Standard form of qudit stabilizer groups

Vlad Gheorghiu1, ∗
1
Department of Physics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA
(Dated: Version of October 22, 2018)
We investigate stabilizer codes with carrier qudits of equal dimension D, an arbitrary integer
greater than 1. We prove that there is a direct relation between the dimension of a qudit stabilizer
code and the size of its corresponding stabilizer, and this implies that the code and its stabilizer are
dual to each other. We also show that any qudit stabilizer can be put in a standard, or canonical,
form using a series of Clifford gates, and we provide an explicit efficient algorithm for doing this.
Our work generalizes known results that were valid only for prime dimensional systems and may be
useful in constructing efficient encoding/decoding quantum circuits for qudit stabilizer codes and
better qudit quantum error correcting codes.
arXiv:1101.1519v1 [quant-ph] 7 Jan 2011

PACS numbers: 03.67.Mn, 03.67.Pp

I. INTRODUCTION stabilizer codes over prime-power finite fields [12]), and

the proof that such a form exists is one of the main results
Quantum error correction is an important part of of the current article.
various schemes for quantum computation and quan- The reminder of the paper is organized as follows.
tum communication, and hence quantum error correcting Sec. II contains definitions of the generalized Pauli group
codes, first introduced about a decade ago [1–3] have re- and some quantum gates used later in the paper. It
ceived a great deal of attention. For a detailed discussion also defines rigorously qudit stabilizers and their corre-
see Ch. 10 of [4]. Most of the early work dealt with codes sponding stabilized subspaces (or codes), together with
for qubits, with a Hilbert space of dimension D = 2, but an alternative algebraic notation that we employ later.
qudit codes with D > 2 have also been studied [5–15]. Sec. III contains our main results: a “size” theorem that
They are of intrinsic interest and could turn out to be of relates the size of the stabilizer group to the dimension of
some practical value. its stabilized subspace, followed by a “structure” theorem
The stabilizer formalism introduced by Gottesman in that shows that any qudit stabilizer can be brought to
[16] for D = 2 (qubits) provides a compact and pow- a standard form through a series of elementary quantum
erful way of generating quantum error correcting codes gates. Finally, Sec. IV contains a summary, conclusions,
and extends the notion of linear classical error correcting and some open questions.
codes [17] to the quantum domain. The stabilizer for-
malism has been generalized to cases where D is prime
II. PRELIMINARY REMARKS AND
or a prime power, see e.g. [6, 12, 18, 19]. For composite
DEFINITIONS
D things are more complicated and there is no imme-
diate and natural way of generalizing the notions. Our
approach is to use generalized Pauli operators and stabi- A. The generalized Pauli group on n qudits
lizers defined in the same way as in the prime case, see
e.g. [13, 15]. This has the virtue that many (although We generalize Pauli operators to higher dimensional
not all) results that are valid in the prime dimensional systems of arbitrary dimension D in the following way.
case can be extended without too much difficulty to the The X and Z operators acting on a single qudit are de-
more general composite case. fined as
An important problem in the theory of stabilizer codes D−1
X D−1
X
is what is their structure. Is there any “canonical” way Z= ω j |ji hj| , X= |ji hj + 1| , (1)
of representing an arbitrary stabilizer code? If yes, can j=0 j=0
one use this fact for implementing various quantum error-
and satisfy
correcting tasks? For prime D it turns out that there is
such a standard form, see e.g. Ch. 10.5.7 of [4], and this X D = Z D = I, XZ = ωZX, ω = e2πi/D , (2)
allows for a better understanding of the error-correcting
where the addition of integers is modulo D, as will be
capabilities of the stabilizer code and also provides an
assumed from now on. For a collection of n qudits we use
efficient way of constructing encoding/decoding circuits
subscripts to identify the corresponding Pauli operators:
for such stabilizer codes. For composite D we are not
thus Zi and Xi operate on the space of qudit i. The
aware of any such standard form (except for the case of
Hilbert space of a single qudit is denoted by H, and the
Hilbert space of n qudits by Hn , respectively. Operators
of the form
[email protected] ω λ X x Z z := ω λ X x1 Z z1 ⊗ X x2 Z z2 ⊗ · · · ⊗ X xn Z zn (3)
∗
1 1 2 2 n n
 2

