See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329780369

Quantum plasmonics with multi-emitters: application to stimulated Raman

adiabatic passage

Article  in  The European Physical Journal D · December 2018

DOI: 10.1140/epjd/e2018-90322-5

CITATIONS READS

16 310

7 authors, including:

Benjamin Rousseaux David Dzsotjan

Institut d'Optique Graduate School RPTU - Rheinland-Pfälzische Technische Universität Kaiserslautern Landau
29 PUBLICATIONS   359 CITATIONS    26 PUBLICATIONS   525 CITATIONS   

SEE PROFILE SEE PROFILE

Gérard Colas des Francs Antonino Messina

University of Burgundy Università degli Studi di Palermo
181 PUBLICATIONS   4,605 CITATIONS    313 PUBLICATIONS   3,039 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Controling the emission of quantum dots by plasmonic structure View project

Quantum Plasmonics View project

All content following this page was uploaded by Gérard Colas des Francs on 14 July 2019.

The user has requested enhancement of the downloaded file.

Eur. Phys. J. D (2018) 72: 223
https://doi.org/10.1140/epjd/e2018-90322-5 THE EUROPEAN
PHYSICAL JOURNAL D
Regular Article

Quantum plasmonics with multi-emitters: application

to stimulated Raman adiabatic passage
Alessia Castellini 1,2 , Hans Rudolf Jauslin 3 , Benjamin Rousseaux 4,5 , David Dzsotjan 6 ,
Gérard Colas des Francs 3 , Antonino Messina 2,7 , and Stéphane Guérin 3,a
1
Dipartimento di Fisica e Chimica dell’Università di Palermo, via Archirafi 36, 90123 Palermo, Italy
2
INFN Sezione di Catania, Catania, Italy
3
Laboratoire Interdisciplinaire Carnot de Bourgogne, CNRS UMR 6303, Université Bourgogne Franche-Comté,
BP 47870, 21078 Dijon, France
4
Department of Physics, Chalmers University of Technology, 412 96 Göteborg, Sweden
5
Department of Microtechnology and Nanoscience – MC2, Chalmers University of Technology, 412 96 Göteborg,
Sweden
6
Wigner Research Center for Physics, Hungarian Academy of Sciences, Konkoly-Thege Miklos ut 29-33,
1121 Budapest, Hungary
7
Dipartimento di Matematica e Informatica dell’Università di Palermo, via Archirafi 34, 90123 Palermo, Italy

Received 2 July 2018 / Received in final form 22 October 2018

Published online 21 December 2018
c EDP Sciences / Società Italiana di Fisica / Springer-Verlag GmbH Germany, part of Springer Nature,
2018

Abstract. We construct a mode-selective effective model describing the interaction of the localised surface
plasmon polaritons (LSPs) supported by a spherical metal nanoparticle (MNP) with N quantum emitters
(QEs) in an arbitrary geometric arrangement. Simplifying previously presented procedures, we develop a
formulation in which the field response in the presence of the MNP can be decomposed into orthogonal
modes, expanding the Green tensor of the system in the spherical vector harmonics basis and using the
generalized global Löwdin orthogonalization algorithm. We investigate the possibility of using the LSPs
as mediators of an efficient control of population transfer between two QEs. We show that a Stimulated
Raman Adiabatic Passage (STIRAP) configuration allows such a transfer via a decoherence-free dark state
for a specific range of angular distances between the QEs, when they are located very close to the MNP.
The transfer is otherwise blocked. We explain this blockade by the destructive superposition of all the
plasmonic modes.

1 Introduction the electromagnetic local density of states (LDOS) of

Plasmonics offers new scenarios for classical and quantum
manipulations of light at the nanoscale [1,2]. It can be
implemented via metallic nano-particles (MNPs) support-
ing propagating and localised surface plasmon polaritons
(P-SPPs and L-SPPs, respectively) at dielectric/metallic
interfaces [3,4]. Plasmonic quantum electrodynamics is the
natural generalization of the principles of quantum optics
for cavity quantum electrodynamics (cQED), enabling
to treat the coupling between quantum emitters (QEs)
and the modes of the SPP field localised in subdiffrac-
tion volumes [5,6]. A model has been initially proposed
[7–10], where all the electromagnetic excitations (cor-
responding to L-SPPs) of the system are described by
collective field operators, exploiting the theory of quan-
tization of the electromagnetic field in dispersive and
absorbing media [7,11–13]. When the QEs are placed very
close to the MNP’s surface (typically less than 10 nm),
a
e-mail: sguerin@u-bourgogne.fr
the L-SPP is enhanced and can lead to a strong cou-
pling [10,14]. The role of the lower and higher order
L-SPP modes in the interaction with QEs depends on
their positions to the MNP [10]. The quasi-degenerate
higher order modes can collectively act, in the interaction
with QEs, as a pseudomode rather than a Markovian bath
[15]. Effective few-states Hamiltonians have been phe-
nomenologically proposed to describe the weak and strong
coupling regimes of layer or ring configurations of N QEs
with the dipolar mode and with this pseudomode [15,16].
CQED-like Hamiltonian models have been derived via the
expansion of the noise operator and the Green tensor in
the spherical vector harmonics basis [14]. Mode-selective
L-SPP field operators, i.e. associated with the different
plasmonic resonances, substitute thus the collective elec-
tromagnetic field operators to describe the interaction of
the different L-SPP modes with QEs [14].
These plasmonic platforms and their nanometric dimen-
sions stimulated the interest for their use as devices
for classical and quantum information processes [17,18].
Page 2 of 14
However, the plasmonic quality factors are much smaller
than the ones of cavities, which prevents in principle
SPPs from carrying information over longer distances. It
has been shown recently that a free-decoherence quantum
channel linking two QEs, which do not interact directly,
can be created exploiting the mediation of L-SPPs of a
spherical MNP [19]. Stimulated Raman adiabatic Pas-
sage (STIRAP) [20,21] can allow population exchange or
entanglement between two QEs aligned at the same side
of the sphere [19]. One can mention very recent demon-
strations of STIRAP in cQED with super–conducting
qubits [22,23].
In this paper, we construct a cQED-like effective
Hamiltonian model of N QEs coupled to L-SPPs of
a spherical MNP. Considering that the Green’s tensor
approach can be applied to arbitrarily shaped particles,
we simplify the procedure presented in [14], decomposing
into orthogonal modes only the Green tensor associated
with the field response in presence of the nanosystem,
with the spherical symmetry as an example. The math-
ematical technique used in this paper improves preceding
models [15,16] and methods [14] in generalizing and apply-
ing the global Löwdin orthonormalization [24] of a set of
mode-selective bright L-SPP field operators. This gives us
a microscopic model involving only the relevant degrees
of freedom and compact analytical expressions for the
mode-selective QE/L-SPP coupling constants.
By completing the analysis presented in [19], we exploit
the derived model by numerical simulations of a STIRAP
process between two QEs, targeting an adiabatic popula-
tion transfer between two states of the dressed system [25].
We analyse the dependence of the transfer efficiency with
respect to the angular distance between the emitters and
to the distance from the metallic surface via the number
of effectively involved plasmonic modes. The population
transfer is achieved efficiently when the QEs are located
on the same side of the MNP. The existence of a blockade
of the population transfer for two QEs located at different
sides of the MNP is explained by a destructive interference
of the plasmonic modes.
This article is organized as follows: In Section 2 we
construct the continuous and discrete effective mode-
selective Hamiltonian models describing the interaction
of N QEs with a spherical MNP, for any possible overlap
of the plasmonic modes excited by different emitters. In
Section 3, we exploit the derived discrete effective model
to show under which geometric and parametric condi-
tions a STIRAP process can be numerically implemented.
Finally, in Section 4, we summarize the results and
perspectives.
2 Effective models for the QE/L-SPP system
We want to describe the dynamics of a system composed
of (see Fig. 1)
– a hybrid plasmon-polariton system, which originates
from the interaction of the free electromagnetic field
with the normal modes associated with the collec-
tive surface oscillations of the conduction electrons
in a spherical MNP. The latter is characterized by
Eur. Phys. J. D (2018) 72: 223
Fig. 1. (a) Global system composed of a set of QEs placed at
a distance d (r) from the metallic surface (center) of a MNP
of radius R. The interaction of each QE with the MNP excites
different LSPP modes (represented by the red clouds on the
MNP’s surface). The angular distance between any two QEs is
indicated by φ. (b) Scheme of the 3-level Λ structure of each
QE, composed of two metastable states, |gi and |f i, and of an
excited state |ei. The transition |ei − |gi is coupled with the
MNP.
a radius R and by a dielectric permittivity m (ω).
The excitation associated with the global system is
annihilated by the bosonic field operator f̂ω (r) [7].
– N identical 3-level QEs, each of which is composed
of two metastable states, |gi and |f i, and an excited
one |ei. Each QE is in Λ configuration, i.e. Eg ≤
Ef < Ee , and is placed at a position ri from the
center of the MNP, corresponding to a distance di ,
i = 1, . . . , N , from its surface.
The distance between the QEs is such that their direct
coulombic interaction is negligible compared to that medi-
ated by the plasmonic field [10,14,16] because of a high
Purcell factor. Each QE is coupled to the MNP through
the transition |ei − |gi. The transition |f i − |gi is forbid-
den and the transition |ei − |f i will be used only via a
control field.
We first omit the metastable state |f i, which will be
involved in the control process in Section 3. We start from
a rotating wave approximation model [7–10,14]:
Z +∞ Z N
X
Ĥ = dω ~ω d3 r f̂ω† (r) · f̂ω (r) + (l)
~ωeg σ̂ee
0 l=1
(1)
N Z +∞
X
(l)
− σ̂eg ⊗ dω d(l) +
eg · Êω (rl ) + H.c.,
l=1 0
(l)
with the emitter operators σ̂ij = |iihj|, deg the dipole
moment at the transition e–g of the lth emitter, and where
the components of the elementary excitation operator
f̂ω (r) obey the following commutation relations:
[fˆωi (r), fˆωj 0† (r0 )] = δ(ω − ω 0 )δ(r − r0 )δij . (2)
The first term of (1) represents the free energy of all the
plasmonic modes supported by the MNP. The last one is
the interaction energy of the lth QE, at the position rl ,
with the plasmonic electric field Êω (rl ) = Ê+
ω (rl ) + H.c.
Eur. Phys. J. D (2018) 72: 223 Page 3 of 14

