Contextuality supplies the magic for quantum computation

Mark Howard,
1, 2,
Joel Wallman,
2,
Victor Veitch,
2, 3,
and Joseph Emerson
2,
1
Department of Mathematical Physics, National University of Ireland, Maynooth, Ireland
2
Institute for Quantum Computing and Department of Applied Mathematics,
University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1
3
Department of Statistics, University of Toronto,
100 St. George St. Toronto, Ontario, Canada M5S 3G3
(Dated: January 20, 2014)
Quantum information enables dramatic new advantages for computation, such as Shors factor-
ing algorithm [1] and quantum simulation algorithms [2]. This naturally raises the fundamental
question: what unique resources of the quantum world enable the advantages of quantum infor-
mation? There have been many attempts to answer this question, with proposals including the
hypothetical quantum parallelism [3] some associate with quantum superposition, the necessity
of large amounts of entanglement [4], and much ado about quantum discord [5]. Unfortunately
none of these proposals have proven satisfactory [68], and, in particular, none have helped resolve
outstanding challenges confronting the eld. For example, on the theoretical side, the most general
classes of problems for which quantum algorithms might oer an exponential speed-up over classical
algorithms are poorly understood. On the experimental side, there remain signicant challenges to
designing robust, large-scale quantum computers, and an important open problem is to determine
the minimal physical requirements of a useful quantum computer [9, 10]. A framework identify-
ing relevant resources for quantum computation should help clarify these issues, for example, by
identifying new ecient simulation schemes for classes of quantum algorithms and by clarifying the
trade-os between the distinct physical requirements for achieving robust quantum computation.
Here we establish that quantum contextuality, a generalization of nonlocality identied by Bell [11]
and Kochen-Specker [12] almost 50 years ago, is a critical resource for quantum speed-up within the
leading model for fault-tolerant quantum computation, magic state distillation [1315]. We prove
our results by nding the exact value of the independence number in an innite family of graphs,
which is a particular instance of an NP-hard problem.
Contextuality was rst recognized as an intrinsic fea-
ture of quantum theory via the Bell-Kochen-Specker no-
go theorem, which points to the impossibility of explain-
ing the statistical predictions of quantum theory in a nat-
ural way. The contextuality of quantum theory means
that a particular outcome of a quantum measurement
cannot be understood as revealing the pre-existing de-
nite value of some underlying hidden variable [16]. In-
stead, any assignment of a denite outcome must depend
upon the full details of the measurement (i.e., the con-
text), even though the probability of the outcome is in-
dependent of this context.
A key observation is that the nonlocality of quantum
theory is a special case of contextuality where the unex-
pected context dependence is on the choice of measure-
ments performed on a remote physical system. When
only local quantum operations and shared randomness
(LOSR) are available, non-locality emerges as a quanti-
able resource in communication complexity [17] and prac-
tical developments such as device-independent quantum
key distribution [18]. Hence a useful approach to iden-
tify the resources required for quantum computation is
to rst identify operational restrictions that are relevant
in the context of quantum computation. Locality restric-
tions are not enforced in the setting of quantum circuits,
so it is unlikely that nonlocality or entanglement is the
precise resource relevant to quantum speed-up. In fact,
it has been recently shown that a large amount of en-
tanglement is neither necessary nor sucient for an ex-
ponential computational speed-up [8]. Here we consider
the framework of fault-tolerant (FT) stabilizer quantum
computation (QC) [19] which provides one of the most
promising routes to achieving robust universal quantum
computation thanks to the discovery of high-threshold
codes in 2D geometries [20] (see, e.g., [21] for a review).
In this framework, only a subset of quantum operations
the stabilizer operations can be achieved via a fault-
tolerant encoding. These dene a convex subtheory of
quantum theory, the stabilizer subtheory, which is not
universal and in fact admits an ecient classical simula-
tion [22]. The stabilizer subtheory can be promoted to
universal QC through a process known as magic state
distillation [1315] which consumes a large number of re-
source states, but the set of states which can enable uni-
versal QC through the distillation process is not known,
and identifying optimal methods and rates for distillation
is a signcant open problem.
Here we show that quantum contextuality plays a criti-
cal role in characterizing the suitability of quantum states
for magic state distillation. Our approach builds on the
recent work of Cabello, Severini and Winter (CSW) [23]
that has established a remarkable connection between
contextuality and graph-theory. The CSW framework
provides a set of experimentally testable noncontextual-
ity inequalities that apply to arbitrary generalized prob-
abilistic theories, including both post-quantum theories
a
r
X
i
v
:
1
4
0
1
.
4
1
7
4
v
1


[
q
u
a
n
t
-
p
h
]


1
6

J
a
n

2
0
1
4
2
and operationally-restricted subsets of quantum theory,
such as the restriction to stabilizer operations relevant
to FT stabilizer computation. It is in this latter context
that we establish that the boundary for the set of quan-
tum states currently known to be unsuitable for magic
state distillation [13, 24, 25] is dened by a set of non-
contextuality inequalities. The scope of our results diers
depending on whether we consider a model of computa-
tion using systems of even prime dimension (i.e. qubits)
or odd prime dimension (i.e. qudits). Whereas in both
cases we can prove that violating a non-contextuality in-
equality is necessary for quantum-computational speed-
up via MSD, in the qudit case we are able to prove that
a state violates a noncontextuality inequality if and only
if it lies outside of the known boundary for MSD.
Graph-based contextuality.Consider a set of n binary
tests, which can be represented in quantum mechanics by
a set of n rank-1 projectors
1
, . . . ,
n
. Two such tests
are compatible, and so can be simultaneously performed
on a quantum system, if and only if the projectors are
orthogonal. Associated with the set of tests, dene a
witness operator as
=
n

i=1

i
. (1)
In a noncontextual hidden variable (NCHV) model, only
one outcome can occur when a measurement of two or-
thogonal projectors is performed, so we require that a
value of 1 will be assigned to at most one of any pair of
commuting projectors. We denote the NCHV maximum
of the expectation of as
NCHV
max
and this quantity is
given by [23] the independence number of an associated
exclusivity graph, (), wherein each vertex corresponds
to a projector and two vertices are adjacent (connected)
if the corresponding projectors are compatible i.e.,