will be referred to as Pauli products, where λ is an integer Pauli operator Sq F

in ZD and x and z are n-tuples in ZnD , the additive group Z Zq X
of n-tuple integers mod D. For a fixed n the collection
X X q̄ Z D−1
of all possible Pauli products (3) form a group under
operator multiplication, the Pauli group Pn . If p is a TABLE I. The conjugation of Pauli operators by one-qudit
Pauli product, then pD = I is the identity operator on gates F and Sq (q̄ is the multiplicative inverse of q mod D).
Hn , and hence the order of any element of Pn is either
D or else an integer that divides D. While Pn is not
Pauli product CNOTab SWAPab CPab
Abelian, it has the property that two elements commute
up to a phase: Ia ⊗ Zb Za ⊗ Zb Za ⊗ Ib Ia ⊗ Zb
Za ⊗ Ib Za ⊗ Ib Ia ⊗ Zb Za ⊗ Ib
p1 p2 = ω λ12 p2 p1 , (4) Ia ⊗ Xb Ia ⊗ Xb Xa ⊗ Ib ZaD−1 ⊗ Xb
Xa ⊗ Ib Xa ⊗ XbD−1 Ia ⊗ Xb Xa ⊗ ZbD−1
with λ12 an integer in ZD that depends on p1 and p2 .
TABLE II. The conjugation of Pauli products on qudits a and
b by two-qudit gates CNOT, SWAP and CP. For the CNOT
B. Generalization of qubit quantum gates to gate, the first qudit a is the control and the second qudit b
higher dimensions the target.

In this subsection we define some one and two qudit

gates generalizing various qubit gates. The qudit gener- with
alization of the Hadamard gate is the Fourier gate
(CNOTba )† = (CNOTba )D−1 = (Ia ⊗F2b )CNOTba (Ia ⊗F2b ).
D−1 (10)
1 X jk
F := √ ω |ji hk| . (5) Finally we define the generalized Controlled-phase or CP
D j=0 gate as

For an invertible integer q ∈ ZD (i.e. integer for which D−1

X D−1
X
there exists q̄ ∈ ZD such that q q̄ ≡ 1 mod D), we define CPab = |ji hj|a ⊗ Zbj = ω jk |ji hj|a ⊗ |ki hk|b .
a multiplicative gate j=0 j,k=0
(11)
D−1
X The CP and CNOT gates are related by a local Fourier
Sq := |ji hjq| , (6) gate, similar to the qubit case
j=0
CNOTab = (Ia ⊗ Fb )CPab (Ia ⊗ Fb )† , (12)
where qj means multiplication mod D. The requirement
that q be invertible ensures that Sq is unitary; for a qubit since F maps Z into X under conjugation (see Table I).
Sq is just the identity. The gates F, Sq , SWAP, CNOT and CP are unitary
For two distinct qudits a and b we define the CNOT operators that map Pauli operators to Pauli operators
gate as under conjugation, as can be seen from Tables I and II.
They are elements of the so called Clifford group on n
D−1 D−1
X X qudits [20, 21], the group of n-qudit unitary operators
CNOTab := |ji hj|a ⊗Xbj = |ji hj|a ⊗|ki hk + j|b , that leaves Pn invariant under conjugation, i.e. if O is
j=0 j,k=0 a Clifford operator, then ∀p ∈ Pn , OpO† ∈ Pn . From
(7) Tables I and II one can easily deduce the result of con-
the obvious generalization of the qubit Controlled-NOT, jugation by F, Sq , SWAP, CNOT and CP on any Pauli
where a labels the control qudit and b labels the target product.
qudit. Next the SWAP gate is defined as
D−1
X C. Qudit stabilizer codes
SWAPab := |ki hj|a ⊗ |ji hk|b . (8)
j,k=0
Relative to this group we define a stabilizer code C to
It is easy to check that SWAP gate is hermitian and does be a K ≥ 1-dimensional subspace of the Hilbert space
indeed swap qudits a and b. Unlike the qubit case, the satisfying three conditions:
qudit SWAP gate is not a product of three CNOT gates,
but can be expressed in terms of CNOT gates and Fourier C1: There is a subgroup S of Pn such that for every s
gates as in S and every |ψi in C

