a

r
X
i
v
:
1
4
0
9
.
3
8
8
4
v
1


[
q
u
a
n
t
-
p
h
]


1
2

S
e
p

2
0
1
4
The Global versus Local Hamiltonian Description
of Quantum Input-Output Theory
John Gough
1,
1
Aberystwyth University, Aberystwyth, SY23 3BZ, Wales, United Kingdom
(Dated: September 16, 2014)
The aim of this paper is to derive the global Hamiltonian form for a quantum system and bath,
or more generally a quantum network with multiple quantum input eld connections, based on the
local descriptions. We give a new simple argument which shows that the global Hamiltonian for a
single Markov component arises as the singular perturbation of the free translation operator. We
show that the Fermi analogue takes an equivalent form provided the parity of the coecients is
correctly specied. This allows us to immediately extend the theory of quantum feedback networks
to Fermi systems.
I. INTRODUCTION
The quantum stochastic calculus was introduced by
Hudson and Parthasarathy [1] in 1984 as a frame-
work to construct explicit dilations of quantum dynam-
ical evolutions (semigroups of completely positive norm-
continuous identity preserving maps) generalizing the
usual It o theory. Here the system Hilbert space is ten-
sored with a Fock space over L
2
C
n(R) where n enumerates
the number of input noise channels. It was quickly re-
alized [3] that the quantum evolutions could in fact be
interpreted as a singular perturbation of the free dynam-
ics corresponding to the second quantization of the shift
along L
2
C
n(R). In particular, one could take the line R
to physically be an innite transmission line along which
Bose quanta are propagating at constant speed, and in-
teracting with the system (located at the origin) instan-
taneously as the pass through. It was a long standing
program to nd the form of the associated Hamiltonian,
which when viewed in the interaction picture with respect
to the free shift dynamics, gave the general quantum open
dynamics, as well as a large class of classical stochastic
models. This was rst done by Chebotarev [4] for the
case of commuting coupling coecients, and by Grego-
ratti [5] for the general bounded operator case. The re-
quirement of boundedness was later dropped [6]. The do-
main of the Hamiltonian is determined through a bound-
ary condition at the origin on the vectors. The actual
expression for the associated Hamiltonian is then asym-
metric in the formal creation and annihilation operator
densities, which makes the interpretation non-obvious for
physicists. However, this has been remedied by an alter-
native approach pioneered by von Waldenfels [7] which
exploits a quantum white noise formulation, the corre-
sponding kernel calculus (that is, a Bose version of the
Berezin calculus due to Maassen), and an explicit con-
struction of deciency spaces needed to then construct
the self-adjoint Hamiltonian. This latter approach con-
forms most naturally to the specication of the associated

jug@aber.ac.uk
Hamiltonian as understandable to a theoretical physics
readership.
We shall refer to this situation of a single system inter-
acting in a singular (=Markov) fashion with a quantum
eld moving along a transmission line as a local Hamilto-
nian model. More generally we can consider several sys-
tems at various points on the transmission line, or more
generally a network of transmission lines with systems
at the vertices. The theory of quantum feedback net-
works has been developed recently to study such quan-
tum mechanical systems connected by various arrange-
ments of quantum eld inputs [8]. Here the notion of a
global Hamiltonian was introduced in [8] in order to con-
struct tractable models of a quantum feedback network.
For a single component, this reduces to the Hamiltonian
obtained by Chebotarev [4] and Gregoratti [5], and we
present a simple derivation of this object in section II
below. The network theory has been applied to quantum
optics with a view to developing closed-loop quantum
control systems [9]- [17].
The separate components are modeled as quantum
open systems [1], see also [2], or equivalently as quantum
input-output systems [18]. In a network, connections are
made by feeding the output of one system in as input
to another, or even to the same component again. In
a physically realistic model, this will involve time-lags,
however it often useful to take the limit of instantaneous
connections. Moreover the components themselves are
assumed to be capable of scattering inputs, in particular
acting as beam-splitters.
The has been interest amongst theoreticians in recent
years in using solid-state devices as in place of optical
elds [19] - [25]. In suitable regimes the models resem-
ble the optical case with the obvious exception that the
elds are now Fermionic rather than Bosonic. In sec-
tion III, we give the Fermi analogue and show that the
global Hamiltonian method applies equally well to this
situation. Indeed, provided the various coupling terms
meet the physical conditions regarding Fermionic parity,
the essential rules governing the construction of networks
(the series product [9], the concatenation rule and feed-
back reduction rule [8]) are identical to the Bose input
case.
2
The outline of this paper is that we rst review the
background theory of open quantum systems adopting
a quantum white noise convention. We make some ten-
tative links with the theory of singular perturbations of
unbounded below Hamiltonians (which is of relevance to
the free Hamiltonian generating the shift). In section 2
we give an alternative rationale behind the form of the
associated local Hamiltonian, and outline how this may
be generalized to the case of a global Hamiltonian for a
quantum feedback network. In section 3, we treat the
extension to Fermi noise models, and indicate that the
formal expressions should be the same as the Bose case
provided that the coecients have specic parity with
respect to the Z
2
-grading of Fermi Fock space. Finally
in section 4 we indicate how, in the limit of zero time
delay in a network edge, we obtain a modular reduction
in the models.
A. Dynamical Perturbations
If we are given a pair V
0
(t) and V (t) of strongly contin-
uous one-parameter groups, that is V
0
(t+s) = V
0
(t)V
0
(s)
and V (t +s) = V (t)V (s), then we may view V as a per-
turbed dynamics with respect to the free dynamics of V
0
by transforming to the interaction picture via the wave
operator
U(t) = V
0
(t)