NCHV
max
= () (2)
The independence number of a graph, (), describes the
size of the largest set of vertices from such that no two
elements of the set are adjacent. The maximum quan-
tum mechanical value of can be obtained by varying
over projectors satisfying the appropriate commutation
relations and over all quantum states. This quantity is
bound above by the Lovasz number of the exclusivity
graph i.e.,

QM
max
(), (3)
where can be calculated as the solution to a semi-
denite program. Graphs for which () < () indicate
that appropriately chosen projectors
i
and states
may reveal quantum contextuality by violating the non-
contextuality inequality
Tr() () . (4)
For generalized probabilistic theories (GPT), an impor-
tant class of post-quantum theories, the maximum
value of is given by the fractional packing number of
the exclusivity graph

() i.e.,

GPT
max
=

(). (5)
Note that if () <
QM
max
=

(), then the optimal

choice of quantum state and projectors is maximally con-
textual, in that no greater violation of the noncontextu-
ality inequality can be obtained in any GPT.
The stabilizer formalism.The stabilizer formalism for
p-dimensonal systems is dened using the generalized X
and Z operators
X[j = [j + 1 Z[j =
j
[j , (6)
where = exp(
2i
p
). The set of Weyl-Heisenberg dis-
placement operators is dened as
D
p
= D
x,z
=
2
1
xz
X
x
Z
z
: x, z Z
p
, (7)
where 2
1
is the multiplicative inverse of 2 in the nite
eld Z
p
= 0, 1, . . . , p 1. For p = 2, one can replace

2
1
with i in Eq. (7) to recover the familiar qubit Pauli
operators. The Cliord group C
p,n
is dened to be the
normalizer of the group D
n
p
(i.e., the group generated
by the set of displacement operators), that is,
C
p,n
= U |(d
n
) : UD
n
p
U

= D
n
p
, (8)
and the set of stabilizer states is the image of the com-
putational basis under the Cliord group C
p,n
.
The stabilizer polytope is the convex hull of the set of
stabilizer states. For a single particle, the stabilizer poly-
tope [27] is dened by the following set of simultaneous
inequalities
T
STAB
=
_
: Tr(A
q
) 0, q Z
p+1
p
_
(9)
where A
q
= I
p
+

p+1
j=1

qj
j
and
qj
j
is the projector onto
the eigenvector with eigenvalue
qj
of the jth operator in
the list D
0,1
, D
1,0
, D
1,1
, . . . , D
1,p1
(the eigenbases of
these operators form a complete set of mutually unbiased
bases).
Magic state distillation (MSD).MSD with stabilizer
codes is one of the most promising methods of achiev-
ing universal FTQC [1921]. An MSD protocol consists
of the following steps: (i) Prepare n copies of a suitable
(see later) input state, i.e.,
n
in
(ii) Perform a Cliord
operation on
n
in
(iii) Perform a stabilizer measurement
on all but the rst m registers, postselecting on a desired
outcome. With appropriate choices of stabilizer opera-
tions, the resulting output state in the rst m registers,

m
out
, is puried in the direction of a magic state [, so
that [
out
[ > [
in
[. This process can be reiter-
ated until
out
is suciently pure, at which point the
3
resource
out
is used up to approximate a non-Cliord
operation (via state injection), e.g., the /8 gate or its
qudit generalizations [15, 26]. Supplementing stabilizer
operations with the ability to perform such gates enables
fault-tolerant and universal QC.
FIG. 1. A 2-dimensional slice through qutrit state space:
Three distinct regions in the space of Hermitean operators Q
describing quantum state space (density operators), PSIM cor-
responding to ancillas known to be eciently simulable (and
hence useless for quantum computation via Magic State Dis-
tillation) and PSTAB describing mixtures of stabilizer states;
the strict inclusion PSTAB Q

PSIM identies a large class

of bound magic states [24].
For which states
in
does there exist an MSD routine
purifying
out
towards a non-stabilizer state? A large
subset of quantum states have been ruled out by virtue
of the fact that ecient classical simulation schemes are
known for noiseless stabilizer circuits supplemented by
access to an arbitrary number of states from the polytope

in
T
SIM
[22, 24, 25]. This polytope T
SIM
of the known
simulable states is prescribed by [27, 28]
T
SIM
=
__
: Tr(A
r
) 0, r Z
3
2
_
p = 2,
_
: Tr(A
xa+z

b
) 0, x, z Z
p
_
p > 2
(10)
where a = [1, 0, 1, . . . , p 1] and

b = [0, 1, 1, . . . , 1]
[29]. Note that T
SIM
= T
STAB
for qubits (giving an
octahedron inscribed within the Bloch sphere) whereas
T
SIM
T
STAB
is a proper superset for all other primes.
In future we refer to the set of facets enclosing T
SIM
as
/
SIM
= A
r
[p = 2 : r Z
3
2
, p ,= 2 : r = xa + z