SWAPab = CNOTab (CNOTba )† CNOTab (F2a ⊗ Ib ), (9) s|ψi = |ψi (13)

 3

C2: The subgroup S is maximal in the sense that every Gate X-part Z-part
s in Pn for which (13) is satisfied for all |ψi ∈ C SWAPab Interchange columns Interchange columns
belongs to S.
a and b a + n and b + n
C3: The coding space C is maximal in the sense that any Sq,a Multiply column a by Multiply column a + n
ket |ψi that satisfies (13) for every s ∈ S lies in C. invertible integer q −1 by invertible integer q
(CNOTab )m Substract m times Add m times column
If these conditions are fulfilled we call S the stabilizer column a from column b b + n to column a + n
of the code C. That it is Abelian follows from the commu-
tation relation (4), since for K > 0 there is some nonzero TABLE III. Conjugation by the above quantum gates corre-
|ψi satisfying (13). spond to elementary column operations on the X and Z parts
Note that one can always find a subgroup S of Pn of the parity-check matrix of a stabilizer code. For the CNOT
satisfying C1 and C2 for any subspace C of the Hilbert gate, the first qudit a is the control and the second qudit b
space, but it might consist of nothing but the identity. the target. The integer exponent m means CNOT applied m
Thus it is condition C3 that distinguishes stabilizer codes times (or, equivalently, the m-th power of the CNOT gate).
from nonadditive codes. A stabilizer code is uniquely
determined by S as well as by C, since S determines C
through C3, so in a sense the code and its stabilizer are analysis can be done using the Theorem 1 of Sec. III,
dual to each other. which implies for this example that the size of the stabi-
lizer group must be equal to 4, hence it must be generated
by a single generator of order 4 or two generators each
D. Stabilizer generators and equivalent algebraic of order 2. By inspection it is easy to rule out the first
descriptions of qudit stabilizer codes case, so indeed one must use 2 generators.
A conjugation of a stabilizer group by a Clifford op-
Any stabilizer group can be compactly described using eration will change the stabilizer group to an isomor-
a set of group generators. A generator corresponds to a phic group. This will correspond to a column opera-
specific Pauli product and can be completely specified, tion on the parity-check matrix of the stabilizer, together
see (3), by a phase λ and two n-tuples in ZnD , x and z. with a transformation of the phase vector. On the other
A collection of k generators can therefore be represented hand, the generator description of a stabilizer group is
by a k-component phase vector over ZD (that contains not unique: one can multiply a generator by another one
all k phases) and a k × 2n parity-check matrix over ZD and still get the same group. This kind of operation cor-
with rows corresponding to the stabilizer generators. For responds to a row operation on the parity-check matrix,
example, the stabilizer again keeping in mind that in general the phase vector
will modify. From now on for the sake of simplicity we
S = hω 2 X13 Z22 , X22 i (14) will ignore the phase vector, although in real applications
one has to keep track of the phases.
corresponds to the phase vector (2, 0) and parity-check The following represent what we call elemen-
matrix tary row/column operations: a) interchanging of
! rows/columns, b) multiplication of a row/column by
3 0 0 2 an invertible integer, c) addition of any multiple of a
S= . (15)
0 2 0 0 row/column to a different row/column. The column op-
erations can be realized by conjugations of the stabilizer
The angular brackets in (14) means “group generated by the Clifford operations in Table III, and the row op-
by”, i.e. the group obtained by all possible products of erations just ensure that the stabilizer group remains the
the group generators. We call the left k × n block of the same, i.e. the new set of generators generate the same
parity-check matrix the X-block, and the right k × n the stabilizer group and not a smaller one.
Z-block, since they describe the X and Z parts of the
stabilizer generators, respectively.
Note that if D is a prime number, any stabilizer group III. SIZE-STRUCTURE THEOREMS
can be described using no more than n generators. How-
ever, in composite dimensions one can have more that n The following theorem generalizes to composite D a
generators but no more than 2n. For example, in D = 4, well-known result for prime D that relates the size of the
the n = 1 qudit stabilizer S = hX 2 , Z 2 i is generated by stabilizer group to the dimension of its stabilizer sub-
2 (and not√1) elements and specify the stabilizer state space. Although the composite D result of our next the-
(|0i + |2i)/ 2. There is no way of representing this state orem may have been known by the community (see e.g.
using only 1 generator; Z 2 by itself stabilizes both |0i the claim near the end of Sec. 3.6 of [16]), we have not
and |2i, hence everything in their span, so the condition yet seen a proof of it.
C3 is not satisfied, i.e. the coding space is not maximal.
The same kind of analysis holds for X 2 . A more rigorous Theorem 1 (Size). Let C be an n-qudit stabilizer code
 4

