Non-Markovian Quantum State Diffusion for Spin Environments

Valentin Link and Walter T Strunz

Institut für Theoretische Physik, Technische Universität Dresden, D-01062, Dresden, Germany

Kimmo Luoma
Laboratory of Quantum Optics, Department of Physics and Astronomy,
University of Turku, FI-20014, Turun yliopisto, Finland∗
(Dated: May 1, 2023)
We introduce an exact open system method to describe the dynamics of quantum systems that
are strongly coupled to specific types of environments comprising of spins, such as central spin
systems. Our theory is similar to the established non-Markovian quantum state diffusion (NMQSD)
theory, but for a spin bath instead of a Gaussian bath. The method allows us to represent the
arXiv:2203.02417v2 [quant-ph] 28 Apr 2023

time-evolved reduced state of the system as an ensemble average of stochastically evolving pure
states. We present a comprehensive theory for arbitrary linear spin environments at both zero and
finite temperatures. Furthermore, we introduce a hierarchical expansion method that enables the
numerical computation of the time evolution of the stochastic pure states, facilitating a numerical
solution of the open system problem in relevant strong coupling regimes.

I. INTRODUCTION and numerical methods, and have applications in nuclear

Open quantum system theory is concerned with in-
vestigating the dynamics of a quantum system in con-
tact with a large environment. This presents a chal-
lenging computational problem that requires the use of
suitable approximations and efficient numerical meth-
ods. As research in this field advances, there has been a
steady increase in the ability to address complex prob-
lems where the system is strongly coupled to structured
environments [1]. However, as the strength of the system-
environment interaction increases, the validity of the
standard assumptions that underlie most open system
methods may become questionable. Conversely, relaxing
these assumptions may open up new possibilities for us-
ing open system theory to address problems that were
previously only accessible through many-body methods
[2–5].
A crucial requirement for the majority of open quan-
tum system methods is that the response of the envi-
ronment to the system is Gaussian. This is a well-
founded assumption because most commonly studied
quantum environments are either linearly coupled Gaus-
sian bosonic systems, as is typical for instance in quan-
tum optics [6], or they consist of a large number of weakly
coupled degrees of freedom so that basic central limit
theorem arguments apply, yielding an effective Gaussian
response [7, 8]. However, a Gaussian description may
loose validity for a variety of strongly coupled many-body
systems, even if their structure suggests that they could
be treated as open systems [9]. One such example are
central spin models, which describe the interaction be-
tween a central spin and an ensemble of independent en-
vironmental spins. These models have been extensively
studied within many-body theory using both analytical
∗ ktluom@utu.fi
magnetic resonance spectroscopy, sensing, and solid-state
quantum information platforms [10–12]. Central spin
problems have a suitable structure for open system the-
ory, with the environmental spins being identified as a
bath for the central spin. However, standard open sys-
tem methods cannot be applied to these systems because
a bath consisting of spins does not follow Gaussian statis-
tics. This holds for almost all established methods such
as the hierarchy of equations of motion (HEOM) [13], the
quasi-adiabatic path integral (QUAPI/TEMPO) [14, 15],
or time-evolving density matrices using orthogonal poly-
nomials (TEDOPA) [16]. Only recently, some new ap-
proaches have been proposed that enable a treatment of
non-Gaussian baths in an open system framework, an ex-
tension of HEOM [17, 18] and a process tensor method
based on a matrix product operator representation [19].
It is the goal of this work to generalize another open
system theory, the non-Markovian quantum state diffu-
sion (NMQSD) [20], in order to describe open system dy-
namics with spin baths. In contrast to other approaches
that work with mixed states, NMQSD allows for a com-
putation of open system dynamics via stochastic sam-
pling of pure state trajectories. In the case of Gaussian
baths, quantum state diffusion is known to have certain
technical and analytical advantages over mixed state de-
scriptions. In particular, it is based on a propagation of
state vectors rather than operators, which can be a sig-
nificant advantage in certain systems [21]. The method
has been utilized successfully both for simulation of non-
Markovian dynamics [22–25], and as a versatile tool to
derive non-Markovian master equations or to treat inte-
grable problems [26, 27]. In this paper we derive a non-
trivial generalization of this framework for open system
problems with spin baths, such as central spin problems,
at zero and finite temperature. We derive exact evolution
equations for the NMQSD trajectories by utilizing a pro-
jection formalism that was introduced in Ref. [28]. Fur-
ther, we show that it is possible to numerically compute
 2

the solution of this equation with a hierarchy method, In the following we always consider an interaction pic-
thus allowing us to simulate the reduced system dynam- ture with respect to the free Hamiltonian of the bath
ics in spin bath problems via Monte-Carlo sampling of spins, which results in the following Hamiltonian
stochastic pure state trajectories.
N
The remainder of this work is structured as follows. We X
~ · Oλ (t)J~λ .
first introduce the open system model in Sec. II and dis- H(t) = HS + gλ L (4)
λ=1
cuss differences to the well known Gaussian bath model.
In the technical part of this work, we first give a brief In this expression, the orthogonal matrices Oλ (t) ∈
review on spin coherent states in Sec. III, as these will SO(3) describe the bare rotation of the spins around the
play an essential role in the following. Readers familiar z-axis
with spin coherent states may directly switch to Sec. IV,  
where we develop the extension of NMQSD theory, the cos ωλ t sin ωλ t 0
main result of this work. As a demonstration we present Oλ (t) = − sin ωλ t cos ωλ t 0 . (5)
an application to the solvable pure dephasing model in 0 0 1
Sec. V. In Sec. VI we introduce a method which allows us
to numerically solve the dynamics of individual NMQSD Assuming a pure initial state of the system and a zero
trajectories and thus, in principle, to compute the re- temperature spin bath, the full system and bath state
duced dynamics of open systems with spin baths. Fi- |Ψ(t)i obeys unitary evolution from the Schrödinger
nally, we present our conclusions in Sec. VII. Various equation
longer derivations are provided in the appendix.
∂t |Ψ(t)i = −iH(t)|Ψ(t)i. (6)

