Professional Documents
Culture Documents
Abstract— In this work, we consider the preservation of a mea- The aforementioned scenario shares similarities with noisy
surement for quantum systems interacting with an environment. channels in information theory [3]. After messages are
Namely, a method of preserving an optimal measurement over encoded, sequences are transmitted and then corrupted during
a channel is devised, what we call channel coding of a quantum
measurement in that operations are applied before and after the transmission by a noisy channel. In information theory,
a channel in order to protect a measurement. A protocol that the problem of unwanted interactions with an environment is
preserves a quantum measurement over an arbitrary channel is resolved by channel coding, in which messages are prepared
shown only with local operations and classical communication in longer sequences with additional bits in order to contain
without the use of a larger Hilbert space. Therefore, the protocol some redundancy on purpose such that the redundant bits are
is readily feasible with present day’s technologies. Channel coding
of qubit measurements is presented, and it is shown that a used to detect and correct errors that have appeared during
measurement can be preserved for an arbitrary channel for both the transmission.
i) pairs of qubit states and ii) ensembles of equally probable In quantum information theory, in a similar vein, quantum
states. The protocol of preserving a quantum measurement is systems can be protected by exploiting more resources in the
demonstrated with IBM quantum computers. state preparation. As quantum states are described by linear,
Index Terms— Quantum state discrimination, quantum com- non-negative, and unit-trace operators in a Hilbert space,
munication, quantum channels, quantum protocols, quantum quantum states can be prepared in a subspace, also called
simulations. a code space, of a larger Hilbert space. The complementary
subspace is used to detect and correct errors that occurred in
I. I NTRODUCTION the code space. Consequently, states prepared on the support
of the code space are protected from interactions with an
Q UANTUM systems are generally fragile in that they
often interact with an environment [1], [2]. One of the
consequences is that a designed quantum information task
environment. The scheme has been referred to as quantum
error correction [4], [5] or noiseless subsystems [6].
becomes noisy. For instance, when quantum states are stored So far, in both cases of classical and quantum scenarios
in a memory, they may interact with an environment so that that deal with unwanted interactions with an environment,
the resulting noisy states are finally read out by a mea- the goal is to preserve systems prepared in sequences or states,
surement. In a communication scenario, interactions with an respectively. In quantum theory, we recall the significance of
environment take place while quantum states are transmitted: an optimal measurement to read out which state a system
a state sent by a party is noisy and then measured by a is prepared in. The role of a quantum measurement is illus-
receiver. trated as follows. Suppose that there are two sets of states,
SZ = {|0, √|1} and SX = {|+, |−}, where |± =
Manuscript received July 1, 2019; revised December 15, 2019; accepted
January 6, 2020. Date of publication January 24, 2020; date of current (|0 ± |1)/ 2. Although both contain a pair of orthogonal
version April 3, 2020. This work was supported by National Research Foun- states, the perfect distinguishability can be achieved only
dation of Korea (NRF-2019M3E4A1080001), an Institute of Information and when a measurement is performed in a correct basis: the
Communications Technology Promotion (IITP) grant funded by the Korean
government (MSIP), (Grant No. 2019-0-00831), University IT Research Z basis for the set SZ and the X basis for the set SX .
Center (ITRC) Program (IITP-2020-2018-0-01402), NASA Academic Mis- Measurements in the Z basis for the states SX , or X to
sion Services (contract number NNA16BD14C), and Iniziativa Specifica states SZ , give no information about the pair of orthogonal
341 INFN-DynSysMath. (Corresponding author: Spiros Kechrimparis.)
Spiros Kechrimparis and Joonwoo Bae are with the School of Electrical states.
Engineering, Korea Advanced Institute of Science and Technology (KAIST), It is clear that if states are protected by a channel, so is
Daejeon 34141, South Korea (e-mail: spiros@kaist.ac.kr; joonwoo.bae@ an optimal measurement preserved for the states. However,
kaist.ac.kr).
Chahan M. Kropf is with the Sezione di Pavia, Istituto Nazionale di Fisica the preservation of a measurement does not necessarily imply
Nucleare, I-27100 Pavia, Italy, also with the Dipartimento di Matematica that quantum states should be protected completely. For
e Fisica, Università Cattolica del Sacro Cuore, I-25121 Brescia, Italy, and instance, suppose that two states {|0, |1} are sent through
also with the Interdisciplinary Laboratories for Advanced Materials Physics,
Università Cattolica del Sacro Cuore, I-25121 Brescia, Italy. a channel as follows,
Filip Wudarski is with the Quantum Artificial Intelligence Laboratory
(QuAIL), Exploration Technology Directorate, NASA Ames Research Center, |0 → (1 − p)|00| + p|11|,
Mountain View, CA 94035 USA, also with the USRA Research Institute
for Advanced Computer Science (RIACS), Mountain View, CA 94043 USA, |1 → (1 − p)|11| + p|00| (1)
and also with the Institute of Physics, Faculty of Physics, Astronomy and
Informatics, Nicolaus Copernicus University, 87-100 Torun, Poland. for p ∈ [0, 1]. An optimal measurement for state discrimination
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org. remains the same as a measurement in the Z basis both before
Digital Object Identifier 10.1109/JSAC.2020.2969034 and after a channel use. The measurement in the Z basis
0733-8716 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
440 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 3, MARCH 2020
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
KECHRIMPARIS et al.: CHANNEL CODING OF A QUANTUM MEASUREMENT 441
where qx denotes the a priori probability that a state ρx A. The Preservation of States
is prepared. In addition, we write an ensemble of equally Quantum states can be preserved if there exists a perfect
probable states as follows, recovery map, i.e., R such that R ◦ N = id. Correctable
(id) errors on code states can be characterized by the condition of
S0 = {1/n, ρx}nx=1 . (3) quantum error correction. Let P denote the projection onto
a code space for a noisy channel N described by Kraus
Throughout, S (id) means an ensemble of states prepared in a
operators {Ej }. Then, a perfect recovery exists if it holds that
noiseless environment, and S (N ) the resulting ensemble by a
P Ei Ej P = αij P for some constants αij . This is equivalent
channel N .