b . (11)
In Fig. 2 we plot the geometric relationship between arbi-
trary quantum states, and sets of states contained within
T
SIM
and T
STAB
for the case of qutrits.
By the results of [24, 30], the set of states T
SIM
co-
incides exactly with the set of states that are nonnega-
tively represented within a distinguished quasiprobabil-
ity representationa discrete Wigner function (DWF)
[31, 32]. Are the states in the set T
SIM
, the set excluded
from MSD by the known ecient simulation schemes, the
complete set of non-distillable states? We address this
fundamental question by demonstrating a remarkable re-
lationship between non-distillability, non-negativity and
non-contextuality. For qudits of odd prime dimension,
a state is non-contextual under the available set of
measurementsstabilizer measurementsif and only if
it lies in the polytope T
SIM
. The same construction
applied to qubits also identies all T
SIM
as non-
contextual. These result establish that contextuality is
a necessary resource for universal quantum computation.
However, for qubits it is no longer true that T
SIM
is
non-contextual for dierent constructions, that is, for dif-
ferent sets of stabilizer measurements. Therefore, while
contextuality alone is plausibly a sucient resource for
universal quantum computation for qudits of odd prime
dimension, identifying the set of sucient resources for
the qubit setting remains an open problem.
Results.We will prove that all states / T
SIM
exhibit
state-dependent contextuality with respect to stabilizer
measurements. Rearranging the denition of A
r
given in
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
FIG. 2. Our construction applied to 2 qubits: Each of
the 30 vertices in this graph corresponds to a 2-qubit
stabilizer state; connected vertices correspond to orthogonal
states. A maximum independent set (representing mutually
non-orthogonal states) of size () = 8 is highlighted. As de-
scribed in Thm. 1, this value of identies all states / PSIM
as exhibiting contextuality with respect to the stabilizer mea-
surements in our construction.
4
Eq. (9) gives
p+1

j=1

s
j
Zp
sj=rj

sj
j
= pI
p
A
r
, (12)
that is, the set of projectors
sj=rj
j
is a set of projectors
whose sum,
r
, is such that
Tr(
r
) > p Tr(A
r
) < 0. (13)
The left hand side of this equivalence is a witness for con-
textuality if and only if the independence number of the
associated graph
r
satises (
r
) = p as in Eq. (4). It
can be shown that no set of single-qudit stabilizer mea-
surements identies any single-qudit state as exhibiting
contextuality [33], that is, Tr() () for all . By
using additional entangled stabilizer states in our con-
struction we will show that exhibits contextuality with
respect to stabilizer measurements if and only if / T
SIM
Our construction uses a dierent set of projectors for
each facet A
r
. For a xed facet A
r
, we dene a set of
separable projectors

r
sep
=
sj=rj
j
[kk[ : 1 j p + 1, s
j
, k Z
p

(14)
that is, take the p(p
2
1) separable projectors consisting
of all tensor products of projectors in Eq. (12) for the
rst qudit and computational basis states for the second
qudit. We also dene the set
ent
to be the set of all
two-qudit entangled projectors.
The sum of the combined set of separable and entan-
gled projectors
r
=
r
sep

ent
is

r
= (p
3
I
p
A
r
) I
p
(15)
so that for any state 1
p
of the second system (even
the maximally mixed state) we have
Tr
_

r
( )

p
3
Tr
_
A
r

0. (16)
Forming the exclusivity graph
r
of
r
and applying
the results of CSW identies the left hand side of Eq. (16)
as a witness for the contextuality of . We prove the
following theorem in the Appendix, which shows that
the inequality on the left-hand-side of Eq. (16) is indeed
a noncontextuality inequality.
Theorem 1. The independence number of the exclusivity
graph associated with
r
is (
r
) = p
3
for all A
r
/
SIM
so that, relative to our construction, exactly the states
/ T
SIM
are those that exhibit contextuality. For qudits
of odd prime dimension there does not exist any construc-
tion using stabilizer measurements that characterizes any
T
SIM
as contextual, so that the conditions for contex-
tuality and the possibility of quantum speed-up via magic
state distillation coincide exactly. Furthermore

2qudit
max
= (
r
) =

(
r
) = p
3
+ 1, (p > 2) (17)
which means maximally contextual states saturate the
bound on contextuality associated with post-quantum gen-
eralized probabilistic theories.
While (
r
) = p
3
also holds for qubits (the exclusivity
graph
r
and independent set is depicted in Fig. 2) it no
longer holds that states T
SIM
are noncontextual with
respect to any set of stabilizer measurements. Because a
dierent collection of two-qubit stabilizer measurements
can characterize T
SIM
as contextual, this forces us
to reconsider the operational signicance of contextual-
ity in the qubit setting. For example, the 24 entangled
two-qubit projectors comprise a rank-1 demonstration
of the Peres-Mermin (PM) magic square [34, 35]. The
associated witness operator and graph parameters are

PM
= 6I
4
and
(
PM
) = 5,
PM

2qubit
= (
PM
) = 6, (18)
where the expectation value of any two qubit state is
sucient to achieve Tr(
PM
) = 6 > (
PM
) = 5.
For completeness we note that the maximally contex-
tual qubit state relative to
r
is given by [TT[ where
[T is the magic state introduced in [13] and is com-
pletely arbitrary. The operator norm of
r
constructed
via two-qubit stabilizer states gives
r