In the following, we derive the effective model for many assumption that the separation into modes of the Green
emitters using a mode-selective quantization in the Green function (5) is such that the coefficients (7) satisfy the
tensor formalism, similarly as in [14], but in a more direct orthogonality relation:
way, without expanding the elementary excitation opera- Z
tors f̂ω (r) in the spherical vector harmonics basis. Only ∗
d3 r gω,n (rl , r) · gω0 ,n0 (rl0 , r) = 0, for n 6= n0 . (8)
the Green tensor is expanded in this basis. This gives
an Hamiltonian in terms of creation and annihilation
operators of a plasmon polariton in each L-SPP mode This is the case, for example, for a system with spher-
(dipolar, quadrupolar,...) excited by the different emit- ical symmetry [14]. The motivation for the separation
ters. The presence of many emitters leads to overlapping into modes is that, under suitable conditions, the system
non-orthogonal modes that have to be orthogonalized into has resonances that strongly couple with the emitters at
bright and dark modes according to the effective interac- particular frequencies. The resonant frequency dominates
tion between the plasmon polaritons and the QEs. We the dynamics for the corresponding
R mode, and thus the
describe a global orthogonalization procedure based on integral over the frequencies dω appearing in the Hamil-
Löwdin’s methods [26–28]. tonian can be approximated by a discrete sum, where
Formally, the dark/bright decomposition can be sum- each term corresponds to a mode n. The approximation
marized as follows: We identify and eliminate their is completed by the inclusion of an exponential relaxation
contribution from the total Hamiltonian: term for each mode that takes into account effectively the
continuum of frequencies around the resonant one.
0 0 0
Ĥ = Ĥbright + Ĥdark + ĤQEs + Ĥbright/QEs . (3) We introduce the mode-selective L-SPP bright operators
[14,19]
0
Since we have [Ĥ, Ĥdark ] = 0, the dark states are not cou- Z
pled with the other states and are not affected by the
âω,n (rl ) = d3 r hω,n (rl , r) · f̂ω (r), (9)
interaction with the emitters, which allows one to drop
0
the Ĥdark term in (3):
with the vectors hω,n (rl , r) ∈ L2 (R3 , C3 ):
0 0
Ĥ = Ĥbright + ĤQEs + Ĥbright/QEs . (4)
hω,n (rl , r) = gω,n (rl , r)/κω,n (rl ), (10)
2.1 Mode-selective quantization in the Green tensor
formalism and
Z
The electric field in (1) can be expressed in terms of the ∗
|κω,n (rl )|2 = d3 r gω,n (rl , r) · gω,n (rl , r), (11a)
inner product of the dyadic Green’s function Gω (rl , r)
with the elementary excitation f̂ω (r) [7,14,16]. We consider 1 ω2 h i
= d · Im ¯ (r , r ) · d∗ , (11b)
Ḡ
the case where the Green function can be decomposed as eg ω,n l l eg
~π0 c2
a sum over a discrete index:
¯ (r, r0 ) =
X
¯ (r, r0 ), satisfying the normalization condition:
Ḡω Ḡ ω,n (5)
n Z
d3 r hω,n (rl , r) · h∗ω,n0 (rl , r) := ( hω,n0 (rl ), hω,n (rl )))
as it is the case, e.g., with a multipole expansion in systems
with spherical symmetry [14,19], where the index n is the = δnn0 , (12)
radial harmonic index. It is referred below to as mode
index. As a result, the interaction part of equation (1) via the identity for non-magnetic materials [7,29]:
can be decomposed into the modes as
Z 2
¯ ∗ (r, r ) = c Im Ḡ
h i
N X Z Z ¯ (r , r)Ḡ
dr ε00ω (r)Ḡ ¯ (r , r ) .
ω,n 1 ω,n 2 2 ω,n 1 2
ω
X
(l)
Ĥint =−i~ σ̂eg ⊗ dω d3 r gω,n (rl , r) · f̂ω (r),
n
(13)
l=1
(6)
with the vector gω,n (rl , r) defined by the inner product of We have defined the scalar product ( h1 , h2 ) in L2 (R3 , C3 ),
the dipole moment with the dyadic Green’s function: i.e. between two complex vectors which depend on r, as
r Z
1 ω 2 p 00 ¯ ( h1 , h2 ) := d3 r h2 (r) · h∗1 (r).
gω,n (rl , r) = ω (r) d(l)
eg · Ḡω,n (rl , r)
(14)
~π0 c2
r (7)
1 ω 2 p 00 ¯ T (l)
= ω (r) Ḡω,n (rl , r) · deg . The phase of κω,n (rl ) can be chosen arbitrarily. The
~π0 c2 operators âω,n (rl ) satisfy the commutation relations:
00ω (r) is the imaginary part of the electric permittiv- h i
ity. For the following development, we make the general âω,n (rl ), â†ω0 ,n0 (rl ) = δnn0 δ(ω − ω 0 ). (15)
Page 4 of 14 Eur. Phys. J. D (2018) 72: 223

Using the above definitions, one can write equation (1) as 2.2.2 A single emitter