to the condition that the resulting states after a channel are
The problem of optimal state discrimination aims to find
supported by subspaces that are pairwise orthogonal, i.e., dis-
the maximal probability of making a correct guess on average
tinguishability can be preserved [4], [5]. The preservation of
when one of the states in an ensemble is given, called the
states can be also obtained by finding a subspace of a larger
guessing probability. It also finds an optimal measurement
Hilbert space such that distinguishability of states is unaffected
that attains the guessing probability [13]–[20]. For the sake
by a noisy channel, such as decoherence-free subspace [23],
of generality, suppose that states are sent through a chan-
or more generally decoherence-free subsystems [24]. Note that
nel N . The guessing probability for the ensemble S (N ) can be
in all case, it is essential to exploit a larger Hilbert space.
written as,
The aforementioned methods of preserving quantum states
can be generally characterized by the information preserving
(N )
pguess = max qx tr[N [ρx ]Mx ] (4)
M
x
structure [25], [26]. Namely, quantum states {ρx }nx=1 are pre-
served by a channel N if and only if the following conditions
where M denotes a set of POVM elements. are satisfied: for all q ∈ (0, 1),
The standard problem of optimal state discrimination cor-
responds to the case S (id) , for which the guessing probability ∀x, y, qρx −(1−q)ρy 1 = qN [ρx ] − (1 − q)N [ρy ]1 . (6)
(id) (id) (N )
is denoted by pguess . It is clear that pguess ≥ pguess for a The information preserving structure can be used to construct
channel N [21], i.e., distinguishability does not increase under the code states which remain perfectly distinguishable after a
a quantum channel. Note also that optimal POVM elements channel use. Note also that, once states are perfectly recovered,
are not always non-zero, i.e., no-measurement is sometimes the measurement for the ensemble S (id) can also be used after
optimal [22]. a channel, see Eq. (5).
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
442 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 3, MARCH 2020
C. States Versus a Measurement as Preserving Resources state after channel coding of a channel C (N ,S) [N ](ρ) is given
One can, however, observe that the condition Eq. (7) shares as follows,
some similarities with the information preserving structure in trX UP X (N ⊗ idE [VP X (ρ ⊗ IX )VP†X ])UP† X (9)
Eq. (6). Technically, taking the L1 norm in the OMP condition
with unitary transformations UP X and VP X on the system
in Eq. (7) for κ = 1, one can obtain the condition of preserving
of a sender and a receiver (P ) and an extra system (X).
states in Eq. (6) with a specific set of a priori probabilities
We summarize that channel coding of a measurement with
{qx }nx=1 . If states can be preserved such that the condition in
respect to a given channel N establishes a channel C (N ,S) [N ]
Eq. (6) holds true for arbitrary probabilities {qx }nx=1 , there
which is OMP for an ensemble S. As it is shown in the
exists a recovery operation that finds the states before a
framework of a supermap in Eq. (9), a transformation among
channel use [25], [26]: ∃R such that R ◦ N [ρy ] = ρy , for
channels corresponds to an LOCC protocol of applying unitary
ρy ∈ span{ρx }nx=1 . This shows that, when a set of states can
transformations before and after a channel use.
be preserved by a channel, i.e., Eq. (6) is fulfilled, it follows
We here take a depolarization channel in Eq. (8) as an OMP
that an optimal measurement can also be preserved, i.e., the
channel of interest. In this case, a supermap in Eq. (9) can be
OMP condition is satisfied, by applying a recovery operation.
realized with unitaries,
However, the OMP condition with κ < 1 cannot be reduced
to the information preserving structure for the preservation of UP† X = VP X = Ux ⊗ |xx|
states. This means that even if the preservation of quantum x
states is unsuccessful, it is possible to preserve an optimal where {Ux } forms a unitary 2-design [28] and {|x} denotes
measurement. a set of orthonormal basis. The supermap is then equivalent to
the well-known result called channel twirling that transforms
a channel to a depolarization map by using LOCC only [29].
