7472 J. Phys. Chem.

C 2010, 114, 7472–7479

Optical Forces in Plasmonic Nanoparticle Dimers†

Vladimir D. Miljković,‡ Tavakol Pakizeh,‡,§ Borja Sepulveda,| Peter Johansson,‡,⊥ and

Mikael Käll*,‡
Department of Applied Physics, Chalmers UniVersity of Technology, S-412 96 Göteborg, Sweden,
Department of Electrical Engineering, K.N. Toosi UniVersity of Technology, 16314 Tehran, Iran,
Nanobiosensors and Molecular Nanobiophysics Group, Research Centre in Nanoscience and
Nanotechnology (CIN2) CSIC-ICN, 08193 Bellaterra, Barcelona, Spain, and School of Science and Technology,
Örebro UniVersity, S-701 82 Örebro, Sweden
ReceiVed: NoVember 30, 2009; ReVised Manuscript ReceiVed: January 26, 2010

We present calculations of the optical forces between two metal nanospheres forming a hybridized
plasmonic dimer. We consider homo- and heterodimers and investigate different plane wave illumination
configurations. The forces between the particles are calculated using full Mie theory combined with the
Maxwell stress tensor (MST) formalism, as well as by approximate methods, such as the Lorentz force
(LF) approach taken in the dipole limit and calculations based on an optical potential. We show that the
simplified calculation schemes can lead to serious errors in the case of strongly interacting particles and
low damping. In particular, we find that equilibrium configurations, corresponding to vanishing optical
forces, only are possible for homodimers illuminated in the end-fire configuration and for heterodimers,
although multipolar effects and damping radically reduce the repulsive interactions in the latter
case.

1. Introduction as recently demonstrated by trapping of silver15 and gold

As first proposed by Martin Moskovits in 1978,1 surface- colloids.16 It has even been shown that polarization dependent
enhanced Raman scattering (SERS) originates from an electro- gradient forces and torques can be used to rotate and spin
magnetic (EM) field enhancement generated by localized surface elongated nanoplasmonic systems, including nanoparticle
plasmon resonances (LSPRs).2 SERS is, however, not the only dimers.17 However, field enhancement and localization as-
application enabled by plasmonic field enhancement. For sociated with LSPR can also be utilized to dramatically
example, light emission from tunnel junctions3 and scanning increase the strength and spatial resolution of the gradient
tunneling microscopes (STM)4 also rely on a field enhancement force acting on small dielectric objects. This concept was
effect brought about by plasmons. The rapid development of first investigated for the case of sharp metal tips18–20 and
nanotechnology and EM simulation capacity over the past recently exploited using lithographically prepared nanoplas-
decade has resulted in the development of a vast range of novel monic surfaces.21–25 In this contribution, however, we discuss
nanoplasmonic applications, many of which essentially build LSPR enhanced optical forces between nanoparticles. We
on the discovery of SERS.5–7 Examples include LSPR refractive focus on the prototypical near-field coupled nanoplasmonic
index sensing,8,9 photodynamic therapy,10 nanoscopic antennas,11,12 system well-known from the SERS field,26–29 i.e., two
and metamaterials.13 One area of increasing interest is to utilize interacting metal nanospheres forming a dimer. The optical
plasmonic field enhancement to amplify optical forces. forces induced by and acting on such a dimer were studied
Optical forces are well-known from the field of laser tweezers, theoretically by Xu and Käll, who found a substantial
developed by Ashkin and co-workers in the 80s as a method to plasmon enhanced attractive force at short surface-to-surface
trap and manipulate microscopic dielectric objects using a tightly separations.30 This result was also corroborated by extended
focused laser beam.14 The technique is based on the so-called EM simulations studies by several groups,31–42 as well as by
gradient force, which in the small particle limit can be SERS experiments on optically aggregated silver colloids.43–45
rationalized in terms of an optical potential U ) -〈p · E〉. Here,
Here we revisit the dimer system with the objective of
p ) RE is the dipole moment induced in the particle by the
clarifying some remaining issues, in particular concerning
spatially varying laser field E(r), leading to a force F(r) )
the accuracy and implementation of the dipole approximation
R∇E2(r) directed toward the point of highest intensity and with
strength proportional to the particle polarizability R. Classical in the context of forces between coupled particles and the
single beam laser tweezers based on the gradient force can be question whether optical forces can lead to an equilibrium
used to manipulate also metal nanoparticles supporting LSPRs, distance between interacting nanoparticles.46

†
Part of the “Martin Moskovits Festschrift”.
* To whom correspondence should be addressed. E-mail: kall@
chalmers.se.
2. Methods
‡
Chalmers University of Technology. 2.1. Maxwell Stress Tensor and Mie theory (MST-Mie).
§
K.N. Toosi University of Technology.
|
CSIC-ICN. In Mie theory, a monochromatic electric field just outside a
⊥
Örebro University. spherical particle can be written29
10.1021/jp911371r  2010 American Chemical Society
Published on Web 02/17/2010
Optical Forces in Plasmonic Nanoparticle Dimers J. Phys. Chem. C, Vol. 114, No. 16, 2010 7473