V (t).
We note that U(t) inherits unitarity and strong conti-
nuity, but is not a group. Instead we have the so-called
cocycle property
U(t + s) =
t
(U(s))U(t),
where
t
(x) = V
0
(t)

XV
0
(t).
By Stones theorem, both V
0
and V possess self-adjoint
(Hamiltonian) innitesimal generators K
0
and K respec-
tively:
i

V
0
(t) = K
0
V
0
(t), i

V (t) = KV (t).
We say that K is a regular perturbation of K
0
if =
K K
0
denes an operator with dense domain. In this
case, U(t) will be strongly dierentiable and
i

U(t) = (t)U(t)
where the time-dependent Hamiltonian is
(t) =
t
().
In situations where is not densely dened, we will have
a singular perturbation and U(t) will not generally be
strongly dierentiable.
B. Quantum Stochastic Evolutions
Quantum stochastic evolutions driven by quantum in-
put processes b
i
(t) have been studied in quantum optics
for some time. We may view these processes a as singu-
lar operators acting formally on the Hilbert space with
the Fock space F over C
n
L
2
(R). For F, we have
a well-dened amplitude
1
, i
1
; ;
m
, i
m
[ which is
completely symmetric under interchange of the m pairs
of labels (
1
, i
1
), , (
m
, i
m
), and this represent the am-
plitude to have m quanta with a particle of type i
1
at

1
, particle of type i
2
at
2
, etc. We have the following
resolution of identity on F
I =

m=0
(
_
d
1
d
m
)(
n

i1=1

n

im=1
)
[
1
, i
1
; ;
m
, i
m

1
, i
1
; ;
m
, i
m
[.
The annihilator input process b
i
(t) is then realized as

1
, i
1
; ;
m
, i
m
[b
i
(t)
=

m + 1 t, i;
1
, i
1
; ;
m
, i
m
[.
The annihilation operators, together with their formal
adjoints the creator operators b
i
(t)

satisfy the singular

canonical commutation relations
[b
i
(t), b
j
(s)] =
ij
(t s). (1)
1. The Time Shift
Let us introduce the following operator on the Fock
space
K
0
=
n

j=1
_

dt b

(t)
j
j

t
b(t)
j
(2)
which is the second quantization of the one-particle op-
erator i

t
. This is clearly a self-adjoint operator and the
unitary group V
0
(t) = e
itK0
it generates is just the time
shift:

1
, i
1
; ;
m
, i
m
[V
0
(t)
=
1
+ t, i
1
; ;
m
+ t, i
m
[.
The free evolution
t
() = V
0
(t)

()V
0
(t) will translate
the input processes in time:

(b
i
(t)) = b
i
(t + ),

(b

i
(t)) = b

i
(t + ).
2. The Local Hamiltonian
Let us x a system space h and consider a singular
perturbation on h F of the form
= E
ij
b

i
(0)b
j
(0) + E
i0
b

i
(0) + E
0j
b
j
(0) + E
00
, (3)
with E

ij
= E
ji
, E

i0
= E
0i
and E

00
= E
00
. We obtain a
time-dependent Hamiltonian by means of the time shift
(t) =
t
()
= E
ij
b

i
(t)b
j
(t) + E
i0
b

i
(t) + E
0j
b
j
(t) + E
00
.
3
The solution to the formal equation

U(t) = i(t)U(t),
with initial condition U(0) = I, may be expresses as the
Dyson series expansion
U(t) =

n=0
(
1
i
)
n
_
n(t)
(
n
) (
1
)
where we encounter integration over the simplices
n
(t)
of times t
n
> >
1
0. The formal series may
be rewritten as the chronologically ordered exponential
which, for convenience, we may denote as
U(t) =

T exp
1
i
_
t
0
()d. (4)
We observe that U(t) is a -cocycle since
U(t + s) =

T exp
1
i
_
t+s
t
()d

T exp
1
i
_
t
0
()d
=
t
(

T exp
1
i
_
s
0
()d)

T exp
1
i
_
t
0
()d
=
t
(U(s)) U(t).
3. Wick Ordered Form
Let us briey indicate how to convert U(t) to Wick
order [26], [27] [28]. Starting from the integro-dierential
equation U(t) = 1 i
_
t
0
(s)U(s)ds, we have
[b
i
(t), U(t)] = i
_
t
0
[b
i
(t), (s)]U(s)ds
= i
_
t
0

ij
(t s)E
jk
b
k
(t) + E
j0
U(s)
=
i
2
E
ij
b
j
(t)U(t)
i
2
E
i0
U(t).
Here we assumed that [b
i
(t), U(s)] = 0 for t > s, since
U(s) depends only on the noise up to time s. We also
adopted the convention that the -function contributes
on half-weight due to the upper limit of the integral.
This implies that b
i
(t)U(t) = [(1 +
i
2
E)
1
]
ij
[U(t)b
j
(t)
i
2
E
j0
U(t)] and we may use this to set the equation

U(t) = i(t)U(t) to Wick order, to obtain

U(t) = b
i
(t)

(S
ij

ij
)U(t)b
j
(t) + b
i
(t)