D. Channels Preserving a Measurement
Let us now present a class of channels that preserve a B. Realization
(id)
measurement. For ensembles of equally probable states S0 1) Channel Twirling: Let T denote a twirling operation for
in Eq. (3), the following channels are OMP. a channel N as follows,
Remark: A channel Nσ : ρ → (1 − η)ρ + ησ for a fixed
T N [ρ] = dμ(U )U † N [U ρU † ]U ,
(id)
state σ is OMP for any ensemble S0 . (10)
One can easily find that a channel Nσ satisfies the OMP
condition in Eq. (7). Clearly, channels Nσ do not preserve where the average is performed over the Haar measure,
states. the uniform measure in the space of unitary operators. The
This shows that a measurement can be preserved if a channel consequence is that the resulting map corresponds to a
can be manipulated such that it is transformed to an OMP one depolarization,
Nσ For this purpose, a particular choice of the state σ = I/d I
is useful, where d denotes the dimension of a Hilbert space. T N [ρ] = DηN [ρ] = (1 − ηN )ρ + ηN , (11)
d
This introduces a depolarization channel as follows, where ηN is determined by a channel N . Note also that DηN
is a quantum channel for (1 − ηN ) ∈ [−1/(d2 − 1), 1].
Dη [ρ] = (1 − η)ρ + ηI/d. (8)
To perform channel twirling, one has to implement appli-
A depolarization channel is then OMP for any ensemble S0 .
(id) cations of unitary transformations, called a unitary 2-design.
A unitary 2-design is a set of unitary transformations in
a d-dimensional Hilbert space, denoted by a set W =
IV. C HANNEL C ODING OF A M EASUREMENT {Uk }k=1...N , such that the following is satisfied [30]. For any
A. The Formalism quantum channel N , it holds that
1 †
We now formulate channel coding of a quantum measure- N
ment that aims to preserve an optimal measurement over a Ui N [Ui ρUi† ]Ui = dU U † N [U ρU † ]U.
N i=1 U(d)
channel. The main idea is to manipulate a given channel such
that an OMP channel can be established between a sender Then, channel twirling can be realized by random applications
and a receiver. The framework of a supermap that formulates of elements in a unitary 2-design.
transformations among channels each other [27] is useful to 2) Channel Coding of a Measurement for Code States:
describe channel coding of a measurement. Therefore, channel coding of a measurement for an ensemble
(id)
Definition Channel Coding of a Measurement: For a quan- of equally probable states S0 can be implemented by
tum channel N : S(H) → S(H) and an ensemble S of interest applying unitaries from a unitary 2-design before and after
to be sent through the channel, let C (N ,S) denote a supermap a channel. The schematic is also shown in Fig. 1. Channel
that works as N → C (N ,S) [N ]. Then, a supermap C (N ,S) coding of a measurement is summearized as follows.
is called channel coding of a measurement if the channel A Protocol of Channel Coding of a Measurement:
(id)
C (N ,S) [N ] is OMP for the ensemble S. 1) For a state ρ ∈ S0 , an element Uj ∈ W is randomly
It has been shown that a deterministic supermap can be chosen from a unitary 2-design W and applied to the
implemented as follows [27]: for a state ρ ∈ S, the resulting state before a channel N .
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
KECHRIMPARIS et al.: CHANNEL CODING OF A QUANTUM MEASUREMENT 443
2) The sender and the receiver communicate their selection 1) Minimal Unitary 2-Design: For qubit cases, a subgroup
of unitaries. of the tetrahedral group of rotations [31] that forms a unitary
3) After the channel transmission, the receiver applies its 2-design with a conjectured minimal [30] cardinality of 12
inverse Uj† to a resulting state N [Uj ρUj† ]. is chosen. Then, these elements can be found explicitly as
4) By randomizing the resulting states, an optimal mea- follows:
(id)
surement for the ensemble S0 is also optimal for its
(N ) W = {U1 , U2 · · · , U12 } (13)
resulting ensemble S0 for any quantum channel N .
where,
C. Enhancement by Preserving a Measurement U1 = I, U2 = −iX, U3 = −iY, U4 = −iZ
We stress that with the protocol, a measurement once 1 1−i −1−i 1 1+i 1 − i
U5 = , U6 = ,
prepared for an ensemble S0
(id)
remains ever optimal no 2 1 − i 1+i 2 −1 − i 1 − i
matter what interactions a system suffers from an environment. 1 −1 − i − 1 − i 1 −1+i 1−i
U7 = , U8 = ,
We also remark that channel coding can improve distinguisha- 2 1 − i − 1+i 2 −1 − i − 1−i
(id)
bility of states over a channel. To be precise, let pguess denote
1 −1+i − 1+i 1 −1 − i 1+i
(id) (N )
the guessing probability for an ensemble S0 , pguess for the U9 = , U10 =
2 1+i − 1 − i 2 −1+i − 1+i
(N ) (T N )
ensemble S0 , and pguess for the ensemble S0
(T N )
. It is clear
1 1+i −1+i 1 1 − i 1+i
that U11 = , U12 =
2 1+i 1 − i 2 −1+i 1+i
(T N )
guess ≥ pguess and pguess ≥ pguess
p(id) (N ) (id)
and X, Y , and Z are the Pauli matrices. With the latter unitary
2-design, the protocol in Sec. IV can realize channel coding
since distinguishability does not increase under a channel. of a measurement for ensembles of equal a priori probabilities
However, for some channels N , it appears that distinguisha- (id)
S0 , as it transforms a channel to a depolarization map.
bility can be improved,
An alternative and widely used unitary 2-design is the Clifford
(N ) N) group that contains 24 elements up to phase factors.