E) ∑ [a1lmjl(kr) + b1lmhl(kr)]Xlm+ the force on the dipole proportional to the gradient of the
incident electric field and the first term after the minus sign
lm
describes the force that the incident magnetic field causes.
k-1∇ × {[a2lmjl(kr) + b2lmhl(kr)]Xlm} (1)
Mie theory together with the Maxwell stress tensor formalism
gives us a rigorous scheme for calculating the optical force
The first term, with a1lm and b1lm, is the magnetic multipole between spherical particles to any given level of accuracy, as
contribution and the second term is the electric multipole determined by the total number of multipoles included in the
contribution. The terms involving the spherical Bessel functions calculation. Nevertheless, it is an advantage to have alternative
jl describe regular (i.e., nonsingular at r ) 0), incident waves, methods that are simpler to handle in various contexts. Next
and the terms with spherical Hankel functions, hl, scattered, we look at a number of such alternatives.
outgoing waves. Xlm denotes the vector spherical harmonics and 2.2. Maxwell Stress Tensor and the Coupled Dipole
k ) nω/c the wave vector, with n defining the refractive index Approximation (MST-CDA). The optical response of a
of the material surrounding the particle. The strengths of the collection of interacting nanoparticles can be estimated from
outgoing scattered waves are related to those of the incident the CDA method,49,50 which is based on solving a system of
waves through sphere response functions sM(E) l ) b1(2)lm/a1(2)lm. coupled dipole equations
The incident wave coefficients depend on, and can be deter-
mined from, the strength of the incoming plane wave and the
magnitude of the waves scattered from the other particle or pi ) RiEi ) Ri(Einc,i - ∑ Aijpj) (5)
particles in the interacting system.29 In the case of a particle j*ì
that only responds as an electric dipole, the only outgoing waves
that occur are the ones described by the coefficients b21m with
Here pi represent the point dipoles, located at positions ri
m ) 0, (1. These coefficients are directly related to the induced
and with magnitudes governed by the isotropic particle polar-
dipole moments px, py, and pz.
izabilities Ri. The total electric field Ei consists of the incident
We calculate the optical force acting on a given particle by
field Einc,i ) E0 exp(ikri - iωt) and the retarded induced fields
integrating the Maxwell stress tensor over a surface surrounding
from all the other dipoles j. Aij is a 3 × 3 matrix (for i * j)
that particular particle. We have

A(rij) )
FR ) ∫S TRβnβ dS
[ ]
(2)
exp(ikrij) 2 ikrij - 1
k (n̂ijn̂ij - 13) + (3n̂ijn̂ij - 13) (6)
rij rij2
where nβ denotes a component of a normal vector n̂ to the
integration surface, and the Maxwell stress tensor is given by
where rij ) |rij|, rij ) ri - rj, n̂ij ) rij/rij, and 13 is the 3 × 3
1 1 identity matrix. By defining Aii ≡ R-1 i , the scattering problem
TRβ ) [
εE E + BRBβ - (εE · E + B · B)δRβ
4π R β 2 ] reduces to a system of 3N complex linear equations
(3)
Np
When the field is expressed in terms of vector spherical
harmonics, the force can be calculated analytically; that is, eq
∑ Aijpj ) Einc,i (7)
j)1
2 does not have to be evaluated through direct numerical
integration. This is most easily done by employing methods from
quantum mechanics designed to deal with spherical symmetry, where Np is the total number of dipoles in the system.
such as the Wigner-Eckhardt theorem.47 For the z component A dipole polarizability is assigned to each spherical nano-
of the force, this leads to particle according to the so-called modified long-wavelength
approximation (MLWA)51

[
2 lmax l
〈Fz〉 )
ε
8πk2
∑ ∑ ∑ lm{a*τl+1,mbτlm + b*τl+1,maτlm + R)
R0
(8)
τ)1 l)1 m)-l
2 k2
1 - ik3R0 - R0
 (l +(2lm ++ 1)(2l
√l(l + 2) 1)(l - m + 1)
2b*τl+1,mbτlm} - 3 R
l+1 + 3)

]
lmax l

∑ ∑ Re{a*1lmb2lm + b*1lma2lm + 2b*1lmb2lm} l(l +m 1) (4)

R0 ) R3
ε - n2
(9)
l)1 m)-l
ε + 2n2
There are similar expressions for the x and y components of
the force, see for example the work of Barton et al.48 Although Here R is the radius of the sphere, which is composed of a
expression 4 looks rather formidable, it still reveals an underly- material with frequency dependent dielectric constant ε, and
ing, fairly simple, structure. The force originates from products R0 is the quasi-static Clausius-Mossotti polarizability. The use
of multipoles of the same type with l differing by 1, or from of eq 8 rather than eq 9 in the CDA calculation partly
products of different multipoles with the same l. compensates for finite size effects included in the full dipolar
In the important approximation of a particle only responding Mie expansion coefficients.
as an electric dipole, only two of the six terms in eq 4 contribute In the case of two interacting particles, Np ) 2, the solution
to the force: the first term on the left-hand side corresponds to of eq 7 is simply
7474 J. Phys. Chem. C, Vol. 114, No. 16, 2010 Miljković et al.