L
i
U(t)
L

i
S
ij
U(t)b
j
(t) (
1
2
L

i
L
i
iH)U(t), (5)
where S = [S
ij
] is the Cayley transform E = [E
ij
],
S =
1
i
2
E
1 +
i
2
E
(6)
and therefore unitary, while
L
i
= i
_
1
1 +
i
2
E
_
ij
E
j0
,
H = E
00
+
1
2
E
0i
_
Im
1
1 +
i
2
E
_
ij
E
j0
(7)
with H self-adjoint.
4. Ito Quantum Stochastic Dierential Form
Introducing integrated elds
B
i
(t)

=
_
t
0
b
i
()

d, B
i
(t) =
_
t
0
b
i
()d, (8)

ij
(t) =
_
t
0
b
i
()

b
j
()d, (9)
called the creation, annihilation and gauge processes, re-
spectively, it is possible to dene quantum stochastic It o
integrals with respect to these elds. Conditions for the
existence and uniqueness of solutions is given and, for
a xed system space h are given in [1]. The equation
(5) is readily interpreted as the It o quantum stochastic
dierential equation
dU(t) = (S
ij

ij
)d
ij
(t) + L
i
dB
j
(t)

i
S
ij
dB
j
(t) (
1
2
L

i
L
i
+ iH)dtU(t). (10)
With the already deduced conditions that S = [S
ij
] be a
unitary matrix and H self-adjoint, (10) gives the general
equation satised by a unitary adapted quantum stochas-
tic process. The rather non-Hamiltonian appearance of
the equation is a result of the fact that we have the fol-
lowing non-trivial products of It o dierentials:
dB
i
dB

j
=
ij
dt, dB
i
d
jk
=
ij
dB
k
, (11)
d
ij
dB

k
=
jk
dB

i
, d
ij
d
kl
=
jk
d
il
. (12)
The It o convention for dierentials, that is where the in-
crement dY (t) is taken as the future pointing increment
Y (t + dt) Y (t) and X(t)dY (t) is understood at the
innitesimal level as X(t)[Y (t +dt) Y (t)]. As an alter-
native we could use the Stratonovich convention which is
to take the midpoint rule
X(t) dY (t) = X(t +
1
2
dt)[Y (t + dt) Y (t)]
X(t)dY (t) +
1
2
dX(t)dY (t).
The Stratonovich quantum stochastic dierential equa-
tion corresponding to (10) is then
dU = iE
ij
d
ij
+ E
i0
dB

i
+ E
0j
dB
j
+ E
00
dt U.
5. The Langevin Equations
For a system operator X, we set
j
t
(X) = U(t)

XU(t).
4
By reference to the quantum It o rules, we see that, for
S = I, we have that
dj
t
(X) = U(t)

XdU(t) + dU(t)

X I U(t)
+dU(t)

XdU(t)
= j
t
([X, L
i
])dB
i
(t)

+ j
t
([L

i
, X])dB
i
(t)
+ j
t
(/X)dt
where we encounter the Lindblad generator
/X =
1
2
L

i
[X, L
i
] +
1
2
[L

i
, X]L
i
i[X, H].
For S ,= I, we have the general Langevin equation
dj
t
(X) = j
t
(S

ki
XS
kj

ij
X)d
ij
(t)
+j
t
(S

ji
[X, L
j
])dB
i
(t)

+ j
t
([L

i
, X]S
ij
)dB
j
(t)
+ j
t
(/X)dt.
6. Input-Output Relations
The output processes B
out
i
(t) are dened by the iden-
tity
B
out
i
(t) = U(t)

I B
i
(t) U(t), (13)
which in dierential form takes the following form
dB
out
i
(t) = j
t
(S
ij
) dB
j
(t) + j
t
(L
i
) dt. (14)
We see that the dierential of the output is a unitary
rotation of the input by the matrix S in the interaction
picture, plus a drift term corresponding to the coupling
L in the interaction picture. The local version of this
relation is then
b
out
i
(t) = j
t
(S
ij
) b
j
(t) + j
t
(L
i
). (15)
From the dening relation (13) we see that the output
processes satisfy the canonical commutation relations.
C. Unitary QSDEs as Singular Perturbations
The stochastic process U(t) dene by either formally
by (4) or mathematically as a solution to (10), is strongly
continuous, but due to the presence of the noise elds
dB

i
, dB
j
and d
ij
not strongly dierentiable. Here we
see that the local interaction given by (3) is a singular
perturbation of the generator of the time-shift (2).
We remark that nevertheless U(t) is a -cocycle and
that if we now dene V (t) by
V (t) =
_
V
0
(t) U (t) , t 0;
U (t)