Explicitly, the zero temperature bath state is simply the

II. SPIN BATH MODEL
The spin bath model we like to consider consists of an
arbitrary quantum system S interacting linearly with an
ensemble of N independent bath spins J~λ . The Hamilto-
nian reads explicitly
N N
X X
H = HS + ~ · J~λ +
gλ L ωλ (j − Jλz ),
λ=1 λ=1
where HS is the system Hamiltonian and L ~ is a vector of
hermitian operators (coupling operators) in the system
Hilbert space. The dot product denotes the standard
euclidian scalar product. Further, we assume that the
spins have a fixed length j = n2 where n ∈ N (note that
H preserves Jλ2 ). The spin operators Jλa satisfy the usual
algebra
[Jλa , Jλb ] = iεabc Jλc
and spin operators acting on different spins commute.
Throughout this paper, roman indices indicating the spin
directions x, y, z are written as upper indices for nota-
tional convenience, and summation is implied when such
an index appears twice. For fixed length spins an addi-
tional relation J~λ · J~λ = j(j + 1) holds, and the Hilbert
state where all spins are pointing up |0i ≡ |j, ji⊗N , and
the full initial condition is |Ψ(0)i = |ψ(0)i|0i with an ar-
bitrary pure system state |ψ(0)i. We show later that the
finite temperature problem can be reduced to zero tem-
perature by introducing additional stochastic thermal
fluctuations to the Hamiltonian. Solving the Schrödinger
equation (6) is hard, because the size of the system and
bath Hilbert space grows exponentially with the number
(1)
of bath spins. Open system methods aim to resolve this
issue by finding a description of the reduced dynamics
within the system Hilbert space, taking the form of a
closed theory for the reduced system state
ρS (t) = trE |Ψ(t)ihΨ(t)|. (7)
One can compare the spin bath model (4) to the Gaus-
sian bath model that is usually considered in open quan-
tum systems. It is well known that any environment
(2) with a Gaussian influence functional can be represented
in terms of independent bosonic modes, so that the gen-
eral open system Hamiltonian for a stationary Gaussian
bath can be written as [33]
Z X
H = HS + dω Li (gik (ω)e−iωt bk (ω) + h.c.), (8)
i,k

space of a single spin is spanned by the 2j + 1 eigenstates

where [bk (ω), b†l (ω 0 )] = δkl δ(ω − ω 0 ). Crucially, the bath
of J z
coupling operators
Jλz |j, miλ = m|j, miλ . (3) Z X
Hamiltonian (1) includes all common spin bath models, Bi (t) = dω (gik (ω)e−iωt bk (ω) + h.c.) (9)
k
for instance condensed phase environments or decoher-
ence models for quantum computing [4, 10, 11, 18]. Vari- obey Gaussian statistics in the sense that all higher mo-
ants of model (1) have been studied previously with dif- ments can be obtained via Wick’s theorem from the sec-
ferent methods such as the multi configurational time- ond moment
dependent Hartree-Fock method [29, 30], a dissipation
formalism [31], analytically [4, 5] or perturbatively [32]. hvac|Bi (t)Bj (s)|vaci = αij (t − s), (10)

where αij (t) is the bath correlation function and |vaci is
the bare vacuum state of the bath. The spin bath model
cannot be mapped to a Gaussian bath model because
higher order cumulants of the bath coupling operators
N ~ ni = j|~ni.
~
X ~n · J|~
B(t) = gλ Oλ (t)J~λ (11)
λ=1
do not vanish (this is a direct consequence of the nontrival
algebra (2)). For a strong system and bath coupling it
may not be sufficient to treat the spin bath within a
Gaussian approximation in order to accurately describe
the system dynamics, see for instance the analysis in
Ref. [9].
Non-Markovian quantum state diffusion (NMQSD) is
an established theoretical framework to describe dynam-
ics due to Gaussian environments, i.e. Hamiltonian (8).
It is a generalization of trajectory methods, such as quan-
tum state diffusion or quantum jumps, to non-Markovian
systems [34]. NMQSD provides an unraveling of the sys-
tem dynamics in terms of a stochastic process in the sys-
tem Hilbert space |ψ(t)i, such that the average repro-
duces the mixed system state
ρ(t) = M [|ψ(t)ihψ(t)|] . (12)
The expectation value M [...] above is to be understood
with respect to all stochastic realizations of the process
|ψ(t)i, and can in practice be evaluated using Monte
Carlo sampling of trajectories. This paper presents a gen-
eralization of the NMQSD method to address spin bath
problems described by the equation (4). Unlike most es-
tablished open systems methods, this new approach can
handle spin baths in an exact manner. A detailed deriva-
tion of the theory is provided below. The main results
include an exact unraveling of the reduced system state
in terms of normalized pure states (Eq. (26)) obeying a
closed evolution equation (Eq. (32)).
III. PROPERTIES OF SPIN COHERENT
STATES
The microscopic derivation of NMQSD relies on coher-
ent states, which are also a crucial element in the general-
ization to spin baths that we aim to establish. Therefore,
it is important to briefly introduce the essential proper-
ties of spin coherent states. Spin coherent states can be
introduced in various ways, depending on the chosen pa-
rameterization [35–37]. In general, a spin coherent state
refers to a rotation of the eigenstate of J z with a maxi-
mum eigenvalue of m = j
|ψSC i = R|j, ji. (13)
R is a unitary rotation operator. This implicit definition
can be made more explicit by realizing that |ψSC i is an
eigenstate of the rotation of J z
RJ z R† |ψSC i = RJ z |j, ji = j|ψSC i. (14)
3
Since R is a rotation there exists a unit vector ~n ∈ S 2
~ Thus, up to a phase, any spin
such that RJ z R† = ~n · J.
coherent state can be characterized by a point on the
sphere |ψSC i ≡ |~ni, with
(15)
For a unique definition we have to assume some con-
vention, or gauge, in order to specify the phase of
these states. This can be achieved for example in the
Bargmann parameterization of spin coherent states, re-
lated to the stereographic projection of the sphere onto
the complex plane by z = tan θ2 eiφ , where θ and φ are
the polar angels pointing towards ~n. In this parame-
terization a (unnormalized) spin coherent state can be
explicitly written as [35, 37]
~
− hz|J|zi
|zi = ezJ |j, ji, j~n = . (16)
hz|zi
This represents an analogy to bosonic Bargmann coher-
ent states [38]. The above definition already specifies a
phase of the states. However, the spin coherent state
|j, −ji, corresponding to ~n = −~ez which is orthogonal to
|j, ji, is not included, as this would correspond to |z| = ∞
and the phase at this point is ill-defined. In fact, due to
the nontrivial topology of the space of pure spin coher-
ent states, it is impossible to find a smooth global phase
gauge [39]. While the overlap of two spin coherent states
depends on the phases, the absolute value of this overlap
is gauge independent and given as
2j
1 + ~n · ~n0

0 2
|h~n|~n i| = , (17)
2
where |~ni and |~n0 i are assumed to be normalized. While
any set of 2j +1 distinct coherent states already spans the
full spin Hilbert space [40], the standard overcomplete-
ness relation from the general group theoretical construc-
tion of coherent states is most commonly used [41]
Z
1= dΩ|~nih~n|, (18)
S2
with the proper uniform measure on S 2
2j + 1 3 2j + 1
dΩ = d n δ(|~n| − 1) = dφ dθ sin θ. (19)
4π 4π
Similar to NMQSD for bosonic baths, this identity rela-
tion proves to be an elegant tool for developing a stochas-
tic unraveling of open system dynamics from the micro-
scopic model of system and bath.
IV. SPIN NMQSD THEORY
We consider model (1) and wish to establish a stochas-
tic theory similar to NMQSD based on spin coherent
states and in particular on the completeness relation (18).
 4

