Decoherence in one-dimensional Quantum Walk

Mostafa Annabestani∗
Dept. of Physics, Basic Sciences Faculty, Tarbiat Modarres University, Tehran, Iran

Seyed Javad Akhtarshenas†

Dept. of Physics, University of Isfahan, Isfahan, Iran
Quantum Optics Group, University of Isfahan, Isfahan, Iran

Mohamad Reza Abolhassani‡

(Dated: October 11, 2009)
In this paper we study decoherence in the quantum walk on the line. We generalize the method
arXiv:0910.1986v1 [quant-ph] 11 Oct 2009

of decoherent coin quantum walk, introduced by Brun et al [Phys. Rev. A 67, 32304 (2003)]. Our
analytical expressions are applicable for all kinds of decoherence. As an example of the coin-position
decoherence, we study the broken line quantum walk and compare our results with the numerical
one. We also show that our analytical results reduce to the Brun formalism when only the coin is
subjected to decoherence.

PACS numbers: 03.67.-a, 03.67.Mn, 03.65.Ud

I. INTRODUCTION
The quantum walk (QW) is the quantum analogue of
the classical random walk (CRW). Notable differences
between the QW and the CRW are the quadratic depen-
dency of variance on the number of steps and the complex
oscillatory probability distribution in the QW instead of
the linear variance dependency and the binomial proba-
bility distribution in the CRW. These differences between
the QW and the CRW have been used to present sev-
eral quantum algorithms in order to solve some specific
problems [1, 2, 3, 4, 5, 6] with performances better than
the best known classical versions. Recently Childs has
shown that the universal computation can be performed
by QW [7] and it is, therefore, another witness of QW
importance. Two types of QW have been introduced as
the quantum mechanical extension of CRW: discrete [8]
and continuous time [9]. Both have the same result in
our problems but finding the relation between them was
an open problem for several years. Recently this rela-
tion has been found [10, 11]. Another classification of
QW is based on the network over which the walk takes
place. The QW on a line is the simplest possible configu-
ration [8, 12, 13, 14], but other topologies such as cycles
[15, 16], two-dimensional lattices [17, 18, 19, 20, 21], or
n-dimensional hypercubes [22, 23] have also been inves-
tigated.
Quantum entanglement is another important property
of quantum mechanics which has recently attracted much
attention in view of its connection with the QW. The
effect of entanglement of the coin subspace on the QW
∗ Electronic
address: Annabestani@modares.ac.ir
† Electronicaddress: akhtarshenas@phys.ui.ac.ir
‡ Dept. of Physics, Basic Sciences Faculty, Tarbiat Modarres Uni-
versity, Tehran, Iran
[14, 19, 24], entanglement between the coin and position
subspaces [18, 25, 26] and the QW as the entanglement
generators [27, 28] are examples of these studies. Beside
of all theoretical studies, experimental implementation
and realization of the QW is another interesting subject
for researchers [29, 30, 31, 32, 33, 34].
In the experimental implementation, the environment
effects will be so important, because in practice, the
preparation of pure quantum states without interaction
with the environment is impossible, and the environment
can disturb the quantum states and fades the quantum
properties. Therefore it is very important to formulate
and quantify the influence of decoherence on the QW and
several valuable researches have been done about the de-
coherent QW [20, 31, 35, 36, 37, 38, 39, 40]. Due to
the complexity of the analytical calculations, in most of
the previous studies the numerical calculations have been
used in investigation of the effects of decoherence on the
QW. Although analytical expressions have been driven
for some particular cases such as the QW with a coin
subject to decoherence [37] or the weak noises case [41],
the general analytical formulas with the wide range of
use are not found yet. The aim of this paper is to gener-
alize the approach of Brun et al [37] and show that such
generalization is applicable for all kinds of decoherence
in the one-dimensional QW.
This work is organized as follows. Section II gives a
brief review on the one-dimensional QW and decoher-
ence. Section III is devoted to drive analytical expres-
sions for the first and second moments in the presence
of general decoherence. Our results have then been used
in section IV to analyze the coin-position decoherence for
which separation between the coin noise and the position
noise is impossible and, as an example, an analytical cal-
culation on the broken line noise has been presented. An
analysis of the coin decoherence for which only the coin is
subjected to decoherence is also presented in this section
and it is shown that our results reduce to the Brun et
 2

al formula for the coin decoherence. We summarize our Therefore the quantum walking is defined by
results and present our conclusions in section V.
|Ψ (t + 1)i = Uw |Ψ (t)i → |Ψ (t)i = Uw t |Ψ (0)i. (5)
In ideal case that one can isolate the system from the
II. BACKGROUND
environment perfectly, the time evolution of the system
takes place coherently and the state of the walker re-
From the various methods of studying the effects of mains pure after t steps of the walking. But in the real
decoherence on the quantum systems, the Kraus repre- world where the interaction between the system and the
sentation is one of the widely accepted method [42]. Let environment is unavoidable, the purity of the system is
us define HW as the Hilbert space of our system (Walker) decreased and the state of the walker becomes mixed. In
and HE as the Hilbert space of environment spanned by this case we should consider the evolution of the whole
{|en i}m
n=0 where m is the dimension of the environment’s system, i.e, system+environment, defined on the Hilbert
Hilbert space. In practice, it is not possible to completely space
isolate the system from the environment, therefore in or-
der to find the time evolution of the system we should H = HE ⊗ Hp ⊗ Hc . (6)
consider the time evolution of the whole system (sys-
tem+environment), and obtain the state of the system By definition of the Kraus operators, one step of the walk-
by tracing out over the environment’s degrees of freedom, ing can be written as follows
i.e. m
X
ρ (t + 1) = En ρ (t) En† . (7)
ρsys = TrEnv U ρ U † .


