Notes on Open System Quantum Dynamics

Contents

I. Introduction to Markovian dynamics 1

A. Open quantum systems 1

1. Denitions 1

2. Basic assumptions of Markovian dynamics 2

B. Quantum dynamical maps 2

1. CPT maps 2

2. Denition of complete positive 2 For both the systems we dene a Hilbert space (HS and

3. Equal statements of complete positivity 2 HE ), an Hamiltonian (HS and HE ) and a partial trace
operation (T rS and T rE ). The total system is given by
4. Example: CPT maps of a qubit 3
the tensor product of the two subsystem
C. Positive maps and entanglement 3

1. Denition of entanglement - pure states 3 H = HS ⊗ H E

2. Denition of entanglement - mixed states 4
This allows the existence of entangled states. In the total
3. PPT criterion 4
system we have a Hamiltonian
4. Entanglement witnesses 5

D. Dynamical semigroups and Lindblad theory 6 H = HS + HE + HI

1. Lindblad theorem 6

2. Proof of Lindblad theorem in nite and a total trace operation

dimensional case 7
T r = T rS T rE (1)

II. Non Markovian Quantum Dynamics -

Projection Operator Techniques 8
We dene mixed states density matrices as oper-
ators in H which are:
A. Projection operators 8

B. Nakajima-Zwanzig (NZ) Master Equation 8

• self-adjoint: ρ† = ρ
C. Time-convolutionless (TCL) Master Equation 9

D. General classication of master equations 10 • positive: ρ≥0

• of unitary trace: T rρ = 1
III. Correlated projection operator 11

A. General requirements 11
We remind that positive-deniteness of operators can be
B. Representation theorem 11 expressed by the condition
C. NZ and TCL expansion with correlated
projections 12 ∀ψ hψ | ρ | ψi ≥ 0
D. Generalized Lindblad structure 12
or equivalently requiring every eigenvalues of ρ to be pos-
itive.
From mixed states DM we can dene reduced density

I. INTRODUCTION TO MARKOVIAN matrices for the system

DYNAMICS
ρS ≡ T rE ρ
A. Open quantum systems
Starting from this denition we can show that ρS has the
same properties as ρ.
1. Denitions One of the goal of the theory is to describe the dynam-
ics of ρS in a closed form. Considering the dynamics of

An open quantum system is a system coupled to an- the total system as a whole, we can obtain the dynamics

other quantum system (which is generally a large one, of the system from

like a heat reservouir, which represents the interaction of

our system of interest with the environment). ρS (t) = T rE ρ(t)
 2

2. Basic assumptions of Markovian dynamics

n o (b)
= T r Ut† Ut (ρS (0) ⊗ ρE (0)) =

1. At time t=0 the total state is given by

(c)
ρ(0) = ρS (0) ⊗ ρE (0) = T r {ρS (0) ⊗ ρE (0)} = 1

where in (a) we have used the cyclic property of

thus the initial total state is a product state. This
trace (which in this case hold just for the total
assumption will lead to a linear homogeneous mas-
trace, since the argument contains operators which
ter equation.
mix the systems and the environment); in (b) the

2. The total dynamics of ρ is the time evolution of a unitarity of Ut and in (c) the unitarity of a total

closed system one and is given by system DM trace.

2. Φt is a positive map, i.e. maps positive operators

ρ(t) = Ut ρ(o)Ut†
to positive operators

where Ut is a unitary operator given by

ρS ≥ 0 =⇒ Φt ρS ≥ 0 ∀t ≥ 0
−ıHt
Ut = e
3. Φt is completely positive (we'll see the denition
In an equivalent way, we demand ρ(t) to satisfy Von later on)
Neumann's equation
Thus Φt is a trace preserving and completely positive
d map. Such a map is called in dierent many ways:
ρ(t) = −ı [H, ρ(t)] quantum operation (even though this name in gen-
dt
erally given also to non trace-preserving maps), CPT
From these two basic assumptions we nd that the exact maps (where CPT just stand for Completerly Positive
evolution of ρS is given by and Trace-preserving) or quantum channel

n o
ρS (t) = T rE Ut (ρS (0) ⊗ ρE (0)) Ut† (2)
2. Denition of complete positive

B. Quantum dynamical maps Consider the tensor product of the space HS with an
auxiliary hilbert space Cn
1. CPT maps H = H ⊗ Cn n = 2, 3, 4 . . .

Consider the initial environmental state ρE (0) as xed. In this space we can dene a trivial extension of Φ
Eq. 2 denes a linear map which is

ρS (0) 7→ ρS (t) = Φt ρS (0) (Φ ⊗ In ) (A ⊗ B) = (ΦA) ⊗ B

Note that Φt is a superoperator, i.e. a linear map in where In is the identity in Cn . This denition given on
the space of operators on HS . the space of product states can be extended on H by
Let's consider Φt from an axiomatic point of view. Φt continuity
has some general properties which follows from its de-  
nition: X X X
(Φ ⊗ In )  Ai ⊗ Bj  = (ΦAi ) ⊗ Bj
1. Φt preserves self-adjointness and trace i j ij