We will first consider a zero temperature bath and elab- When inserted in (21), this yields an average of normal-
orate on the finite temperature case later. Our starting ized trajectories with time independent weights
point is the partial trace with respect to the environment
expressed in terms of coherent states: trE |Ψ(t)ihΨ(t)| =
n0 (t)|Ψ(t)ihΨ(t)|~
h~ n0 (t)i (26)
! Z
N
dµ(~n) .
Z
n0 (t)ih~
n0 (t)|Ψ(t)i
Y
trE |Ψ(t)ihΨ(t)| = dΩλ h~ n|Ψ(t)ihΨ(t)|~ni (20) hΨ(t)|~
λ=1
In equation (21) each pure state trajectory belongs to a
The bold vector ~ n = (~nλ ) is a shorthand notation for fixed set of coherent state labels and has a time depen-
the coherent state labels of all bath spins, and dΩλ is as dent weight. Through the transformation (25) resulting
in (19). The system Hilbert space vectors h~ n|Ψ(t)i can in equation (26) the coherent state labels now follow the
be used as a definition for pure state trajectories. How- evolution of the characteristic curves of the environment
ever, they are not normalized which is a severe issue for Q-function, but the weight of each trajectory becomes
Monte Carlo sampling of trajectories. After a short time, time independent. For a zero temperature bath initial
a large majority of trajectories may have a vanishing con- the initial conditions of the coherent state labels ~nλ are
tribution to the average whereas single, rare trajectories distributed according to the non-Gaussian measure
contribute with a significant weight. To find an average
over normalized states we divide by the norm to find YN
!
dµ(~
n) = dΩλ Q(0, ~ n)
trE |Ψ(t)ihΨ(t)| =
λ=1
N
! (27)
h~
n|Ψ(t)ihΨ(t)|~
ni (21)
Z N 2j
1 + nzλ
Y 
dΩλ Q(t, ~
n) ,
Y
hΨ(t)|~
nih~
n|Ψ(t)i = dΩλ .
λ=1 2
λ=1
where we defined the function Q(t, ~ n) =
hΨ(t)|~nih~n|Ψ(t)i. In fact, this squared norm of Defining an average M [...] as the integration with respect
the system Hilbert space vectors h~ n|Ψ(t)i is the spin to dµ(~
n) we can obtain the reduced state of the system
Q-function (phase space distribution) of the reduced via
state of the environment. To proceed, we can derive  0
h~
n (t)|Ψ(t)ihΨ(t)|~ n0 (t)i

an evolution equation for this function. The derivation ρ(t) = M , (28)
is provided explicitly in appendix A. Due to the linear hΨ(t)|~ n0 (t)ih~
n0 (t)|Ψ(t)i
coupling and linear bath Hamiltonian, the Q-function
as in Eq. (12). Remarkably, this allows us to determine
satisfies a Liouville equation
the reduced state with Monte Carlo sampling of normal-
N
X ized pure states (‘stochastic states’) in the system Hilbert
n) = −
∂t Q(t, ~ ~λ·A
∇ ~ λ (t, ~nλ )Q(t, ~
n), (22) space, because every sample contributes to the mixed
λ=1 state evolution with the same weight.
∂
It remains to find an evolution equation for the
with ∇aλ = ∂na and the ‘drift’ term NMQSD trajectories |ψ(t, ~ n)i = h~ n0 (t)|Ψ(t)i. For this
λ
a projection formalism in terms of a Feshbach partition-
~ λ (t, ~nλ ) = gλ Oλ (t)T hLi(t)
A ~ × ~nλ . (23) ing can be used. This idea was introduced in Ref. [28].
It is based on finding a projection operator P (t) in the
~
hLi(t) = hΨ(t)|~ ~ n|Ψ(t)i/hΨ(t)|~
niLh~ nih~n|Ψ(t)i is the ex- full Hilbert space such that it extracts a single stochastic
pectation value of the coupling operators with respect to state and maps the initial state to itself. In our case the
the system state h~ n|Ψ(t)i. As should be, the character- requirements are explicitly
istic curves of this first order partial differential equation
remain on the sphere. The solution is given by transport P (t)|Ψ(t)i ∝ |ψ(t, ~
n)i, P (0)|0i = |0i. (29)
of the initial condition along the characteristic curves
As shown in Ref. [28], applying a standard Feshbach
~n˙ 0λ (t) = A
~ λ (t, ~n0λ (t)), ~n0λ (0) = ~nλ . (24)
method with such a projector gives a closed evolution
One can identify the classical equation of motion for the equation for the stochastic state. When defining this pro-
bath spins due to coupling to the system state h~ n|Ψ(t)i. jector in the spin case, we encounter the problem that
It is surprising that these classical equations appear in the spin coherent state |~ n0 (t)i is not smoothly defined
0
this exact theory even though the bath evolution is not for all values of ~nλ (t), because no smooth global phase
Gaussian. Using the method of characteristics we can gauge exists. To circumvent this issue, we define a time-
explicitly write the Q-function as dependent rotation operator as
Z N
! N
Y X
Q(t, ~
n) = dΩ00λ δ(~n00λ − ~n0λ (t)) Q(0, ~
n00 ). (25) ~
∂t R(t) = −ihLi(t) · gλ Oλ (t)J~λ R(t), (30)
λ=1 λ=1
 5

with initial condition R(0) = 1. Since R(t) is a rotation Note that this termal state can also be expressed in terms
it acts on the spin operators as R† (t)J~λ R(t) = Sλ (t)J~λ , of a distribution of coherent states by using the so-called
where Sλ (t) ∈ SO(3) are orthogonal matrices. Note that P representation
this rotation propagates the coherent state |~ ni to the Z N
!
n0 (t)i = R(t)|~
shifted state |~ ni or equivalently Sλ (t)~nλ =
Y
ρβ = dΩλ Pβ (m)|~ mih
~ m|,
~ (35)
~n0λ (t). Because the dynamics of the phase of the shifted λ=1
state is smoothly specified by the rotation operator R(t),
the phase singularity issue is circumvented. We note in where m~ = (m~ λ ) denotes a set of coherent state labels
passing that by writing |~ n0 (t)i we abuse the notation for the bath spins. The P -function can be explicitly
because the phase of this state is not determined by the computed and reads
orientations ~nλ alone. With this construction we can
make the following choice for a time dependent projector N βωλ −2j−2
1 Y βωλ 1 + eβωλ
 