(1)
n=0
Here U acts both on the system and environment Hilbert For t steps, we can write
spaces. Without loss of generality we assume that the
m m X
m
state of the whole system is ρ = ρ0 ⊗ |env0 ihenv0 |. So we X X
can write Eq. (1) as ρ (t) = ... Ent ...En2 En1 ρ (0) En† 1 En† 2 ...En† t .
nt =0 n2 =0 n1 =0
m m
X
†
X (8)
ρsys = hen | U |env0 i ρ0 henv0 | U |en i = En ρ0 En† It is worth to note that the Eq. (8) is general and the
n=0 n=0
Kraus operators En include the whole information of all
(2)
types of evolution. It follows therefore that the coin op-
where En = hen | U |env0 i, n = 0, 1, · · · , m, are the so
erator, the translation operator and the environment ef-
called Kraus operators. These operators preserve the
m fects all are embedded in En and we didn’t assume any
En† En = I.
P
trace condition, i.e. restriction yet.
n=0
The Kraus operators satisfy an important constraint
known as completeness relation [42] which arise from the
fact that the trace of ρ (t + 1) must be equal to one, i.e.
A. Decoherent one-dimensional QW !
X
†
1 = Tr (ρ (t + 1)) = Tr En ρ(t)En (9)
QW is the quantum version of the CRW where instead
n
of the coin filliping, we use the coin operator to make su- !
perposition on the coin space, and instead of the walking X
†
= Tr En En ρ(t)
in CRW we use the translation operator to move quan-
n
tum particle according to the coin’s degrees of freedom.
In one-dimensional QW we have two degrees of free- Since this is true for all ρ then
dom in the coin space Hc , spanned by {|Li, |Ri}, and infi- X
nite degrees of freedom in the position space Hp , spanned En † En = I. (10)
by {|ii i = −∞, · · · , ∞}. The whole Hilbert space of the n
walker is defined as the tensor product of the coin space To make any progress we should therefore find the Kruse
Hc and the position space Hp , i.e. H = Hp ⊗ Hc . In operators for our system defined in Eq. (2), and use
one step of the QW we first make superposition on the the Eq. (8) in order to obtain the final state ρ(t). Ev-
coin space with the coin operator Uc and after that we idently, the En are operators that act on the system
move the particle according to the coin state with the (coin+position) Hilbert space, and therefore we can write
translation operator S as follows the general form of En as follows
Uw = S.(I ⊗ Uc ) (3) X X (n)
En = ax,x′ ,i,j |x′ i hx| ⊗ |ii hj| (11)
where x,x′ i,j
(n)
X XXX
S= |x + 1i hx| ⊗ |Ri hR| + |x − 1i hx| ⊗ |Li hL|. (4) = ax,l,i,j |x + li hx| ⊗ |ii hj|
x x l i,j
 3

where x, l = −∞, · · · , ∞ and i, j = {L, R}. III. FORMALISM

In the following section we show that a reasonable sug- The Fourier transformation is a powerful tool for an-
gestion on the coefficient a(n) of Eq. (11) enables us to alytical investigation of the one-dimensional QW, where
derive useful analytical expression for the first and second is defined as follows
moments of position.
Zπ
dk −ikx
|xi = e |ki. (16)
2π
−π
B. Environment effects (decoherence)
Using this transformation we can write Eq. (11) in k
In this section we briefly discuss about the interpreta- space as
tion of noisy evolution and the Kraus representation. As-
dkdk′ −ilk −ix(k−k′ )
ZZ
(n)
E˜n =
XX
|ki k′ ˛ ⊗ |ii hj|.
˙ ˛
sume that the pi is the probability that the ith unknown ax,l,i,j 2
e e
4π
reason affects the state of the system. In the other word, x,l i,j

the evolution operators Ai act on the system with the (17)