N
For a single QE (at the position r1 ), we construct the dark
Z +∞ Z
operator by subtracting from the field operator f̂ω (r) its
X
Ĥ = dω ~ω 3
d r f̂ω† (r)f̂ω (r) + (l)
~ωeg σ̂ee
0 l=1
projection on the subspace generated by the orthogonal
N
set of operators âω,n (r1 ) in a similar way of a step in a
Z +∞
X
(l)
X Gram–Schmidt orthogonalization procedure [14,19]:
− i~ σ̂eg ⊗ dω κω,n (rl )âω,n (rl ) + H.c.,
l=1 0 n X h i
(16) d̂ω (r) := f̂ω (r) − âω,n (r1 ) â†ω,n (r1 ), f̂ω (r) , (23a)
n
where each emitter interacts with one effective L-SPP
X
= f̂ω (r) − âω,n (r1 )h∗ω,n (r1 , r). (23b)
field. The operator â†ω,n (rl ) is associated with the creation n
of a quantized plasmon by the emitter at the position rl .
The free plasmonic term includes dark modes, i.e. The dark operators satisfy the following properties:
modes that are not involved in the coupled dynamics. In
the following we construct the dark operators d̂ω (r) that h i h i
will be eliminated from the model. âω,n (rl ), d̂†ω (r) = âω,n (rl ), d̂ω (r) = 0, (24a)
Z X
2.2 Identification and elimination of the dark modes d3 r f̂ω† (r) · f̂ω (r) = â†ω,n (r1 )âω,n (r1 )
n
Z
2.2.1 Hilbert space structure generated by the
+ d3 r d̂†ω (r) · d̂ω (r). (24b)
creation–annihilation operators fˆi (r)
The linear decomposition (9) of the general form
Omitting the dark modes that are not populated by
Z 3 Z
the interaction with the emitter, we derive the reduced
X Hamiltonian from the above decomposition as anticipated
â = d3 r h(r) · f̂ (r) = d3 r hi (r)fˆi (r), (17)
in (4):
i=1
Z +∞
with h(r) ∈ L2 (R3 , C3 ) defines a vector space Vf , on which
X
Ĥ = dω ~ω â†ω,n (r1 )âω,n (r1 ) + ~ωeg σ̂ee
(1)
we can define a scalar product with the commutator: 0 n
Z +∞
(25)
( â1 , â2 ) := [â2 , â†1 ] = ( h1 , h2 ) .
X
(1)
(18) − σ̂eg ⊗ dω κω,n (r1 )âω,n (r1 ) + H.c.
0 n
Since each element â is uniquely represented by the
coefficient We can remark that the above construction of bright and
dark modes can also be done with global mode operators,
hi (r) = ( fˆi (r), â)) = [â, fˆi† (r)], (19) i.e. without separating them with respect to the index n.
Using the concepts defined in Section 2.2.1, we can rein-
Vf is isomorphic to L2 (R3 , C3 ) and we use the same nota- terpret the construction as follows: The bright operators
tion for the scalar product in (18). The operators fˆi (r) âω,n (r1 ) span a subspace Vâ of Vfˆ, of which they are an
form an orthonormal basis of the space Vfˆ. orthonormal basis. The dark mode operators are in the
With the indices of the present problem, each operator orthogonal complement of Vâ .
âω,n (rl ) can be uniquely represented as
2.2.3 Many emitters: orthonormalization of the operators
( fˆω0 (r), âω,n ) = [âω,n (rl ), f̂ω† 0 (r)] = hω,n (rl , r)δ(ω − ω 0 ), âω,n (ri )
(20) The generalization of this procedure is not direct for
with the commutation relations N > 1 emitters since the set of operators âω,n (ri ) is not
orthogonal. To solve this problem, we construct a set of
[âω,n (ri ), â†ω0 ,n0 (rj )] = δnn0 δ(ω − ω 0 )µi,j
ω,n , (21) bright operators that are mutually orthonormal by taking
suitable linear combinations of the âω,n (ri ):
where
N
µi,j
X
ω,n = ( hω,n (rj ), hω,n (ri ))) (22) b̂(j)
ω,n =
j,i
βω,n âω,n (ri ), j = 1, . . . , Nind , (26)
i=1
is the overlap (cross density of states, CDOS [30]) of the
modes labelled by n and ω excited by the ith and jth. where Nind ≤ N is the number of linearly independent
j,i
The operators âω,n (rl ) are thus not orthogonal in the QE- operators âω,n (ri ) and the coefficients βω,n are chosen
index. such that the new operators satisfy the orthonormality
Eur. Phys. J. D (2018) 72: 223 Page 5 of 14

condition: technique has the advantage that it gives us a global algo-

h i rithm, i.e. it considers simultaneously all the vectors to be
(j)† 0 orthonormalized. The fact formulated in Section 2.2.1 that
b̂(i)
ω,n , b̂ω 0 ,n0 = δnn0 δ(ω − ω )δij , (27a)
h i the commutators of boson creation–annihilation operators
(j) can be interpreted as scalar products allows us to apply
b̂(i)
ω,n , b̂ω 0 ,n0 = 0. (27b)
directly the results of the Löwdin orthogonalization to the
They can be constructed by the Gram–Schmidt method construction of the bright and dark boson operators.
[14] or by other orthonormalization procedures, as we Defining the following N × N matrix,
discuss below.  
In order to implement the orthonormalization proce- −1/2 −1/2
D−1/2;ω,n := diag λ1,ω,n , . . . , λNind ,ω,n , 0, ..., 0 , (31)
dure, we have to identify first the number of linearly
independent field âω,n (ri ). In [14], it has been shown that
we implement the Löwdin’s canonical orthonormalization,
if |µi,j
ω,n | = 1 with i > j, multiplying the relation (21) by obtaining a new set of bright operators
µi,j∗
ω 0 ,n0 leads to
Bω,n = Aω,n Tω,n D−1/2;ω,n , (32)
âω,n (ri ) = µi,j
ω,n âω,n (rj ). (28)
through the one row arrays
In this way we can identify pair of linearly dependent field
operators. There is also the possibility that one operator
h i
Bω,n := b̂(1)
ω,n , . . . , b̂(Nind )
ω,n , 0, . . . , 0 , (33)
can be expressed as linear combination of two or more
other ones. To identify the number Nind , we define for the Aω,n := [âω,n (r1 ), . . . , âω,n (rN )] . (34)
N L-SPP effective field operators {âω,n (rl )}, the N × N
overlap matrix Mω,n (see Appendix A): (j)
Each Löwdin operator b̂ω,n can be expressed in terms of
the old ones as follows:
µ2,1 ··· µN,1
 
1 ω,n ω,n
 1,2 .. .. ..  (
−1/2 PN i,j
. . λj i=1 Tω,n âω,n (ri ), j ∈ [1, Nind ],
 µω,n .  (j)
Mω,n := 
 .
. (29) b̂ω,n =
 .. .. .. 
N,N −1  0, j ∈ [Nind + 1, N ].
. . µω,n
−1,N (35)
µ1,N
ω,n ··· µN
ω,n 1
We can express the old operators âω,n (ri ) in terms
of Löwdin’s operators as a step of the singular value
The overlap matrix is Hermitian, so we can diagonalize it decomposition of Bω,n [24] (see Appendix B),
through a N × N unitary matrix Tω,n :
1/2 †
† Aω,n = Bω,n Dω,n Tω,n , (36)
Tω,n Mω,n Tω,n ≡ Dω,n
= diag (λ1,ω,n , . . . , λNind ,ω,n , 0, . . . , 0) . (30) by inverting (35).

