2positive semi-deﬁnite. The outcome of the measure-ments on the environment is recordedas a “current”
y
=
C
ˆx
+
d
w
dt
,where
C
=
√
2
U
1
/
2
¯
C
and¯
C
jk
=
(
δ
2
j
−
1
,
k
+
δ
2(
j
−
N
)
,
k
)for
j
,
k
∈
[1
,...,
2
N
]. Finally, d
w
is avectorofrealWienerincrements satisfying d
w
d
w
⊤
=
½
d
t
.The
conditional
evolution of the moments under suchcontinuous measurements can also be derived by stan-dard techniques (Itˆo calculus). It amounts to a di
ﬀ
usiveequation with a stochastic component for the ﬁrst mo-ments
ˆ
x
, and to a
deterministic
Riccati equation for thesecondmoments[6]. Inourreasoningstofollow, wewillnotmakeuseofthedetailsofsuchequationsdirectly. Wewill be interested in stable systems, and will determinethe maximal entanglement achievable at steady state.Hence, all we need to remark is that the set of CMs
{
σ
∞
}
that are
stabilising solutions
[9] of the Riccati equation forthe second moments can be characterised as follows [6]:
A
σ
∞
+
σ
∞
A
T
+
½
≥
0
.
(4)TogetherwithInequality(1),thisrelationshipcompletelydetermines the set of stabilising solutions of our condi-tional dynamics.The ﬁnal ingredient to be added is the possible de-pendence of the linear drive
u
(
t
) on the history of themeasurement record
y
(
s
) for
s
<
t
, which a
ﬀ
ects bothﬁrst and second moments of the
unconditional
, ‘average’,evolution (whereas second moments of the
conditional
states are una
ﬀ
ected by the linear driving), and closesthe control loop. We will denote the unconditional state by
̺
. Note that, for our class of dynamics,
̺
is a statisti-cal mixture of states with the same conditional CM
σ
∞
,obeying Inequality (4), and varying ﬁrst moments. ForGaussian states, this implies that
̺
can be obtained froma Gaussian state
̺
0
with CM
σ
∞
and vanishing ﬁrst mo-ments by local operations and classical communicationalone:
̺
=
L
(
̺
0
), where
L
is some LOCC map.The typical aim of control over some time interval isto minimise the expected value of a
cost function
[1,9].
Our cost function will be the entanglement of Gaussianmulti-mode steady states for bipartitions of 1 versus(
N
−
1) modes and ‘bi-symmetric’ bipartitions (
i.e.,
in-variant under the permutation of local modes). Such anentanglement can be quantiﬁed by the logarithmic neg-ativity
E
N
, which is in turn determined, for a Gaussianstate with CM
σ
, as
−
log
2
˜
ν
−
, where ˜
ν
2
−
is the smallesteigenvalue of (
σ
˜
Ω
σ
˜
Ω
T
), being˜
Ω
the (skew-symmetric)partial transposition of
Ω
ν
−
is
not
aquadratic cost function (
i.e.
, it is not linear in
σ
). Thisis why, albeit dealing with linear systems with Gaussiannoise, we cannot resort to optimisation methods mutu-ated from classical LQG control problems [6].
General results –
Hereafter,we present our main ﬁndingsas three lemmas leading to a ﬁnal proposition. Proofs of these statements may be found in appendix.Our investigation starts o
ﬀ
from a corollary of Ref.[12]:
Lemma 1 (Bound on smallest symplectic eigenvalue)
The smallest partially transposed symplectic eigenvalue
˜
ν
−
of a generic CM
σ
is bounded from below as follows
˜
ν
2
−
≥
λ
↑
1
λ
↑
2
,
(5)
λ
↑
1
and
λ
↑
2
being the two smallest eigenvalues of
σ
.
Next, the uncertainty principle entails:
Lemma 2 (Uncertainty relation for CMs’ eigenvalues)
Let
{
λ
↑
j
}
and
{
λ
↓
j
}
be, respectively, the
2
N increasingly-ordered and decreasingly-ordered eigenvalues of an N-modeCM
σ
. Then one has:
λ
↑
j
λ
↓
j
≥
1 for 1
≤
j
≤
N
.
(6)As an immediate corollary of Lemma 2, one obtains
λ
↑
1
λ
↑
2
≥
1
λ
↓
1
λ
↓
2
.
(7)
Lemma 3 (Bound on eigenvalues of steady state CMs)
Let
σ
∞
be a conditional CM at steady state obtained undercontinuous Gaussian measurements, pure losses and a Hamiltonian matrix H. The product of the two largesteigenvalues
λ
↓
1
and
λ
↓
2
of
σ
∞
is bounded as follows:
λ
↓
1
λ
↓
2
≤
1
α
↑
1
α
↑
2
,
(8)
where
{
α
↑
j
}
are the (strictly positive) eigenvalues of
(
−
A
−
A
⊤
)
in increasing order, and A
=
12
(
Ω
H
−
½
)
.
The chain of Inequalities (5),(7) and (8)leads to
˜
ν
2
−
≥
α
↑
1
α
↑
2
,
(9)which constrains the maximal logarithmic negativityachievable for states conditioned by Gaussian measure-ments. Further, and more generally, one has:
Proposition 1 (Maximal unconditional entanglement)
The logarithmic negativity E
N
(
̺
)
of any
1
versus
(
N
−
1)
modes or bi-symmetric bipartition of an unconditional steadystate achievable by continuous Gaussian feedback and lineardriving is bounded by:E
N
(
̺
)
≤
max
0
,
−
12log
2
(
α
↑
1
α
↑
2
)
.
(10)
Applications–
Clearly, the eigenvalues
{
α
↑
j
}
can be analyt-ically or numerically determined for general quadraticHamiltonians, describing a wide variety of systems of practical interest (bosonic atoms, trapped ions, nano-mechanical resonators and Josephson junctions, just to