pguess < p(T
guess . (12)
Let Np denote a Pauli channel for a qubit state ρ as follows,
When an ensemble of code states is identified, channels for
Np [ρ] = p0 ρ + px XρX + py Y ρY + pz ZρZ
which distinguishability can be improved by channel coding
of a measurement can be characterized. For instance, the tech- where p = (p0 , px , py , pz ). We also write collections of
niques of optimal state discrimination, e.g., [13]–[20], can be unitarties in Eq. (13) as Ui = {Ui , Ui+1 , Ui+2 , Ui+3 } for
applied to find the guessing probability, and then channels i = 1, 5, 9. Taking three unitaries from the sets, U ∈ U1 ,
satisfying the relation in Eq. (12) can be classified. In the V ∈ U5 , and W ∈ U9 , channel twirling on a Pauli channel
next section, examples of such channels are provided. In fact, can be achieved for an ensemble of equiprobable qubit states.
in a wide range of channels distinguishability can be improved Hence, three unitaries suffice to realize channel twirling of a
by channel coding of a measurement. Finally, we emphasize Pauli channel.
that that resources to realize the protocol of channel coding of 2) Validity of Channel Coding of a Measurement: Channel
a measurement contain LOCC only: local unitaries are only coding of a measurement is useful only when a measurement
applied before and after a channel use, which are feasible with for optimal detection is valid. Note that no-measurement is
current technologies. sometimes optimal when states are too noisy [22]. Then,
an optimal discrimination is obtained by simply making a
V. C HANNEL C ODING OF A Q UBIT M EASUREMENT guess according to a priori probabilities without a detection
Channel coding of a measurement is applied to qubit event by a measurement. This strategy is here referred to as a
cases. We begin by identifying a minimal unitary 2-design trivial measurement. Channel coding of a measurement is not
in order to efficiently realize channel twirling in practice with valid if a channel T N leads to an ensemble S (T N ) for which
a minimal number of unitaries. Its subset is also identified a measurement is trivial.
to perform channel twirling for Pauli channels. Then, there Furthermore, a caveat is also that from Eq. (11), the preser-
are two cases to additionally consider in the preservation of a vation of a measurement does not work if a resulting depo-
measurement. First, if states are so noisy after a channel that a larization map has 1 − ηN < 0 since states of an ensemble
null measurement is optimal, a measurement is not preserved S (id) are less probable than their complement ones after a
by a channel. In this case, no measurement is optimal but a channel use. Note that for qubits, we have 1−ηN ∈ [−1/3, 1].
random guess according to the a priori probabilities is the In this case, as we will show below, it is possible to have an
best strategy. The condition that channels do not lead to a optimal measurement such that existing POVM elements are
null measurement is characterized. Next, some channels do only re-labeled. Since a set of POVM elements is identical
not lead to a null measurement but ask a rearrangement of before and after a channel, we say that a measurement is
the labels of the POVM elements. From a practical point preserved but it is essential to re-label them.
of view, we conclude that a measurement is preserved as a Therefore, for the preservation of a measurement, it is
measurement setting remains the same. It is then necessary to important to find when channel twirling leads to a null
re-label POVM elements for optimal detection of states. measurement or requires a new set of labels to existing
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
444 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 3, MARCH 2020
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
KECHRIMPARIS et al.: CHANNEL CODING OF A QUANTUM MEASUREMENT 445
C. A Measurement for Multiple States Is not Preserved by Fig. 2. In a quantum algorithm, the initial state |0 ⊗ · · · ⊗ |0 evolves
Channel Twirling by a series of quantum gates and the measurement in the Z basis is applied.
Suppose that one of the qubits contains unverified noise N after the circuit due
For more than two states, the protocol of channel coding of to interactions with a local environment. While ancillary systems cannot be
a measurement in Sec. IV fails to preserve a measurement exploited for detecting and correcting errors, channel coding of a measurement
provides a suboptimal strategy for optimal detection of desired measurement
if a priori probabilities are unequal. We here provide a outcomes. For the qubit under noise, local operations are applied before and
counter-example that channel twirling fails to preserve a after the channel. The measurement in the Z basis remains optimal. The box
measurement. is simulated on the IBM quantum computer, see Figs. (3) and (4).
Let us consider a set of modified trine states with unequal
a priori probabilities S (id) = {qx , ρx }3x=1 , with q1 = 2q2 =
2q3 = 1/2. Trine states can be characterized as three states
that are equally spaced in a half-plane of a Bloch sphere. Let r
denote the Bloch vector of a qubit state ρ. Then, we consider
modified trine√ states with Bloch vectors √ r1 = (1/2, 0, 0),
r2 = (−1/2, 3/2, 0) and r3 = (−1/2, − 3/2, 0), i.e., the Fig. 3. A quantum circuit for channel coding of a qubit measurement is
first state is not pure. For the ensemble, an optimal measure- shown. The interaction between a system qubit q1 and an environment (ani-
ment can be found explicitly [33] as follows clla) qubit q0 in Eq. (20) is realized with two CNOT gates, so that the channel
for the system is described by NY . By default all qubits are initialized in |0,
so that a unitary operation Ui is required to prepare other states, for instance
0.44 0.44 0.28 − 0.22 − 0.17 i
, , |1, |+, and |−. Analogously, since measurements are always done in the
0.44 0.44 −0.22 + 0.17 i 0.28 Z-basis, a unitary transformation Um is required to prepare POVM elements
other than |00| or |11|. The channel NX is realized by replacing the block
0.28 − 0.22+0.17 i
. inside the dashed box with a single CNOT gate. The gates Utwirl denote the
−0.22 − 0.17 i 0.28 application of one of the elements in a unitary 2-design, Eq. (13).