z1−e
Pβ (m)
~ = e + mλ .
n|R† (t)
R(t)|0ih~ Z 2 2
λ=1
P (t) = . (31) (36)
h~
n|0i
Since this function is positive the thermal initial state
This operator satisfies all desired properties required in
can be realized by averaging over pure state dynamics
the stochastic Feshbach theory. P (t) is well defined by
corresponding to coherent initial states [23, 28]. Note
construction because spin coherent states orthogonal to
that coherent initial states are just rotations of the zero
the initial state |0i have zero probability in the average
temperature bath state. A fixed rotation of the bath
(26). It is a projector for all times P (t)2 = P (t) and the
spins does not modify the linear form of the Hamiltonian,
relations (29) hold. Using this operator, the Feshbach
so that this can be captured by our theory. To be more
theory can be readily applied to find evolution equations
explicit, let Mβ [...] denote the average of coherent state
for the stochastic state. The derivation is provided in
labels m
~ with respect to the thermal P -function, so that
appendix B and the resulting equation reads
Mβ [|mih
~ m|]~ = ρβ . Then, the full system and bath state
∂t |ψ(t)i = can be obtained in the following way as an average over
dynamics with coherent initial states
− iHS |ψ(t)i
~ † (t) . (37)
 
N ρ(t) = Mβ U (t)|ψ(0)ihψ(0)| ⊗ |mih ~ m|U
~
X h~nλ |J~λ |j, ji
− i∆L(t) · gλ Oλ (t)Sλ (t) |ψ(t)i (32) U (t) denotes the time evolution operator for the full
h~nλ |j, ji
λ=1 system-and-bath state. By moving to a rotated frame,
Z t
the coherent initial conditions can be set to the zero
− ds K(t, s)|ψ(s)i,
0 temperature bath state. In particular, let R(m)~ be a
rotation operator which satisfies R(m)|0i
~ = |mi.
~ Then
where we introduced the abbreviation ∆L(t) ~ = L ~ − the Hamiltonian in the rotated frame is transformed as
~
hLi(t). The explicit form of the kernel operator K(t, s) ~ = R† (m)H(t)R(
H(t, m) ~ m)
~ and reads
is given in the appendix. In this equation the non- N
X
Markovian nature of the bath is apparent via the convo- H(t, m)
~ = HS + ~ · Oλ (t)M (m
gλ L ~ λ )J~λ , (38)
lution term which is not local in time. In general, the ker- λ=1
nel operator is difficult to determine except within per-
turbation theory. In fact, while the projector approach where M (m ~ λ ) ∈ SO(3) satisfies M (m ~ λ )~ez = m
~ λ . In
is very transparent and elegant, the hierarchy expansion summary, to realize a thermal initial condition of the
discussed in a later section may be more relevant in prac- bath one can propagate the state with zero temperature
tice, as it can be directly employed to numerically solve initial condition for a set of rotated Hamiltonians (38),
for the stochastic state. where the rotation is drawn randomly according to the P -
Finally, we want to show how to describe thermal ini- function (36). In practice this can be achieved by Monte
tial states within the same framework. At finite temper- Carlo sampling of the labels m ~ λ . Crucially, the form
ature kB T = 1/β the initial state of the bath in Hamil- of Hamiltonian (38) is identical to (4). Thus, all results
tonian (1) is given by from the zero temperature case can still be utilized. Since
NMQSD is a stochastic method itself, such a stochastic
N realization of the thermal initial state poses no further
1 O −βωλ (j−Jλz )
ρβ = e (33) complication [23].
Z
λ=1

and the partition function reads V. PURE DEPHASING

Y 1 − e−βωλ (2j+1)
Z= . (34) To demonstrate the validity of the spin NMQSD theory
1 − e−βωλ we first consider a solvable model, namely pure dephasing
λ
 6

in a spin bath, i.e. Hamiltonian (1) with [Lx , HS ] = 0 1.00

and Ly = Lz = 0 [18, 42]. In this model the evolution (a) no. samples
0.75 1
of the full system and bath state can easily be derived
10
analytically, as shown in appendix C. Therefore, we can 0.50 100
use this as a test for the unraveling. For pure dephasing

hσx i
exact
we do not need to solve the intricate equation of motion 0.25
(32) to find the stochastic states. Instead, we can simply
0.00
extract these states directly from the exact system and
bath state, which reads −0.25
X N
O 0 1 2 3 4
|Ψ(t)i = e−iεn t hεn |ψS (0)i|εn i |ψλ,n (t)i. (39) εt
n λ=1
1.50
The summation above is with respect to all system en- (b) 1 no. samples

distance to exact state

ergies HS |εn i = εn |εn i and the states |ψλ,n (t)i are spin 1.25 1
coherent states of the individual environment spins. The 10
1.00
exact expression for a single NMQSD trajectory in the 100
model can be obtained from projection of the full state 0.75 1000
onto the co-moving coherent state h~ n0 (t)| (see appendix
0.50
C). The simplest nontrivial example that we can con-
sider is a two level system coupled to a spin bath. The 0.25
Hamiltonian reads
0.00
N
X N
X 0 1 2 3 4
H = εσz + 2σz gλ Jλx + ωλ (j − Jλz ), (40) εt
λ=1 λ=1
FIG. 1. Pure dephasing dynamics (Hamiltonian (40), j =
where σz = −|0ih0| + |1ih1|. To specify the parameters of 1
1/2) in a spin bath with parameters ωλ , gλ from a discretized
the bath we use a discretization of a continuous spectral Ohmic SD (Eq. (42), s = 1, α = 0.5ε, ωc = 10ε) with N = 100
density [9]. The spectral density of the bath is given by frequencies chosen linear
√ from 0 to 60ε. The initial state of the
system is (|0i + |1i)/ 2. (a) Estimated evolution of hσx i for
N different numbers of NMQSD trajectories in the average (26).
1X 2
J(ω) = gλ δ(ω − ωλ ), (41) (b) Convergence of spin NMQSD to the exact solution. The
π Hilbert-Schmidt distance between the exact reduced state and
λ=1
the ensemble average vanishes for a large number of trajecto-
which is the fourier transform of the bath correlation ries.
function α(t). As an example we choose couplings gλ
and frequencies ωλ according to a discretized Ohmic en-
vironment with spectral density The idea is similar to the hierarchy of pure states (HOPS)
approach in Gaussian NMQSD [22–24, 44, 45] or the re-
J(ω) = 2παωe−ω/ωc Θ(ω), (42) lated hierarchy of equations of motions (HEOM) method
[13, 46–48]. We would like to emphasize from the begin-
see Ref. [43] for details on discretization of continuous ning that the hierarchy expansion for a spin bath could
spectral densities. With our example we confirm that the potentially demand considerably more resources in com-
ensemble average of spin NMQSD trajectories converges parison to Gaussian baths. This is primarily because
to the exact reduced state as expected, see Fig. 1. For individual spins cannot be combined to form a single,
the parameters considered here, 1000 samples give good larger spin (unlike bosonic modes).
convergence. Single trajectories show a typical pinning To derive the hierarchy it is useful to switch to
behavior known for QSD of pure dephasing: Trajectories the Bargmann parameterization of spin coherent states
localize close to an eigenstate of the system Hamiltonian. (Eq. (16)). For Bargmann states the following relations
For an initial state with equal amplitude in each eigen- are easy to prove:
state, the trajectories localize with equal probability on
all eigenstates so that on average the populations are 1
Jλx |zi = jzλ |zi + (1 − zλ2 )∂zλ |zi
conserved but the coherences decay [34]. 2
y 1 (43)
Jλ |zi = −ijzλ |zi + (−1 − zλ2 )∂zλ |zi
2i
VI. HIERARCHY EXPANSION
Jλz |zi = j|zi − zλ ∂zλ |zi