2qubit
max
= (15 +

3)/2 8.366 which is less than that which can be ob-

tained in arbitrary GPTs,

2qubit
max
< (
r
) <

(
r
) p
3
+ 1, (p = 2). (19)
Discussion and conclusion. We have identied a set of
noncontextuality inequalities that characterize the set of
states that are necessary (and conjectured to be sucient
[36]) for universal QC using magic state distillation [13].
A natural question to consider is whether contextual-
ity is a resource in any scheme for UQC. Understand-
ing and resolving the disparity in behaviour between sys-
tems of even and odd prime dimension is an important
open problem in this direction. Besides the circuit model,
another natural framework to consider is measurement-
based quantum computation [10] for which contextuality
(in the form of nonlocality for xed measurements) has
already been shown to quantify the resources required to
perform certain computations [37, 38].
Acknowledgements.M.H. was nancially supported
by the Irish Research Council (IRC) as part of the Em-
power Fellowship program, and all authors acknowledge
nancial support from CIFAR and the Government of
Canada through NSERC.
Appendix
Here we prove the theorem presented in the main text.
We do not explicitly prove () = p
3
for p = 2 as this
requires a modied technique. However, it is easy to
5
verify using appropriate software [39], or by verifying that
the largest independent set of the graph in Fig. 2 is 8.
Henceforth, consider p an odd prime. Taken together,
Theorems 2 and 3 imply Theorem 1 of the main text.
Theorem 2. The exclusivity graph for the set of pro-
jectors
r
has graph parameters (
r
) p
3
and

QM
max
= (
r
) =

(
r
) = p
3
+ 1.
Proof. We will begin by proving that both and

are
upper-bounded by p
3
+ 1 using the graph theoretical in-
equalities
()
QM
max
()

() (), (20)
where () N is the clique cover number, an integer
corresponding to the minimum number of cliques needed
to cover every vertex of . The clique cover number
cannot be greater than the number of distinct bases in

r
, which contains p + 1 separable bases and p
3
p
entangled bases. Therefore ()

() p
3
+ 1.
For qudits with p > 2 there exist quantum states
such that Tr(A
r
) = 1 [31], which by Eq. (16) gives
Tr
_

r
( )

= p
3
+ 1 as a lower bound on

QM
max
.
With reference to Eq. (20) we immediately see that the
nal four quantities are all equal to p
3
+ 1.
We now give a lower bound for (), with the aid of the
phase-space formalism for stabilizer projectors described
in, e.g., Refs. [24, 31, 40]. By denition
r

NCHV
max
=
(
r
) = (
r
), where is the graph-complement of
and () is the size of the largest subset of vertices of
such that every pair of vertices is connected. The graph

r
can be dened as that in which projectors
i
and
j
are connected if and only if they do not commute.
In the phase-space formalism, two stabilizer projectors
do not commute if and only if their phase-space represen-
tations share at least one common point. Hence (
r
)
is lower bounded by the maximum number of projectors
from our set
r
that pass through any point u in phase
space i.e.,
(
r
) max
uZ
4
p

i

r
:
i
u
_

. (21)
where u
i
means that the phase space point u is
contained in the phase space line corresponding to state

i
. A two-qudit phase point operator A
u
with u Z
4
p
is
given explicitly by
A
u
= A
u1a+u2

b
A
u3a+u4

b
(22)
with a and

b dened as in Eq. (10). Since
Tr (
i
A
u
) =
_
1 if u
i
0 if u ,
i
, (23)
we have, by Eq. (21) and Eq. (23)
(
r
) max
uZ
4
p

i

r
:
i
u
_

max
uZ
4
p
Tr
_

r
A
u
_
max
uZ
4
p
Tr
__
(p
3
I
p
A
r
) I
p

A
u
_
p
3
min
(u1,u2)Z
2
p
Tr
_
A
r
A
u1a+u2

b
_
p
3
, (24)
where the second last line follows from Tr(A
u
) = 1 and
the last line from Tr(A
r
A
s
) = (r s) with s = u
1
a +
u
2

b.
Theorem 3. The exclusivity graph
r
associated with
our construction has independence number (
r
) p
3
.
To prove Theorem 3, note that the independence num-
ber of the exclusivity graph of a set of rank-1 projec-
tors
k
is the size of the largest subset of mutually
nonorthogonal elements of
k
. Consequently, any in-
dependent set can contain at most one element of each
orthonormal stabilizer basis, of which there are p
3
+1 in
our construction.
All graphs obtained by applying our construction to
operators A
r
/
SIM
are isomorphic since /
SIM
=
_
D
x,z
A
[0,0,...,0]
D

x,z
, x, z Z
p
_
[31]. We call our con-
struction applied to A
[0,0,...,0]
our canonical construction.
We will show that for any maximally independent sets
I
sep
and I
ent
of separable and entangled projectors re-
spectively, the only way for I
sep
I
ent
to be a maximally
independent set is for I
sep
to contain a zero eigenstate.
Since I
sep
cannot contain zero eigenstates in the canon-
ical construction, this proves that () < p
3
+ 1 (which
implies () p
3
since () is an integer) for the canon-
ical construction and hence for all A
r
/
SIM
.
To achieve this goal, we need to determine when ele-
ments of separable and entangled bases can be orthog-
onal. To this end, we parametrize the Cliord group
in terms of the semi-direct product of the symplectic
SL(2, Z
p
) and the Pauli groups [29].
In this parameterization of the Cliord group, Cliord
elements are written as
C = D
x,z
U
F
, (25)
where D
x,z
is as dened in the Eq. (7) and
F =
_