Note that none of the POVM elements is zero, i.e., a null
detector is not applied. No ancillary system is needed. For instance, a measurement
Suppose that the ensemble is sent through a channel N in quantum key distribution can be preserved [34], where it is
with the protocol of channel coding in Sec. IV such that a also shown that security can be improved.
depolarization channel is obtained with 1 − ηN = 2/3. For The preservation of a measurement can also be applied in
the ensemble S (T N ) , an optimal measurement can be found the realization of quantum algorithms with noisy intermediate
as follows scalable quantum (NISQ) devices, in which imperfections
0.46 0.46 0.27 − 0.23 − 0.14 i exist such that some qubits are under a high-precision control
, ,
0.46 0.46 −0.23 + 0.14 i 0.27 and some are not. In a quantum algorithm, a measurement
is prepared in the Z basis, MZ = {|00|, |11|}, which
0.27 − 0.23+0.14 i
. would remain for optimal detection by channel coding of a
−0.23 − 0.14 i 0.27
measurement. Thus, the preservation of a measurement can
Thus, it is shown that an optimal measurement is not preserved be used as a suboptimal strategy to improve the probability of
with the protocol of channel coding. Note that the a priori obtaining the desired measurement outcomes. As it is depicted
probabilities are not equal. in Fig. 2, our result can be applied to protecting a measurement
on a single qubit state. In the following subsection, it is
VI. A PPLICATIONS TO Q UANTUM C OMPUTING demonstrated that a single qubit measurement can be protected
A. Practical Applications in NISQ devices and, by doing so, the probability of optimal
detection can also be improved.
Channel coding of a measurement can be applied whenever
one aims to preserve a measurement for quantum systems that
suffer from unwanted interactions with an environment. The B. Demonstration on the IBM Quantum Computer
scheme we propose here is feasible with current technologies We here present a proof-of-principle demonstration of chan-
as three unitaries are applied before and after a channel. nel coding of a qubit measurement with the IBM quantum
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
446 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 3, MARCH 2020
Fig. 4. Guessing probabilities for channels NX and NY in Eq. (17) are shown: a) a channel NX for a pair of orthogonal states SZ and b) a channel
NY for the BB84 states SBB84 . The solid lines are theoretical prediction in Eqs. (21) and (22). The guessing probabilities after a channel are in the red
color. After twirling a channel, they are in the blue color. Circles are guessing probabilities from the ibmqx2. The circle size is larger than the statistical
measurement noise. The dashed line is made by assuming the noise model in Eq. (23) η = 0.05 for a) and η = 0.15 for b), which shows a reasonably good
explanation about the source of errors in IBM quantum computers.
computer ibmqx2 using the Qiskit package [35]. We specif- In order to implement channel coding of a qubit measure-
ically consider ensembles of a pair of orthogonal states ment, the minimal unitary 2-design W in Eq. (13) is realized
SZ = {|0, |1} and the four-states in the Bennett-Brassard with the unitary gate U2 (φ, λ) = U3 (π/2, φ, λ) as
1984 (BB84) protocol [36] SBB84 = {|0, |1, |+, |−}, both
U2 (0, π/2), U2 (π, π/2), U2 (0, 3π/2), U2(π, 2π/2),
with equal a priori probabilities. We consider flip channels
U2 (π/2, π), U2 (π/2, 0), U2 (3π/2, π), U2 (3π/2, 0) ,
NR [ρ] = (1 − pf )ρ + pf RρR (17)
together with I, −iX, −iY , and −iZ. Note that for the
where pf ∈ [0, 1] is a flipping probability and R = X, Y ensemble SZ , a measurement in the basis Z is optimal. With
with Pauli matices X and Y , which arise as a result of the this measurement, for a Pauli channel NX we have (see also
interactions of the system qubit ρ with an environment. Eq. (4))
The aforementioned channels can be realized with a quan-
1 |1−2pf | NX ) 1 |3−4pf |
tum circuit with two qubits – q1 for the system, and q0 the p(N X)
guess = + , and p(T
guess = + . (21)
ancillary qubit mimicking an environment – see Fig. 3. To this 2 2 2 6
end, we need gate operations, Pauli X, Hadamard H, and For the ensemble SBB84 a random measurement in the X and
controlled-not (CNOT) gate denoted by CX . Note that with Z basis is optimal. With the measurement, we have
the Qiskit notation, the most general single-qubit operation is 1 |1−2pf | NY ) 1 |3 − 4pf |
written as p(N Y)
guess = + , and p(Tguess = + . (22)
4 4 4 12
cos(θ/2) −eiθ sin(θ/2) To implement the twirling protocol, we collect the data for
U3 (θ, φ, λ) = iφ . (18)
e sin(θ/2) eiθ+iφ cos(θ/2) 8000 (maximum allowed is 8192) shots for each unitary matrix
In order to obtain the channel NX on the system q1, the envi- in the unitary-2-design W , which is applied before and its
ronment q0 is prepared in the state conjugate after the channel NR , and perform the averaging in
post-processing.
|aE = U3 (2α, 0, 0)|0E = cos(α)|0E + sin(α)|1E , As a proof-of-principle, this is equivalent to random applica-
tions of unitary 2-design before and after a channel use. Thus,
where the angle α corresponds to the flipping probability as
channel coding of a measurement is implemented. We show
cos(α)2 = 1 − pf . Then, when the system is prepared in a
in Fig. 4 the guessing probabilities Eq. (4) measured on the
state |ψ, the interaction with the environment is described as
5-qubit quantum machine ibmqx2. We verified that the results
√
CX |aE |ψ = 1 − pf |0E |ψ + pf |1E X|ψ. (19) from the classical quantum circuit simulator ’qsam_simulator’
are in perfect agreement with Eqs. (21) and (22). This certifies
The channel NX is obtained by tracing out the environment that the circuits are simulating correctly the quantum channel
qubit q0, which in practice is done by measuring q1 only. and the twirling protocol.