R1(Einc,1 - R2A12Einc,2) 〈Fi〉 ) 〈Finc,i〉 + 〈Fint,i〉 (15)

p1 ) (10)
1 - R1R2A12A21
Hoekstra et al.55 derive the following useful expressions for
the two components
with an analogous expression for p2.
With the dipole moments on the two interacting particles at
hand, we can calculate the total electric field anywhere in space 1
〈Finc,i〉 ) Re[ik(pi* · Einc,0) exp(ik · ri)] (16)
according to 2

E(r) ) Einc(r) + A(r - r1)p1 + A(r - r2)p2

(11)
〈Fint,i〉 ) ∑ 21 Re(Fij) (17)
j*ì

The first term on the right represents the incident field, while The interaction component is in turn determined by pairwise
the second and third terms are the fields from the induced dipoles forces according to
on particle 1 and 2, respectively. We note that all three terms
have to be included in a MST calculation, even if only the
interparticle interaction is of interest.52 The magnetic fields Fij ) exp(ikrij){[(p*i · pj)n̂ij + pi*(n̂ij · pj)+(pi* · n̂ij)pj -
needed to evaluate the stress tensor are found from the Maxwell
equation B ) (c/iω)∇ × E, by evaluating the curl of eq 11.
Thus, using the MST method, the total optical force on a single
(
5(pi* · n̂ij)n̂ij(n̂ij · pj)] -
k2
rij 2
3ik 3
- 3 + 4 +
rij rij )
particle can be obtained by numerically integrating over a virtual
sphere enclosing that particular particle. With E and H being
the total electric and magnetic fields in spherical coordinates,
(
[(pi* · pj)n̂ij-(pi* · n̂ij)n̂ij(n̂ij · pj)]
ik3
rij
k2
- 2
rij ) (18)

respectively, the full expression reads48

Here Einc,0 is the amplitude of the incoming plane wave,
rij ) |rij|, rij ) ri - rj, and n̂ij ) rij/rij.
〈Fi〉 )
1
8π
Re 0 ∫ ∫
2π
0
π 1 2
2 {
[n (ErE*r - EθE*θ - EφE*)+
φ
It is straightforward to use eqs 15-18 to obtain the force
between coupled point dipoles, with moments given by eq 10.
(HrH*r - HθH*θ - HφH*)]r̂
φ + (n ErE*θ + HrH*)θ̂
2
θ + This then constitute the LF-CDA method. However, the same

}
formalism can be used in a discrete dipole approximation (DDA)
(n2ErE*φ + HrH*)φ̂
φ r2 sin(θ) dθ dφ (12) approach, which is based on a division of the interacting particles
into discrete mesh elements. We use a free version of the DDA
code developed by Draine and co-workers49,50,56 together with
2.3. Lorentz Force in the CDA (LF-CDA) and the
the force calculation methodology developed by Hoekstra et
Discrete Dipole Approximation (LF-DDA). The MST-CDA
al.55 Briefly, using the DDA code, we solve for the dipole
method is a “brute force” method that is comparatively costly
moments that build up the two particles in the dimer. Then,
due to the numerical integrations involved. A much more
using eqs 15-18, we find the total force on each dipolar mesh
effective way of utilizing the dipole moments obtained from
element in both particles. The total averaged radiation forces
CDA is to directly use the Lorentz force. As we shall see in the
due to all mesh elements on each particle are then obtained by
results section, the two methods yield identical results. However,
summing up all elementary force contributions comprising
we point out that the LF-CDA method has to be implemented
different particles.33–35
with care so as not to yield erroneous results.
In both the LF-CDA and LF-DDA methods, it is important
The Lorentz force on a dipole pi in an electromagnetic field
to stress that the gradients in eq 14 refer to a given field
is33,52–55
configuration with constant dipole moments. In other words,
these spatial derivatives do not include contributions that results
1
Fi ) Re(pi · ∇i)Re(Ei) + Re
c
dpi
dt ( )
× Re(Bi) (13)
from changes in dipole moments occurring as the result of a
displacement of a particle.55 In the case of two interacting
particles, this means that the first derivative of the total field
on particle 1, E1 ) Einc,1 + A(r12)p2, has to be calculated as
where Ei and Bi are the total electric and magnetic fields,
respectively, c is the speed of light in vacuum and t denotes
time. The incident field at the location of dipole i is specified ∂E1 ) ∂Einc,1 + ∂[A(r12)p2] ) ∂Einc,1 + [∂A(r12)]p2
as a plane wave, with Einc,i ) Einc,0 exp(ik · ri). For harmonic (19)
fields, time averaging of eq 13 results in
with A(r12) from eq 6. In a recent publication based on LF
1 calculations,46 Käll and co-workers mistakenly included force
〈Fi〉 ) Re[(pi* · ∇i)Ei + ikpi* × Bi] (14) contributions resulting from changing dipole moments. This lead
2
to the erroneous result that the optical interaction force between
two identical spheres, illuminated in the parallel polarization
where the asterisk denotes complex conjugation and k ) |k| is configuration, vanished under certain circumstances, thus leading
the magnitude of the wave vector. Following refs 54 and 55, to an equilibrium distance between the two particles. As we
the time-averaged force can be split into two components, one shall see in the results section, it turns out that an equilibrium
due to the incident field and the other due to the interaction separation for a homodimer only can occur for the end-fire
between the different dipoles, i.e. due to the induced dipole fields illumination configuration. However, equilibrium separations
Optical Forces in Plasmonic Nanoparticle Dimers J. Phys. Chem. C, Vol. 114, No. 16, 2010 7475