We can prove that the number of non-zero eigenval- Nind

ues is the number of linearly independent operators (see 1/2
X
i,j∗ (j)
âω,n (ri ) = λj,ω,n Tω,n , b̂ω,n , i ∈ [1, N ]. (37)
Appendix A). Implementing, for example, the Gram– j=1
Schmidt orthogonalization procedure, we can obtain an
orthonormal set of Nind bright operators in terms of
If the N operators {âω,n (ri )} are linearly independent, in
{âω,n (ri )} with i = 1, . . . , Nind [14]. To obtain the ana-
equation (30), then no eigenvalue is zero. According to
lytical expressions of the coupling constants, we have to
the expressions (32) and (35), Löwdin’s method gives us
derive, from the knowledge of the overlap matrix, the
N orthonormal mode-selective bright L-SPP operators. If
expressions of the operators {âω,n (rNind +1 ), . . . , âω,n (rN )}
Nind < N , at least one of the eigenvalues is zero and the
in terms of {âω,n (r1 ), . . . , âω,n (rNind )}, and substitute
new set in (35) shows a reduced dimension. The expression
them in the interaction Hamiltonian in terms of the new
(37) can be directly substituted in the interaction term
orthonormal ones.
of (16) (without the knowledge of the expressions of the
operators {âω,n (rNind +1 ), . . . , âω,n (rN )}):
2.2.4 Löwdin orthonormalization of the operators âω,n (rl )
and singular value decomposition N
X Z +∞ XN
X ind
(i)
The Gram–Schmidt method is a sequential technique of Ĥint = −i~ σ̂eg ⊗ dω κi,j (j)
ω,n b̂ω,n + H.c.
i=1 0 n j=1
orthonormalization, i.e. the mth new operator can be
obtained after having derived the m − 1 ones. In this (38)
(j)
paper, we generalize the Löwdin’s canonical orthonor- The coupling constants of each bright field b̂ω,n with each
malization [24] to the case in which N initial operators emitter are expressed in terms of the eigenvalues and of the
are not necessarily linearly independent. Moreover, this eigenvectors of the overlap matrix Mω,n in the following
Page 6 of 14 Eur. Phys. J. D (2018) 72: 223

compact form: 2.2.6 Löwdin orthonormalization for N = 2

For the two L-SPP effective field operators âω,n (r1 ) and
1/2
κi,j i,j∗
ω,n = κω,n (ri )λj,ω,n Tω,n . (39) âω,n (r2 ), the overlap matrix
 
Identifying (35) with (26), we obtain 1 µ2,1
ω,n
Mω,n =   (45)
j,i −1/2 i,j µ1,2
ω,n 1
βω,n = λj Tω,n . (40)
can be diagonalized by a unitary matrix
2.2.5 Construction of the dark operators
1 µ2,1 1 µ2,1
 
The bright operators are the only ones that appear in the ω,n ω,n
interaction term of the Hamiltonian (38). The final step √ −√
 2 |µ1,2
ω,n | 2 |µ1,2
ω,n | 

is to express the free term of the Hamiltonian in terms Tω,n =  (46)
of the bright operators and another set of operators, the
 
 1 1 
dark operators, which are not coupled to the emitters and √ √
that are orthogonal to the bright operators. Once we have 2 2
(j)
the set of bright operators {b̂ω,n } that are orthonormal to such that
each other, the construction of the dark operators takes
the same general form as what we did for the single emitter †
case. We define the dark operators as follows: Tω,n Mω,n Tω,n = Dω,n := diag (λ1,ω,n , λ2,ω,n ) (47)

XN
Xind with the eigenvalues
d̂ω (r) = f̂ω (r) − b̂(i) (i)†
ω,n [f̂ω (r), b̂ω,n ]. (41)
n i=1 λ1,ω,n = 1 + |µ1,2
ω,n |, λ2,ω,n = 1 − |µ1,2
ω,n |. (48)

The bright and dark operators satisfy the following prop- We define the matrix D−1/2;ω,n that, if Nind = N = 2, has
erties: the following form:
 
−1/2 −1/2
(i) [b̂(i)†
ω,n , d̂ω (r)] = 0, [b̂(i)
ω,n , d̂ω (r)] = 0, (42) D−1/2;ω,n = diag λ1,ω,n , λ2,ω,n ; (49)
Z ∞ Z
(ii) H0 = dω ~ω d3 r f̂ω† (r) · f̂ω (r)
0
otherwise,
Z ∞ XN
X ind  
−1/2
= dω ~ω b̂(i)† (i)
ω,n b̂ω,n D−1/2;ω,n = diag λ1,ω,n , 0 . (50)
0 n i=1
Z ∞ Z
+ dω ~ω d3 r d̂†ω (r) · d̂ω (r). (43) Implementing the Löwdin’s canonical orthonormalization,
0 we obtain a new set of bright operators:
" #
The final mode-selective continuous microscopic model 1 1 µ2,1
ω,n
can be written as follows: b̂(1)
ω,n =√ q âω,n (r2 ) + 1,2 âω,n (r1 ) ,
2 1 + |µ1,2 | |µω,n |
ω,n
Z ∞ XN
Xind N
X (51a)
Ĥ = dω ~ω b̂(i)† (i)
ω,n b̂ω,n +
(i)
~ωeg σ̂ee
0 n i=1 i=1
(44)
N Z +∞ XNind " #
1 1 µ2,1
X X
(i)
− i~ σ̂eg ⊗ dω κi,j (j)
ω,n b̂ω,n + H.c. b̂(2) =√ q
ω,n
âω,n (r2 ) − 1,2 âω,n (r1 ) ,
ω,n
i=1 0 n j=1 2 1 − |µ1,2 | |µω,n |
ω,n
(51b)
We remark that, in practice, in order to write the effective
Hamiltonian we only need to determine the coupling con-
stants κi,j j,i satisfying the orthonormality condition (27). If the
ω,n . The coefficients βω,n (40) are only needed for
the theoretical justification of the separation of the bright two operators are linearly independent, |µ1,2
ω,n | =
6 1 and
−1/2
and the dark modes. λ2,ω,n 6= 0, the Löwdin’s method gives us two new
The Löwdin’s method provides simple compact for- orthonormal mode-selective L-SPP operators. If Nind = 1,
1,2
mulas using an algorithm that is more stable than the |µω,n | = 1 [14], the second eigenvalue is zero: the technique
Gram–Schmidt algorithm in numerical implementations, returns a single new operator.
allowing in principle the treatment of a large number of Expressing the old effective operators âω,n (ri ) in terms
emitters. of the new one(s) and substituting them in the interaction
Eur. Phys. J. D (2018) 72: 223 Page 7 of 14

term of equation (16), we obtain the coupling constants:

µ1,2
ω,n 1
q
κ1,1
ω,n = κω,n (r1 ) √ 1 + |µ1,2
ω,n |, (52a)
|µ1,2
ω,n | 2

µ1,2
ω,n 1
q
κ1,2
ω,n = −κω,n (r1 ) 1,2 √ 1 − |µ1,2
ω,n |, (52b)
|µω,n | 2

1
q
κ2,1
ω,n = κω,n (r2 ) √ 1 + |µ1,2
ω,n |, (52c)
2

1
q
κ2,2
ω,n = κω,n (r2 ) √ 1 − |µ1,2
ω,n |. (52d)
2

We can see that if |µ1,2ω,n | = 1, i.e. the two original non-

orthogonal operators are not linearly independent, the
theory can be formulated only in terms of the first oper-
(1)
ator b̂ω,n and the coupling constants linking the two QEs
to the second field, κ1,2 2,2
ω,n and κω,n , are automatically zero,
as shown by (52b) and (52d).