(n)
corresponding probabilities pi . In the following we assume that the coefficients are ax,l,i,j
Now we are interested in the walker, though the uni- not dependent on the coordinate x. This means that the
tary evolution takes place in the Hilbert space of the probability of translation in the position space depends
system+environment, given in Eq. (6). Furthermore the only on the length l of the translation, not on the posi-
environment could be the rest of the universe, so deter- tion x where the translation takes place. In view of this
mining the exact reasons needs the investigation of the constraint on the a(n) , we are able to derive an analytical
whole universe which is impossible. expression for the first and second moments of the posi-
tion, which is applicable for a large family of decoherence
Fortunately, in practice, understanding of the exact in the one-dimensional QW. In this regime, the Eq. (17)
form of the environment is not required. Let us assume takes the following form
that we have r different operators Ai , i = 1, · · · , r where
dkdk′ −ilk `
ZZ
each of them acts on the system with the probability pi . E˜n =
XX (n)
al,i,j e δ k − k′ |ki k′ ˛ ⊗ |ii hj|
´ ˙ ˛
If the environment be in the state |ei i then the operator i,j
2π 2
l
Ai acts on the system. Therefore we can imagine the r- (18)
dimensional Hilbert space for the environment, spanned where we have used the orthonormalization relation
by {|ei i}ri=1 , and the following initial state for the envi-
e−ix(k−k ) = 2πδ (k − k ′ ) .
X ′
ronment (19)
x
√ √ √
|env0 i = p1 |e1 i + p2 |e2 i + ... + pr |er i . (12) By integration on k ′ and changing the order of integration
and summation we get
It is clear that the probability of finding the environ-
ment in the state |ei i is pi . Therefore we can write dk
Z
˜
En = |ki hk| ⊗ Cn (k) (20)
the unitary transformation of the whole system (envi- 2π
ronment+system) as follows
where
U = |e1 i he1 |⊗A1 +|e2 i he2 |⊗A2 +...+|er i her |⊗Ar . (13) XX (n)
Cn (k) = al,i,j e−ilk |ii hj|. (21)
l i,j
So the Kraus operators will be
√ Now by writing the general form of ρ0 in the k space as
Ei = hei | U |env0 i = pi Ai . (14)
dkdk ′
ZZ
ρ0 = |ki hk ′ | ⊗ |ψ0 i hψ0 | (22)
Accordingly, the Eq. (7) gives the density matrix after 4π 2
the first step as
the first step of walking from Eq. (7) becomes
r r
dkdk ′
ZZ
Ei ρEi† = pi Ai ρA†i .
X X
ρ′ =
X
(15) ρ′ = |ki hk ′ | ⊗ Cn (k) |ψ0 i hψ0 | Cn† (k ′ )
4π 2
i=1 i=1 n
dkdk ′
ZZ
As we see, this is exactly our expectation from the = |ki hk ′ | ⊗ Lk,k′ |ψ0 i hψ0 | (23)
influence of noise because it gives ρ′ as a mixture of 4π 2
different evolutions with the corresponding probabilities where Lk,k′ is a superoperator defined by
pi . With Eqs. (12) and (13) we can represent the effects
of noise by the Kraus operators.
X
Lk,k′ Õ = Cn (k) ÕCn† (k ′ ) . (24)
n
 4

Therefore after t steps, the state of the walker and the Using the orthonormalization relation (19), we get for
probability of finding the walker in position x are, respec- the moments hxi and hx2 i
tively

dkdk ′ −i dδ (k ′ − k)
ZZ ZZ
|ki hk ′ | ⊗ Ltk,k′ |ψ0 i hψ0 | dkdk ′ Tr Ltkk′ ρ0


ρ (t) = (25) hxi = (32)
4π 2 2π dk
1 d2 δ (k ′ − k)
ZZ

2
and dkdk ′ Tr Ltkk′ ρ0 .


x = ′
2π dk dk
dkdk ′ In order to carry out these integrations we need the fol-
ZZ
hx|kihk ′ |xiTr Ltkk′ ρ0


p (x, t) = 2
(26) lowing equations
4π
dkdk ′ −ix(k′ −k)
ZZ
Tr Ltkk′ ρ0 .


= 2
e
4π
!
d   X dCn (k)
Tr Lkk′ Õ = Tr ÕCn† (k ′ ) (33)
Note that the completeness relation given in Eq. (10) dk n
dk
implies that !
d   X dCn† (k ′ )
Tr Lkk′ Õ = Tr Cn (k) Õ
dk ′ dk ′
X
Cn† (k)Cn (k) = I. (27) n
n

It follows from this condition on the coin operator that where according to Eq. (21)
the superoperators Lk,k are trace preserving, i.e.
!
dCn (k) X X (n)
lal,i,j e−ilk |ii hj|
 
= −i (34)
X
†
Tr Lk,k Õ = Tr Cn (k) ÕCn (k) (28) dk
n i,j l

dCn† (k ′ )
! X X  (n) ∗ ′
l al,i,j eilk |ji hi|.
 
= i
X
= Tr Cn† (k) Cn (k) Õ = Tr Õ dk ′
n i,j
l

and therefore
    Since the superoperator Lk,k′ acts on the density matrix
Tr Lm
k,k Õ = Tr Õ (29) which is positive and Hermitian, we can write the Eq.
(33) as follows
which are true for arbitrary operator Õ.
Evidently if we know the operators Cn (k), the state d
and the probability distribution of the system can be ob- Tr (Lkk′ ρ) = Tr (Gkk′ ρ) (35)
dk
tained by direct using of Eqs. (25) and (26). But unfortu-
d
Tr (Lkk′ ρ) = Tr G † k′ k ρ


nately it is these equations which are difficult to handle,
and therefore we turn our attention on the moments of dk ′
this distribution. The mth moment of the probability
distribution p (x, t) is defined by where
X
hxm i = xm p (x, t). (30) X dCn (k)
x Gk,k′ Õ = ÕCn† (k ′ ). (36)
n
dk
Inserting Eq. (26) in Eq. (30) we get

1 X m
ZZ
dkdk ′ e−ix(k −k) Tr Ltkk′ ρ0 .
′
m Finally by carrying out the integrations of Eq. (32) and


hx i = x
4π 2 x putting everything together, we arrive at the following
(31) relations for the first and second moments
 5