With a hierarchical expansion the equation of motion These allow us to write the action of spin operators
for the stochastic state (32) can be systematically solved. in a differential form. We can redefine the norm of
 7

the stochastic states in order to express them with the finds that higher order derivatives with respect to the
Bargmann states hz|R† (t)|Ψ(t)i ≡ |ψ(t, ~ n)i. Taking the Bargmann labels occur. We label these ‘auxiliary states’
time derivative and making use of the relations (43) yields with a multi-index k ∈ NN
0 as
a differential equation for the stochastic state including
derivatives with respect to the state label N
Y
!
k
|ψ (t, ~
n)i = ∂zk∗λ hz|R† (t)|Ψ(t)i. (45)
∂t |ψi = λ
λ=1
− iHS |ψi
zλ∗
 
N To find the evolution equation for all auxiliary states we
can simply take the corresponding zλ∗ derivatives of (44).
X
~·
− i ∆L gλ jOλ Sλ izλ∗  |ψi
λ=1 1 Then, the following commutations can be used to move
 1 ∗ 2
 the derivatives to the right hand side
2 (1 − (zλ ) )
XN
~·
− i ∆L 1
gλ Oλ Sλ − 2i (−1 − (zλ∗ )2 ) ∂zλ∗ hz|R† |Ψi
λ=1 −zλ∗ [∂zkλ∗ , zλ∗ ] = k∂zk−1
∗ ,
λ
(46)
(44) [∂zkλ∗ , (zλ∗ )2 ] = 2kzλ∗ ∂zk−2
∗ + k(k − 1)∂zk−1
∗ .
λ λ

In this equation we have omitted to write the dependen-

cies on time explicitly. It turns out that the first two
terms are exactly the first two terms of (32), thus the
last term is the intricate memory term. Deriving an evo-
lution equation for the derivative terms ∂zλ∗ hz|R† |Ψi one
One obtains a hierarchy of equations of motion where
PN
states of ‘order’ |k| = λ=1 kλ couple only to states one
order above or below. The equations of motion read ex-
plicitly:

zλ∗
 
N
X
~·
∂t |ψ k i = − iHS |ψ k i − i ∆L gλ (j − kλ )Oλ Sλ izλ∗  |ψ k i
λ=1 1
 1 ∗ 2
   (47)
2 (1 − (zλ ) ) 1
N
~·
X h 1 i
− i ∆L gλ Oλ Sλ − 2i 1
(−1 − (zλ∗ )2 ) |ψ k+eλ i + kλ (j − (kλ − 1))  i  |ψ k−eλ i
−zλ∗ 2 0
λ=1

In this equation, eλ = (δλγ ) is a unit vector. The k = 0 parameters or extremely strong coupling.
state of this hierarchy is the stochastic state which we For a proof of concept we computed a nontrivial ex-
aim to determine. The full hierarchy is as large as the full ample problem which cannot be solved analytically. In
Hilbert space of system and bath, i.e. exponentially large. particular, we considered a two level system HS = εσx /2
Therefore, the hierarchy must be truncated at a low or- coupled to a bath of two-level systems (j = 1/2) with
der to be numerically feasible, as in the case of HOPS. ~ = σz ~ex , i.e. the Hamiltonian
L
Formally, the truncation of the hierarchy at finite order
|k| corresponds to a perturbative expansion in gλ2 to the X 1 
ε X
same order. Note, however, that the auxiliary states with H = σx + σz gλ Jλx + ωλ − Jλz . (48)
2 2
kλ = 2j+1 do not couple to the lower states and thus they λ λ

remain zero and the hierarchy truncation becomes exact

As is common for this model [18], we choose couplings
after kλ = 2j. This shows that a truncated hierarchy
gλ and frequencies ωλ randomly from a uniform distri-
clearly goes beyond a perturbation series. The simplest
bution. In the spin-NMQSD simulation we used the
sensible truncation for nontrivial spin bath problems is
second order hierarchy expansion. For comparison we
the second order truncation |kmax | = 2. At second order
computed the exact dynamics with the recently intro-
the hierarchy has 1/2(N + 2)(N + 1) ∝ O(N 2 ) auxil-
duced ACE method (Automated Compression of Envi-
iary states, which is easily feasible also for larger values
ronments) [19]. This method is based on a representa-
of N . A first order truncation does not make sense, as
tion of the discretized influence functional in terms of a
this would be consistent with Redfield theory where the
tensor network, which has to be compressed iteratively
spin bath could be replaced by a bosonic bath in the first
with standard MPO compression techniques in order to
place. Furthermore, in case of an extremely large number
ensure a feasible evaluation of the system dynamics. For
of spins N , for which the second order truncation would
the strong system-bath coupling that we choose in our ex-
become intractable, we do not expect deviations from a
ample, ACE requires a large numerical effort reflected in
Gaussian approximation of the bath except for fine tuned
a large bond dimension that is needed for a sufficiently