2.3 Discrete effective model

A next step consists of getting rid of the dependence on
the continuous parameter ω and of truncating the infinite
Hilbert space of the system.
It has been proved in [14] that the QEs-L-SPP coupling
constants can be very well approximated by a Lorentzian
function for each plasmonic mode:
r
γn gn (ri )
κω,n (ri ) = gn (ri )Ln (ω) = , (53)
2π ω − ωn + iγn /2 Fig. 2. Local density of states associated with the different
excited LSPP modes, for a radially polarized emitter placed at
where Ln (ω) indicates the Lorentzian function, γn is its a distance d = 2 nm (upper frame) and d = 7 nm (lower frame),
half-width, i.e. the lossy rate of the nth mode, and ωn is its respectively, from the surface of a spherical silver MNP of
resonance frequency. Exploiting this coupling shape (53), radius R = 8 nm (with plasmonic resonances in the frequency
it is possible to get rid of the dependence on the continuous range ω ' (2.7 − 3) eV [19]).
(i)
parameter ω by defining Nind effective operators b̂n [14]:
where the eigenvalues and the eigenvectors of the overlap
Z +∞
matrix are considered at the resonance frequency of the
b̂(i)
n = dωLn (ω)b̂(i)
ω,n , (54)
0
mode n: Mωn ,n .
The effective model (55) can be truncated since not all
allowing us to derive a discrete effective model describ- the L-SPP modes play a role in the coupled dynamics.
ing the interaction of each QE with bosonic plasmonic Indeed, the LDOS of some modes at certain distances from
resonances: the MNPs surface, i.e. at the QEs-positions, is negligible.
Two example of this are plotted in Figure 2.
XN
Xind
γn (i)† (i) X
N
(i)
Consequently, we truncate the model keeping n ∈ [0, n0 ],
Ĥd = ~(ωn − i )b̂n b̂n + ~ωeg σ̂ee where n0 is the harmonic index associated with the last
2
n i=1 i=1 L-SPP mode effectively involved in the coupled dynam-
(55)
N
X XN
X ind ics. For n > n0 , the LDOS at the position of the QE is
(i)
− i~ σ̂eg ⊗ [gni,j b̂(j)
n − H.c.] approximately zero.
i=1 n j=1 In the case of N = 2, if we assume the initial state
|e, g; [0]i (indicating that the first emitter is in its excited
with state, the second one is in its ground state, and both plas-
monic fields are in their ground state, i.e. |[0]i = |0, 0i),
1/2
gni,j = gn (ri )λj,ωn ,n Tωi,j∗
n ,n
, (56) or more generally an initial state coupled to |e, g; [0]i, the
Page 8 of 14 Eur. Phys. J. D (2018) 72: 223

coupled. The MNP, which is coupled with both emitters,

plays the same role of the intermediate lossy excited state
|2i but in a tensored product representation for the two
atoms and the plasmonic dressing field.
For two 3-state QEs we have to introduce two additional
states in the basis (57), |f, g; [0]i and |g, f ; [0]i. We assume
that the state |Ψ(t)i of the global hybrid system is initially
prepared as follows:

|Ψ(t → −∞)i = |f, g; [0]i. (59)

We seek a control setup to obtain a final population

Fig. 3. (a) Allowed and forbidden transitions in a standard exchange of the two QEs, i.e. |Ψ(t → +∞)i = |g, f ; [0]i.
3-level system in which STIRAP is usually applied. (b) Allowed We consider the STIRAP configuration with the pump
and forbidden interactions in the analysed QEs-LSPs system. and Stokes pulses, as shown in Figure 4a (in a system of
units such that ~ = 1):

single excitation subspace spanned by the following basis, XN ind

γn (i)† (i) X
N
(i)

X
(i)
Ĥd = (ωn − i )b̂n b̂n + ωeg σ̂ee + ωf g σ̂f f
B01 ≡ {|e, g; [0]i, |[g]; 11 , 0i, . . . , |[g]; 1n0 , 0i, n i=1
2 i=1
(57)
|[g]; 0, 11 i, . . . , |[g]; 0, 1n0 i, |g, e; [0]i}, N
X XN
X ind
(i)
−i σ̂eg ⊗ (gni,j b̂(j)
n − H.c.)
is involved in the dynamics. The state |[g], 1 , 0i refers n0 i=1 n j=1
to the condition in which both emitters are in their states (1) (2)
P (t)e−iωP t σ̂ef + S(t)e−iωS t σ̂ef + H.c. .
 
+
|gi, i.e. |[g]i = |g, gi, and there is one excitation in the
n0 th mode of the first plasmonic field. The effective matrix (60)
Hamiltonian in the basis (57) is
In the resonant case, ΩP = ωS = ωef , the complete matrix
Hamiltonian reads in the dressed basis
Ĥd =
0 ··· gn1,1 ··· ··· gn1,2 ··· 0 {|f, g; [0]i, B01 , |g, f ; [0]i},
 
(61)
 .. .. .. 
 . . 0 ··· 0 . 
.. ..
 
 1,1∗ γn 2,1∗ 
 H=
 gn ∆n − i . . gn 

 . 2 
0 P ··· 0 ··· ··· 0 ··· 0 0

 . .. ..  1,1 1,2
 . . 0 ··· 0 . 
  P 0 ··· gn ··· ··· gn ··· 0 0 
,
 
 .  . .. .. ..
.. ..  .. 
 .  .
.

 . 0 ··· 0 . .   . . 0 ··· 0 . .
 
 
.. ..
 1,2∗ .. .. γn γn
   
1,1∗ 2,1∗
gn2,2∗   0 gn ∆n − i . . gn 0
 
∆n − i

 gn . . 2


 . 2  
 . .. .. .. ..


 .. .. ..   .
 . . . 0 ··· 0 . .

0 ··· 0 . .  

,
 . .. .. .. ..
gn2,1 2,2

0 ··· ··· ··· gn ··· 0  .
 . . 0 ··· 0 . . .


(58)
 
 .. .. γn 
1,2∗ 2,2∗
 0 gn . . ∆n − i gn 0
 

 2 
with the detuning from the resonance of the mode n: ∆n =  .
 . .. .. .. .. 
.