between dissimilar particles are in principle possible for both

the parallel and perpendicular polarization configurations.
2.4. Potential Energy and Gradient Force Calculations.
Gradient force and optical potential calculations probably
constitute two most popular and well-known methods for
estimating optical forces in the small particle limit. However,
as we shall see in the results section, both methods can lead to
serious errors if applied to plasmonic systems.
The potential energy experienced by a dipole i can be
calculated as57–59
Figure 1. Illumination configurations for the two-particle (dimer)
1 1 1 system in the presence of a plane wave. The wave-vector of the incident
Ui ) - 〈pi · Ei*〉 ) - 〈RiEi · Ei*〉 ) - Re(Ri)|Ei | 2 wave is denoted by k and its polarization by E.
2 2 4
(20)

Here Ei ) Einc,i + A(rij)pj again represents the total electric

field on dipole i, including contribution from all other dipoles
as obtained from a CDA or DDA calculation. The optical force
on dipole i can then be calculated as a gradient of the potential
energy (20),

1
FU ) -∇Ui ) Re(Ei* · ∇pi + pi · ∇Ei*) )
4
1 1
Re(Ei* · ∇pi) + Re(Ri)∇|Ei | 2 (21)
4 4

The last term in eq 21 corresponds to the gradient force Fgrad.

However, the first term should be set to zero because it involves
spatial derivatives of the dipole moments
An alternative way of introducing a potential energy would
be through a line integral of the total optical force
Figure 2. Spectra of the optical force between two gold spheres with
radii R1 ) R2 ) 5 nm for an intensity of the incident field of 1 W/µm2.
UF ) - ∫∞r F dr
0
(22)
The center-to-center separation distances between the nanoparticles are
d ) 12 nm (black lines) and d ) 14 nm (red lines). The polarization
of the incident field is either parallel (a) or perpendicular (b) to the
dimer axis. The optical forces are calculated using the following
or as the integral of the gradient force (if the nongradient force methods: MST-Mie (lmax ) 13, thick solid lines); LF-DDA (dashed
contribution is small) lines); LF-CDA (thin solid lines); MST-CDA (stars); and MST-MieDip
(lmax ) 1, diamonds).

Ugrad ) - ∫∞r Fgrad dr

0
(23)
the parallel (Figure 2a) and the perpendicular (Figure 2b)
polarization configurations. Referring to Figure 1, we display
Provided that the force only depends on the position of the the z component of the force on particle 2; that is, a positive
particle and the motion is restricted to one dimension, this yields value means that the force between the particles is repulsive,
an unambiguous definition of a potential, even though the total and vice versa. We use three different calculation methods in
optical force, i.e. including scattering contributions, is not the dipole limit (LF-CDA, MST-CDA and MST-MieDip) and
conservative. two methods (MST-Mie and LF-DDA) that include higher
multipole responses. Results are displayed for two different
3. Results and Discussion interparticle distances, d ) 12 and 14 nm, corresponding to
We analyze optical forces between two metal nanospheres surface-to-surface separations of 2 and 4 nm, respectively.
illuminated in three different polarization configurations, as All curves in Figure 2 are dominated by the strong dipolar
shown in Figure 1. The surrounding medium is water (n ) 1.33) plasmon resonance at ∼2.3 eV. The force is attractive for the
and the dielectric constants of the particles correspond to silver parallel configuration and repulsive for the perpendicular
or gold.60 The intensity of the incident plane wave is set to 1 configuration. This difference can be understood from the
W/µm2. In the following, we will display results for both a distribution of induced charges, as discussed in the following
symmetric case, i.e., both particles have the same radii (R1 ) paragraph. For parallel polarization, the magnitude of the force
R2 ) 5 nm), and asymmetric cases for which particle 1 has increases strongly with decreasing distance d between particles.
radius R1 ) 10 nm and particle 2 has a radius in the range This is a direct consequence of the increasing field-enhancement
R2 ) 20-40 nm. between the particles.30 At the same time, the spectral position
3.1. Optical Forces between Identical Nanospheres. We of the force maximum red-shifts due to plasmon hybridization.
begin by analyzing the force acting on one of the particles in a For the perpendicular configuration, on the other hand, the
homodimer composed of two 5 nm radius Au spheres. Figure coupling between the elementary dipoles leads to a slight blue
2 shows the optical interparticle force versus photon energy for shift of the force maximum with decreasing d.
7476 J. Phys. Chem. C, Vol. 114, No. 16, 2010 Miljković et al.