π t
dk X 
Z
Tr Gk Lm−1


hxit = i k |ψ0 ihψ0 | (37)
−π 2π m=1
π t m−1
dk X X
Z n    o
′ ′ ′ ′
Tr Gk† Lkm−m −1 Gk Lkm −1 |ψ0 ihψ0 | + Gk Lm−m −1
Gk† Lkm −1 |ψ0 ihψ0 |

2
x t = k
−π 2π m=1 ′
m =1
π t
dk X
Z
Tr Jk Lm−1


+ k |ψ0 ihψ0 |
−π 2π m=1

where we have defined

n-1 n n+1
†
 
dGk,k 
′  X dCn (k) dC † (k ′ ) 
n
Jk = = Õ (38)
dk  ′ dk dk ′ 
 
n
k =k k′ =k (a)
and have used Lk ≡ Lk,k and Gk ≡ Gk,k for the sake of n-1 n n+1
simplicity. It is worth to note that thus obtained mo-
ments in Eq. (37) are a generalization of the moments
given by Brun et.al in [37], in the sense that they are (b)
applicable for all kinds of decoherence. n-1 n n+1

IV. CALCULATION AND RESULT (c)

In this section our task is to show that the method is
applicable for all possible kinds of decoherence. To this
aim, in the first subsection we consider the broken line
decoherence as an example of the coin-position decoher-
ence for which we cannot separate the coin noise from
the position noise. In the second subsection, we restrict
ourselves to the case that the coin is subjected to deco-
herence but the position is free of noise, and show that
our method reduces to the result of Brun et al given in
[37].
A. coin-position decohrence
FIG. 1: (Figure from [40]) Possible situations for site n of the
line when there are: (a) no broken links, (b) the link to the
left of the site is broken, and (c) both links are broken. The
arrows indicate the direction of the probability flux associated
to the L,R components.
Let us rename S in Eq. (4) to S1 as the translation
operator for the case that we don’t have any broken line.
We also define the translation operators S2 , S3 , S4 for
the cases that the left, right and both lines are broken,
respectively, i.e.

|x + 1i hx| ⊗ |Ri hR| + eiθ1 |x − 1i hx| ⊗ |Li hL|

P
S1 =
In this subsection we investigate the effects of coin- x
|x + 1i hx| ⊗ |Ri hR| + eiθ2 |xi hx| ⊗ |Ri hL|
P
position decoherence on the QW. As an example of our S2 =
x
approach we find the diffusion coefficient of the one-
|xi hx| ⊗ |Li hR| + eiθ3 |x − 1i hx| ⊗ |Li hL|
P
S3 =
dimensional QW in the presence of broken line noise. x

This model has been studied by Romanelli et al [40], |xi hx| ⊗ |Ri hL| + eiθ4 |Li hR| .
P
S4 =
where the authors have calculated the diffusion coeffi- x
cient numerically. They have introduced broken line de- (39)
coherence for the one-dimensional QW in such a way The situations corresponding to the above translation op-
that the connection between neighbors of current posi- erators are demonstrated in Fig.1. Note that we have
tion breaks with probability p, i.e. the walker doesn’t used the general phase for each translation operator.
walk with probability p. It turns out that with proba- Later we will see that we need these generic phases in
bility (1 − p)2 both links are connected and the walker order to have the correct physical phenomena.
proceeds normally. On the other hand with probability As mentioned before, the environment state deter-
p(1 − p) (res. p2 ) one line (res. both lines) is broken and mines the broken line situation. For instance if the envi-
the walker is returned back (res. is stopped) (see Fig.1). ronment is in the state |e1 i, then S1 acts on the system
and so on. Therefore, according to the Eq. (12) we can
 6

define the initial state of the environment as follows the affine map approach. Since the Lk is linear we can
p represent it as a matrix acting on the space of two-by-two
|env0 i = (1 − p) |e1 i + p (1 − p) (|e2 i + |e3 i) + p |e4 i . operators. To do this we note that one can represent any
(40) two-by-two matrix by a four-dimensional column vector
Also from Eqs. (3) and (13), the general unitary trans- as
formation of the system-environment is as follows
r0
 
U = |e1 i he1 | ⊗ S1 (I ⊗ H) + |e2 i he2 | ⊗ S2 (I ⊗ H) (41)  r1 
Õ = r0 I + r1 σ1 + r2 σ2 + r3 σ3 ≡   (49)
+ |e3 i he3 | ⊗ S3 (I ⊗ H) + |e4 i he4 | ⊗ S4 (I ⊗ H) r2
r3
where we have assumed the Hadamard walk, i.e. Uc = H.
Therefore according to Eq. (13) Ai = Si (I ⊗ H), and we where
find from Eq. (14) the Kraus operators as 1  
ri = Tr σi Õ . (50)
X 2
E1 = (1 − p) |x + 1i hx| ⊗ |Ri hR| H
x Here we have defined σ0 = I, and σi (i = 1, 2, 3) are the
usual Pauli matrices.
+eiθ1 |x − 1i hx| ⊗ |Li hL| H (42)
p X Now, in order to find the superoperators Lk , Gk , Gk† and
E2 = p (1 − p) |x + 1i hx| ⊗ |Ri hR| H Jk , we represent the action of them on an arbitrary two-
x by-two matrix Õ as follows
+eiθ2 |xi hx| ⊗ |Ri hL| H (43)
1 0 0 0 r0
  
p X
E3 = p (1 − p) |xi hx| ⊗ |Li hR| H 2
0 0 e f + p   r1 
Lk Õ ≡  (51)
x 0 0 −f + p2 e   r2 
+eiθ3 |x − 1i hx| ⊗ |Li hL| H (44) 0 1 − 2p −2g −2h r3
X
|xi hx| ⊗ |Ri hL| H + eiθ4 |Li hR| H .(45)