with stabilizer S. Then the 1-qudit stabilizer generated by S = hZ 2 i in D = 4.

It is obvious that S stabilizes a K = 2-dimensional code
K × |S| = Dn , (16) C=span{|0i, |2i}, and the size of the stabilizer is |S| = 2.
Whenever D = 2 (qubits) it was shown in [16] (see also
where K is the dimension of C, |S| is the size (or order) Ch. 10.5.7 of [4] for a detailed discussion) that any stabi-
of the stabilizer group S and D is the dimension of the lizer code can be put into a “standard form” or “canoni-
Hilbert space of one carrier qudit. cal form”, and this is very useful for constructing encod-
Proof. Define ing/decoding quantum circuits for stabilizer codes. This
result can be generalized at once to prime D. However,
1 X for composite dimensions, it is not so obvious how to do
P := s. (17)
|S| the generalization, and the main technical difficulty is
s∈S
that ZD is a ring (and not a field) and therefore some
We will first show that P is the projector onto C. integers are not invertible. However, in the following
It follows at once that Theorem we show that one can still apply a technique
similar to a Gaussian elimination over ZD and put any
P = P † = P 2, (18) composite D stabilizer code into a standard form similar
to the one of prime D case.
where the equalities follow from the group property of S,
so P is an orthogonal projector. Let |ψi be an arbitrary Theorem 2 (Standard form). Let C be a K dimen-
vector that belongs to the stabilizer code C. Then s|ψi = sional n-qudit stabilizer code with stabilizer S generated
|ψi ∀s (see the condition C1 (13) that a stabilizer code by k 6 2n generators and with corresponding parity-check
must satisfy), which together with (17) yields matrix S of size k × 2n. Then S is isomorphic through a
conjugation by a Clifford operation to another stabilizer
P |ψi = |ψi. (19) S ′ , with parity check matrix S’ in standard form

Therefore the subspace W onto which P projects includes r n−r r n−r

z}|{ z}|{ z}|{ z}|{ !
C, C ⊂ W. Let us now choose an arbitrary |φi ∈ W. r{ M 0 Z1 Z3
Then S’ = , (24)
k − r{ 0 0 Z2 Z4
1 X
|φi = P |φi = s|φi. (20) where the dimensions of the block matrices are indicated
|S|
s∈S by curly brackets.
Multiply (20) on the left by some arbitrary t ∈ S and use Here M = diag(m1 , . . . , mr ), with 1 6 r 6 n, is a diag-
the group property of S to get onal matrix with all mj 6= 0 divisors of D. The matrices
Z1 and Z2 satisfy
1 X 1 X
t|φi = tP |φi = ts|φi = s|φi Z1 M = MZT1 mod D, (25)
|S| |S|
s∈S s∈S
Z2 M = 0 mod D, (26)
= P |φi = |φi. (21)
where T in the exponent denotes the transpose, and the
Since t was arbitrary we arrived at the conclusion that matrix Z4 is a diagonal rectangular matrix, with diagonal
t|φi = |φi, ∀t ∈ S, (22) elements divisors of D. The notation 0 denotes the zero-
block matrix.
which proves that |φi belongs to the stabilizer subspace
Proof. The key ingredient of the proof is the Smith nor-
C, and this implies that W ⊂ C. Hence P projects strictly
mal form: through a sequence of elementary row/column
onto C (and not on some larger subspace that includes
operations mod D (see the discussion at the end of
C). Its trace is just the dimension K of C,
Sec. III, a matrix M over ZD can be converted to the
1 n Smith normal form [22, 23] (see also Sec. IV.B of [15] for
Tr(P ) = K = D . (23) an example)
|S|
where we have used that fact that all Pauli products in M’ = V · M · W, (27)
(17) are traceless except the identity (that must belong to
the sum, since S is a group), of trace Dn . This concludes where V and W are invertible (in the mod D sense)
the proof. square matrices, and M’ is a diagonal rectangular ma-
trix, with diagonal elements divisors of D. The matrix
When D is a composite integer the dimension of the V represents the row operations and W the column op-
stabilizer code does not have to be a power of D any erations.
more (as was the case in the prime D case), but can be Note that in our case all necessary column operations
any divisor of Dn . As an illustrative example, consider can be realized by the corresponding gates in Table III,
 5