Figure 3. Same plots as in Figure 2, but for the case of two 5 nm

silver particles and only including MST-Mie (lmax ) 13) and LF-CDA
calculations.

Regarding the differences between the various calculation

methods, we first note that the calculations based on the dipole
approximation all fall on the same curve. However, there is a
clear difference between the dipole approximation calculations,
on the one hand, and the MST-Mie (here with lmax ) 13) and
LF-DDA calculations, on the other. This difference becomes
more pronounced for smaller interparticle separations because Figure 4. Optical forces between two silver nanoparticles (R1 ) R2 )
electrostatic contributions to the induced fields that are neglected 5 nm) as a function of the center-to-center separation distance for
in the dipole limit then become more pronounced. These different photon energies. The polarization is parallel to the dimer axis.
contributions are taken into account by including a large number Fz and UFz denote the interparticle force and potential, respectively, in
the point dipole limit (CDA-LF method). FU and U are calculated from
of multipoles in the MST-Mie calculation and by using a small the classical expression for the potential energy of a dipole in an applied
enough mesh size in the LF-DDA method. In particular, the field. Fgrad and UFgrad correspond to gradient force calculations. The
dipole calculations underestimate the field-enhancement, and potentials are normalized to the thermal energy kBT, with T ) 300 K.
thereby the attractive force, in the gap between the particles
for the parallel polarization configuration. For the perpendicular
polarization, on the other hand, the dipole models overestimate dielectric properties of silver, which is more polarizable than
the repulsive force between the particles. This is a result of gold and also exhibits a considerably smaller optical damping
neglecting attractive multipole couplings in the dipole limit, for in the frequency range of interest. This leads to higher dipole
example dipole-quadropole interactions. We shall also point out moments on the particles, which in turn are further enhanced
that there is a small but clear difference between the MST-Mie by the particle-particle interaction. A further consequence of
and LF-DDA calculations. In the present case, we use a mesh this is that the discrepancy between the results from the LF-
size of 0.20 nm (about 130 000 mesh elements), which is not CDA and MST-Mie methods becomes more distinct. The
enough to completely represent all relevant multipoles in a case importance of higher multipoles is obvious from Figure 3a,
where the interparticle separation is comparable with the mesh where for d ) 12 nm, the optical force exhibits a higher-order
size. In order to get improved agreement with the MST-Mie mode around 3.3 eV.
theory, we would have to increase the number of mesh elements, Figure 4 shows optical forces versus separation for two silver
and thus the calculation time, beyond what is feasible. It is then particles illuminated in the parallel configuration at three
clear that the MST-Mie theory is both the most exact and the different photon energies. The “correct” optical force in the
fastest method for calculating the interparticle forces we are dipole limit, Fz, is calculated using the LF-CDA method. This
considering here, i.e. between spherical particles. Regarding the is compared to the gradient force Fgrad obtained from the last
dipole methods used, we have previously mentioned that MST- term in eq 21 and FU, which is calculated as a spatial derivative
CDA is the least useful, because it is much more time- of the classical expression for the potential energy of a dipole
consuming compared to LF-CDA and MST-MieDip. In the in an applied field, i.e., eq 20. The latter energy is shown as an
following, we will therefore only use LF-CDA for calculations inset and denoted U, but we also show the potential energies
in the dipole limit, while the full MST-Mie method is used to that correspond to Fz and Fgrad, which are obtained as integrals
obtain exact results. according to eq 22 and 23, respectively.
To further illustrate the difference between the dipole limit Figure 4a shows an out-of resonance case (pω ) 2 eV), where
and full Mie theory, Figure 3 shows the same type of results as the optical force between the two particles is rather weak. The
Figure 2 but for the case of silver particles. From a qualitative agreement between the “correct” force Fz and the gradient force
point of view, the results are similar to the ones found for gold. Fgrad is then more or less perfect, as is the case for the potentials
Notice, however, that the magnitude of the force now is UFz and Ugrad in the inset. The small deviations compared to
measured in nN rather than pN. This is of course a result of the the potential U and the optical force FU for small separation
Optical Forces in Plasmonic Nanoparticle Dimers J. Phys. Chem. C, Vol. 114, No. 16, 2010 7477