†
ρ†S = ρS =⇒ (Φt ρS ) = Φt ρS Φ

Now we can state the denition: is completely posi-
∀t ≥ 0 ⇐⇒Φ ⊗ In ≥ 0 ∀n
tive
T rS ρS = 1 =⇒ T rS (Φt ρS ) = 1
Remark: If HS is nite dimensional and dim HS = m,
As an example we briey review the proof of the Choi's theorem say that for an operator to be completely
trace-preserving property: positive it's sucient to check the denition for 2≤n≤
n o m.
eq(2)
T rS (Φt ρS ) = T rS T rE Ut (ρS (0) ⊗ ρE (0)) Ut† =
3. Equal statements of complete positivity
eq(1)
n o (a)
= T r Ut (ρS (0) ⊗ ρE (0)) Ut† = All the following proposition are equivalent:
 3

1. Φ is completely positive and trace-preserving Note that in this case purity is conserved

2. Φ admit a Kraus decomposition, i.e. it action upon d 2

an operator can be written as
|~v | = 0
dt
Ai ρS A†i Ai A†i = IS
X X
ΦρS = with From a physical point of view, such map corresponds

i to a completely reversible transformation. This kind of

transformation is a CPT map.
3. Its action upon a DM ρS can also be written as b. Depolarization channel Another kind of CPT
n o map is a depolarization channel
ΦρS = T rE Ut (ρS ⊗ ρE ) Ut†
1
ΦρS = p (T rρS ) I + (1 − p) ρS 0≤p≤1
where ρE HE and Ut is a
is a density matrix in 2
unitary operator on the space HS ⊗HE (stinespring
A Bloch vector ~v is transformed according to
dilation theorem)
   
v1 (1 − p)v1
Thus we have this very important results: the denition
~v =  v2  → v~0 =  (1 − p)v2 
of CPT maps is a necessary and sucient characteriza-
v3 (1 − p)v3
tion of maps which can represent the dynamics of a open
system. Thus this transformation is a shrinkage of the Bloch
sphere.
We can show that a depolarization channel is a CPT
4. Example: CPT maps of a qubit map by showing that is sum of CPT maps.
c. Transposition Transposition Φ=T is a positive
A qubit is a 2-state system. Its general DM is given trace-preserving map which is NOT a CPT map. In fact
by the formula we'll show in the following that transposition is positive
but not completely positive.
ρS = 1/2 (IS + ~v · ~σ ) Furthermore, it could be shown that transposition is
essentialy the only map of such kind for qubit, in the
where ~σ is a vector whose components are Pauli matrices
sense that every other non completely positive map can
     
0 1 0 −ı 1 0 be written in a simple way in terms of transposition.
σ1 = σ2 = σ3 =
1 0 ı 0 0 −1 For a bloch vector, transposition corresponds to the
transformation
and ~v is a R3 vector in the so-called bloch sphere, i.e. a    
sphere of radius 1 centered in the origin (in other words, v1 v1
|~v | ≤ 1). ~v =  v2  → v~0 =  −v2 
The pure states are represented by a vector ~v which v3 v3
lie on the boundary of the sphere, since for such states
From a physical point of view, transposition is a time
we have |~v | = 1. Another remarkable point is the origin,
reversal transformation.
since for |~
v | = 0 we have the innite temperature state
or maximum entropy state

1/20

C. Positive maps and entanglement
ρS =
0 1/2
1. Denition of entanglement - pure states
Now consider some examples of map of qubit.
a. Unitary transformation Consider a time evolu-
In the following we focus on a bipartite system, i.e. a
tion which satisfy Liouville's equation:
system that can be decomposed as tensor product of two
d hilbert spaces
ρS = −ı [HS , ρS ]
dt
H = H1 ⊗ H2
For each time t, the DM is just related to the initial DM
by a unitary transformation: For a state vector |ψi we can write a unique Schmidt
decomposition
Φt ρS = Ut ρS Ut† E E
(1) (2)
X
3
|ψi = αi χi ⊗ χi
Considering the R representation of a qubit, this is just
i
a rotation of the Bloch sphere around a xed axis:
where the Schmidt coecients αi are real and positive,
d ~ ~
~ × v(t)
n
(1)
Eo n
(2)
Eo
v(t) = β while χi and χi are bases respectively in
dt
 4

H1 and H2 (they depend on |ψi and are called Schmidt From the denition of separable state, pi are positive
bases, or local bases in Quantum Information). (1) (2)
numbers. At the same timeΦρi and ρi are positive
From this decomposition of a state vector we can dene operators, since DM are positive operators and Φ is pos-
the Schmidt numer as the number of non-zero Schmidt itive by hypothesis. Thus we have
coecients. Thus we have the following denitions
(Φ ⊗ I) ρ ≥ 0 
• A state is said to be separable (or not entangled)
if has Schmidt number equal to 1. We can see from
the formula of Schmidt decomposition that such a If Φ is completely positive this theorem is is obvious,
state is a product state. since the thesis holds for any state ρ. However, if Φ
is positive but not CP this theorem is very interesting
• If Schmidt number is > 1 then the state is not
since it expresses a necessary criterion for separability
separable, or entangled.
and a sucient criterion for entanglement. In fact the
• In particular if all Schmidt coecient are nonzero contrapositive proposition is
and equal, we say that the state is maximally en-
tangled. (Φ ⊗ I) ρ  0 =⇒ ρ entangled

E.g. Bell basis for a system of two qubits is a basis of Note that this is just a sucient condition, since there
entangled vectors are entangled ρ which give a positive operator (Φ ⊗ I) ρ.
√ √ To get a necessary and sucient condition we have to
|ψ1 i = 1/ 2 (|+−i + |−+i) |ψ2 i = 1/ 2 (|+−i − |−+i) take much stronger hypotheses:
√ √
|ψ3 i = 1/ 2 (|++i − |−−i) |ψ4 i = 1/ 2 (|++i + |−−i)
Theorem (1b). Let Φ be a positive map. ρ is sep-
arable if and only if (Φ ⊗ I) ρ ≥ 0 for all positive maps
2. Denition of entanglement - mixed states Φ.