and, more important, they do not mix the X and Z parts Also for prime D it was proven in [12] that any stabi-
of the parity-check matrix. Therefore, without being con- lizer group S is Clifford equivalent to another stabilizer
cerned with what happens to the Z part of the parity- S ′ generated only by Z’s, S ′ = hZ1 , Z2 , . . . , Zk i 2 . This
check matrix S, we can put its X part in the Smith nor- result is not true for composite D’s, and one easy to see
mal form (again we stress that this can be done because counterexample is the 1-qudit stabilizer S = hX 2 , Z 2 i in
the Z part to not interfere with the X part), and ar- D = 4, mentioned before in Sec. II D.
rive at a parity-check matrix of the form (24). Next by IV. CONCLUSIONS AND OPEN QUESTIONS
another series of Clifford gates acting only on the last
(n − r) qudits one can further put the Z4 matrix in its We studied stabilizer codes with carrier qudits of com-
Smith normal form, without modifying the X part of posite dimension D. We proved a size theorem that re-
the parity-check matrix (which is already in Smith nor- lates the size of the stabilizer group to the dimension of
mal form), since there are only zeros on the last n − r its stabilized code. Furthermore, we have shown that
columns of the X part; note that the row operations are any stabilizer code can be put in a standard (canonical)
done on the last k − r rows, and again do not modify the form through a series of Clifford gates, and we provided a
X part of the parity-check matrix. Since the elementary constructive algorithm. Our result generalizes what was
row operations do not change the stabilizer group and known in the prime D case and may be useful in con-
the elementary column operations correspond to Clifford structing efficient encoding/decoding quantum circuits,
gates, see Table III, our whole transformation from S to following the procedures outlined in [16].
S ′ is a conjugation by a Clifford operation. Our approach was based on the generalized Pauli group
Finally note that conjugation by Clifford operations introduced by (1) and (2). However, for composite di-
do not change the commutation relations. It is easy to mensions, this is not the only way of introducing Pauli
deduce from (3) that two Pauli products described by operators. An alternative way is to split the dimension in
(x|z) and (x’|z’) commute if and only if its prime-power factors which will then induce a natural
splitting of the carrier qudits in subsystems of prime-
x · z’ = z · x’ mod D, (28) power dimensions. In each of these subsystems then one
can define Pauli operators using finite fields (any finite
where the dot represents the usual inner product of two field is isomorphic to a prime-power canonical represen-
vectors in ZD , e.g. the sum of the products of individual tation), as done e.g. in [12]. Although this is the scope
components. Using (28) we observe at once that the final of future work, we think it may be useful since in a sense
set of generators commute if and only if (25) and (26) “decouples” the stabilizer into prime-power subsystems,
hold, and this concludes the proof. and the latter can be put into standard forms as done
in [12]. One can then use previously known results for
It is proved in [23] that a M × N matrix can be re- stabilizers over finite fields to study various properties of
duced to the Smith form in only O(M θ−1 N ) operations composite D stabilizers, and this may help building more
from ZD , where θ is the exponent for matrix multiplica- efficient quantum error correcting codes.
tion over the ring ZD , i.e. two M × M matrices over ZD Finally one may ask if there exist alternative standard
can be multiplied in O(M θ ) operations from ZD . Using forms of qudit stabilizer codes, perhaps more useful that
standard matrix multiplication θ = 3, but better algo- the one presented here. We do not know if such forms
rithms [24] allow for θ = 2.38. This ensures that our exist, and searching for them may be worthwhile.
procedure outlined above is computationally efficient.
Whenever D is prime its only non-zero divisor is 1,
hence M is just the identity matrix and Z4 has only 0’s ACKNOWLEDGMENTS
and 1’s on the diagonal. From (25) and (26), Z1 must
be symmetric and Z2 must be the zero matrix, hence our The research described here received support from the
standard form reduces to the one known for prime D’s Office of Naval Research and from the National Science
[16]. 1 Foundation through Grant No. PHY-0757251.