in the positive z direction. The total force is always positive

and does not change much when the particle separation changes.
This means that the two particles, by the influence of the light
pressure, float together in the direction of the incident k wave
vector, i.e., in the positive z direction. However, as we see from
Figure 5b, the binding force can be zero. Physically, this means
that two particles float together in the z direction with a constant
relative distance between them.
When the distance between particles is large (d ) 80 nm),
the MST-Mie and LF-CDA methods, as expected, agree
perfectly. For smaller separation distances, the LF-CDA method
does not predict an equilibrium position. While for large
separation distances the equilibrium distance is a result of
retardation effects, for small separations the optical force
becomes zero because of higher-mode interactions. Note that
the forces on particles 1 and 2 (Figure 5a1-a2) point toward
each other for d ) 12 nm, while for d ) 80 nm they are both
directed in the positive z direction. To summarize, for small
separation distances, the main contribution to the total force
Figure 5. Spectra of the optical forces between two equal silver comes from the interaction between dipoles Fz ≈ Fint (eq 17)
nanoparticles (R1 ) R2 ) 5 nm) in the case of an end-fire illumination resulting in an attractive force, whereas for large separation
configuration. The intensity of the incident light is 1 W/µm2. The center- distances the total force Fz ≈ Finc (eq 16) points in the direction
to-center separation distances are d ) 12 nm (left column: (a1-c1))
and d ) 80 nm (right column: (a2-c2)). The optical forces are of the incident k wave vector, while also retardation effects
calculated using the Mie-MST (lmax ) 13) method (solid lines) and the become important.
LF-CDA method (crosses). The first row (a1-a2) shows the total optical 3.2. Optical Forces between Nonidentical Nanospheres.
forces in the z direction on each particle (black curves represents forces We now turn to the more realistic case of heterodimers. To
on particle 1 and red color represents forces on particle 2). The second clearly illustrate the effect of symmetry breaking, we need to
and third rows show the binding (b1-b2) and the total (c1-c2) forces avoid broadening due to interband transitions. We therefore first
between both particles, respectively.
use a Drude function, with plasma frequency ωp ) 8 eV and
damping Γ ) 0.1 eV, to describe the dielectric response of the
distances occur because of the derivatives of the dipole moments
particles. Referring to Figure 1, we study a case in which particle
that are hidden in eq 20.
1 has a fixed radius of R1 ) 10 nm while the radius of particle
Figure 4b highlights two interesting effects that occur for
2 varies between 10 and 40 nm. Figure 6 (a1 and b1) shows
illumination with photons close to resonance (pω ) 3.2 eV).
the optical force on particle 1 in the z direction, with positive
The first is the rather large deviations of U and FU from the
values corresponding to attraction, for the parallel and perpen-
correct values of the potential, UFz, and the optical force Fz.
dicular illumination configurations, respectively.
Moreover, the potential energy shows an equilibrium position
between the two particles, something that is obviously not For the homodimer (R1 ) R2 ) 10 nm) in the parallel
correct. The second effect is a deviation of the gradient force (perpendicular) polarization configuration, the force is attractive
Fgrad from the total force Fz. Since the imaginary part of the (repulsive) just as for the 5 nm radii spheres studied in the
polarizability is rather large at or near resonance, it is no longer preceding section. However, by increasing the radius of particle
correct to identify the optical force with just the gradient force. 2, we see that the optical force now can change sign as a
Notice also that, as a result of the resonance, the optical force function of photon energy for both polarization configurations.
is much stronger than in Figure 4a. Specifically, in the case of parallel polarization we see a new
In Figure 4c the photon energy is increased to pω ) 3.4 eV, repulsive (negative) peak develop at high energies when the
which corresponds to a negative value of the real part of the radius of particle 2 increases. Similarly, for perpendicular
polarizability. Still the optical force between the particles is polarization, we see a new attractive (positive) peak develop at
attractive also in this case. A negative value of the polarizability energies lower than the original hybridized dipole resonance
means only that dipole moments induced in the particles are of the homodimer. Both peaks are also seen in extinction spectra,
oriented opposite to the incident field. But since both particles shown in Figure 6 (a3 and b3) for completeness, but the relative
are still polarized in the same way relative to each other, the amplitudes are very different compared to the force spectra. This
force between them is also now attractive. This is further is because the latter are dominated by optical near-fields, while
discussed and explained in more detail when we deal with the only far-fields contribute to the extinction spectra.
case of differently sized particles. The behavior of the optical force can be qualitatively
The results in Figure 4 show that one should be careful when interpreted and explained by considering the phase differences
using the potential energy or the gradient force in the dipole between the dipoles excited on particle 1 and 2, see Figure 6
approximation. This is especially true near resonance and in (a2 and b2). For both polarization configurations, the dipoles
case of materials with small damping, such as silver. oscillate out of phase whenever the phase difference is greater
Although the equilibrium distance seen for a homodimer in than π/2. This means that the charge interactions revert
Figure 4b turned out to be an artifact, illumination in the end- compared to the homodimer case, one obtains a repulsive force
fire geometry actually can lead to a vanishing interparticle force. for the parallel case and an attractive force for perpendicular
Figure 5 shows the optical forces on each of the two silver polarization. The reason why we at all can see a substantial
particles in the z direction (a1-a2), together with binding phase difference between the two particles is of course that the
(b1-b2), and total forces (c1-c2), for the end-fire configuration. resonance positions change with size. Specifically, an increasing
A positive value means that the corresponding force is directed radius leads to a red shift, which implies that the hybridized
7478 J. Phys. Chem. C, Vol. 114, No. 16, 2010 Miljković et al.