E4 = p
0 i (p − 1) ig ih r0
x
  
0 0 f −e   r1 
All of these operators are in the form given by Eq. (11). Gk Õ ≡  (52)

0 0 e f  r2 
For instance, the coefficients a(1) for E1 are
i (p − 1) 0 −h g r3
(1) (1−p) (1)
a1,R,R = √
2
a−1,L,L = −eiθ1 (1−p)
√
2
(1) (1−p) (1) (46) 1−p 0 0 0 r0
a−1,L,R = eiθ1 (1−p)
  
a1,R,L = √
2
√
2
 0 0 −e −f   r1 
Jk Õ ≡  (53)
where can be used in Eq. (21) and obtain C1 as 0 0 f −e   r2 
0 1−p 0 0 r3
 −ik
e−ik

(1 − p) e
C1 = √ . (47) and
2 ei(k+θ1 ) −ei(k+θ1 )
Gk† Õ = Gk∗ Õ. (54)
Similarly, we can find the other Ci as follows
r where the last is obtained, simply, from the Hermiticity
(1 − p) e−ik + eiθ2 e−ik − eiθ2
 
C2 = (48) of the Pauli matrices. Here e, f, g and h are functions of
2 0 0 p and k, defined by
r  
(1 − p) 0 0
C3 = e(p, k) = (p − 1)2 sin (2k)
2 1 + ei(k+θ3 ) 1 − ei(k+θ3 ) 2
  f (p, k) = (p − 1) cos (2k)
p 1 −1
C4 = √ iθ iθ . g(p, k) = p (1 − p) sin (k)
2 e 4 e 4
h(p, k) = p (1 − p) cos (k) . (55)
These Ci must satisfy the completeness relation given
by Eq. (27). After some calculations we find that for With this representation wecan calculate
 the moments
any θ1 , θ4 but only when θ2 − θ3 = π, the Eq. (27) given by Eq. (37). Since Tr Lk Õ = 2r0 and the Lk is
is satisfied. In the following, for simplicity, we choose trace preserving, so it doesn’t change r0 . Therefore the
θ1 = θ3 = θ4 = 0 and θ2 = π. only nontrivial result arise from the following three-by-
According to Eq. (24), with these Ci in hands, we three submatrix
can find the superoperator Lk,k′ . But as we know, we
f + p2
 
have to calculate the mth power of this superoperator 0 e
which is not an easy task to handle. To get around this Mk =  0 −f + p2 e . (56)
complexity, we follow the method of Ref. [37] and use 1 − 2p −2g −2h
 7

For finding hxi in Eq. (37) we should take trace, accord- where σ 2 = hx2 i − hxi2 . Clearly the time independent
ingly, only the first row of Gk is important term doesn’t contribute to D, therefore we focus on find-
#  ing the second moment in Eq. (37). To begin with, let
r1 us first calculate the last term of hx2 it , i.e.
Z π " t
−1 X
Mkm−1  r2 .


hxit = dk (p − 1) g h
π −π m=1 r 3
(57) π t
dk X 
Z
Tr Jk Lm−1


We can use the geometric progression to simplify the k |ψ0 ihψ0 | . (60)
summation in this equation. −π 2π m=1
 
−1
Z π h i r1

−1 
hxit ≈ dk (p − 1) g h (I − Mk ) r2 . Since only the first row of Jk (Eq. (53)) makes nonzero
π −π r 3 result in the trace, therefore
(58)
Note that we omit the Mkt in this equation because all
eigenvalues of Mk obey 0 < |λ| < 1 and Mkt → 0 in long 1/2

π t
time limit. All t dependence has vanished, therefore in dk
Z X ′  r 
1 − p 0 0 0 Lkm −1  1  = (1 − p) t


long time limit the first moment is time independent. r2
−π π m=1
The matrix (I − Mk ) is exactly invertible and calcula- r3
tion of hxit is straightforward but our interest is finding (61)
the diffusion coefficient with below definition where we put r0 = 1/2 because of the normalization con-
1 ∂σ 2 dition of |ψ0 i. Now we turn our attention to the first
D= lim (59) term of hx2 it in Eq. (37 ), i.e.
2 t→∞ ∂t

πt m−1
dk X X
Z n    o
Tr Gk† Lkm−m −1 Gk Lkm −1 |ψ0 ihψ0 | + Gk Lm−m Gk† Lkm −1 |ψ0 ihψ0 |
′ ′ ′ ′
−1
= k
−π 2π m=1 m′ =1