Consider the states space S(H) = {ρ|ρ ≥ 0, T rρ = 1}. Since there is no systematic way to generate any pos-
A state ρ ∈ S(H1 ⊗H2 ) is said to be separable (or not en- itive map, this necessary and sucient criterion is of no
tangled, or classically correlated) if it can be represented practical utility.
as

(1) (2)
X
ρ= pi ρi ⊗ ρi (3)
3. PPT criterion
i

where
P pi is a probability distribution (i.e. pi ≥ 0 and Now consider a very important example of a positive
pi = 1). Thus ρ is separable if it can be written as map that is not CP: the transposition. Let's show this
convex linear combination of product state. as a theorem:
Such a representation is highly non-unique, and the Theorem (2). 1) T is a positive map
problem of deciding if a given state is separable is a really 2) T is not a CPT map
hard problem. E.g. the sum of the Bell basis vector, Proof. 1) The trasposition is dened only in relation to
which are highly entangled states, is a separable state a basis, but the nal results doesn't depends on the choice

4
of the basis. Thus consider a generic positive operator B
1X 1 whose matrix representation is
|ψi i hψi | = I(1) ⊗ I(2)
4 i=1 4
Bij = hi | B | ji
The problem of entanglement in mixed states is con-
nected to the CPT maps: The positivity implies that

X
Theorem (1a). Φ beLet a positive map. If ρ is Bij vi? vj ≥ 0
separable =⇒ (Φ ⊗ I) ρ ≥ 0. ij

Proof. Let ρ be a separable state. Then by denition

for any vector v. The matrix element of the transposed
operator BT is
(1) (2)
X
ρ= pi ρi ⊗ ρi X X (a) X
T ?
i Bij vi vj = Bji vi? vj = Bij vj? vi
ij ij ji
and from the linearity of Φ⊗I
X  (1)  Where in (a) we have just exchanged the summation
(2)
(Φ ⊗ I) ρ = pi Φρi ⊗ ρi Bij vj? vi
P
index i and j. ji is just the scalar product
 5

hv ? | B | v ? i, and thus is positive by hypothesis. Thus we of just one subsystem. This operator give raise to a suf-
have cient criterion for entanglement called PPT (positive
X partial transposition) criterion. Namely
T ?
Bij vi vj ≥ 0 ∀v
ij (T ⊗ I)ρ  0 =⇒ ρ entangled

2) To prove the second proposition we just have to Note that in the proof of the second proposition we
nd a DM ρ such that (T ⊗ I)ρ has a negative eigenvalue. have obtained a very remarkable results: for any entan-
Let's consider a pure states gled pure state partial transposition is not completely
positive. Thus we can conclude that PPT criterion is
ρ = |Ψi hΨ| not only sucient but also necessary for entangled pure
states.
where |Ψi is an entangled state, i.e. it has a Schmidt The PPT criterion is necessary in an other particular
decomposition with at least two nonzero Schmidt coe- case:
cients:
X (1) (2) Theorem (3). The PPT criterion is necessary and
|Ψi = αi |χi i ⊗ |χi i sucient for all 2 ⊗ 2 and 2 ⊗ 3 systems. For all the other
i
systems it is not necessary, i.e. there are entangled PPT
states.
Let's assume that αk and αl are two nonzero Schmidt
coecients, with k 6= l. By substitution, we get
The idea behind this theorem is that in such cases any
(1) (1) (2) (2) positive map can be written as
X
ρ= αi αj |χi i hχj | ⊗ |χi i hχj |
ij
Φ = T1 + T2 ◦ T
Now we can compute the image of ρ under partial traspo-
sition, which by linearity of (T ⊗ I) is where T1 and T2 are completely positive maps, whileT
is transposition. Thus if (T ⊗ I)ρ ≥ 0, then for any other
ρ0 = (T ⊗ I) ρ = positive map (Φ ⊗ I)ρ ≥ 0. From theorem (1b), we get
that ρ is separable.
Another example of a map which is positive but not
 
(1) (1) (2) (2) completely positive is
X
= αi αj T |χi i hχj | ⊗ |χi i hχj | =
ij
ΦB = (T rB)I − B

(1) (1) (2) (2)

X
= αi αj |χj i hχi | ⊗ |χi i hχj | 4. Entanglement witnesses
ij

Now we show that the state We dene an entanglement witness as an operator W

  that is
√ (1) (2) (1) (2)
|φi = 1/ 2 |χk i ⊗ |χl i − |χl i ⊗ |χk i
• self adjoint W = W† on H = H1 ⊗ H 2
is an eigenstate of negative eigenvalue
• Tr{W ρsep } ≥0 for all separable states
 