For an observable M of interest (in particular POVMs) on the Comments on the Simulation: The results in Fig. 4 obtained
system, the expectation is given by in March in 2019 show a good agreement with the theoretical
† prediction in Eqs. (21) and (22). A certain loss of probability
tr[M NX [|ψψ|]] = tr[(I ⊗ M ) CX |aE a| ⊗ |ψψ|CX ]
can be observed that can be modelled by shot noise on the
Similarly, the channel NY is obtained by implementing the measurement outcome,
two-qubit controlled-Y gate CY as follows (see also Fig. 3)
tr(M ρ) → tr(M [(1 − η)ρ + ηI/2]), (23)
√
CY |aE |ψ = 1 − pf |0E |ψ + pf |1E Y |ψ, (20)
as indicated by the dashed line. The same measurements were
where CY = CX (I ⊗ H)CX (I ⊗ H) and subsequent tracing performed in June 2019. It turns out that the shot noise was
out the environment qubit. significantly higher (η = 0.65 instead of η = 0.05 for SZ ,
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
KECHRIMPARIS et al.: CHANNEL CODING OF A QUANTUM MEASUREMENT 447
η = 0.57 instead of η = 0.15 for SBB84 ). This seems consistent arbitrary ensembles. Next, for further extension beyond a
with a recent study of the noise sources on the IBM quantum single qubit state, it is significant to find a minimal unitary
computers [37] where a loss of the overall norm of the Bloch 2-design or its approximation for multiple qubits. In addition,
vectors due to shot noise is identified as one of the major it is interesting to find how tightly schemes of preserving
sources of error. The same measurements were performed in states and a measurement are related to each other. In future
June on ibmqx4 and ibmq_16_melbourne machines. For the investigations, it would be also interesting to apply channel
ibmq_16_melbourne, the shot noise was η ≈ 1 so that no coding of a measurement in a realistic and practical application
significant results could be extracted. For the ibmqx4, pguess such as few-qubit quantum algorithms.
for T NR was systematically larger than for NR , in agreement
with the theory. However, the measured pguess values strongly ACKNOWLEDGMENT
fluctuated around the theoretical values. These observations
cannot be accounted for by the estimates of the gate fidelities The authors acknowledge the use of IBM QISKIT for this
reported by IBM Q Experience. work. The views expressed are those of the authors and do
In an attempt to find the sources of the errors, we measured not reflect the official policy or position of IBM. The authors
the state of the environment qubit q0 to control the actual value acknowledge support from the NASA Advanced Exploration
of pf . Some fluctuations were observed, which are correlated Systems program and the NASA Ames Research Center.
with the fluctuations in pguess on ibmqx4, but can only partially The views and conclusions contained herein are those of the
explain the deviations from the theory. Consequently, further authors and should not be interpreted as necessarily repre-
understanding of the device’s imperfections is of paramount senting the official policies or endorsements, either expressed
importance for channel coding, and for quantum information or implied, of the U.S. Government. The U.S. Government is
processing in general, on available quantum hardware. authorized to reproduce and distribute reprints for Governmen-
tal purpose notwithstanding any copyright annotation thereon.
VII. C ONCLUSION
R EFERENCES
We have formulated and presented a framework of pre-
[1] H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems.
serving a measurement for quantum systems interacting with New York, NY, USA: Oxford Univ. Press, 2002.
an environment. We show that an LOCC protocol with local [2] R. Alicki and K. Lendi, Quantum Dynamical Semigroups and Applica-
unitaries only can realize the preservation of an optimal tions (Lecture Notes in Physics). Berlin, Germany: Springer, 2007.
[3] R. W. Hamming, “Error detecting and error correcting codes,” Bell Syst.
measurement without further resources such as a larger Hilbert Tech. J., vol. 29, no. 2, pp. 147–160, Apr. 1950.
space, contrasting to the case of preserving states. A general [4] P. W. Shor, “Scheme for reducing decoherence in quantum computer
framework for channel coding of a measurement is presented memory,” Phys. Rev. A, Gen. Phys., vol. 52, no. 4, pp. R2493–R2496,
Jul. 2002.
as a supermap that transforms a channel to an OMP one. [5] A. R. Calderbank and P. W. Shor, “Good quantum error-correcting codes
In particular, it is shown that channel twirling implements exist,” Phys. Rev. A, Gen. Phys., vol. 54, no. 2, pp. 1098–1105, Jul. 2002.
the preservation of a measurement for ensembles of equally [6] D. Kribs, R. Laflamme, and D. Poulin, “Unified and generalized
approach to quantum error correction,” Phys. Rev. Lett., vol. 94,
probable states. For qubit ensembles, the protocol of preserv- May 2005, Art. no. 180501.
ing a measurement is investigated in detail and is found that [7] C. E. Shannon, “A mathematical theory of communication,” Bell Syst.
it works for i) any pair of qubit states and ii) ensembles of Tech. J., vol. 27, no. 3, pp. 379–423, 1948.