1/2
 
Z π t m−1
dk ′
  ′  r 
Gk − Gk† Lkm −1  1 
X X
−1
0 (p − 1) g h Lm−m


= −i (62)
−π π k r2
m=1 m′ =1
r3

where in the second line we use the facts that only the Therefore we can easily write
first row of Gk and Gk† makes nonzero result in the trace
and that the first row of Gk is pure imaginary.   
f k, p, r~′

1/2
 
Since the Lk is trace preserving and leaves r0 un-  r 
  ′
Gk − Gk† Lkm −1  1  = i  0
 
 . (65)
changed, we write r2  0 
r3 p−1
1/2 1/2
   
′  r   r  ′ Now putting everything together, we get for the second
Lkm −1  1  =  1′  . (63) moment hx2 it
r2 r2
r3 r3′
π
dk
Z
hx2 it = (1 − p) t + Γ (k, p, t) (66)
−π π
Also from Eqs. (52) and (54) we find the exact form of
Gk − Gk† as
where (we drop out arguments for simplicity)

0 (p − 1) g h r0 0
    
t m−1
0 0 0 0   r1   0 
 
Gk − Gk†
X X ′
0 (p − 1) g h Lkm−m −1 


Õ = 2i  . Γ= .

0 0 0   r2  0 
m=1 m′ =1
(p − 1) 0 0 0 r3 p−1
(64) (67)
 8
 
Note that the f k, p, r~′ doesn’t appear in Γ, since
 
f k, p, r~′
 
′
0 (p − 1) g h Lkm−m −1  0

 
 = 0. (68)
 0 
0

Same as before we can use the matrix Mk for calculat-

ing Eq. (67)
# 
0
" t m−1

X X ′
Γ = (p − 1) g h Mkm−m −1  0 
m=1 m′ =1 p−1
(69)
where
t m−1 t FIG. 2: The K(p) versus p
−1
X X ′ X
Mkm−m −1 = (I − Mk ) (I − Mkm ) (70)
m=1 m′ =1 m=1
n o The expression of D (p) in Eq. (73) is exactly same as the
= (I − Mk )−1 t − (I − Mk )−1 M . numerical result of Ref. [40] in which the authors have
proposed a constant value for K and have estimated K ≈
By inserting this into Eq. (69), the calculation of hx2 it 0.4 by linear regression. But our analytical calculations
will be straightforward. show that the coefficient K is a function of p (Eq. (74))
Our interest is finding D here, so only the time de- and ranges from 0.19 to 0.5 as p goes from 0 to 1 (see Fig.
pendent terms will be important for us. By inserting 2 ). The numeric prediction of Romanelli et al implies
the time dependent term of Eq. (70) into Eq. (66) and that for p=0.44 the diffusion coefficient is exactly the
definition of D given in Eq. (59), we have same as the unbiased classical random walk, i.e. 1/2, but
our analytical calculations indicate that this probability
1 ∂σ 2 is less than the numeric prediction and is about p=0.417.
D = lim
2 t→∞ ∂t This means that for p < 0.417 the diffusion coefficient is
 Z π 
1 dk ∂Γ greater than the classical one and the quantum walker
= (1 − p) + . (71)
2 −π π ∂t spreads more faster than the classical one. The origin of
this difference is behind of the fact that the numerical
−1
The matrix (I − Mk ) is exactly invertible but it’s ele- method can never be applied in problems with infinite
ments are a little long to appear here. Fortunately this steps.
matrix appear in the form of
 
∂Γ 0 B. coin decoherence

−1
= (p − 1) g h (I − Mk )  0  (72)
∂t p−1
In this subsection we show that in the coin decoher-
i.e., it is sandwiched between two vectors. It is not diffi- ence, the general formalism of Eq. (37) reduces to the
cult to calculate Eq. (71) to have formalism of Brun et al presented in [37].
Let us suppose that before each step of walking, the
(n)
(1 − p)2
 
1 operators Dc acts on the coin subspace with probability
D = (1 − p) + [1 − I (1 − p)]
2 p pn . Then according to Eq. (13) the An are
1−p
= K (p) (73)
   
p An = S. (I ⊗ H) . I ⊗ Dc(n) = S. I ⊗ HDc(n) . (76)

where By plugging it into Eq. (14) and using the explicit form
1 of S we get
K (p) = {1 − (1 − p) I(x) (1 − p)} (74)
2 √ X
En = pn |x + 1i hx| ⊗ |Ri hR| HDc(n)
and x
π
dk cos (k) + x + |x − 1i hx| ⊗ |Li hL| HDc(n) . (77)
Z
I (x) = 2 .
−π 2π cos (k) x + cos (k) x + 2x2 − 2x + 1
(n)
(75) Now by writing the operator HDc in the basis
 9
 
{|Ri , |Li} †
Clearly Gk,k ′ Õ = i Lk,k′ Õ Z, and the Jk in Eq. (38)
X will be
HDc(n) = (n)
γr,s |ri hs| (78)
r,s={R,L}

the Eq. (77) takes the following form †


dGk,k ′ Õ 
  
√ X X (n) Jk Õ = = Z Lk Õ Z. (84)
dk 

En = pn γR,s |x + 1i hx| ⊗ |Ri hs|
k′ =k
x s={R,L}
(n)
+ γL,s |x − 1i hx| ⊗ |Li hs| .(79)