[1] P. W. Shor, Phys. Rev. A 52, R2493 (1995). [2] E. Knill and R. Laflamme, Phys. Rev. A 55, 900 (1997).
[3] A. M. Steane, Phys. Rev. Lett. 77, 793 (1996).
[4] M. A. Nielsen and I. L. Chuang, Quantum Computation
and Quantum Information, 5th ed. (Cambridge Univer-
1 Except the fact that we also perform column operations by Clif- sity Press, Cambridge, 2000).
ford conjugations, which further simplifies the standard form of [5] E. M. Rains, IEEE Trans. Inf. Theory 45, 1827 (1999).
[16]. [6] A. Ashikhmin and E. Knill, IEEE Trans. Inf. Theory 47,
2 Their result is more general and holds for any finite field, pro- 3065 (2001).
vided one redefines the Pauli operators in (1) accordingly. [7] D. Schlingemann and R. F. Werner, Phys. Rev. A 65,
 6

012308 (2001). [17] F. J. MacWilliams and N. J. A. Sloane, The Theory

[8] Dirk Schlingemann, “Stabilizer codes can be realized as of Error-Correcting Codes (North-Holland Mathematical
graph codes,” E-print arXiv:quant-ph/0111080. Library, Amsterdam, 1977).
[9] D. Schlingemann, “Logical network implementation for [18] A. Ketkar, A. Klappenecker, S. Kumar, and P. K.
cluster states and graph codes,” E-print arXiv:quant- Sarvepalli, IEEE Trans. Inf. Theory 51, 4892 (2006).
ph/0202007. [19] X. Chen, B. Zeng, and I. L. Chuang, Phys. Rev. A 78,
[10] M. Grassl, T. Beth, and M. Roetteler, Int. J. Quantum 062315 (2008).
Inf. 2, 55 (2004). [20] D. Gottesman, “Fault-Tolerant Quantum Computation
[11] V. Arvind, P. P. Kurur, and K. R. Parthasarathy, with Higher-Dimensional Systems,” E-print arXiv:quant-
“Non-stabilizer quantum codes from abelian subgroups ph/9802007.
of the error group,” E-print arXiv:quant-ph/0210097. [21] E. Hostens, J. Dehaene, and B. D. Moor, Phys. Rev. A
[12] M. Grassl, M. Roetteler, and T. Beth, “Efficient Quan- 71, 042315 (2005).
tum Circuits for Non-Qubit Quantum Error-Correcting [22] M. Newman, Integral Matrices (Academic Press, New
Codes,” E-print arXiv:quant-ph/0211014. York, 1972).
[13] S. Y. Looi, L. Yu, V. Gheorghiu, and R. B. Griffiths, [23] A. Storjohann, in ISSAC ’96: Proceedings of the 1996
Phys. Rev. A 78, 042303 (2008). international symposium on Symbolic and algebraic com-
[14] D. Hu, W. Tang, M. Zhao, Q. Chen, S. Yu, and C. H. putation (ACM, New York, NY, USA, 1996) pp. 267–274.
Oh, Phys. Rev. A 78, 012306 (2008). [24] D. Coppersmith and S. Winograd, in STOC ’87: Pro-
[15] V. Gheorghiu, S. Y. Looi, and R. B. Griffiths, Phys. Rev. ceedings of the nineteenth annual ACM symposium on
A 81, 032326 (2010). Theory of computing (ACM, New York, NY, USA, 1987)
[16] D. Gottesman, “Stabilizer codes and quantum error cor- pp. 1–6.
rection,” E-print arXiv:quant-ph/9705052.