(1) (1) (2) (2)
X
ρ0 |φi = αi αj |χj i hχi | ⊗ |χi i hχj | |φi = • Tr{W ρent } <0 for at least an entangled state
ij
The existence of such operators is guaranteed by Hahn-
Banach theorem.
X αi αj  (1) (2) (1) (2)

√ A very interesting fact is that exists a one-to-one cor-
= |χj i δik ⊗ |χi i δjl − |χj i δil ⊗ |χi i δjk =
ij
2 respondence between entanglement witness and positive
maps which are not CP. This correspondence is called
Jamiolkowski Isomorphism:
αi αj  (1) (2) (1) (2)

= √ |χl i ⊗ |χk i − |χk i ⊗ |χl i = −αk αl |φi
2 Φ ⇐⇒ W = (Φ ⊗ I)(|Ψi hΨ|)

since αk and αl are strictly positive by denition, |φi where |Ψi is a xed maximally entangled state. E.g., the
is an eigenstate of negative eigenvalue, which proves the entanglement witness associated to the transposition is
non-positivity of the operation (T ⊗ I).  the swap operator
We call (T ⊗ I) partial transposition. This operator
corresponds to the (physically unfeasible) time reversal W |φi ⊗ |χi = |χi ⊗ |φi
 6

D. Dynamical semigroups and Lindblad theory a. Example 1. two level system with photon emission
Consider a two level system: a ground state |gi and an

Let's consider the quantum dynamical maps Φt . We excited state |ei, separated by an energy ω0 . We call σ+
say that Φt satisfy Markov condition if the excitation operator, namely the annihilator the cre-
ator for state |gi and the creator for state |ei. Its adjoint
Φt+s = Φt ◦ Φs is the de-excitation operator σ− , which annihilate the
In such hypothesis the maps Φt has the formal struc-
state |ei and create the state |gi. The master equation
for a density matrix in two level system is
ture of a semigroup, with identity Φt=0 . A semigroup is
a group without inverse, in fact even though we can build  
d 1
the operators Φt with t < 0, in general they are neither ρS = −ı [HS , ρS ]+γ0 (N +1) σ− ρS σ+ − {σ+ σ− , ρS } +
dt 2
positive nor completely positive.
Unitary transformation as a subset of {Φt } are the
only transformation which have completely positive in-
 
1
verse. The non-invertibility of non unitary dynamical +γ0 N σ+ ρS σ− − {σ− σ+ , ρS }
2
maps is connected to the irreversibility of the dynamical
processes that they describe. where
Thanks to the structure of semigroup we can introduce  
the generator of the semigroup L and write
ω0 ω0/2 0
HS = σ3 =
2 0 −ω0/2
Φt = eLt
The rst term describe the non-dissipative evolution
This equation, together with
of the system. The dissipator is composed by two terms,

ρS (t) = LρS (0) the rst describe the de-excitation of the system accom-
panied along with photon emission, while the second de-
yields the quantum markovian master equation scrive the excitation due to photon absorption.

d This equation is in Lindblad form, since it can be ob-

ρS (t) = LρS (t) tained from eq. (4) by setting
dt
This equation is central in the theory, since it describes
p p
A1 = γ0 (N + 1)σ− and A2 = γ0 N σ+
the evolution of the system without an explicit introduc-
tion of the degrees of freedom of the environment. Ob- The constant γ0 is the spontaneous emission rate, while
viously this equation depends on the environment, that N is the Bose-Einstein statistical distribution
enter in the generator L determining its structure.
The physical condition for such an equation to hold is
1
N=
that eβω0 −1

τE  τR b. Example 2. how to check if a master equation can

be put in Lindblad form Let's consider a bit more com-
where τR is the relaxation time of system S. plex equation

d
1. Lindblad theorem ρS = −ı [HS , ρS ] − γ0 M̃ σ+ ρS σ+ − γ0 M̃ ? σ− ρS σ− +
dt

Theorem (4). L is a generator of a completely pos-  

1
itive and trace preserving dynamical semigroup if and +γ0 Ñ σ+ ρS σ− − {σ− σ+ , ρS } +
only if the master equation can be written in the form
2

d
ρS (t) = −ı [HS , ρS (t)] +
 
dt 1
+γ0 (Ñ + 1) σ− ρS σ+ − {σ+ σ− , ρS }
2
X 1n † o
+ Ai ρS (t)A†i − Ai Ai , ρS (t) (4) In this case it's not immediate to say whether the master
i
2 equation can be put in Lindblad form or not. For this
purpose there is a general procedure:
where HS = HS† .
While the rst term of the generator is equal to the 1. Chose a complete orthogonal set of operators in
generator of Von Neumann equation (and so describes the sense of Hilbert-Schmidt scalar product (i.e.
(A, B) =Tr A† B

an unitary evolution), the second term is responsible for ):
the dissipative behaviour, hence is called dissipator.
Now we can consider some simple example of Lindblad 1
Fi i = 0, 1, 2...N 2 − 1 with F0 = √ I
form. N
 7

such that We can determine the generator L from

d 1
n o
T r Fi† Fj = δij and T r {Fi } = 0 i = 1...N 2 − 1 LρS = Φt ρS = lim [Φε ρS − ρS ]
dt t=0
ε→0 ε
performing the calculation
In our example we can choose F1 = σ− , F2 = σ+
2
and F3 = √1 σ3 NX −1  o
1n †
2
LρS = −ı [HS , ρS ]+ aij Fi ρS (t)Fj† − Fj Fi , ρS (t)
2
2. Express the master equation in the form ij=1

d where (aij ) is a submatrix of (cij ), hence is positive, to.