V
0
(t) , t < 0.
then V (t) is a strongly continuous unitary group and
therefore admits an innitesimal generator K.
Surprising as it may seem, the quantum stochastic pro-
cess U(t) may be considered as the wave-operator for a
quantum dynamics with Hamiltonian K with respect to
the free dynamics of the time shift generated by K
0
. The
relation
K = K
0
+
however has only a formal meaning as the is singular
with respect to K
0
.
It has been a long standing problem to characterize the
associated Hamiltonian K for these models [3]. The ma-
jor breakthrough came in 1997 when A.N. Chebotarev
solved this problem for the class of quantum stochas-
tic evolutions satisfying Hudson-Parthasarathy dieren-
tial equations with bounded commuting system coe-
cients [4]. His insight was based on scattering theory of a
one-dimensional system with a Dirac potential, say, with
formal Hamiltonian
k = i + E
describing a one-dimensional particle propagating along
the negative x-axis with a delta potential of strength E
at the origin. (In Chebotarevs analysis the -function
is approximated by a sequence of regular functions, and
a strong resolvent limit is performed.) The mathemat-
ical techniques used in this approach were subsequently
generalized by Gregoratti [5] to relax the commutativity
condition. More recently, the analysis has been further
extended to treat unbounded coecients [6].
Independently, several authors have been engaged in
the program of describing the Hamiltonian nature of
quantum stochastic evolutions by interpreting the time-
dependent function (t) as being an expression involving
quantum white noises satisfying a singular CCR [26],[27],
[29]. This would naturally suggest that should be in-
terpreted as a sesquilinear expression in these noises at
time t = 0.
The generator of the free dynamics k
0
= i is not semi-
bounded and the -perturbation is viewed as a singular
rank-one perturbation. Here methods introduced by Al-
beverio and Kurasov [30],[31],[32] may be employed to
construct self-adjoint extensions of such models, which
we show in the next section for a wave on a 1-D wire.
In this note we revisit the one-dimensional model con-
sidered by Chebotarev, though consider this as a problem
of nding a suitable self-adjoint for the singular second
quantized Hamiltonian and present a intuitive argument
leading to the correct from of K.
II. GLOBAL HAMILTONIAN AS SINGULAR
PERTURBATION OF THE TIME SHIFT
GENERATOR
A. Single quantum on a 1-D wire
We begin it a model of a single quantum mechanical
particle moving in one-dimension with free Hamiltonian
H
0
= vp = iv

x
.
5
The evolution is just the translation of the wave-
function at velocity v along the negative x-axis:
x[(t) = (x + vt).
For simplicity we take = v = 1, so that x is arc-time
along the wire.
We shall consider the singular perturbation consisting
of a -kick a the origin:
H = H
0
+ (x).
For the particle coming in from the right, it will feel an
impulse as it passes the origin which will have the nett
eect of introducing a jump discontinuity. The singular
part of the Schrodinger equation, i

= H is then
i[(0
+
) (0

)] +
(0
+
) + (0

)
2
,
where we have the momentum impulse (proportional to
the jump in at the origin) and the average value picked
out by the -function. To obtain a self-adjoint extension,
we argue that this term vanishes exactly, and this implies
the boundary condition
(0

) = s(0
+
).
where
s =
1
i
2

1 +
i
2

.
At present, we are considering the class of square-
integrable functions with a possible jump discontinuity
at x = 0 for which the derivative function exists away
from zero and is again square-integrable. We note the
following integration by parts formula for functions ,
in this class:
_

(i

x
) =
_
(i

x
)

(0
+
)(0
+
) + i

(0

)(0

).
Let us denote the bras 0

[ and

0[ =
1
2
0
+
[ +
1
2
0

[
dened on this class by
0

[ = (0

),

0[ =
(0
+
) + (0

)
2
(16)
then we have
[i

x
= i

x
[ +[j
where the jump term is
j = i[0
+
0
+
[ + i[0

[
i[0
+
0

0[ + i[

00
+
0

[,
where 0
+
0

[ = (0
+
) (0

). The operator i

x
,
understood in the current distributional sense, is clearly
not symmetric on this class of functions, however we do
have [k
0
= k
0
[ where
k
0
= i

x
+ i[

00
+
0

[. (17)
From von Neumanns theory of self-adjoint extensions
of symmetric operators, it is well known that all self-
adjoint extensions of the momentum operator dened on
the 1-dimensional line with the origin removed are deter-
mined by a boundary condition (0

) = s(0
+
) where
s is unimodular, as is the case here. Evidently, this cap-
tures in, a very simple setting, the type of scattering
that we see in (6). We now show that this problem can
be second-quantized without too much diculty.
B. Indenite number of identical (Bose) quanta on
a 1-D wire
To deal with discontinuities at zero, we introduce the
averaged noise

b
i
(0) =
1
2
b
i
(0
+
) +
1
2
b
i
(0

). (18)
Following our remarks leading to (17), we split K
0
=
_
+

b(x)

i

x
b(x)dx =

K
0
+ J where

K
0
=
n

j=1
(
_
0

+
_
+
0
)b
j
(x)

i

x
b
j
(x)dx,
J = i
n

j=1

b
j
(0)

[b
j
(0
+
) b
j
(0

)]. (19)
1. Pure Scattering
We take the singular potential to be the second quan-
tization of the -function written in an explicitly sym-
metric manner:
= E
ij

b
i
(0)

b
j
(0).
Then
i

= (K
0
+ )
=

K
0
+

b
i
(0)

[i[b
i
(0
+
) b(0

)] + E
ij

b
j
(0)]
and again asking for the singular part (that is, the coe-
cient of

b
i
(0)

) to vanish leads to the boundary condition

b
i
(0

) = [
1
i
2
E
1 +
i
2
E
]
ij
b
j
(0
+
) = S
ij
b
j
(0
+
).
We see that we have free propagation by translation along
the incoming and outgoing wires. At the origin we have
the boundary condition b
i
(0