_
(26)
is an element of the symplectic group SL(2, Z
p
), that is,
the entries of F are elements of Z
p
and det F = 1, and
U
F
=
_
1

p1
j,k=0

1
(k
2
2jk+j
2
)
[jk[ ,= 0

p1
k=0

k
2
[kk[ = 0
(27)
6
where =
2
1
.
Counting all choices of x, z and F, there are [(
p
[ =
p
3
(p
2
1) elements of the single-qudit Cliord group.
In what follows, we will treat the Pauli and the sym-
plectic components separately. For Pauli operators we
have D

x,z
= D
(x|z)
, while for symplectic gates, C

F
=
C
F
1. We then have
C
F
D
x,z
C

F
= D
Fx,z
. (28)
An important feature of the two-qudit Cliord group
that enables the following proof is that the set of two-
qudit entangled states is exactly the set
[x, z, F := (D
x,z
U
F
I)[ ([ =

j
[jj/

p)
(29)
of states Jamiolkowski isomorphic to to the single-qudit
Cliord group. Moreover, F labels an orthonormal en-
tangled basis, while the Pauli component selects an ele-
ment of the basis [41].
We dene BP to be the group of computational-basis-
preserving gates contained within SL(2, Z
p
),
BP :=
_
C
,
=
_
0

1
_
: Z

p
, Z
p
_
, (30)
where Z

p
= 1, . . . , p 1. The left cosets of BP map
the computational basis to the dierent mutually unbi-
ased bases (MUBs) for a single qudit. To see that this
must be the case, note that each left coset maps the com-
putational basis to the same basis, since for any F, F

in
a coset, we can write F

= FC
,
for some , (by de-
nition of a coset), and hence
k[U

F
U
F
[l = k[U
C,
[l (k l) . (31)
where denotes the Kronecker delta. Therefore, since
there are p + 1 MUBs for a qudit of dimension p and
p + 1 left cosets of BP (since [SL(2, Z
p
)[/[BP[ = p + 1),
they must be in one-to-one correspondence.
This one-to-one correspondence between MUBs and
cosets shows that the set of entangled bases that are not
mutually unbiased (non-MU) with respect to a a sepa-
rable basis (U
F
b
[k)[l are precisely the entangled bases
that are Jamio lkowski isomorphic to the left coset of BP
containing F
b
. To see this, consider the inner product
[(k[U

F
b
)l[[x, z, F[
2
= p
1
[k[U

F
b
D
x,z
U
F
[l[
2
= p
1
[k[D
F
1
b
x,z
U
F
1
b
F
[l[
2
(32)
for some xed F. Using Eq. (27), the above inner product
will be p
2
unless F
1
b
F = C
,
for some , .
In general, we consider left-coset representatives with
trace 2. For ease of notation, we label the coset repre-
sentatives as
F
b
=
_
1 b
0 1
_
(b ,= ) (33)
F

=
_
2 1
1 0
_
. (34)
Note that this labeling gives a dierent labeling of the
MUBs then that of the main text.
By explicit calculation, we then nd that the entangled
states that are non-MU with respect to a given separable
basis satisfy
[k[U

F
b
l[x, z, F
b
C
,
[
2
=
_

_
1
p
if x bz = k l and b ,=
1
p
if z = k l and b =
0 otherwise,
(35)
for all b Z
p,
:= Z
p
, k, l Z
p
and C
,
BP.
The constraints from Eq. (35) (applied to appropri-
ate parameterizations of I
sep
and I
ent
) are the only con-
straints required for I
sep
and I
ent
to be mutually non-
commuting sets. The individual sets I
sep
and I
ent
will
be independent if and only if they contain no mutually
orthogonal elements. We now determine when two en-
tangled states are not orthogonal.
Lemma 4. Two entangled stabilizer states [x
1
, z
1
, F
1
C
1

and [x
2
, z
2
, F
2
C
2
are orthogonal if and only if Tr F = 2
and

F
z ,= (1
F
)x (36)
where
F = C
1
1
F
1
1
F
2
C
2
=
_

F

F

F

F
_
(37)
and
_
x
z
_
= C
1
1
F
1
1
_
x
2
x
1
z
2
z
1
_
. (38)
Proof. We rst note that
[x
1
, z
1
, F
1
C
1
[x
2
, z
2
, F
2
C
2
[ = [[x, z, F[ , (39)
where F = C
1
1
F
1
1
F
2
C
2
and
_
x
z
_
= C
1
1
F
1
1
_
x
2
x
1
z
2
z
1
_
. (40)
From Ref. [41], the above inner product can only be zero
if Tr F = 2, so assume that Tr F = 2.
Case 1:
F
,= 0. Since det F = 1 and Tr F = 2, we
can write
F =
_

F

F

1
F
(
F
1)
2
2
F
_
(41)
7
Expanding out the inner product gives
[x, z, F = p
1

kZp

kz
k x[U
F
[k
p
3/2

kZp

(
1
F
x(
F
1)z)k
p
1/2
(
1
F
x(
F
1) z) (42)
where the last line follows from the standard identity

kZp

ak
= p(a) a Z
p
. (43)
Case 2:
F
= 0. If
F
= 0 and Tr F = 2, then the
requirement det F = 1 gives
F =
_
1 0