ρS (t) = −ı [HS , ρS (t)] + Thus we can diagonalize it, obtaining the master equa-
dt
tion in Lindblad form.
b. Given a generator in Lindblad form show that Φt =
2
NX −1 
1n † o eLt is a CPT map To show that Φt is trace preserving,
+ aij Fi ρS (t)Fj† − Fj Fi , ρS (t) we can just get
ij
2
T rS {LρS } = 0
where the matrix (aij ) is called Kossakowski ma-
by direct computation, exploiting the cyclic property of
trix.
trace.
In our example we have
The proof that Φt is completely positive is more di-
  cult. Let's divide L in two terms L0 and L1
Ñ + 1 M̃ ? 0
(aij ) = γ0  M̃ Ñ 0 1 Xn † o
L0 ρS = −ı [HS , ρS ] − Ai Ai , ρS
0 0 0 2 i

Ai ρS A†i
X
3. If Kossakowski matrix is positive, the master equa- L 1 ρS = (5)
tion can be put in Lindblad form. We just have i
to nd the basis that diagonalize the Kossakowski
we shall treat the problem by considering the pertur-
matrix and express the master equation in this new
bative expansion of Φt obtained by considering L1 as a
basis.
perturbation of L0 . By assumption, L is the generator
Thus in our case the master equation can be put in
of the semigroup Φt , i.e.
Lindblad form if and only if
d
  2 Φt = LΦt (6)
Ñ Ñ + 1 − M̃ ≥0 dt
If we dene Φ0t as the solution of the equation

d 0
2. Proof of Lindblad theorem in nite dimensional case Φ = L0 Φ0t
dt t
0
with initial condition Φ0 = I, we can express the solution
a. Given a dynamical semigroup Φt = eLt show that
by Dyson's equation
L must be of the given Lindblad form Since Φt is a CPT
ˆ t
map, we know that it can be represented as
Φt = Φ0t + dt1 Φ0t−t1 L1 Φt1
X X 0
Φt ρ S = Ωα (t)ρS Ω†α (t) Ωα (t)Ω†α (t) = IS
which can be written in a series representation as
α α
Φt = Φ0t + Φ0t ∗ L1 Φ0t + Φ0t ∗ L1 Φ0t ∗ L1 Φ0t + . . .
if we expand the operators Ωα as a sum of a basis (in a
Hilbert-Schmidt sense) From this expansion, we nd that the maps Φt are

X CP if and only if L1 and Φ0t are CP. L1 is CP, since its

Ωα = (Fi , Ωα ) Fi expression (eq. 5) is a Kraus representation. To show

i that Φ0t is CP too, we rewrite its generator as

 
†
we get L0 ρS = −ı Hef f ρS − ρS Hef f
2 P †
NX −1 ı
with Hef f = HS − i Ai Ai . Even though Hef f isn't
cij (t)Fi ρS Fj†
2
Φt = 0
self-adjoint, anyway we can write Φt as
ij=0
†
Φ0t ρS = eL0 t ρS = Uef f (t)ρS Uef f (t) Uef f (t) = eiHef f t
X ?
cij (t) = (Fi , Ωα (t)) (Fj , Ωα (t)) c(t) ≥ 0 This is a Kraus decomposition (of just one term) and

α thus Φ0t is CP. 

 8

II. NON MARKOVIAN QUANTUM DYNAMICS B. Nakajima-Zwanzig (NZ) Master Equation

- PROJECTION OPERATOR TECHNIQUES
The idea behind this method is to solve eq. (8) and to
A. Projection operators substitute the solution into eq. (7).
A formal solution of eq. (8) gives
We'll consider projectors P as linear maps ˆ t
0
ρ 7→ ρ = P ρ Qρ(t) = G(t, 0)Qρ(0) + dt1 G(t, t1 )QL(t1 )P ρ(t1 )
0

which are superoperators on the state space of total sys-

tem. The key property of such superoperators is idem- t
ˆ 
potency, i.e. G(t, t1 ) = T exp dt2 QL(t2 )
← t1
2
P =P
with T being the chronological time ordering operator.
←
For each projection, we can dene a complementary Substituting this solution in eq. (7) gives the (rather
projection, namely complicated) Nakajima-Zwanzig master equation

Q≡I−P ˆ t
d
P ρ(t) = dt1 K(t, t1 )P ρ(t1 )+P L(t)P ρ(t)+J (t)Qρ(0)
which is idempotent, too. Furthermore, we have dt 0

