
Relative BER

0.04

0.02

0
0.55
0.38
0.5 0.36
0.34
0.32
0.3
0.45

h¢ h

FIG. 2: Relative error rates, i.e. error rate/total rate, for Bell state analysis as a function of
beamsplitter ratios close to the optimum values of η ′ = 1/2 and η = 1/3.

9
 ^ 1 ^
vc
B1 h1 = 3
vc O

^
cH
^
cH O

c^ V c^
V1
^
cV O

^ ^ h2 = 13 ^
cVm
v1
^
v2
cV2 B2
^
tHm

^ ^
^
tH
B3 h3 = 12 B4 h4 = 1
2
tH O

^
t V tV O

vt^ B5 h5 = 13 ^
tVm

v3
^ ^
vt O

FIG. 3: Schematic diagram of the coincidence CNOT gate including the effects of mode matching.
The mismatch is represented by splitting cV into two orthogonal modes cV 1 and cV 2 . Ancillary
modes v1 , v2 and v3 interact with the propagating mismatch.

0.8

0.7

0.6
Error probability

0.5
Bell states

0.4

0.3

0.2

0.1

0 0.2 0.4 0.6 0.8 1

Mode matching, x

FIG. 4: Error probability as a function of mode matching for the four Bell states.

10