F
1
_
. (44)
Expanding out the inner product gives
[x, z, F = p
1

kZp

kz
k x[U
F
[k (45)
= p
1

kZp

kz+2
1
k
2
k x[k ,
(46)
which will be nonzero if and only if x = 0 (unless = 0,
in which case z = 0 also).
Lemma 5. Let
I
sep
=
rj
j
[l
j
l
j
[ : j = 1, . . . , p + 1
be a maximal independent set of separable projectors.
Then there only exists an independent set I
ent
of p(p
2
1)
stabilizer projectors such that I
sep
I
ent
is an independent
set if I
sep
contains a zero eigenstate, that is, if r
j
= 0 for
some j.
Proof. Assume that I
sep
and I
ent
are maximal indepen-
dent sets such that I
sep
I
ent
is also independent. We
parametrize I
sep
and I
ent
as (projectors onto the states)
I
sep
= (U
F
b
1)[k
b
l
b
: b Z
p,

I
ent
= [x
b
,
, z
b
,
, F
b
C
,
: b Z
p,
(47)
for some choice of k
b
, l
b
, x
b
,
, z
b
,
: b Z
p,
,
Z

p
, Z
p
. Note that this parameterization of single-
qudit stabilizer bases is dierent from the parameteriza-
tion in the main text, but is more suitable for subsequent
calculations using the Jamio lkowski isomorphism.
The crucial point is that the set of zero eigenstates
in the two parameterizations is the same, so that we can
prove the lemma by showing that there exists some b such
that k
b
= 0. To show that the set of zero eigenstates are
the same, we need to show that for all j, [
0
j
= [
0
b
for
some b, where [
0
j

0
j
[ =
0
j
is the labeling of stabilizer
states in the main text.
We trivially have [
0
1
= [
0
0
= [0 since F
0
= I. We
now show that [
0
b
= U
F
b
1
[0 = [
0
b
1
for b ,= 0, .
Note that [
0
j
is the unique state such that D
1,j1
[
j
=
[
j
for j = 1, . . . , p + 1. For b ,= 0, , we then have
D
1,b
[
0
b+2
= U
F
b
1
U

F
b
1
D
(1|b)
U
F
b
1
[0
= U
F
b
1
D
F
1
b
1
(1|b)
[0
= U
F
b
1
D
(0|b)
[0
= U
F
b
1
[0
= [
0
b+2
, (48)
as required. The exact same argument shows that [
0
2
=
F

[0 = [
0

.
Having shown that the sets of zero eigenstates in the
two parameterizations is the same, that is,

0
j
: j = 1, . . . , p + 1 = [
0
b

0
b
[ : b Z
p,
, (49)
we now obtain constraints on I
sep
and I
ent
for I
sep
I
ent
to be an independent set.
First note that for I
sep
to be an independent set, we
must have l
b
= l
0
=: l for all b. We will show that k
b
= 0
is required for some value of b in order for I
sep
I
ent
to
be an independent set, which completes the proof.
First note that by Lemma 4, requiring
x
0
,
, z
0
,
, C
,
[x
b
,
, z
b
,
, F
b
C
,
, = 0 (50)
is equivalent to requiring
z
b
,
= z
0
,
, (b ,= )
x

,
= x
0
,
+ z
0
,
z

,
, (b = ) , (51)
where we have used
C
1
,
F
b
C
,
=
_

_
_
1 +
1
b
2
b

2
b 1
1
b
_
(b ,= )
_
2 +
1

2
( + )
2

_
(b = ).
(52)
We now consider the requirement that the sets I
sep
and
I
ent
are pairwise mutually nonorthogonal. By Eq. (35),
k
b
l
b
[(U

F
b
1)[x
b
,
, z
0
,
, F
b
C
,
, = 0 (53)
is equivalent to
x
b
,
= bz
0
,
+ k
b
l , (b ,= )
z

,
= l k

=: z

, (b = ) , (54)
which completely characterizes the restrictions on I
ent
such that it contains no elements orthogonal to any ele-
ment of a xed I
sep
.
8
This then completely species every parameter except
z
0
,
and k
b
: b Z
p,
. To specify the remaining
parameters, we will have to impose further constraints
on the elements of I
ent
to ensure I
ent
contains no pairs of
mutually orthogonal elements.
To do this, note F
b
F
c
= F
b+c
for b, c Z
p
and
Tr F
c
C
2,(2c)
1 = 2, so again by Lemma 4
x
b
1,
, z
0
1,
, F
b
C
1,
[x
b+c
2,
, z
0
2,
, F
b+c
C
2,
, = 0 , (55)
where

= 2
1
(2c)
1
, is equivalent to
k
b+c
k
b
= l c(2z
0
2,
z
0
1,
) (56)
for all b, Z
p
, c ,= 0, , where we have used Eq. (51)
and (54). If the right-hand-side of Eq. (56) is nonzero for
any value of c or , then k
b
Z
p
and so I
sep
contains
a zero eigenstate.
Therefore the only way for I
sep
to not contain a zero
eigenstate given that I
sep
I
ent
is an independent set is
if k
b
= k for all b Z
p
and
l = c(2z
0
2,
z
0
1,
) c ,= 0, . (57)
This can be solved to obtain
z
0
1,
= z
0
1,0
l
z
0
2,
= z
0
2,0
l (58)
We now consider the case where b = in Eq. (55),
which implies that
x

1,0
, z

1
, F

[x
c
2,
, z
0
2,
, F
c
C
2,
, = 0 , (59)
for c ,= 2 where

= 2
1
c 1 is equivalent to
(c 1)k

= 2z
0
2,0
z
0
1,0
, (60)
which can only hold independently of c if k

= 0.