Figure 8. Force spectra for heterodimers (R1 ) 10 nm, R2 ) 10 - 40

nm, dedge ) 2 nm) composed of Ag and Au spheres calculated with the
MST-Mie (lmax ) 30) method in the parallel (a1-a2) and perpendicular
(b1-b2) polarization configuration. The arrows mark energies for field
Figure 6. Spectra of the optical interparticle force calculated using
plots in Figure 9.
the CDA-LF method and a Drude dielectric function for parallel (a1)
and perpendicular (b1) polarization configurations, respectively. Particle
1 has radius R1 ) 10 nm, particle 2 varies between R2 ) 10-40 nm,
and the edge-to-edge distance is fixed to dedge ) 2 nm. Figures (a2) high-energy regions, leading to attraction, even though the
and (b2) show the phase differences ∆θ ) θ1 - θ2 between the dipole polarizabilities have different signs in the two energy ranges,
moments induced in particles 1 and 2. Extinction spectra of the dimers and repulsion can only occur in the intermediate energy range.
calculated with the CDA method are shown in (a3) and (b3) for parallel The same argument explains why attraction can occur in the
and perpendicular polarizations, respectively. intermediate energy region for perpendicular polarization.
The behavior of the coupled system of particles in Figure 6
is of course much more complex than what is illustrated in
Figure 7, where we have assumed that the resonances refer to
single isolated particles. In particular, Figure 7 can be misleading
in cases when one of the particles in the heterodimer is much
larger than the other one. If this is the case, the scattered field
from the large particle (particle 2 in Figure 6) becomes so large
compared to the incident field that it forces the dipole on the
small particle to oscillate in phase. Accordingly, with increasing
radius of particle 2, the optical force between the particles
becomes attractive over a much broader energy range, irrespec-
tive of polarization configuration. This is clearly seen for the
R2 ) 40 nm case in Figure 6. Nevertheless, the overall picture
sketched in Figure 7 still seems to be valid, but the question is
Figure 7. Schematic illustration of induced dipole charges in a how much it is affected by introducing multipoles and the actual
heterodimer illuminated with polarization parallel (upper row) and metal dielectric functions of Ag and Au.
perpendicular (lower row) to the dimer axis. The curves illustrate the Figure 8 shows force calculations for the same particle sizes
real parts of the polarizabilities for the large and the small particle
versus energy. as in Figure 6, but with silver (a1-a2) and gold (b1-b2)
dielectric functions and using the MST-Mie (lmax ) 30) theory.
A comparison with Figure 6 reveals rather drastic differences,
peak at low energy and parallel polarization is dominated by in particular at higher photon energies, where multipolar effects
the large particle in the pair, while the small peak at high energy becomes dominant in the case of Ag and interband damping
is dominated by the 10 nm particle. To further illustrate this sets in for the Au case. In particular, for the Ag dimer and
effect, we show in Figure 7a schematic illustration of how the parallel polarization case, we see a strong suppression of the
resonance positions, shown through the real part of the dipole “repulsive region”, evident for R2 ) 30-40 nm in Figure 6
polarizabilities, dictate the interparticle force for a heterodimer. (a1). It is only at very high energies (inset in Figure 8 a1),
Three different energy ranges can be distinguished, i.e., corre- indicated by the arrow marked II, for which a weak repulsive
sponding to photon energies below, between and above the two interaction remains. In Figure 9, we show the corresponding
resonance energies. As can be seen in the figure, for parallel field distributions in the gap region, demonstrating a weak
polarization, opposite charges face each other in the low- and repulsion at high energies and a strong attraction near the main
Optical Forces in Plasmonic Nanoparticle Dimers J. Phys. Chem. C, Vol. 114, No. 16, 2010 7479
(10) Gobin, A. M.; Lee, M. H.; Halas, N. J.; James, W. D.; Drezek,
R. A.; West, J. L. Nano Lett. 2007, 7, 1929.
(11) Mühlschlegel, P.; Eisler, H. J.; Martin, O. J. F.; Hecht, B.; Pohl,
D. W. Science 2005, 308, 1607.
(12) Pakizeh, T.; Käll, M. Nano Lett. 2009, 9, 2343.
(13) Smith, D. R.; Pendry, J. R.; Wilthire, M. C. K. Science 2004, 305,
788.
(14) Ashkin, A.; Dziedzic, J. M.; Bjorkholm, J. E.; Chu, S. Opt. Lett.