[8] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum
equally probable states. A counter-example for three states is Information, New York, NY: Cambridge Univ. Press, 2011.
explicitly provided. Channel coding of a qubit measurement [9] M. Wilde, Quantum Information Theory. Cambridge, U.K.: Cambridge
is demonstrated for ensembles of a pair of orthogonal states Univ. Press, 2013.
[10] A. S. Holevo, “Remarks on optimal quantum measurements,” Problemy
and the four states in the BB84 protocol, and can be readily Peredachi Informatsii, vol. 10, no. 4, p. 317, 1974.
applied to practical quantum communication protocols. [11] H. Yuen, R. Kennedy, and M. Lax, “Optimum testing of multiple
Our work sheds a new light in directions of an early-stage hypotheses in quantum detection theory,” IEEE Trans. Inf. Theory,
vol. 21, no. 2, pp. 125–134, Mar. 1975.
quantum information processor, that works with limited [12] C. W. Helstrom, Quantum Detection and Estimation Theory. New York,
resources and restricted controls. For instance, the preservation NY, USA: Academic, 1976.
of quantum states in a noisy environment, which needs ancillas [13] A. Chefles, “Quantum state discrimination,” Contemp. Phys., vol. 41,
no. 6, p. 401, Nov. 2000.
and a high-precision control over a system and ancillas,
[14] J. A. Bergou, U. Herzog, and M. Hillery, “Discrimination of quantum
may not be achieved within the near future. The presented states,” in Quantum State Estimation (Lecture Notes Physics), vol. 649,
framework of preserving an optimal measurement could be M. Paris and J. Řeháček, Eds. Berlin, Germany: Springer, 2004.
a next best and feasible opportunity. Our demonstration with [15] J. A. Bergou, “Quantum state discrimination and selected applications,”
in J. Phys. Conf., vol. 84, Oct. 2007, Art. no. 012001.
IBM quantum computers has shown that by channel coding [16] J. A. Bergou, “Discrimination of quantum states,” J. Mod. Opt., vol. 57,
of a measurement, single-qubit information processing can no. 3, pp. 160–180, Jan. 2010.
readily work against an adversarial environment that may [17] J. Bae and W.-Y. Hwang, “Minimum-error discrimination of qubit states:
Methods, solutions, and properties,” Phys. Rev. A, Gen. Phys., vol. 87,
cause high-rate errors. Our work initiates a new direction Jan. 2013, Art. no. 012334.
toward an early-stage quantum information processing, lead- [18] J. Bae, “Structure of minimum-error quantum state discrimination,” New
ing to a number of questions. First, channel coding of a J. Phys., vol. 15, no. 7, Jul. 2013, Art. no. 073037.
[19] J. Bae and L.-C. Kwek, “Quantum state discrimination and its
measurement for ensembles of arbitrary a priori probabilities applications,” J. Phys. A, Math. Theor., vol. 48, no. 8, Feb. 2015,
is sought, to apply channel coding of a measurement for Art. no. 083001.
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.
448 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 3, MARCH 2020
[20] S. M. Barnett and S. Croke, “Quantum state discrimination,” Adv. Opt. Spiros Kechrimparis was born in Athens, Greece. He received the B.Sc.
Photon., vol. 1, p. 238, Apr. 2009. degree in physics and the M.Sc. degree in theoretical physics, computational
[21] S. Kechrimparis, T. Singal, C. M. Kropf, and J. Bae, “Preserving physics, and astrophysics from the University of Patras, Greece, and the Ph.D.
measurements for optimal state discrimination over quantum channels,” degree in mathematics from the University of York, U.K. He subsequently held
Phys. Rev. A, Gen. Phys., vol. 99, Jun. 2019, Art. no. 062302. a post-doctoral position at the Department of Applied Mathematics, Hanyang
[22] K. Hunter, “Measurement does not always aid state discrimination,” University, South Korea. He is currently a Post-Doctoral Researcher with the
Phys. Rev. A, Gen. Phys., vol. 68, Jul. 2003, Art. no. 012306. School of Electrical Engineering, Korea Advanced Institute of Science and
[23] D. A. Lidar, I. L. Chuang, and K. B. Whaley, “Decoherence-free Technology (KAIST), South Korea. His scientific interests include topics on
subspaces for quantum computation,” Phys. Rev. Lett., vol. 81, p. 2594, the foundations of quantum mechanics and quantum information theory, such
Sep. 1998. as quantum uncertainty, entanglement theory, and state discrimination among
[24] E. Knill, R. Laflamme, and L. Viola, “Theory of quantum error correc- others.
tion for general noise,” Phys. Rev. Lett., vol. 84, no. 11, pp. 2525–2528,
Jul. 2002.