mark.howard@uwaterloo.ca

joel.wallman@uwaterloo.ca

vveitch@uwaterloo.ca

jemerson@math.uwaterloo.ca
[1] P. W. Shor Algorithms for quantum computa-
tion: discrete logarithms and factoring, Foun-
dations of Computer Science, 1994 Proceedings.,
35th Annual Symposium on , pp.124134, (1994).
http://dx.doi.org/10.1109/SFCS.1994.365700
[2] S. Lloyd Universal Quantum Simulators,
Science 273 5278 pp. 10731078 (1996)
http://dx.doi.org/10.1126/science.273.5278.1073
[3] D. Deutsch, Quantum theory, the Church-Turing prin-
ciple and the universal quantum computer. Proceedings
of the Royal Society of London. A. Mathematical and
Physical Sciences Volumne 400(1818) pp. 97117, (1985)
http://dx.doi.org/10.1098/rspa.1985.0070
[4] G. Vidal, Ecient Classical Simulation
of Slightly Entangled Quantum Computa-
tions Phys. Rev. Lett. 91, 147902 (2003)
http://dx.doi.org/10.1103/PhysRevLett.91.147902
[5] A. Datta, A. Shaji and C. M. Caves,
Quantum Discord and the Power of One
Qubit Phys. Rev. Lett.100, 050502 (2008)
http://dx.doi.org/10.1103/PhysRevLett.100.050502
[6] A. M. Steane, A quantum computer only needs
one universe arXiv:quant-ph/0003084 (2000)
http://arxiv.org/abs/quant-ph/0003084
[7] V. Vedral, The Elusive Source of Quantum
Speedup, Found. Physics, 40, 8, pp. 11411154
(2010). http://dx.doi.org/10.1007/s10701-010-9452-
0 A. Brodutch, Discord and quantum computa-
tional resources, Phys. Rev. A 88, 022307 (2013).
http://dx.doi.org/10.1103/PhysRevA.88.022307
[8] M. Van den Nest, Universal Quan-
tum Computation with Little Entangle-
ment Phys. Rev. Lett. 110, 060504 (2013)
http://dx.doi.org/10.1103/PhysRevLett.110.060504
[9] E. Knill and R. Laamme, Power of One Bit of Quan-
tum Information Phys. Rev. Lett. 81, 56725675 (1998).
http://dx.doi.org/10.1103/PhysRevLett.81.5672
[10] H. J. Briegel, D. E. Browne, W. D ur, R. Raussendorf,
M. Van den Nest, Measurement-based quan-
tum computation, Nat. Phys. 19 - 26 (2009).
http://dx.doi.org/10.1038/nphys1157
[11] J. Bell, On the Problem of Hidden Variables in Quan-
tum Mechanics, Rev. Mod. Phys. 38, 447452 (1966).
http://dx.doi.org/10.1103/RevModPhys.38.447
[12] S. Kochen and E. P. Specker, The Prob-
lem of Hidden Variables in Quantum Mechan-
ics, J Math. Mech. 17(1), pp. 59-87 (1968).
http://www.iumj.indiana.edu/IUMJ/FTDLOAD/1968/17/17004/pdf
[13] S. Bravyi and A. Kitaev, Universal quantum
computation with ideal Cliord gates and noisy
ancillas, Phys. Rev. A 71, 022316 (2005).
http://dx.doi.org/10.1103/PhysRevA.71.022316
[14] E. Knill, Quantum Computing with Realistically
Noisy Devices Nature 434, pp. 3944, (2005).
http://dx.doi.org/10.1038/nature03350
[15] E. T. Campbell, H. Anwar and D. E. Browne Magic
state distillation in all prime dimensions using quantum
Reed-Muller codes, Phys. Rev. X 2, 041021, (2012).
http://link.aps.org/doi/10.1103/PhysRevX.2.041021
[16] N. D. Mermin, Simple unied form
for the major no-hidden-variables theo-
rems, Phys. Rev. Lett. 65, 3373, (1990).
http://link.aps.org/doi/10.1103/PhysRevLett.65.3373
[17] H. Buhrman, R. Cleve, S. Massar, and R. de
Wolf, Nonlocality and communication com-
plexity, Rev. Mod. Phys. 82, 1, (2010).
http://link.aps.org/doi/10.1103/RevModPhys.82.665
[18] A. Acn, N. Brunner, N. Gisin, S. Massar, S. Piro-
nio, and V. Scarani, Device-Independent Security of
Quantum Cryptography against Collective Attacks,
Phys. Rev. Lett. 98, 230501 (2007).
[19] D. Gottesman, Theory of fault-tolerant quan-
tum computation, Phys. Rev. A 57, 127 (1998)
http://dx.doi.org/10.1103/PhysRevA.57.127
[20] R. Raussendorf, J. Harrington, and K. Goyal, Annals of
Physics 321, 2242 (2006). http://arxiv.org/abs/ quant-
ph/0510135
9
[21] A. G. Fowler, M. Mariantoni, J. M. Martinis, and
A. N. Cleland, Surface codes: Towards practical large-
scale quantum computation Phys. Rev. A. 86, 032324,
(2012). http://dx.doi.org/10.1103/PhysRevA.86.032324
[22] S. Aaronson and D. Gottesman, Improved simulation
of stabilizer circuits Phys. Rev. A 70, 052328 (2004).
http://link.aps.org/doi/10.1103/PhysRevA.70.052328
[23] A. Cabello, S. Severini and A. Winter. (Non-
)Contextuality of Physical Theories as an Axiom,
arXiv:1010.2163.
[24] V. Veitch, C. Ferrie, D. Gross and J. Emerson, Nega-
tive quasi-probability as a resource for quantum compu-
tation New Journal of Physics 14,11 pp. 113011, (2012).
http://dx.doi.org/10.1088/1367-2630/14/11/113011
[25] A. Mari, J. Eisert, Positive Wigner Functions Ren-
der Classical Simulation of Quantum Computation
Ecient Phys. Rev. Lett. 109, 230503, (2012).
http://link.aps.org/doi/10.1103/PhysRevLett.109.230503
[26] M. Howard and J. Vala, Qudit versions of the
qubit /8 gate Phys. Rev. A. 86, 022316, (2012).
http://dx.doi.org/10.1103/PhysRevA.86.022316
[27] C. Cormick, E. F. Galv ao, D. Gottesman, J. Pablo Paz,
and A. O. Pittenger, Classicality in discrete Wigner
functions Phys. Rev. A. 73, 012301, (2006).
http://dx.doi.org/10.1103/PhysRevA.73.012301
[28] W. van Dam and M. Howard, Noise thresholds for
higher-dimensional systems using the discrete Wigner
function Phys. Rev. A. 83, 032310, (2011).
http://dx.doi.org/10.1103/PhysRevA.83.032310
[29] D. M. Appleby, I. Bengtsson and S. Chaturvedi, Spectra
of phase point operators in odd prime dimensions and
the extended Cliord group J. Math. Phys. 49, 012102,
(2008). http://link.aip.org/link/?JMP/49/012102/11
[30] J. J. Wallman and S. D. Bartlett, Non-negative
subtheories and quasiprobability representations of
qubits Phys. Rev. A. 85, 062121, (2012).
http://dx.doi.org/10.1103/PhysRevA.85.062121
[31] D. Gross, Hudsons theorem for nite-dimensional quan-
tum systems J. Math. Phys. 47, number 12, 122107,
(2006). http://link.aip.org/link/doi/10.1063/1.2393152
[32] W. K. Wootters, The Discrete Wigner Function
Annals of the New York Academy of Sciences,
480 275-282 (1986) http://dx.doi.org/10.1111/j.1749-
6632.1986.tb12431.x K. S. Gibbons, M. J. Homan,
and W. K. Wootters, Discrete phase space based
on nite elds Phys. Rev. A. 70, 062101, (2004).
http://dx.doi.org/10.1103/PhysRevA.70.062101
[33] We rst note that no noncontextuality inequality can be
derived for any set {}
single
of single-qudit stabilizer pro-
jectors, since (
single
) = (
single
) = N where N is
the number of MUBs in {}
single
and (), which sat-
ises