1.0 x Crucially, the stochastic states are normalized so that
y proper convergence with respect to the number of sam-
0.5 z pled states is assured. This kind of importance sampling
ACE
hσi i
0.0
−0.5
−1.0
0 5 10 15 20
εt
FIG. 2. Decay of a two level system in a spin bath (Hamil-
tonian (48)) computed with the hierarchy (47) in the second
order truncation (average of 2000 samples). The bath consists
of 60 spins with uniform randomly chosen frequencies and
couplings ωλ /ε ∈ [0, 3], gλ2 /(N ε2 ) ∈ [0, 1.6]. For comparison,
the exact dynamics was computed with the ACE method [19]
(dashed lines, convergence was reached with a discretization
using 180 time steps and SVD compression with maximum
bond dimension of 180).
accurate tensor network representation of the influence
functional. This is an indication of strong coupling and
a nontrivial structure of the environment. As can be
seen in Fig. 2, the approximate NMQSD solution from
the second order hierarchy truncation is in good agree-
ment with the ACE result. To improve the accuracy of
the NMQSD calculation, a deeper hierarchy has to be
considered which would require more advanced trunca-
tion schemes [45, 49]. Compared to ACE, the hierarchy
method has a more favourable scaling with respect to the
size of the system Hilbert space, since it is based on wave
function propagation instead of density matrices. How-
ever, a detailed analysis of the numerical performance of
the method is beyond the scope of this paper and will be
addressed in future works.
VII. CONCLUSIONS
In this work, we have presented a generalization of the
non-Markovian quantum state diffusion method for ar-
bitrary quantum systems coupled to linear spin baths.
This advancement now enables the treatment of certain
intricate many-body problems such as central spin mod-
els within the NMQSD framework. Our main contribu-
tion is the derivation of a representation for the time-
evolved reduced state of the system as the average of an
ensemble of pure states. These pure states are depen-
dent on specific spin orientations ~nλ that are distributed
according to a non-Gaussian measure and can be sam-
pled using Monte Carlo methods. This approach allows
for an approximate determination of the reduced state
by averaging over a finite set of stochastic pure states.
The labels ~nλ thus take the role of the colored Gaussian
noise process in the Gaussian NMQSD theory [20, 50].
8
requires that the time evolution of the stochastic states
is nonlinear. In fact, the stochastic dynamics includes a
co-moving semi-classical evolution of the bath spins via
Eq. (24). In general, for non-Markovian dynamics, deter-
mining evolution equations is difficult. With a projection
approach we can formally construct such an evolution
equation for the stochastic state, but an intricate mem-
ory kernel remains to be determined. Alternatively, we
propose a hierarchical scheme to numerically propagate
the stochastic states. Although the full hierarchy does
not cure the exponential complexity of the problem at
hand, we demonstrate with a simple example that a low
order truncation is numerically feasible and gives good
agreement with the exact dynamics. If, for instance in
the case of strong coupling, a larger hierarchy is required
to obtain accurate results, one could employ MPS tech-
niques in order to keep the state dimension under control.
Recently, such strategies have been successfully applied
within the hierarchical expansion of Gaussian NMQSD
[21, 51]. As a future perspective, it would be interesting
to benchmark whether the proposed hierarchical scheme
competes with other methods such as ML-MCTDH [52]
or ACE [19] from a numerical perspective. In our view,
the conditions under which a spin bath can be adequately
described within a Gaussian framework [33] have not
been investigated thoroughly in the literature. With the
new range of methods available for treating non-Gaussian
baths, it should now be possible to have a rigorous and
quantitative discussion of this question.
ACKNOWLEDGMENTS
VL acknowledges support from the international Max
Planck research school (IMPRS) of MPI-PKS Dresden.
VL and WTS are grateful for inspiring discussions with
Chang-Yu Hsieh, Jianshu Cao and Shouryya Ray. Ac-
cess to the computer resources of the Finnish IT Center
for Science (CSC) and the FGCI project (Finland) is ac-
knowledged by KL.
Appendix A: Equation of Motion for the Q Function
To derive the equation of motion we crucially use the
rules [53]
1
J a |~nih~n| = jna |~nih~n| + (δ ab − iεabc nc − na nb )∂ b |~nih~n|
2
1
|~nih~n|J a = jna |~nih~n| + (δ ab + iεabc nc − na nb )∂ b |~nih~n|.
2
(A1)
The Q function can be written as Q(t, ~n) =
nih~
tr|~ n|ρSE (t), where ρSE (t) = |Ψ(t)ihΨ(t)| is the full
 9

state of system and environment. The derivation of the from (30) gives (32), where the kernel operator K(t, s)
evolution equation goes as follows: reads
n) = −i tr (|~
∂t Q(t, ~ nih~
n|[H(t), ρSE (t)]) 1  
K(t, s) = − n| Ṙ† (t) − iR† (t)H(t) ·
h~
N   h~
n|0i (B5)
X
gλ εabc nc ∂ b tr |~ ~ a ρSE (t)
n|(Oλ (t)T L)
 
= λ λ nih~ · W (t, s) Q̇(s) − iQ(s)H(s) R(s)|0i.
λ=1
N
X (A2) A perturbation expansion of this operator is obtained
= ~
gλ εabc ncλ ∂λb (Oλ (t)T hLi(t)) a
Q(t, ~
n) upon inserting a perturbation expansion for W (t, s). The
λ=1 lowest nonvanishing order is obtained from W (t, s) ≈ 1.
N
X  
=− ~ λ · Oλ (t)T hLi(t)
gλ ∇ ~ × ~nλ Q(t, ~
n)
λ=1 Appendix C: Pure Dephasing
∂
Above, we have set ∂λa = ∂na and defined
λ The pure dephasing Hamiltonian with a spin bath
reads
nih~
tr|~ ~ SE (t)
n|Lρ hΨ(t)|~ ~ n|Ψ(t)i
niLh~
~
hLi(t) = = . (A3) N
nih~
tr|~ n|ρSE (t) hΨ(t)|~
nih~
n|Ψ(t)i X
H(t) = HS + L gλ (cos ωλ tJλx + sin ωλ tJλy ), (C1)
λ=1
Appendix B: Projection Formalism
where HS and L commute so that the energy eigenstates
HS |εn i = εn |εn i are also eigenstates of the coupling
The derivation of a closed evolution equation for the operator L|εn i = ln |εn i. Projecting the Schrödinger
stochastic state |ψ(t, ~ n0 (t))i is based on the projector equation onto an energy eigenstate then gives a closed
(31). The corresponding orthogonal projector is Q(t) = equation in the bath Hilbert space
1 − P (t). In the following we omit writing the depen-
dence on the labels ~ n. Taking the time derivative of the
stochastic state and inserting an identity 1 = P (t) + Q(t) i∂t hεn |Ψ(t)i =εn hεn |Ψ(t)i
yields N
X
  + ln gλ cos ωλ tJλx hεn |Ψ(t)i
n| Ṙ† (t) − iR† (t)H(t) P (t)|Ψ(t)i
∂t |ψ(t)i = h~ λ=1 (C2)
  (B1) N
n| Ṙ† (t) − iR† (t)H(t) Q(t)|Ψ(t)i.
X
+ h~ + ln gλ sin ωλ tJλy hεn |Ψ(t)i.
λ=1
The first term is easy to evaluate because P |Ψi is pro-
portional to the stochastic state itself. In order to find This corresponds to non-interacting spins so that the
an expression for Q|Ψi we can use the standard Feshbach equation can be easily solved with a product ansatz
technique and formally integrate the evolution equation N
O
  hεn |Ψ(t)i = cn e−iεn t |ψλ,n (t)i. (C3)
∂t Q(t)|Ψ(t)i = Q̇(t) − iQ(t)H(t) |Ψ(t)i λ=1
  (B2)
= Q̇(t) − iQ(t)H(t) (Q(t) + P (t))|Ψ(t)i, The evolution equation for the individual bath spin states
conditioned on the system energy n are linear
which yields
i∂t |ψλ,n (t)i = gλ (cos ωλ tJλx + sin ωλ tJλy )|ψλ,n (t)i (C4)
Q(t)|Ψ(t)i = W (t, 0)Q(0)|Ψ(0)i
Z t   (B3) and, thus, spin coherent states are solutions. The re-
+ dsW (t, s) Q̇(s) − iQ(s)H(s) P (s)|Ψ(s)i, duced state is obtained easily as
0
ρS = trE |Ψ(t)ihΨ(t)|
where we introduced the time evolution operator W (t, s)
N
which is the solution to W (s, s) = 1 and X X (C5)
= |εn ihεm |cn c∗m e−i(εn −εm )t hψλ,m |ψλ,n i.
  nm λ=1
∂t W (t, s) = Q̇(t) − iQ(t)H(t) W (t, s). (B4)
Since HS commutes with H the diagonals of the density
Note that the first term in (B3) drops out because, by matrix in the energy representation are constant, ρS,nn =
construction, the initial state lies in the subspace spanned |cn |2 . The ‘dephasing’ is due to the time dependence of
by P (0), so that Q(0)|0i = 0. Inserting this formal solu- the nondiagonal terms which typically decay to zero for
tion into (B1) and inserting the time derivatives of R(t) large environments.