ωn − ωeg .  . . 0 ··· 0 . . 
2,1 2,2
 
 0 0 ··· gn ··· ··· gn ··· 0 S 
0 0 ··· 0 ··· ··· 0 ··· S 0
3 STIRAP process (62)
where we have used the resonant transformation
3.1 Presentation
H = U † Hd U − iU † ∂t U − ωeg 1 (63)
We analyze the possible quantum channel that exploits
the plasmonic coupling of each emitter with the MNP, with
through which two QEs can exchange their population.
We consider the STIRAP process between two atoms. U = diag(eiωP t , 1, . . . , 1, eiωS t ). (64)
Figure 3 shows a qualitative sketch of the analogy between
the 3-level system (Fig. 3a) in which STIRAP is usually This system can be interpreted as a chainwise-linked
applied and the present system (Fig. 3b). Like the two system [31], but by blocks that include multiple inter-
states |1i and |3i involved in the transfer in Figure 3a, mediate states [32], in which the total Hamiltonian has
the two emitters in Figure 3b are assumed not directly a (2n + 2) × (2n + 2) diagonal central block.
Eur. Phys. J. D (2018) 72: 223
Fig. 4. (a) Multilevel scheme of the allowed transitions char-
acterizing the laser–controlled system composed of two QEs
and the MNP. P (t) and S(t) are the Rabi frequencies asso-
ciated with the interactions with the pump and the Stokes
laser, respectively. (b) Effective 3-state system obtained after
partial adiabatic elimination. The Rabi frequencies P̃ (t) and
S̃(t) quantify the coupling strength of the states |f, g, 0, 0i and
|g, f, 0, 0i, respectively, with |Φ0 i.
3.2 Adiabatic transfer state
In analogy to the cases without losses [31,32], the best
strategy to achieve an efficient resonant STIRAP-type
transfer is to find a dark state, i.e. with no compo-
nent of the lossy excited states, i.e. an eigenstate of
zero eigenvalue, which is sufficiently well adiabatically
separated from the other intermediate states to prevent
non-adiabatic couplings with them. In order to determine
a global dark state immune to losses, we have to find
an eigenstate in the subsytem (58), the inner part of the
Hamiltonian (62), which will be adiabatically eliminated
from the others and which we will couple resonantly via P
and S. This resulting resonant intermediate state should
feature low losses. More general adiabatic states featur-
ing transient contributions from the intermediate states
will be expected to introduce lossy parts and to produce
thus inefficient transfer. A possible but less efficient strat-
egy consists in devising a far-detuned process from the
excited states in order to achieve an adiabatic elimination
of them.
Page 9 of 14
We define φ as the angle between the position vectors of
the two QEs. We anticipate favorable configurations with
an efficient transfer between the two QEs mediated by
plasmons, when they satisfy an overlap |µ12 ωn ,n | ' 1 (for
all n), i.e. for the cases in which the two emitters are
aligned in the same side (φ = 0) or in the two opposite
ones of the sphere (φ = π), almost at the same distance
from the metallic surface. We determine the eigenstates of
the subsytem (58) in these two extreme cases, for φ = 0
and φ = π, for which gn1,2 = gn2,2 = 0. For the analysis,
we first consider a simplified situation for which ∆n = 0
and all the dissipative rates are equal: γn = γ and all the
couplings are equal in absolute value (and real).
For φ = 0, i.e. gn1,1 = g = gn2,1 , we obtain the eigenvalues
(the second one being degenerate):
r  γ 2 i
γ 1h γ
λ0 = 0, λg = −i , λ± = −i ± 8N g 2 − .
2 2 2 2
(65)
We thus obtain a dark state of zero eigenvalue, featuring
thus no loss, which will be targeted as the intermediate
state of the full system. This shows that for P, S  γ,
the states corresponding to λg can be safely adiabatically
eliminated with respect to this state of zero eigenvalue,
which can be referred to as a partial adiabatic elimina-
tion. This is also the case for the states corresponding to
λ± if N g  P, S even if γ is still much larger than g as it
is the case when the strong coupling is not fully achieved.
We conclude that, when the losses are strong, the par-
tial adiabatic elimination is better achieved for larger N
as
√ the distance with the zero eigenvalue state grows as
N . This can be interpreted as a cooperative effect of all
the modes. In a more realistic situation, i.e. with differ-
ent values of g and γ and adding detunings, we can show
that we still obtain a dark state of zero eigenvalue if we
select an appropriate detuning with the emitter and that
the preceding conclusions of partial adiabatic elimination
still hold. The multilevel system can be thus reduced to
an effective 3-state subspace spanned by
{|f, g; [0]i, |Φ0 i, |g, f ; [0]i}, (66)
where |Φ0 i is the preceding dark state of zero eigenvalue of
the subsytem (58) (see Fig. 4b). Its coupling to |f, g; [0]i
is denoted P̃ (t) and its coupling to |g, f ; [0]i S̃(t).
For φ = π, due to the change of signs of the g couplings,
the eigenvalue structure when many modes are involved
is more complicated and zero eigenvalue associated with a
state well separated from the others cannot be found; the
many-mode cooperative effect does not apply. The pop-
ulation transfer will be expected to be either lost in the
metal during the process or blocked to the initial state for
strong losses. In this case a strategy could be to far-detune
the process in order to achieve a full adiabatic elimination
of all the intermediate states.
In the following, we numerically verify these assump-
tions.
Page 10 of 14
Fig. 5. Dynamics of the population transfer from |f, g, 0, 0i to
|g, f, 0, 0i for φ = 0 (middle frame) and φ = π (lower frame)
for a MNP of R = 8 nm interacting with two dipoles of 10 D,
one positionned at 2 nm and the other one at 4 nm from the
metallic surface, and Gaussian profile of the two delayed and
counterintuitively ordered laser pulses (upper frame). The area
of each pulse is Ω0 T = 60 and the delay is τ /T = 0.7, where
T ' 1 ns. The transient population transfer in the excited
states is less than 10−2 .
3.3 Dependence of the STIRAP process
on the angular distance of the QEs
We numerically analyze the dependence of the STIRAP
process on the angular distance of the QEs for a metal-
lic sphere of radius R = 8 nm, characterized by a
Drude dielectric function m (ω) [10], and Gaussian laser
pulses:
2 2
P (t) = Ω0 e−[(t−τ )/T ] S(t) = Ω0 e−[(t+τ )/T ] , (67)
where 2τ is the delay between the two pulses, T is the
pulse width, and Ω0 is the peak Rabi frequency.
Numerical simulations are presented in Figure 5, for
QEs placed very close to the surface (2 and 4 nm, respec-
tively) in order to approach the strong coupling. We have
numerically found that a STIRAP process can be well
implemented for φ = 0 (see Fig. 5b) since more than 90%
of the population can be transferred from the initial state
|f, g; [0]i to the target state |g, f ; [0]i with a negligible
plasmonic population during all the dynamics (giving to
the plasmon polaritons the role of a dark subsystem). On
the other hand, if the QEs are aligned in the opposite
sides of the MNP (φ = π), the population is blocked in
the initial state (see Fig. 5c) as anticipated.
We have numerically investigated the dependence of
the transfer efficiency on the angular distance φ in
Figure 6. We found that, under adiabatic conditions, a
good population transfer (not less than 70%) can be
Eur. Phys. J. D (2018) 72: 223
Fig. 6. Contour plot of the transfer efficiency |hg, f, 0, 0|Ψ(t =
+∞)i|2 as a function of the area Ω0 T of the laser pulses and
of the angle φ ∈ [0, π] between the position vectors of the two
emitters from the center of the MNP.
obtained for very small angles up to π/15. For larger
angles, Figure 6 shows a negligible population transfer.
This shows that the blockade takes place already for a
quite small angle.
3.4 Dependence of the STIRAP process
on the number of plasmonic modes
For a distance between the QEs and the metallic surface
of 2 or 4 nm, we have considered the first 25 plasmonic
modes whose role in the transfer is not negligible. We now
analyze the effect of each plasmonic mode in the popula-
tion transfer by artificially truncating the basis choosing
the number of modes in Figure 7. Figure 7a shows that an
efficient transfer occurs for φ = 0 or φ = π (and around
these angles) when the sole dipolar mode is taken into
account. Involving more and more modes has a detrimen-
tal effect on the transfer for φ = π and also reduces the
angular width of transfer around φ = 0. The blockade of
transfer for φ = π appears as expected when the over-
lap between the modes n of the two emitters is given by
µ12 n 12
ωn ,n = (−1) . On the other hand, for φ = 0, µωn ,n = 1
should result in a cooperative effect of all the modes as
explained above. This would be apparent if the emitters
were placed exactly at the same position, i.e. we would
get in this case a better transfer efficiency for a larger
number of modes. Here, considering a more realistic sit-
uation with distant emitters leads, when φ = 0, to a less
efficient transfer for a larger number of modes, which can
be however compensated by increasing the pulse areas.
The distance between each emitter and the metallic sur-
face determines the number of plasmonic modes effectively
involved in the transfer. The greater the distance, the
smaller the number of the involved modes (see Fig. 2).
Increasing the distance d from the metallic surface up
to 7 nm, for φ = π, weakens the blockade present for
d = 2 nm (see Fig. 8b) and almost 50% of the population
is transferred to the target state (see Fig. 8c).
Eur. Phys. J. D (2018) 72: 223
Fig. 7. Countour plots of the transfer efficiency |hg, f, 0, 0|Ψ(t = +∞)i|2 as a function of the angular distance φ ∈ [0, π] and of
the area of the laser pulses Ω0 T , for the artificial truncation of the number of modes, as indicated in (a)–(d).
4 Discussions and conclusions
We have constructed an effective model describing the
interaction of N QEs with a spherical metal nanoparti-
cle in an arbitrary geometric arrangement of the QEs. We
have separated the different contributions of the L-SPPs
supported by the MNP, in this way improving other effec-
tive models present in the literature [15,16]. Simplifying
preceding procedures [14], we have developed a general
formulation including the decomposition into orthogo-
nal plasmonic modes, generalizing the Löwdin’s canonical
algorithm of orthogonalization [24], with the example of
spherical symmetry.
Completing the analysis of reference [19], we have
shown that a Stimulated Raman Adiabatic Passage con-
figuration allows an efficient population transfer, as an
exchange of population between two QEs, when the QEs
are located very close to the MNP and for a specific range
of angular distances between the QEs, placed on the same
side of the MNP. Outside of this range up to the configu-
ration in which the emitters are positioned each other at
the opposite sides of the MNP, the transfer is blocked. We
have investigated the blockade effect in terms of the num-
bers of plasmonic modes involved in the process. For QEs
at the same side of the MNP, the coupling is magnified
by all the modes. On the other hand, when they are very
close to the MNP (a few nanometers and so in the strong
coupling regime), population transfer is blocked for QEs
Page 11 of 14
located at the opposite sides, which can be interpreted
as the destructive superposition of all the interacting
plasmonic modes. If only the dipolar mode was involved,
the blockade would be prevented and efficient population
transfer would be obtained in the strong coupling regime
also when the emitters are positioned at the opposite sides
of the MNP. However, to magnify the relative role of the
dipolar mode with respect to the other modes requires an
intermediate distance featuring a relatively weak coupling
regime, for which only a partial transfer can be obtained
(not greater than 70%). For a spherical MNP, one cannot
thus find a compromise between strong coupling and
dipolar modes only involved, which would lead to efficient
population transfer. It would be interesting to analitically
verifying our numerical results in terms of intermediate
dressed states, in order to derive general conditions of the
different parameters (e.g. single-photon detunings, pump
and Stokes Rabi frequencies, couplings of the emitters
to the plasmonic modes) for existence of an AT-state
[32] in the presence of plasmonic losses. One can thus
anticipate that ellipsoids or nanorods with an enhanced
dipolar mode coupling would be more appropriate for the
required population transfer.
A generalization to the case of N emitters will offer
the possibility to produce at the nanoscale a N -qubit
processor, via a closest neighbour effective coupling, and
also for the design of quantum simulators using plasmonic
nanoparticles.
Page 12 of 14 Eur. Phys. J. D (2018) 72: 223