() (), is the minimal number of cliques

required to cover every vertex in (a clique is a sub-
graph of in which every vertex is connected to every
other vertex). Any two-qudit contextuality witness com-
prised solely of separable projectors will either (i) reduce
to two independent witnesses if the set of projectors is
the tensor product of two sets of single-qudit projectors,
or (ii) will not provide independent information about
the rst subsystem (i.e., it will give information about
correlations). Therefore, to construct a noncontextuality
inequality that only involves the negativity of the rst
subsystem, we include the complete set of p
3
(p
2
1) two-
qudit entangled stabilizer projectors, {}ent in the con-
struction (smaller choices of sets of entangled bases also
work).
[34] M. Howard, E. Brennan and J. Vala, Quantum Contex-
tuality with Stabilizer States Entropy 15, 6, 23402362,
(2013). http://dx.doi.org/10.3390/e15062340
[35] Peres, A.; Two simple proofs of the Kochen-Specker
theorem J. Phys. A: Math. Gen. 24, L175 (1991).
http://dx.doi.org/10.1088/0305-4470/24/4/003
[36] V. Veitch, S. A. H. Mousavian, D. Gottesman,
J. Emerson The Resource Theory of Stabilizer
Computation arXiv:1307.7171 [quant-ph] (2013).
http://arxiv.org/abs/1307.7171
[37] Robert Raussendorf, Contextuality in
measurement-based quantum computa-
tion, Phys. Rev. A 88, 022322 (2013).
http://link.aps.org/doi/10.1103/PhysRevA.88.022322
[38] M. J. Hoban, J. J. Wallman and Dan E. Browne,
Generalized Bell-inequality experiments and
computation Phys. Rev. A 84, 062107 (2011).
http://link.aps.org/doi/10.1103/PhysRevA.88.022322
[39] P. R. J.

Ostergard, A fast algorithm for the maxi-
mum clique problem Disc. Appl. Math. 120 (1-3)
pp. 197-207, (2002). http://dx.doi.org/10.1016/S0166-
218X(01)00290-6 S. Niskanen and P. R. J.

Ostergard,
Cliquer Users Guide, Version 1.0 Communications
Laboratory, Helsinki University of Technology, Espoo,
Finland, Tech. Rep. T48, (2003).
[40] K. S. Gibbons, M. J. Homan, and W. K. Woot-
ters, Discrete phase space based on nite
elds Phys. Rev. A. 70, 062101, (2004).
http://dx.doi.org/10.1103/PhysRevA.70.062101
[41] W. van Dam and M. Howard, Bipartite entangled sta-
bilizer mutually unbiased bases as maximum cliques of
Cayley graphs Phys. Rev. A. 84, 012117, (2011).
http://dx.doi.org/10.1103/PhysRevA.84.012117