) = S
ij
b
j
(0
+
).
6
2. General Situation
The fact that we are modeling a quantum eld pro-
cess traveling along the wire means that we may consider
more general interactions than just scattering. In partic-
ular, we may include the action on the system due to the
emission or absorption of the eld quanta. To model this
we now take
= E
ij

b
i
(0)

b
j
(0) + E
i0

b
i
(0)

+ E
0j

b
j
(0) + E
00
.
Then
i

= (K
0
+ )
=

K
0
+ E
i0

b
i
(0) + E
00

+

b
i
(0)

i[b
i
(0
+
) b
i
(0

)] + E
ij

b
j
(0) + E
10

We once again ask that the nal term involving

b
i
(0)

(the singular part!) vanishes and this is equivalent to the

algebraic condition:
b
i
(0

) = [
1
i
2
E
1 +
i
2
E
]
ij
b
j
(0
+
) i[
1
1 +
i
2
E
]
ij
E
j0
,
and, substituting in for the It o coecients (7), this is
equivalent to the boundary condition b
i
(0

) = L
i
+
S
ij
b
j
(0
+
). Rearranging then leaves us with

= i

K
0
(
1
2
L

i
L
i
+ iH)L

i
S
ij
b
j
(0
+
).
C. The Global Hamiltonian
We therefore deduce that the form of the Hamiltonian
K is
iK = i

K
0
(
1
2
L

i
L
i
+ iH)L

i
S
ij
b
j
(0
+
),
(20)
on the domain of suitable functions satisfying the bound-
ary condition
b
i
(0

) = L
i
+ S
ij
b
j
(0
+
). (21)
here the suitable functions in question are those on the
joint system and Fock space that are in the domain of the
free translation along the edges (excluding the vertex at
the origin) and in the domain of the one-sided annihila-
tors b
i
(0

). This agrees with the expression found in [4]

and [5].
III. FERMIONIC MODELS
In the Fermi analogue we encounter input processes
a
i
(t) satisfying the singular canonical anti-commutation
relations
a
i
(t), a
j
(s)

=
ij
(t s), (22)
with a
i
(t), a
j
(s) = 0 = a
i
(t)

, a
j
(s)

. The appropri-
ate Hilbert space to describe these objects is the Fermi
Fock space F

consisting of vectors with amplitudes

1
, i
1
; ;
m
, i
m
[ that are completely anti-symmetric
under interchange of the labels (
i
, i
j
). (By the exclusion
principle, the amplitude vanishes if two labels are iden-
tical.) The Fermi annihilator is then dened by
a
i
(t)[
1
, i
1
; ;
m
, i
m
= [i, t;
1
, i
1
; ;
m
, i
m
.
On the domain of suitable test vector F

, we then
dene the singular densities a
i
(0

) and a
i
(0) =
1
2
a
i
(0
+
)+
1
2
a
i
(0

).
The second quantization procedure is similar to the
Bose case and we can immediately introduce the Fermi
analogues of the time shift operators:
K
0
=
n

j=1
_

dt a
j
(t)

i

t
a
j
(t),

K
0
=
n

j=1
(
_
0

+
_

0
)dt a
j
(t)

i

t
a
j
(t),
J = i
n

j=1
a
j
(0)

[a
j
(0
+
) a
j
(0

)].
A. Coupling to the System
The theory of Fermionic quantum stochastic calculus
was developed in the mid-1980s by Hudson and Apple-
baum [33], [34] for Fermi diusions of even parity, and
Hudson and Parthasarathy [35] for the general case. In
applications to physical models we encounter restrictions
on the type of coupling and dynamical evolutions, mean-
ing that the full theory presented in the latter paper is
too broad.
For instance, if the bath is an electron reservoir, then
the specic issue that arises in practice is that the cre-
ation of an electron in the bath necessarily requires the
removal of an electron from the system. This means that
the system must carry Fermi degrees of freedom. An
example of a suitable Fermionic local Hamiltonian is
(t) =
ij
a
i
(t)

a
j
(t) +

+
j
c

a
j
(t) +

j
a
j
(t)

where c

are Fermionic modes of the system and the

ij
,
j
,

are constants. We require that the system

modes satisfy anti-commutation relation c

, c

,
c

, c

= 0 = c

, c

, and also anti-commute with the

bath modes
c

, a
i
(t) = c

, a
i
(t) = 0,
c

, a
i
(t)

= c

, a
i
(t)

= 0.
7
1. Parity Restrictions
We dene the parity operator by
(c

) = c

,
(a
i
(t)) = a
i
(t),
with (XY ) = (X)(Y ) and (X

) = (X)

. An oper-
ator X on the joint system and bath space is said to be of
even parity is (X) = X and of odd parity if (X) = X.
We note that the local Hamiltonian is of even parity,
((t)) = (t),
and this is a natural requirement for all physically realis-
tic models. The most general type of local Hamiltonian
that we shall consider will be of the form
(t) = E
ij
a
i
(t)

a
j
(t) + a
i
(t)

E
i0
+ E
0j
a
j
(t) + E
00
where the E
ij
are operators on the system space neces-
sarily possessing the following denite parities
E
ij
- even
E
i0
= (E
0i
)

- odd
E
00
- even.
We note that a
i
(t)

E
i0
= E
i0
a
i
(t)