Author contribution statement

S.G., H.R.J, and G.Cd.F. conceived the initial idea and
supervised the project. S.G., H.R.J. and A.M. coordinated
the work. A.C., B.R., and D.D. made the theoretical calcu-
lations. All the authors contributed in writing and editing
the manuscript.

Appendix A: Overlap matrix

We consider a set of N vectors |ci i ∈ H, that span a
subspace Hind ⊂ H, whose dimension coincides with the
number Nind ≤ N of independent vectors in the defined
set. We want to determine the dimension of this subspace.
For this purpose, we define a one row array and its formal
adjoint
 
hc1 |
C := [|c1 i, ..., |cN i] , C† :=  ...  , (A.1)
 
hcN |

and we define a rule of multiplication that is similar to

the tensor product of two vectors:
Fig. 8. Dynamics of the population transfer from |f, g, 0, 0i
to |g, f, 0, 0i for φ = π and d = 2 nm, for which 25 modes  
hc1 , b1 i · · · hc1 , bN i
are involved and featuring the blockade (middle frame), and
C† B =  .. .. ..
. (A.2)
 
d = 7 nm, for which only 7 modes are involved (lower frame). . . .
Gaussian profile of the two delayed and counterintuitively hcN , b1 i · · · hcN , bN i
ordered laser pulses (upper frame). The area of each pulse is
Ω0 T = 90 and the delay is τ /T = 0.8, where T ' 15 ns.
We introduce the overlap matrix M , called also metric
matrix or Gram matrix corresponding to the set of vectors
|ci i as
In an experimental point of view, the strong coupling  
hc1 , c1 i · · · hc1 , cN i
regime should correspond to g ∼ γ. In the present sys- .. ..
M := C† C = 
 .. .

(A.3)
tem, even for the small distances considered between the . . .
emitters and the nanoparticle, the strong coupling is not hcN , c1 i · · · hcN , cN i
really achieved since we have for the highest coupling
contant g ∼ 0.02γ with the typical width in this system It is an N × N Hermitian and positive semi-definite matrix
γ ∼ 8 × 1013 s−1 . Considering a pulse duration of 1 ns as shown below.
leads to γT ∼ 8 × 104 and gT ∼ 1600. A typical pulse
area of order Ω0 T ∼ 50 in this problem leads us in the Lemma A.1. Let {|ϕi i} be an arbitrary set of N vectors,
situation γ  g  Ω0 , which is very favorable for popu- then the eigenvalues of the overlap matrix M with ele-
lation transfer when the emitters are closer to each other ments Mij := hϕi , ϕj i are positive or zero. Furthermore,
since a zero eigenvalue intermediate state well adiabati- if the vectors |ϕi i are linearly independent, all the eigen-
cally separated from the others (via the dissipation) can values of M are strictly positive and thus M is invertible.
be found (partial adiabatic elimination) and used for the
STIRAP transfer. But it blocks the transfer when they Proof. Let vk ∈ CN be an eigenvector of M , i.e. M vk =
are at the opposite sides of the nano–sphere since all the λk vk and define the following linear combination:
intermediate states are too strongly separated by the dissi-
pation from the resonance. This corresponds to peak laser X
pulses of Rabi frequency of 500 GHz, i.e. of peak intensity |φk i := |ϕi ivik . (A.4)
5 GW/cm2 for a dipole of 10D. i

Then
This work was supported by the French “Investissements X X X
d’Avenir” program, project ISITE-BFC / I-QUINS (con- hφk , φk i = hϕi , ϕj ivik∗ vjk = vik∗ Mij vjk
tract ANR-15-IDEX-03). We acknowledge additional support ij i j
X X (A.5)
from the Labex ACTION program (ANR-11-LABX-01-01) and = vik∗ λk vjk = λk vik∗ vik = λk |vk |2 ,
PLACORE (ANR-BS10-0007). i i
Eur. Phys. J. D (2018) 72: 223 Page 13 of 14

and thus non-orthogonal linearly independent vectors, both meth-

ods consider simultaneously all of them in the process of
||φk ||2 construction of the orthonormal set [24]. We generalize the
λk = ≥ 0. (A.6) canonical method to the case in which the non-orthogonal
|vk |2
vectors are not necessarily linearly independent.
The strict inequality follows from the fact that the linear We introduce a set {|ci i} of vectors and the corre-
independence of the |ϕi i implies that |φk i =
6 0. sponding overlap matrix M . The vectors |ci i are not
mutually orthogonal, i.e. hci , cj i 6= 0. We want to obtain
The rank of a set of vectors |ci i ∈ H, defined as the from them an orthonormal set {|bi i}. The overlap matrix
dimension of the subspace they span, is equal to M is Hermitian and so it can be diagonalized by a unitary
matrix T
– the maximal number of linearly independent vectors;
– the dimension of the image of the linear map defined T † M T = D = diag(λ1 , ..., λNind , 0, ..., 0), (B.1)
by
where the presence of the zero eigenvalues indicates the
C: CN 7→ H (A.7) possibility that the vectors |ci i are not linearly indepen-
N
X dent. Defining the following matrix,
v 7→ Cv = |ci ivi , (A.8)
i=1 −1/2 −1/2
D−1/2 := diag(λ1 , ..., λNind , 0, ..., 0), (B.2)
i.e. equal to N minus the dimension of the kernel of according to the Löwdin canonical method, we can obtain
this map. an orthonormal set of vectors B:
Lemma A.2. A vector v ∈ CN satisfies the following
condition: B := CT D−1/2 , (B.3)