To find the stochastic state from the full solution we
project the system and bath state onto spin coherent
states
N 0
n0 (t)|Ψ(t)i =
X X
h~ cn e−iεn t |εn i h~n0λ (t)|ψλ,n i,
n λ=1
where the time dependent labels ~n0λ (t) are obtained by
[1] I. de Vega and D. Alonso, Dynamics of non-Markovian
open quantum systems, Rev. Mod. Phys. 89, 015001
(2017).
[2] M. Bortz, S. Eggert, C. Schneider, R. Stübner, and
J. Stolze, Dynamics and decoherence in the central spin
model using exact methods, Phys. Rev. B 82, 161308
(2010).
[3] T. Villazon, P. W. Claeys, M. Pandey, A. Polkovnikov,
and A. Chandran, Persistent dark states in anisotropic
central spin models, Sci. Rep. 10, 1 (2020).
[4] J. Schliemann, A. Khaetskii, and D. Loss, Electron spin
dynamics in quantum dots and related nanostructures
due to hyperfine, J. Phys.: Condens. Matter 15, R1809
(2003).
[5] J. Dukelsky, S. Pittel, and G. Sierra, Colloquium: Ex-
actly solvable Richardson-Gaudin models for many-body
quantum systems, Rev. Mod. Phys. 76, 643 (2004).
[6] D. F. Walls and G. J. Milburn, Quantum Optics
(Springer Berlin Heidelberg, 2008).
[7] N. Makri, The Linear Response Approximation and Its
Lowest Order Corrections: An Influence Functional Ap-
proach, J. Phys. Chem. B 103, 2823 (1999).
[8] P. Fernández-Acebal, O. Rosolio, J. Scheuer, C. Müller,
S. Müller, S. Schmitt, L. P. McGuinness, I. Schwarz,
Q. Chen, A. Retzker, B. Naydenov, F. Jelezko, and M. B.
Plenio, Toward Hyperpolarization of Oil Molecules via
Single Nitrogen Vacancy Centers in Diamond, Nano Lett.
18, 1882 (2018).
[9] M. Bramberger and I. De Vega, Dephasing dynamics
of an impurity coupled to an anharmonic environment,
Phys. Rev. A 101, 012101 (2020).
[10] T. Villazon, P. W. Claeys, A. Polkovnikov, and A. Chan-
dran, Shortcuts to dynamic polarization, Phys. Rev. B
103, 075118 (2021).
[11] J. M. Taylor, C. M. Marcus, and M. D. Lukin, Long-Lived
Memory for Mesoscopic Quantum Bits, Phys. Rev. Lett.
90, 206803 (2003).
[12] P. Fowler-Wright, K. B. Arnardóttir, P. Kirton, B. W.
Lovett, and J. Keeling, Determining the validity of
cumulant expansions for central spin models, arXiv
10.48550/arXiv.2303.04410 (2023), 2303.04410.
[13] Y. Tanimura, Numerically “exact” approach to open
quantum dynamics: The hierarchical equations of mo-
tion (HEOM), J. Chem. Phys. 153, 020901 (2020).
[14] N. Makri and D. E. Makarov, Tensor propagator for itera-
tive quantum time evolution of reduced density matrices.
I. Theory, J. Chem. Phys. 102, 4600 (1995).
[15] A. Strathearn, P. Kirton, D. Kilda, J. Keeling, and
B. W. Lovett, Efficient non-Markovian quantum dynam-
ics using time-evolving matrix product operators - Na-
10
integrating
 