. As before, we
wish to study the unitary
U(t) =

T exp
1
i
_
t
0
()d
which by construction should be of even parity and sat-
isfy the cocycle relation with respect to the free transla-
tion on the Fermionic Fock space.
2. Conversion to Ito Form
As in the Bose case we encounter
[a
i
(t), U(t)] = i
_
t
0
[a
i
(t), (s)]U(s)ds.
The parities of the components of are essential in
computing [a
i
(t), (s)]. For instance, using the anti-
commutation relations and observing that E
jk
commutes
with the bath modes a
i
(t),
[a
i
(t), E
jk
a
j
(s)

a
k
(s)] = +E
jk
a
i
(t)a
j
(s)

a
k
(s)
E
jk
a
j
(s)

a
k
(s)a
i
(t)
= E
jk
a
i
(t), a
j
(s)

a
k
(t)
= E
jk
a
k
(t)(t s).
We in fact see that
[a
i
(t), (s)] =
ij
E
jk
a
k
(t) + E
j0
(t s)
which is structurally identical to the Bose case. The Wick
ordered form is therefore equivalent to (5)

U(t) = a
i
(t)

(S
ij

ij
)U(t)a
j
(t) + a
i
(t)

L
i
U(t)
L

i
S
ij
U(t)a
j
(t) (
1
2
L

i
L
i
iH)U(t),
The coupling operators S, L, H take the same forms as in
(6, 7) though we note that the carry the following denite
parities listed in the table below
Parity Even Odd
Bath Processes
ij
(t) A
i
(t), A
i
(t)

System coecients E
ij
, E
00
E
i0
, E
0j
It o coecients S
ij
, H L
i
B. It o Form
As in the Bose case we may introduce the integrated
elds
A
i
(t) =
_
t
0
a
i
(s)ds, A
i
(t)

=
_
t
0
a
i
(s)

ds

ij
(s) =
_
t
0
a
i
(s)

a
j
(s)ds
which are regular operators on the Fermi Fock space and
which may be extended to operators on the joint system
and bath space in the obvious manner. The operators
A
i
(t) and A
i
(t)

are clearly odd, while

ij
(t) is even.
They lead to a quantum It o table that is exactly the
same as the Bose case (12).
The It o form of the QSDE is therefore
dU(t) = (S
ij

ij
)d
ij
(t) + dA

i
(t)L
i
L

i
S
ij
dA
j
(t) (
1
2
L

i
L
i
+ iH)dtU(t),
and we note the change dA

i
L
i
= L
i
dA

i
.
1. Fermi Input-Output Relations
An application of the It o table shows that Fermi output
elds dened by
A
out
i
(t) = U(t)

A
i
(t)U(t)
will satisfy the dierential relations
dA
out
i
(t) = U(t)

S
ij
U(t)dA
j
(t) + U(t)

L
i
U(t)dt.
While formally identical to the Bose case, we should em-
phasize that the Fermi input-output relation has the ad-
ditional property that both sides of the relation are of
odd parity.
8
2. The Fermi Langevin Equations
We again dene j
t
(X) = U(t)

XU(t). For S = I, the

QSDE reduces to
dU(t) = dA

i
L
i
L

i
dA
i

1
2
L

i
L
i
+ iH)dtU(t)
We now have
dj
t
(X) =U(t)

XdU(t) + dU(t)

XU(t) + dU(t)

XdU(t)
=U(t)

XdA

i
L
i
XL

i
dA
i
X(
1
2
L

i
L
i
+ iH)dt
+ L

i
dA
i
X dA

i
L
i
X (
1
2
L

i
L
i
iH)Xdt
+ L

i
dA
i
XdA

j
L
j
U(t)
and to proceed further we need to take into account the
parity features of X.
For a given operator Z we can write Z as a sum of
even and odd parts by setting Z
even
=
1
2
Z +
1
2
(Z) and
Z
odd
=
1
2
Z
1
2
(Z) to yield Z = Z
even
+ Z
odd
with
(Z) = Z
even
Z
odd
. We see that
ZdA

i
(t) = dA

i
(t)(Z), dA
i
(t)Z = (Z)dA
i
(t).
The Langevin equation therefore becomes
dj
t
(X) = dA

i
(t) j
t
((X)L
i
L
i
X)
+j
t
(L

i
(X) XL

i
) dA
i
(t)
+j
t
(L

i
(X)L
i

1
2
XL

i
L
i

1
2
L

i
L
i
X) dt
ij
t
([X, H])dt
and for S ,= I this generalizes to
dj
t
(X) = j
t
(S

ki
XS
kj

ij
X) d
ij
(t)
+dA

i
(t) j
t
(S

ji
[(X)L
j
L
j
X]) + j
t
([L

i
(X) XL

i
]S
ij
) dA
j
(t)
+j
t
(L

i
(X)L
i

1
2
XL

i
L
i

1
2
L

i
L
i
X) dt
ij
t
([X, H]) dt.
For even operators X this is formally identical to the
Bose Langevin equation, however, for odd X we have
dj
t
(X) = j
t
(S

ki
XS
kj

ij
X)d
ij
(t)
dA

i
(t)j
t
(S

ji
X, L
j
) j
t
(L

i
, XS
ij
)dA
j
(t)
j
t
(
1
2
X, L

i
L
i
+
1
2
L

i
L
i
, X + i[X, H])dt.
3. The Fermi Global Hamiltonian
We can now state the global Hamiltonian K for the
Fermi case:
iK = i