Cv = |0i ⇐⇒ M v = 0, (A.9) where, following the notation introduced in (A.1), the

form of B is
where the vector |0i indicates the vector zero in H and 0
is the vector zero in CN . This means that the kernel of the B = [|b1 i, ..., |bNind i, |0i, ..., |0i] , (B.4)
map C is equal to the kernel of the matrix M .
and C is the one row array defined in (A.1).
Proof. The new set of vectors B satisfies indeed the orthonor-
(⇒): M v = C† Cv and thus Cv = |0i implies M v = 0. mality condition
(⇐): If M v = C† Cv = 0, it implies that
B† B = D−1/2 T † M T D−1/2
N
X = D−1/2 DD−1/2 (B.5)
0 = v · C† Cv = vi∗ hci , cj ivj


i,j=1 = diag(1Nind , 0, ...0),
* + (A.10)
X X 0 0
= vi c i , vj cj = hCv, Cvi, i.e. hbj , bj i = δjj 0 , for all j, j ∈ [1, Nind ].
i j According to (B.3), the vectors |bj i ∈ H can be written
in terms of the original ones in the following explicit form:
and thus Cv = |0i. (
−1/2 PN
Lemma A.3. The number Nind of linearly independent |bj i :=
λj i=1 |ci iTi,j , for j ∈ [1, Nind ],
vectors |ϕi i is equal to the rank of the overlap matrix M , |0i, for j ∈ [Nind + 1, N ].
which is equal to the number of its non-zero eigenvalues. (B.6)
The relations (B.3) and (B.6) can be inverted to express
Proof. The rank of C is equal to N minus the dimension the original vectors in terms of the orthonormal basis:
of the kernel of the map C and, from Lemma A.2, the
dimension of this kernel is equal to the dimension of the C := BD1/2 T † , (B.7)
eigenspace of M corresponding to the eigenvalue 0.
i.e.
Appendix B: Löwdin canonical Nind
orthonormalization for non-linearly 1/2
X
∗
|ci i := λj Ti,j |bj i, i ∈ [1, N ]. (B.8)
independent vectors j=1

Löwdin introduced two global methods of orthonormaliza- We remark that (B.7) has the form of a singular value
tion of a vector: the canonical approach and the symmetric decomposition extended for vectors |ci i in an infinite
one. The adjective global indicates that, given a set of dimensional Hilbert space [33–35]. The canonical Löwdin
Page 14 of 14 Eur. Phys. J. D (2018) 72: 223

orthonormalization can thus be viewed as the inverse of
the singular value decomposition of C.
To prove (B.8), we start from the definition (B.6),
obtaining
(P
N
1/2
λj |bj i := i=1 , |ci iTi,j , for j ∈ [1, Nind ],
|0i, for j ∈ [Nind + 1, N ],
N (2016)
X
= |ci iTi,j , for j ∈ [1, N ],
i=1
where the second equality follows from Lemma A.2: for
j ∈ [Nind + 1, N ], the columns of the overlap matrix are
the eigenstates tj corresponding to the eigenvalue 0, i.e.
M tj = 0, and, by Lemma A.2, Ctj = |0i.
Multiplying both sides of (B.9) by Ti∗0 ,j and summing over
j, we obtain
N N
1/2
X X
|bj iλj Ti∗0 j = |ci i
j=1 i=1
N
X Commun. 8, 14148 (2017)
= |ci iδi,i0 = |ci0 i,
i=1
which is the singular value decomposition (B.7).
References
1. E. Le Ru, P. Etchegoin, Principles of surface enhanced
Raman spectroscopy and related plasmonic effects
(Elsevier, 2009)
2. S.A. Maier, Plasmonics: fundamentals and applications
(Springer, 2009)
3. A.V. Zayats, I.I. Smolyaninov, J. Opt. A 5, S16 (2003)
4. E. Ozbay, Science 311, 189 (2006)
5. P. Ginzburg, Rev. Phys. 1, 120 (2016)
6. G. Colas des Francs, S. Derom, R. Vincent, A. Bouhelier,
A. Dereux, Int. J. Optics 2012, 175162 (2012)
7. L. Knöll, S. Scheel, D.-G. Welsch, QED in dispersing and
absorbing dielectric media, in Coherence and Statistics of
Photons and Atoms (Wiley, New York, 2001), p. 1
8. D. Dzsotjan, A.S. Sørensen, M. Fleischhauer, Phys. Rev.
B 82, 075427 (2010)
9. C. Van Vlack, P.T. Kristensen, S. Hughes, Phys. Rev. B
85, 075303 (2012)
10. J. Hakami, L. Wang, M.S. Zubairy, Phys. Rev. A 89,
053835 (2014)
11. L.G. Suttorp, A.J. van Wonderen, Europhys. Lett. 70, 766
(2004)
12. B. Huttner, S.M. Barnett, Phys. Rev. A 46, 4306 (1992)
13. T.G. Philbin, New J. Phys. 12, 123008 (2010)
14. D. Dzsotjan, B. Rousseaux, H.R. Jauslin, G.C. des
Francs, C. Couteau, S. Guérin, Phys. Rev. A 94, 023818
15. A. Delga, J. Feist, J. Bravo-Abad, F.J. Garcia-Vidal, Phys.
Rev. Lett. 112, 253601 (2014)
16. A. Delga, J. Feist, J. Bravo-Abad, F.J. Garcia-Vidal, J.
(B.9)
Opt. 16, 114018 (2014)
17. D. Chang, A. Sorensen, E. Demler, M. Lukin, Nat. Phys.
3, 807 (2007)
18. C.E. Susa, J.H. Reina, R. Hildner, Phys. Lett. A 378, 2371
(2014)
19. B. Rousseaux, D. Dzsotjan, G. Colas des Francs, H.R.
Jauslin, C. Couteau, S. Guérin, Phys. Rev. B 93, 045422
(2016) [Erratum: Phys. Rev. B 94, 199902 (2016)]
20. K. Bergmann, N.V. Vitanov, B.W. Shore, J. Chem. Phys.
N 142, 170901 (2015)
21. N.V. Vitanov, A.A. Rangelov, B.W. Shore, K. Bergmann,
X
Ti,j Ti∗0 ,j
Rev. Mod. Phys. 89, 015006 (2017)
j=1
(B.10) 22. S.P. Premaratne, F.C. Wellstood, B.S. Palmer, Nat.
23. K.S. Kumar, A. Vepsäläinen, S. Danilin, G.S. Paraoanu,
Nat. Commun. 7, 10628 (2016)
24. R.N. Annavarapu, AJCAM 3, 33 (2013)
25. H. Varguet, B. Rousseaux, D. Dzsotjan, H.R. Jauslin,
S. Guérin, G. Colas des Francs, Opt. Lett. 41, 4480
(2016)
26. P.-O. Löwdin. J. Chem. Phys. 18, 365 (1950)
27. P.-O. Löwdin. Adv. Phys. 5, 1 (1956)
28. P.-O. Löwdin, Linear Algebra for Quantum Theory (John
Wiley, 1998)
29. H.T. Dung, L. Knöll, D.-G. Welsch, Phys. Rev. A 57, 3931
(1998)
30. R. Carminati, A. Cazé, D. Cao, F. Peragut, V.
Krachmalnico, R. Pierrat, Y. De Wilde, Surf. Sci. Rep.
70, 1 (2015)
31. N.V. Vitanov, B.W. Shore, K. Bergmann, Eur. Phys. J. D
4, 15 (1998)
32. N.V. Vitanov, S. Stenholm, Phys. Rev. A 60, 3820 (1999)
33. J.H. Wilkinson, The Algebraic Eigenvalue Problem
(Oxford University Press, 1965)
34. L.N. Trefethen, D. Bau, Numerical Linear Algebra (SIAM,
Philadelphia 1997)
35. W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P.
Flannery, Numerical Recipes: The Art of Scientific Com-
puting, 3rd edn. (Cambridge University Press, 2007)

View publication stats