cos(ωλ t)
~n˙ 0λ (t) = gλ hLi(t)  sin(ωλ t)  × ~n0λ (t). (C7)
(C6)
The initial conditions ~nλ (0) can be drawn from the
proper distribution (27) using inverse transform sampling
for nzλ and uniform sampling of the azimuth angle.
ture Communications, Nat. Commun. 9, 1 (2018).
[16] J. Prior, A. W. Chin, S. F. Huelga, and M. B. Plenio,
Efficient Simulation of Strong System-Environment In-
teractions, Phys. Rev. Lett. 105, 050404 (2010).
[17] C.-Y. Hsieh and J. Cao, A unified stochastic formulation
of dissipative quantum dynamics. I. Generalized hierar-
chical equations, J. Chem. Phys. 148, 014103 (2018).
[18] C.-Y. Hsieh and J. Cao, A unified stochastic formulation
of dissipative quantum dynamics. II. Beyond linear re-
sponse of spin baths, J. Chem. Phys. 148, 014104 (2018).
[19] M. Cygorek, M. Cosacchi, A. Vagov, V. M. Axt, B. W.
Lovett, J. Keeling, and E. M. Gauger, Simulation of open
quantum systems by automated compression of arbitrary
environments, Nat. Phys. , 1 (2022).
[20] W. T. Strunz, Linear quantum state diffusion for non-
markovian open quantum systems, Physics Letters A
224, 25 (1996).
[21] S. Flannigan, F. Damanet, and A. J. Daley, Many-Body
Quantum State Diffusion for Non-Markovian Dynamics
in Strongly Interacting Systems, Phys. Rev. Lett. 128,
063601 (2022).
[22] D. Suess, A. Eisfeld, and W. T. Strunz, Hierarchy of
Stochastic Pure States for Open Quantum System Dy-
namics, Phys. Rev. Lett. 113, 150403 (2014).
[23] R. Hartmann and W. T. Strunz, Exact open quan-
tum system dynamics using the hierarchy of pure states
(hops), Journal of Chemical Theory and Computation
13, 5834 (2017).
[24] P.-P. Zhang and A. Eisfeld, Non-Perturbative Calcula-
tion of Two-Dimensional Spectra Using the Stochastic
Hierarchy of Pure States, J. Phys. Chem. Lett. 7, 4488
(2016).
[25] L. Chen, D. I. G. Bennett, and A. Eisfeld, Simulation of
absorption spectra of molecular aggregates: A hierarchy
of stochastic pure state approach, J. Chem. Phys. 156,
10.1063/5.0078435 (2022).
[26] W. T. Strunz, The Brownian motion stochastic
Schrödinger equation, Chem. Phys. 268, 237 (2001).
[27] W. T. Strunz and T. Yu, Convolutionless Non-Markovian
master equations and quantum trajectories: Brownian
motion, Phys. Rev. A 69, 052115 (2004).
[28] V. Link and W. T. Strunz, Stochastic feshbach projection
for the dynamics of open quantum systems, Phys. Rev.
Lett. 119, 180401 (2017).
[29] H. Wang and J. Shao, Dynamics of a two-level system
coupled to a bath of spins, J. Chem. Phys. 137, 22A504
(2012).
[30] D. Gelman, C. P. Koch, and R. Kosloff, Dissipative
quantum dynamics with the surrogate Hamiltonian ap-
proach. A comparison between spin and harmonic baths,

J. Chem. Phys. 121, 661 (2004).
[31] H.-D. Zhang, R.-X. Xu, X. Zheng, and Y. Yan, Nonper-
turbative spin–boson and spin–spin dynamics and nonlin-
ear Fano interferences: A unified dissipaton theory based
study, J. Chem. Phys. 142, 024112 (2015).
[32] Z. Lü and H. Zheng, Influence of temperature on coherent
dynamics of a two-level system immersed in a dissipative
spin bath, J. Chem. Phys. 131, 134503 (2009).
[33] W. Yang, W.-L. Ma, and R.-B. Liu, Quantum many-body
theory for electron spin decoherence in nanoscale nuclear
spin baths, Rep. Prog. Phys. 80, 016001 (2016).
[34] W. T. Strunz, L. Diósi, and N. Gisin, Open system dy-
namics with non-markovian quantum trajectories, Phys.
Rev. Lett. 82, 1801 (1999).
[35] J. P. Gazeau, Coherent States in Quantum Physics (Wi-
ley, 2009).
[36] I. Bengtsson and K. Życzkowski, Geometry of Quan-
tum States: An Introduction to Quantum Entanglement
(Cambridge University Press, 2006).
[37] J. M. Radcliffe, Some properties of coherent spin states,
J. Phys. A: Gen. Phys. 4, 313 (1971).
[38] V. Bargmann, On a Hilbert space of analytic functions
and an associated integral transform part I, Commun.
Pure Appl. Math. 14, 187 (1961).
[39] R. Mosseri and R. Dandoloff, Geometry of entangled
states, Bloch spheres and Hopf fibrations, J. Phys. A:
Math. Gen. 34, 10243 (2001).
[40] C. Chryssomalakos, E. Guzmán-González, and
E. Serrano-Ensástiga, Geometry of spin coherent states,
Journal of Physics A: Mathematical and Theoretical 51,
165202 (2018).
[41] C. Brif and A. Mann, Phase-space formulation of quan-
tum mechanics and quantum-state reconstruction for
physical systems with lie-group symmetries, Phys. Rev.
A 59, 971 (1999).
[42] D. D. Bhaktavatsala Rao and G. Kurizki, From Zeno
to anti-Zeno regime: Decoherence-control dependence on
the quantum statistics of the bath, Phys. Rev. A 83,
11
032105 (2011).
[43] R. Hartmann, M. Werther, F. Grossmann, and W. T.
Strunz, Exact open quantum system dynamics: Optimal
frequency vs time representation of bath correlations, J.
Chem. Phys. 150, 234105 (2019).
[44] D. Suess, W. T. Strunz, and A. Eisfeld, Hierarchical
Equations for Open System Dynamics in Fermionic and
Bosonic Environments, J. Stat. Phys. 159, 1408 (2015).
[45] P.-P. Zhang, C. D. B. Bentley, and A. Eisfeld, Flexible
scheme to truncate the hierarchy of pure states, J. Chem.
Phys. 148, 134103 (2018).
[46] Y. Tanimura, Stochastic liouville, langevin,
fokker–planck, and master equation approaches to
quantum dissipative systems, Journal of the Physical
Society of Japan 75, 082001 (2006).
[47] Y. Tanimura, Reduced hierarchical equations of motion
in real and imaginary time: Correlated initial states
and thermodynamic quantities, The Journal of Chemi-
cal Physics 141, 044114 (2014).
[48] Z. Tang, X. Ouyang, Z. Gong, H. Wang, and J. Wu,
Extended hierarchy equation of motion for the spin-boson
model, J. Chem. Phys. 143, 224112 (2015).
[49] L. Varvelo, J. K. Lynd, and D. I. G. Bennett, For-
mally exact simulations of mesoscale exciton dynamics
in molecular materials, Chem. Sci. 12, 9704 (2021).
[50] L. Diósi and W. T. Strunz, The non-markovian stochastic
schrödinger equation for open systems, Physics Letters A
235, 569 (1997).
[51] X. Gao, J. Ren, A. Eisfeld, and Z. Shuai, Non-Markovian
stochastic Schrödinger equation: Matrix-product-state
approach to the hierarchy of pure states, Phys. Rev. A
105, L030202 (2022).
[52] H. Wang and M. Thoss, Multilayer formulation of the
multiconfiguration time-dependent Hartree theory, J.
Chem. Phys. 119, 1289 (2003).
[53] K. Merkel, V. Link, K. Luoma, and W. T. Strunz, Phase
space theory for open quantum systems with local and
collective dissipative processes, J. Phys. A: Math. Theor.
54, 035303 (2020).