P Q = QP = 0 where the memory kernel K(t, t1 ) is

In this section we'll mainly consider the standard pro- K(t, t1 ) = P L(t)G(t, t1 )QL(t1 )P
jection
and the inhomogeneity J (t) is
P ρ = T rE ρ ⊗ ρE
J (t) = P L(t)G(t, 0)Q
where ρE is a xed environmental state. We can easily
show that Nakajima-Zwanzig master equation is an exact inho-
mogeneous integro-dierential equation. Note that ex-
• P2 = P cept for the inhomogeneous term, this equation depends
only on the state of the system. On the contrary the
• P is trace preserving, T r {P ρ} = T r {ρ}
inhomogeneous term contains information of the initial
• P is a CP map state of the bath.
To go further we must introduce some semplications
• T rE {P ρ} = T rE {ρ} = ρS
• For an odd number of L's we suppose that
In this section we'll show how projection techniques can
be used to develop closed equation for non markovian
P L(t1 )L(t2 )L(t3 ) . . . L(t2n+1 )P = 0 n = 0, 1, 2 . . .
open quantum dynamics. The starting point will be the
Von Neumann equation in interaction picture for the to-
• The initial state is such that the inhomogeneous
tal state
term vanishes
d
ρ(t) = −ıα [HI (t), ρ(t)] ≡ L(t)ρ(t) Qρ(0) = 0 ⇐⇒ P ρ(0) = ρ(0)
dt
where α is an expansion parameter that will be used to for the standard projectors this simply means that
develop a perturbation treatment of the problem.
Assuming the projection P to be time indipendent ρ(0) = P ρ(0) = ρS (0) ⊗ ρE
d i.e. the initial state is a product state
P ρ(t) = P L(t)ρ(t) = P L(t) (P ρ(t) + Qρ(t)) =
dt
With such assumption Nakajima-Zwanzig equation is far

= P L(t)P ρ(t) + P L(t)Qρ(t) (7) simpler

ˆ t
d
d P ρ(t) = dt1 K(t, t1 )P ρ(t1 )
Qρ(t) = QL(t)ρ(t) = QL(t) (P ρ(t) + Qρ(t)) = dt 0
dt
ˆ t 
= QL(t)P ρ(t) + QL(t)Qρ(t) (8)
K(t, t1 ) = P L(t)T exp dt2 QL(t2 ) QL(t1 )P
← t1
 9

Now we expand K(t, t1 ) in powers of α where T is the anti-chronological time ordering operator.
→
Thus the operator G(t, t1 ) describes a backward prop-
K(t, t1 ) = K2 (t, t1 ) + K4 (t, t1 ) + . . . agation of the system. In this way we get a dierent
expression for Qρ(t), that is
The rst term is of second order (it contains two L's).
Further, as a results of the rst assumption, we have an
(1 − Σ(t)) Qρ(t) = G(t, 0)Qρ(0) + Σ(t)P ρ(t)
even terms series. The rst two terms of the Nagajima-
Zwanzig (NZ) series are
ˆ t

K2 (t, t1 ) = P L(t)L(t1 )P Σ(t) = dt1 G(t, t1 )QL(t1 )P G(t, t1 )

0

ˆ ˆ The superoperator Σ(t) has two properties:

t t2
K4 (t, t1 ) = dt2 dt3 [P L(t)L(t2 )L(t3 )L(t1 )P + • Σ(0) = 0
t1 t1

• Σ(t)|α=0 = 0
− P L(t)L(t2 )P L(t3 )L(t1 )P ] Thus for short times and for coupling not too large, the
−1
inverse operator [1 − Σ(t)] should exists. In such hy-
pothesis, we get
C. Time-convolutionless (TCL) Master Equation
−1 −1
Qρ(t) = [1 − Σ(t)] Σ(t)P ρ(t)+[1 − Σ(t)] G(t, 0)Qρ(0)
TCL method is a dierent approach. The goal is to
nd a time local master equation (i.e. without a memory Substituting this equation in eq. (7), we get TCL Master

kernel). We show with a heuristic argument why that Equation

should be possible.
d
Suppose there exists a set of dynamical maps Φt also P ρ(t) = K(t)P ρ(t) + J (t)Qρ(0)
for non markovian dynamics
dt
where now we have a time-dependent but local generator
ρS (t) = Φt ρS (0)
−1
K(t) = P L(t) [1 − Σ(t)] P
Even though this set hasn't the semigroup structure
(since otherwise it would satisfy the hypothesis of Lind- and an inhomogeneous term
blad theorem), we can dene it anyway. Thus we can −1
write
J (t) = P L(t) [1 − Σ(t)] G(t, 0)Q

d With the usual semplications, the master equation

ρS (t) = Φ̇t ρS (0) = Φ̇t Φ−1 ρS (t) becomes
dt | {zt }
K(t)
d
P ρ(t) = K(t)P ρ(t)
where K(t) is a time-local generator.
dt
Now let's derive such an equation exploiting projection As for NZ eq., we can try a perturbation expansion of
operator techniques. As we have done in deriving NZ the generator K(t) in powers of α. Also in this case we
master equation, we have to solve eq. (8) for Qρ(t). get an even series starting from second order. The rst
As in NZ case, we have two terms of the TLC expansion are

ˆ t ˆ t
Qρ(t) = G(t, 0)Qρ(0) + dt1 G(t, t1 )QL(t1 )P ρ(t1 ) K2 (t) = dt1 P L(t)L(t1 )P
0 0