K
0
(
1
2
L

i
L
i
+ iH)L

i
S
ij
a
j
(0
+
),
(23)
on the domain of suitable functions satisfying the bound-
ary condition
a
i
(0

) = L
i
+ S
ij
a
j
(0
+
). (24)
IV. QUANTUM FEEDBACK NETWORKS
A general quantum feedback network consists of a di-
rect graph with vertices 1 and edges c. At each vertex
we have a quantum mechanical system described by the
triple (S
v
, L
v
, H
v
). To describe the open-loop model, we
may form the concatenation (S, L, H) where
S =
_

_
S
1
0
0 S
2

.
.
.
.
.
.
.
.
.
_

_, L =
_

_
L
1
L
2
0
_

_, H =

vV
H
v
.
The closed loop arrangement comes from feeding output
elds in as input elds as indicated in the network graph.
The global Hamiltonian K for the network will take the
form [9]
iK = i

eE

K
e

vV
_
1
2
L

v
L
v
+ iH
v
_

vV
L

v
S
v
b
v
(+)
where b
v
(+) is the vector of incoming annihilator den-
sities evaluated immediately before vertex v 1. This
must be supplemented by the set of boundary conditions
b
v
() = S
v
b
v
(+) + L
v

for each vertex v 1. Here

K
e
is the generator of free
translation along each particular edge e c. A detailed
account may be found in [8].
A. The Series Product
We illustrate the method next with a derivation of the
series product of [9] using a general argument introduced
in [8]. The basic set up is sketched in Figure 1.
FIG. 1. (color online) Two systems (S
(1)
, L
(1)
, H
(1)
) and
(S
(2)
, L
(2)
, H
(2)
) connected in series.
We may model a pair of systems in cascade by
specifying the local Hamiltonians through the triples
_
S
(i)
, L
(i)
, H
(i)
_
for i = 1, 2. The positions of the sys-
tems are t
1
and t
2
respectively. The global Hamiltonian
9
is then
iK = i
__
t2

+
_
t1
t2
+
_

t1
_
b
j
(t)

i

t
b
j
(t)
(
1
2
L
(1)
j
L
(1)
j
+ iH
(1)
)L
(1)
j
S
(1)
jk
b
k
_
t
+
1
_

(
1
2
L
(2)
j
L
(2)
j
+ iH
(2)
)L
(2)
j
S
(2)
jk
b
k
_
t
+
2
_

with boundary conditions

b
j
_
t

1
_
= S
(1)
jk
b
k
_
t
+
1
_
+ L
(1)
j
,
b
j
_
t

2
_
= S
(2)
jk
b
k
_
t
+
2
_
+ L
(2)
j
.
To obtain the instantaneous feedforward limit we take
t
1
t
2
0. Denoting the common limit as t
0
then
b
j
_
t

0
_
= S
(2)
jk
lim
t2t0
b
k
_
t
+
2
_
+ L
(2)
j

= S
(2)
jk
lim
t1t0
b
k
_
t

1
_
+ L
(2)
j

= S
(2)
jk
lim
t1t0
_
S
(1)
jk
b
k
_
t
+
1
_
+ L
(1)
j

_
+ L
(2)
j

= S
(2)
jk
_
S
(1)
jk
b
k
_
t
+
0
_
+ L
(1)
j

_
+ L
(2)
j
.
The global Hamiltonian then reduces to
iK = i
__
t0

+
_

t0
_
b
j
(t)

i

t
b
j
(t)
(
1
2
L
(1)
j
L
(1)
j
+ iH
(1)
)L
(1)
j
S
(1)
jk
b
k
_
t
+
0
_

(
1
2
L
(2)
j
L
(2)
j
+ iH
(2)
)L
(2)
j
S
(2)
jk
b
k
_
t
+
0
_
.
Substituting in gives
iK = i
__
t0

+
_

t0
_
b
j
(t)

i

t
b
j
(t)
(
1
2
L

j
L
j
+ iH)L

j
S
jk
b
k
_
t
+
0
_

with boundary condition

b
j
_
t

0
_
= S
jk
b
k
_
t
+
0
_
+ L
j

where we have
S = S
(2)
S
(1)
, L = L
(2)
+ S
(2)
L
(1)
,
H = H
(1)
+ H
(2)
+ Im
_
L
(2)
S
(2)
L
(1)
_
.
The rule
_
S
(2)
, L
(2)
, H
(2)
_

_
S
(1)
, L
(1)
, H
(1)
_
= (S, L, H)
determined above is referred to as the series product of
the cascaded system [9].
FIG. 2. (color online) The internal line (blue) is feedback in
as input.
B. General Feedback Reduction
More generally, the reduced model obtained from a
concatenation (S, L, H) obtained by eliminating the edge
(r
0
, s
0
) is shown in Figure 2.
By a similar argument, it is readily seen to be deter-
mined by the operators (S
red
, L
red
, H
red
) where [8]
S
red
sr
= S
sr
+ S
sr0
(1 S
s0r0
)
1
S
s0r
,
L
red
s
= L
s
+ S
sr0
(1 S
s0r0
)
1
L
s0
,
H
red
= H +