This equation is exactly in the form of Eq. (11) with the Inserting Eq. (84) into the last term of Eq. (37) we get
specified values for l, i.e. l = 1, −1, hence
(n) (n) (n) (n)
a1,R,s = γR,s , a−1,L,s = γL,s . (80)
π t
dk X
Z
Therefore according to Eq. (21) we have Tr Jk Lm−1

|ψ0 ihψ0 |


k
−π 2π m=1
(n) (n)
X
Cn (k) = γR,j e−ik |Ri hj| + γL,j eik |Li hj| (81) π t
dk X
Z
Tr Z Lk Lm−1
 

j={L,R} = k |ψ0 ihψ0 | Z
−π 2π m=1
and π t
dk X
Z
dCn (k) = Tr {Lm
k |ψ0 ihψ0 |} = t. (85)
(n) (n) 2π m=1
X
= −i γR,j e−ik |Ri hj| − γL,j eik |Li hj| −π
dk
j={L,R}

= −iZCn (k) . (82)

And finally according to the definition of Lk,k′ in Eq. Finally, putting Eqs. (83) and (85) in the moment expres-
(36) sions of Eq. (37), reduces them to the same expression as
  introduced by Brun et al [37] for only coin decoherence,
Gk,k′ Õ = −iZ Lk,k′ Õ . (83) i.e.

t π
dk
X Z
hxit = Tr {ZLm
k |ψ0 ihψ0 |} (86)
−π m=1 2π
Z π t m−1

2 dk X X n ′
 ′
o n ′
 ′  o
x t = t+ Tr ZLm−mk ZLm
k |ψ0 ihψ0 | + Tr ZLm−m
k Lm
k |ψ0 ihψ0 | Z .
−π 2π m=1 ′ m =1

Therefor the first and second moments given in Eq. (37)
are general in the sense that thay can be applied for
any types of decoherence, including coin-position deco-
herence, coin decoherence and position decoherence.
V. SUMMARY AND CONCLUSION
We have present an analytical method for calculat-
ing the first and second moments of the one-dimensional
QW, which are applicable for all kinds of decoherence.
Our method is a generalization of the Brun et al method,
introduced in [37], where there the authors have pre-
sented the analytical expressions for the first and second
moments in the presence of coin decoherence and have
shown that the transition from quantum to classic hap-
pens even for the weak coin noise.
We have made exact calculations on the broken line
decoherence as an example of the non-separable decoher-
ence of the coin-position system. Our analytical calcula-
tions are in consistent with the previous numerical result
[40]. We have shown that the K in the diffusion coeffi-
cient D is, indeed, a function of p, and ranges from 0.19
to 0.5 as p varies from 0 to 1.
Furthermore our calculation shows that for p = 0.417
the value of D is 1/2, exactly the same as the unbiased