ˆ t  ˆ t ˆ t1 ˆ t2
G(t, t1 ) = T exp dt2 QL(t2 ) K4 (t) = dt1 dt2 dt3 [P L(t)L(t1 )L(t2 )L(t3 )P +
← t1 0 0 0

This time, we set

−P L(t)L(t1 )P L(t2 )L(t3 )P − P L(t)L(t2 )P L(t1 )L(t3 )P +
ρ(t1 ) = G(t, t1 )ρ(t)
−P L(t)L(t3 )P L(t1 )L(t2 )P ]
 ˆ t 
In general the n-th order term of the perturbative ex-
G(t, t1 ) = T exp − dt2 QL(t2 )
→ t1 pansion can be written down easily following a method
 10

derived by Van Kampen. The general expression of the unless the time dierence are less than the relaxation
term Kn (t) is time of the environment

ˆ t ˆ t1 ˆ tn−2
|ti − tj | . τE ∀i, j = 0, 1, . . . n − 1
dt1 dt2 . . . dtn−1 hhL(t)L(t1 ) . . . L(tn−1 )ii
0 0 0
thus the n-th order term can be approximated by
where hhL(t)L(t1 )L(t2 ) . . . L(tn−1 )ii is the ordered cu-
mulant, which has an expression like Kn (t) ∼ αn τEn−1
hhL(t)L(t1 )L(t2 ) . . . L(tn−1 )ii = in particular for even order terms

2(n−1)
K2n (t) ∼ α2n τE2n−1 = α2 τE (ατE )
X
= (−1)q P L(t)L(t1 ) . . . L(ti )P L(tj ) . . . L(tn−1 )P

More precisely, the ordered cumulant is the sum of every The second order, in particular, is

dierent term that can be obtained with the following

1
procedure: K2n (t) ∼ α2 τE = γ =
τR
• write down the expression P LLL
| {z. . . L} P where we have introduced γ, the relaxation rate in 2nd
n f actors
order and its invers τR , the relaxation time in 2nd order.
• insert q factors of P With this denition, we can approximate the even order
term as
P LLL . . . LP → P LLP L . . . LP
 n−1
τE
• give times argument such that (1) the rst Lhas K2n ∼ γ
τR
argument t; (2) in between two P 's the time argu-
ments are ordered from right to left and we can see that the expansion is in powers of τE/τR
. In this form, it is now evident we can apply Markov
It's easy to show that using the usual semplications the
approximation when τE/τR is small enough that we can
ordered cumulant for n = 4 gives the 4-th order TCL
just take the leading term of the expansion, since any
term.
power higher than that is negligible. Thus the standard
Let's compare the second order TLC and NZ master
Markov condition is
equation:
ˆ t τE  τR
d
P ρ(t) = dt1 P L(t)L(t1 )P ρ(t) (T LC)
dt 0

ˆ D. General classication of master equations

t
d
P ρ(t) = dt1 P L(t)L(t1 )P ρ(t1 ) (N Z)
dt 0 The most general form of TCL master equation is

The two master equations dier just by the dependence d

of ρ, which is local in TLC form and non-local in NZ one. ρS (t) = K(t)ρS (t) = ı [HS (t), ρS (t)] +
dt
As we expect, the markovian master equation can be
obtained from TLC and NZ master equation by imposing
 o
suitable approximations. In particular they are 1n †
γi (t) Ai (t)ρS (t)A†i (t) −
X
+ Ai (t)Ai (t), ρS (t)
2
• Born approximation: it's taking just the second or- i
der term of the perturbative expansion
where HS is a time dependent hamiltonian, Ai are time
• Markov approximation: it's the replacement of the dependent system operators and γi (t) are real functions
higher integration limit of the integral by ∞ of time. This form is much more general than Lindblad
form. It guarantees only that K(t) preserves hermiticity
• Rotating wave approximation
and trace, but not positivity.
This form let us dene a very general classication of
Let's conclude our discussion with some general remark
quantum dynamical processes.
on higher order terms. It can be shown that higher order
terms are bounded in time, thus in some sense the per-
c. 1) Markovian HS , Ai and γi are all time indipen-

turbative series uniformly converge with respect to time.

dent and γi is positive

More precisely, we have that By Lindblad theorem we know that this master equa-
tion dene a completely positive dynamical semigroup
hhL(t)L(t1 )L(t2 ) . . . L(tn−1 )ii ≈ 0 Φt
 11

d. 2) Time-dependent Markovian It's an extension From these assumptions we have the following implica-
of Markovian case in which γi is positive but HS , Ai tions
and γi can be time dependent. The set of Φt is not a
semigroup, however one can show that for each time t Φt • P maps product states to product states, namely

is a CPT map. Further one can substitute Φt with the

P (A ⊗ B) = A ⊗ ΛB
more general propagator • P maps separable states to separable states, i.e.
ˆ t  X X
V (t, t1 ) = T exp dt2 K(t2 ) ρ= pi ρiS ⊗ ρiE ⇒ P ρ = pi ρiS ⊗ ΛρiE
← t1
In other words, P doesn't create entanglement. This is
which satisfy
essentialy the reason of the general form chosen.