s: output edge
ImL

s
S
sr0
(1 S
s0r0
)
1
L
s0
,
We comment that the same rule applies to the Fermi
case as well, and in particular that the series product,
and feedback reduction rule ( preserve the correct parity
in table above.
V. CONCLUSION
We have shown that the standard Markov models for
quantum mechanical systems driven by quantum inputs
may be formulated as the free translation of an indef-
inite number of indistinguishable quanta along a one-
dimensional wire with a singular localized interaction at
the point which the system is placed. The approach
works equally well for Boson and Fermion quanta. More
generally we may consider quantum network models with
quanta propagating along the edges and localized quan-
tum mechanical systems located at the vertices.
Under physically motivated assumptions on the parity
of the coupling operator coecients, the Fermi analogue
leads to identical equations for the series product for cas-
caded and direct feedback situations, and more generally
for the feedback reduction formula for closed-loop net-
works involving general feedback relations.
ACKNOWLEDGMENTS
It is a great pleasure to thank Matthew James and
Hendra Nurdin for several stimulating discussions on
quantum feedback networks. He is also indebted to
Alexei Iantchenko for pointing out the work of Albeverio
10
and Kurasov for the rst-quantization scattering prob-
lem. The support of the UK Engineering and Physi-
cal Sciences research council under grants EP/H016708/1
and EP/G020272/1 is gratefully acknowledged.
[1] R.L. Hudson and K.R. Parthasarathy, Commun. Math.
Phys. 93, 301-323 (1984)
[2] K.R. Parthasarathy, An Introduction to Quantum
Stochastic Calculus. Berlin: Birkhauser, 1992.
[3] L. Accardi, Rev. Math. Phys., 2, 127-176, (1990)
[4] A.M. Chebotarev, Math. Notes, 61, No. 4, 510-518,
(1997)
[5] M. Gregoratti, Commun. Math. Phys., 222, 181-200,
(2001)
[6] R. Quezada-Batalla, O. Gonzalez-Gaxiola, Math. Notes,
81, 5-6, 734-752, (2007)
[7] W. von Waldenfels, A Measure Theoretical Approach to
Quantum Stochastic Processes, Lecture Notes in Physics
Volume 878, Springer Berlin Heidelberg (2014)
[8] J. Gough, M.R. James, Commun. Math. Phys., Volume
287, 1109-1132, Number 3 / May, (2009)
[9] J. Gough, M.R. James, IEEE Trans. Automatic Control,
54(11):2530-2544, (2009)
[10] J.E. Gough, R. Gohm, M. Yanagisawa, Phys. Rev. A 78,
062104 (2008)
[11] J.E. Gough, Phys. Rev. A 78, 052311 (2008)
[12] J.E. Gough, S. Wildfeuer, , Phys. Rev. A 80, 042107
(2009)
[13] J.E. Gough, M.R. James, H.I. Nurdin, Phys. Rev. A 81,
023804 (2010)
[14] J. Kerckho, H.I. Nurdin, D.S. Pavlichin, H. Mabuchi,
Phys. Rev. Lett. 105, 040502 (2010)
[15] H.I. Nurdin, M.R. James, and A.C. Doherty, SIAM J.
Control Optim., vol. 48, no. 4, pp. 2686-2718, (2009)
[16] H.I. Nurdin, IEEE Transactions on Automatic Control,
55(4), pp. 1008-1013, (2010)
[17] H. I. Nurdin, , IEEE Trans. Autom. Control, 55(10), pp.
2439-2444, (2010)
[18] C. Gardiner and P. Zoller, Quantum Noise: A Handbook
of Markovian and Non-Markovian Quantum Stochastic
Methods with Applications to Quantum Optics, 2nd ed.,
ser. Springer Series in Synergetics. Springer, (2000)
[19] G.J. Milburn, H.B. Sun, B. Upcroft, Aust. J. Phys. 53,
463-476, (2000)
[20] G.J. Milburn, Aust. J. Phys. 53, 477-487, (2000)
[21] H. Ohno, et al., Nature 408, 944 (2000)
[22] P. Recher, E.V. Sukhorukov, D. Loss, Phys. Rev. Lett.
85, 1962 (2000)
[23] S.A. Wolf, et al., Science 294, 1488-1495 (2001)
[24] J.M. Elzerman, R. Hanson, L.H. Willems van Beveren,
B. Witkamp, L.M.K. Vandersypen, L.P. Kouwenhoven,
Nature 430, 431 (2004)
[25] D.D. Awschalom, M.E. Flatte, Nature 3, 153 (2007)
[26] J. Gough, Theor. Math. Phys. 111, No. 2, 218-233, May
(1997)
[27] J. Gough, Rep. Math. Phys. Vol. 44, 313-338, (1999)
[28] J. Gough, J. Math. Phys., vol. 47, no. 113509, (2006)
[29] L. Accardi, Y.G. Lu, I. Volovich, Quantum Theory and
Its Stochastic Limit, Springer, (2002)
[30] S. Albeverio and P. Kurasov, J. Func. Anal., 148, 152-
169, (1997)
[31] S. Albeverio and P. Kurasov, Integr. Equ. Oper. Theory,
27 , 379-400, (1997)
[32] S. Albeverio and P. Kurasov, London Math. Soc. Lect.
Note Ser. No. 271, Cambridge Univ. Press, (2000)
[33] D. Applebaum, R.L. Hudson, Commun. Math. Phys. 96,
473-496 (1984)
[34] D. Applebaum,: Publ. RIMS, Kyoto University 23, 17-56
(1987)
[35] R.L. Hudson, K.R. Parthasarathy, Commun.Math.Phys.
104, 457-470 (1986)