classical random walk. This specific value of p is critical
in the sense that the transition from quantum to classic
happens at this point. In other word, the quantum walk
with non zero broken line probability, spreads faster than
the classical random walk if the probability of broken
line is not higher than 0.417. For higher values of p, the
lines are broken too frequently and this will prevent the
full diffusion. The critical p, estimated by the numerical
calculation of Ref. [40], is a little higher than the one
obtained here analytically. The difference arises from the
non-ability of numerical method in simulation of the QW
with infinite steps.
We have also studied the case that only the coin is
subjected to decoherence and have shown that the ex-
pressions for the first and second moments lead to the
Brun et al results presented in [37].
The long time behavior of moments and, of course, the
variance can be used as a qualitative measure of classi-
cality. Therefore these calculated moments provides a
powerful tool for investigating the one-dimensional QW
in the presence of any kinds of decoherence. For instance,
finding the particular cases for which dependency of the
variance on the time remains quadratic, i.e. the decoher-
ence does not affect the quantumness, is an application
of these moments. Brun et al [37] have shown that the
[1] N. Shenvi, J. Kempe, and B. Whaley, Phys. Rev. A 67,
52307 (2003), eprint quant-ph/0210064.
[2] A. Ambainis, SIAM Journal on Computing 37, 210
(2007), preprint quant-ph/0311001.
[3] A. Ambainis, J. Kempe, and A. Rivosh, Proc. 16th ACM-
SIAM SODA , p. 1099 (2005), quant-ph/0402107.
[4] A. Ambainis, SIGACT News 35, 22 (2004), preprint
quant-ph/0504012.
[5] A. M. Childs and J. Goldstone, Phys. Rev. A 70, 022314
(2004), preprint quant-ph/0306054.
[6] A. Tulsi, Phys. Rev. A 78, 012310 (2008), quant-
ph/0806.1257.
[7] A. M. Childs, Phys. Rev. Lett. 102, 180501 (2009).
[8] A. Nayak and A. Vishwanath, preprint quant-
ph/0010117.
[9] E. Farhi and S. Gutmann, Phys. Rev. A 58, 915 (1998).
[10] A. M. Childs, preprint quant-ph/0810.0312.
[11] F. Strauch, Phys. Rev. A 74, 030301 (R) (2006), preprint
quant-ph/0606050.
[12] A. Romanelli et al., Phys. A 338, 395 (2004), e-print
quant-ph/0310171.
[13] N. Konno, J. Quant. Inf. Proc. 1, 345 (2002), quant-
ph/0206053.
[14] S. Venegas-Andraca, J. Ball, K. Burnett, and S. Bose,
New J. Phys. 7, 221 (2005), preprint quant-ph/0411151.
[15] M. Bednarska, A. Grudka, P. Kurzynski, T. Luczak, and
A. Wojcik, Physics Letters A 317, 21 (2003), quant-
ph/0304113.
[16] W. Adamczak et al., International Journal of Quantum
Information 5, 781 (2007), quant-ph/0708.2096.
[17] T. T.D. Mackay, S. Bartlett, L. Stephenson, and
B. Sanders, Journal of Physics A 35, 2745 (2002),
10
quadratic time dependency of the variance is preserved
for multiple-coin QW subjected to the coin decoherence,
but their calculation is restricted to the coin decoherence
only. In this paper this restriction is relaxed and the
model allows the presence of any kind of decoherence.
These formulas are useful also in the experimental im-
plementation of the quantum walk. The application of
our results for two kinds of decoherence, i.e. the coin-
only and the coin-position noises, have been done in this
paper. The application on the position decoherence have
been also tested but the full calculations were so long to
appear here. The results of our calculations on the tun-
nelling effect, for which the position is subjected to deco-
herence, and it’s consistency with the previous numerical
work [31] confirm the usefulness of these expressions for
the position decoherence too. Other aspects of the tun-
nelling effects are also under consideration. The paper,
therefore, can be regarded as a further development in
the study of decoherent one-dimensional QW.
Acknowledgments
M.A. thanks G. Abal for valuable discussions.
preprint quant-ph/0108004.
[18] I. Carneiro et al., New J. Phys. 7, 156 (2005), preprint
quant-ph/0504042.
[19] Y. Omar, N. Paunkovic, L. Sheridan, and S. Bose, Phys.
Rev. A 74, 42304 (2006), preprint quant-ph/0411065.
[20] A. Oliveira, R. Portugal, and R. Donangelo, Phys. Rev.
A 74, 12312 (2006).
[21] K. Watabe, N. Kobayashi, M. Katori, and N. Konno,
PRA 77, 062331 (2008), quant-ph/0802.2749.
[22] C. Moore and A. Russell, Quantum walks on the hyper-
cube, in Proceedings of 6th International Workshop on
Randomization and Approximation Techniques (RAN-
DOM 2002), Lecture Notes in Computer Science 2483
(LNCS), edited by J. D. P. Rolim and S. Vadhan, pp.
164–178, Cambridge, MA, 2002, Springer-Verlag, Berlin,
2002, preprint quant-ph/0104137v1.
[23] F. Marquezino, R. Portugal, G. Abal, and R. Donangelo,
PRA 77, 042312 (2008), quant-ph/0712.0625.
[24] C. Liu and N. Petulante, Phys. Rev. A 79, 032312 (2009),
quant-ph/0807.2263.
[25] G. Abal, R. Siri, A. Romanelli, and R. Donangelo, Phys.
Rev. A 73, 042302, 069905(E) (2006), preprint quant-
ph/0507264.
[26] M. Annabestani, M. R. Abolhasani, and G. Abal, (2009),
quant-ph/0901.1188.
[27] S. K. Goyal and C. M. Chandrashekar, quant-
ph/0901.0671.
[28] S. E. Venegas-Andraca and S. Bose, 2009 , quant-
ph/0901.3946.
[29] B. Travaglione and G. Milburn, Phys. Rev. A 65, 032310
(2002).
[30] P. Knight, E. Roldán, and J. Sipe, Op. Comm. 227,

147 (2003).
[31] W. Dür, V. Kendon, and H. Briegel, Phys. Rev. A 66,
52319 (2002), arXiv preprint quant-ph/0207137.
[32] J. Du et al., Phys. Rev. A 67, 042316 (2003), preprint
quant-ph/0203120.
[33] C. Ryan, M. Laforest, J. Boileau, and R. Laflamme,
Phys. Rev. A 72, 062317 (2005), preprint quant-
ph/0507267.
[34] Z. Zhao, J. Du, H. Yang, Z. Chen, and J. Pan, (2002),
arXiv preprint quant-ph/0212149.
[35] J. Kempe, Contemp. Phys. 44, 307 (2003), preprint
quant-ph/0303081.
[36] V. Kendon and B. Tregenna, Phys. Rev. A 67, 42315
(2003), eprint quant-ph/0209005.
[37] T. Brun, H. Carteret, and A. Ambainis, Phys. Rev. A
67, 32304 (2003), arXiv preprint quant-ph/0210180.
11
[38] T. Brun, H. Carteret, and A. Ambainis, Phys. Rev. Lett.
91, 130602 (2003).
[39] C. C. López and J. P. Paz, Phys. Rev. A 68, 052305
(2003).
[40] A. Romanelli, R. Siri, G. Abal, A. Auyuanet, and R. Do-
nangelo, Phys. A 347, 137 (2004), preprint quant-
ph/0403192.
[41] V. Kendon and B. Tregenna, Decoherence in discrete
quantum walks, in Decoherence and Entropy in Complex
Systems, , Lecture Notes in Physics Vol. 633/2003, pp.
253–267, Springer Berlin / Heidelberg, 2003.
[42] M. Nielsen and I. Chuang, Quantum Computation
and Quantum Information (Cambridge University Press,
Cambridge, 2000).