V (t, t1 ) = V (t, t1 )V (t1 , t2 ) t ≥ t1 ≥ t2

B. Representation theorem
e. 3) Non Markovian The more general case, with
H S , Ai and γi time indipendent and γi which can be
Now we want to have a representation of a projection
negative for some times. In this case general conditions
for which the properties from 1. to 3. hold. We don't
that guarantees complete positivity are not known. For
want to use an expansion like Kraus representation, since
this reason it's not possible to predict in general if a non
we want the projection representation to include the stan-
markovian model could yield unphysical results (negative
dard projection as a trivial case.
probability).
Theorem (5). A linear map P has the properties 1.
- 3. if and only if it can be written in the form

III. CORRELATED PROJECTION OPERATOR X

Pρ = T rE {Ai ρ} ⊗ Bi
i
In the derivation of NZ or TCL master equation we
made no assumptions on the projection P other than where {Ai } and {Bi } are linear indipendent sets of ob-
P2 = P . Thus the topic of the following discussion is servables with the properties
the extension of the same projection techniques to other 2
1. T rE {Bi Aj } = δij (P = P)
projection, dierent from standard projection.
P
2. i (T rE Bi )Ai = IE (trace preserving)

A. General requirements
P T
3. i Ai ⊗ B i ≥ 0 (complete
positivity)

We make some general assumptions on the form of the We omit the proof and give some example of projections
projection of such form.
f. Ex. 1 Take A = IE and B = ρE where ρE is a
1. P has property of projection operator: P2 = P xed environmental state. In this way we get the stan-
dard projection
2. P is a quantum operation, i.e. a completely posi-
tive and trace preserving map
P ρ = (T rE ρ) ⊗ ρE
in particular P maps physical states to physical
states g. Ex. 2 Take Ai = Πi where the projectors Πi
come from an orthogonal decomposition of unity
ρ ≥ 0 ⇒ Pρ ≥ 0 X
Πi Πj = δij Πi Πi = IE
i
T rρ = 1 ⇒ T r {P ρ} = 1
and take

note that positivity would be enough for this pur- Πi ρ E Πi

pose (because we are explitely considering total Bi =
T r {Πi ρE }
states), but complete positivity is required for fu-
ture convenience Thus we get

3. P has a form like

X Πi ρ E Πi
Pρ = T rE {Πi ρ} ⊗
i
T r {Πi ρE }
P = IS ⊗ Λ
In some physical applications, ρE may be a Gibbs state
where Λ is a CPT map and a projection in the space and Πi some projections onto certain energy bands of a
2
of environmental states, Λ = Λ structured reservoir.
 12

C. NZ and TCL expansion with correlated D. Generalized Lindblad structure

projections
Consider a master equation of the form
First of all, we call the projections we have studied so
far correlated projections. In fact the image of a state
n
ρ(t) is d X
ρi (t) = Kij (t)ρj (t) ≡ Kit (ρ1 . . . ρn ) i = 1...n
X X dt j=1
P ρ(t) = T rE {Ai ρ(t)} ⊗Bi = ρi (t) ⊗ Bi t
Assume that Ki = Ki is time indipendent.
| {z }
i i
ρi Now consider a density matrix ρ on an extended
n
Hilbert space Hext = HS ⊗ C dened by
which is a (classically) correlated state.
With such a projection, secondo order TCL becomes X
ρ= ρi ⊗ |ii hi|
X Xˆ t
ρ˙i (t) ⊗ Bi = dt1 T rE {Ai L(t)L(t1 )P ρ(t)} ⊗ Bi
i i 0 such that the density matrix has a bloch diagonal struc-
ture
Using T rE {Bi Aj } we can nd

ˆ ρ1 0
 
t
ρ˙i (t) = dt1 T rE {Ai L(t)L(t1 )P ρ(t)} = ρ2
ρ=
 
0 ... 
0 ρn
Xˆ t
= dt1 T rE {Ai L(t)L(t1 )ρj (t) ⊗ Bj } Suppose there exists a Lindblad generator on extended
0
j space such that

Thus we have a set of coupled master equations for the ! !

components ρi (t). We briey write this set of equations
X X
L ρi ⊗ |ii hi| = Ki (ρ1 . . . ρn ) ⊗ |ii hi|
as
i i
d X
ρi (t) = Kij (t)ρj (t)
dt j !
X X
Lt
The corresponding second order NZ master equation ρi ⊗ |ii hi| = e ρi (0) ⊗ |ii hi|
has form like i i
ˆ t
d X
ρi (t) = dt Kij (t, t1 )ρj (t1 ) In this way we would have embedded a non markovian
dt j 0 dynamics into a Lindblad dynamics on extended space,
which preserves the bloch diagonal structure.
One interesting thing is that there are dierent initial
Such Lindblad generator on extended space exists if
conditions for TCL and NZ not to have the inhomoge-
and only if the propagator Ki can be written as
neous term. In fact for a correlated projection the con-
dition
Ki (ρ1 . . . ρn ) = −ı H i , ρi +
 

Qρ(0) = 0 ⇐⇒ P ρ(0) = ρ(0)

become
X 1 h ij† ij i

X + Rλij ρj Rλij† − Rλ Rλ , ρi
ρ(0) = ρi (0) ⊗ Bi 2
jλ

This shows that the initial state can be correlated with-

out having inhomogeneity in the master equation. with H i = H i† and arbitrary Rλij