[25] R. Blume-Kohout, H. K. Ng, D. Poulin, and L. Viola, “Characterizing Chahan M. Kropf was born in Bern, Switzerland, in 1988. He received
the structure of preserved information in quantum processes,” Phys. Rev. the B.Sc. degree in physics from the University of Neuchâtel (CH)
Lett., vol. 100, Jan. 2008, Art. no. 030501, in 2009, the M.Sc. degree in physics from the Ruprecht Karl Uni-
[26] R. Blume-Kohout, H. K. Ng, D. Poulin, and L. Viola, “Information- versity of Heidelberg (DE) in 2012, and the Ph.D. degree in physics
preserving structures: A general framework for quantum zero-error infor- from the Albert-Ludwigs University of Freiburg (DE) in 2017. His Ph.D.
mation” Phys. Rev. A, Gen. Phys., vol. 82, Dec. 2010, Art. no. 062306. dissertation was on Effective dynamics of disordered quantum systems.
[27] G. Chiribella, G. M. D’ariano, and P. Perinotti, “Transforming quan- He is currently working as a Post-Doctoral Researcher with the INFN
tum operations: Quantum supermaps,” Europhys. Lett., vol. 83, no. 3, Sezione di Pavia and with the Università Cattolica del Sacro Cuore,
p. 30004, Aug. 2008. Brescia, Italy. His main research interests are quantum technologies:
[28] C. Dankert, R. Cleve, J. Emerson, and E. Livine, “Exact and approximate quantum transport, quantum communication, quantum computing, and
unitary 2-designs and their application to fidelity estimation,” Phys. Rev. quantum machine learning. In 2014, he received the doctoral fellowship from
A, Gen. Phys., vol. 80, Jul. 2009, Art. no. 012304. the German Academic Foundation. In October 2017, he was awarded the
[29] M. Horodecki, P. Horodecki, and R. Horodecki, “General teleportation INFN Post-Doctoral Fellowship in the research focus DYNSYSMATH from
channel, singlet fraction, and quasidistillation,” Phys. Rev. A, Gen. Phys., the section of Pavia for two years.
vol. 60, no. 3, pp. 1888–1898, Jul. 2002.
[30] D. Gross, K. Audenaert, and J. Eisert, “Evenly distributed unitaries: Filip Wudarski was born in Poland, in 1987. He received the B.Sc. degree
On the structure of unitary designs,” J. Math. Phys., vol. 48, no. 5, in chemistry and the M.Sc. and Ph.D. degrees in physics from Nicolaus
May 2007, Art. no. 052104. Copernicus University, Torun, Poland, in 2009, 2011, and 2015, respectively.
[31] I. Bengtsson and K. Zyczkowski, Geometry of Quantum States: An He was a National Institute for Theoretical Physics (NITheP) Post-Doctoral
Introduction to Quantum Entanglement. Cambridge, U.K.: Cambridge Fellow with the University of KwaZulu-Natal, Durban, South Africa, from
Univ. Press, 2006. 2015 to 2017. He received the Mobility Plus IV Grant from the Polish
[32] G. Weir, S. M. Barnett, and S. Croke, “Optimal discrimination of single- Ministry of Higher Education to join the Quantum Optics and Statistics Group,
qubit mixed states,” Phys. Rev. A, Gen. Phys., vol. 96, Aug. 2017, Freiburg, as a Visiting Researcher, from 2017 to 2019. Since March 2019,
Art. no. 022312. he has been a Research Associate with the University Space Research
[33] D. Ha and Y. Kwon, “Complete analysis for three-qubit mixed- Association and a member of the Quantum Artificial Intelligence Laboratory
state discrimination,” Phys. Rev. A, Gen. Phys., vol. 87, Jun. 2013, Group at NASA Ames Research Center. His main area of research interests are
Art. no. 062302. quantum computing (gate based and annealers), and theoretical and statistical
[34] S. Kechrimparis et al., “Measurement-protected quantum key distrib- modeling of noise in quantum circuits.
ution,” 2019, arXiv:1912.00768. [Online]. Available: https://arxiv.org/
abs/1912.00768
[35] G. Aleksandrowicz et al., “Qiskit: An open-source framework for Joonwoo Bae received the Ph.D. degree from the Universitat de Barcelona and
quantum computing,” Tech. Rep., 2019, doi: 10.5281/zenodo.2562110. ICFO-The Institute of Photonic Sciences, Barcelona, in 2007. He has worked
[36] C. H. Bennett and G. Brassard, “Quantum cryptography: Public key with the Korea Institute for Advanced Study (KIAS), the Centre for Quantum
distribution and coin tossing,” in Proc. IEEE Int. Conf. Comput., Syst. Technologies (CQT), Singapore, the ICFO-Institute of Photonic Sciences,
Signal Process., New York, NY, USA, Jan. 1984, pp. 175–179. the Freiburg Institute for Advanced Studies (FRIAS) as a Junior Fellow, and
[37] Y. Chen, M. Farahzad, S. Yoo, and T.-C. Wei, “Detector tomography on Hanyang University. He is currently with the School of Electrical Engineering,
IBM 5-qubit quantum computers and mitigation of imperfect measure- Korea Advanced Institute of Science and Technology (KAIST). His research
ment,” 2019, arXiv:1904.11935. [Online]. Available: https://arxiv.org/ interests contain secure quantum communication, entanglement applications,
abs/1904.11935 open quantum systems, quantum foundations, and their practical applications.
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on May 11,2022 at 05:54:08 UTC from IEEE Xplore. Restrictions apply.