1986, 11 (5), 288–290.
(15) Prikulis, J.; Svedberg, F.; Käll, M.; Enger, J.; Ramser, K.; Goksör,
M.; Hanstorp, D. Nano Lett. 2004, 4 (1), 115–118.
(16) Hansen, P. M.; Bhatia, V. K.; Harrit, N.; Oddershede, L. Nano
Lett. 2005, 5 (10), 1937–1942.
(17) Tong, L. M.; Miljkovic, D. V.; Käll, M. Nano Lett. 2010, 10, 268–
273.
(18) Novotny, L.; Bian, R. X.; Xie, X. S. Phys. ReV. Lett. 1997, 79,
Figure 9. Electric field plots for the R1 ) 10 nm, R2 ) 30 nm Ag 645.
heterodimer for the energies marked with arrows in Figure 8. The plots (19) Okamoto, K.; Kawata, S. Phys. ReV. Lett. 1999, 83, 4534 Chaumet
show the real parts of the normalized electric field in Y and Z directions P. C. .
in the plane X ) 0 from MST-Mie theory (lmax ) 30). Solid black (20) Rahmani, A.; Nieto-Vesperinas, M. Phys. ReV. Lett. 2002, 88,
lines represent the edges of the particles. 123601.
(21) Quidant, R.; Petrov, D.; Badenes, G. Opt. Lett. 2005, 30 (9), 1009–
1011.
(22) Rhighini, M.; Zelenina, A. S.; Girard, C.; Quidant, R. Nat. Phys.
dipole resonance (arrow marked I in Figure 8 a1). The situation 2007, 3, 477–480.
is similar for the perpendicular polarization configuration and (23) Righini, M.; Girard, C.; Quidant, R. J. Opt. A: Pure Appl. Opt.
2008, 10, 093001.
the Ag case, i.e. attractive interactions start to dominate as soon (24) Grigorenko, A. N.; Roberts, N. W.; Dickinson, M. R.; Zhang, Y.
as the homodimer symmetry is broken and it is only at high Nat. Photonics 2008, 2, 365–370.
energy (arrow marked IV and inset in Figure 8 a2) for which a (25) Juan M. L.; Gordon R.; Pang Y.; Eftekhari F.; Quidant R. Nat.
Photonics 2009, doi: 10.1038/nphys1422.
weak repulsion remains. Turning to the case of Au dimers and (26) Xu, H. X.; Bjerneld, E. J.; Käll, M.; Borjesson, L. Phys. ReV. Lett.
parallel polarization, Figure 8 (b1), we in fact find a complete 1999, 83 (21), 4357–4360.
cancelation of the “repulsive region”, but in this case due to (27) Xu, H. X.; Aizpurua, J.; Käll, M.; Apell, P. Phys. ReV. E 2000, 62
strong interband damping. On the other hand, for perpendicular (3), 4318–4324.
(28) Xu, H. X.; Wang, X. H.; Persson, M. P.; Xu, H. Q.; Käll, M.;
polarization (Figure 8 b2), one do see a rather clear transition Johansson, P. Phys. ReV. Lett. 2004, 93 (24), 243002.
from an attractive to a repulsive force for the heterodimers with (29) Johansson, P.; Xu, H. X.; Käll, M. Phys. ReV. B 2005, 72 (3),
R2 ) 30 and 40 nm, but the effect is not very large. 035427.
(30) Xu, H. X.; Käll, M. Phys. ReV. Lett. 2002, 89 (24), 246802.
(31) Chaumet, P. C.; Nieto-Vesperinas, M. Phys. ReV. B 2001, 64,
4. Summary and Conclusions 035422.
(32) Hallock, A. J.; Redmond, P. L.; Brus, L. E. Proc. Natl. Acad. Sci.
We have calculated optical interparticle forces between two U.S.A. 2005, 102 (5), 1280–1284.
small metal spheres. The calculations show that the optical force (33) Wong, V.; Ratner, M. A. J. Opt. Soc. Am. B 2006, 23 (9).
is always attractive (repulsive) for a homodimer illuminated in (34) Wong, V.; Ratner, M. A. Phys. ReV. B 2006, 73, 075416.
(35) Wong, V.; Ratner, M. A. J. Opt. Soc. Am. B 2007, 24 (1).
the parallel (perpendicular) polarization configuration. However, (36) Zelenina, A. S.; Quidant, R.; Nieto-Vesperinas, M. Opt. Lett. 2007,
in the end-fire configuration and in the case of heterodimers it 32 (9), 1156–1158.
is in principle possible to get a transition from an attractive to (37) Chu, P.; Mills, D. L. Phys. ReV. Lett. 2007, 99, 127401.
a repulsive force as a function of interparticle distance, thus (38) Chu, P.; Mills, D. L. Phys. ReV. B 2008, 77, 045416.
(39) Ng, J.; Tang, R.; Chan, C. T. Phys. ReV. B 2008, 77, 195407.
enabling an equilibrium distance. We show that calculations in (40) Li, Z.; Käll, M.; Xu, H. X. Phys. ReV. B 2008, 77, 085412.
the dipole limit are extremely useful as a tool for physical (41) Novotny L.; Henkel C. Opt. Lett. 2008, 33 (9).
understanding. However, multipoles generally contribute at- (42) Nome, R. A.; Guffey, M. J.; Scherer, N. F.; Gray, S. K. J. Phys.
Chem. A 2009, 113, 4408–4415.
tractive interactions that cannot be neglected at small interpar- (43) Svedberg, F.; Li, Z. P.; Xu, H. X.; Käll, M. Nano. Lett. 2006, 6
ticle separations. We have also found that the often-used dipole (12), 2639–2641.
potential energy and gradient force equations only are valid for (44) Svedberg, F.; Käll, M. Faraday Disc. 2006, 132, 35–44, Tong LM.
weakly interacting dipoles. (45) Righini, M.; Gonzalez, M. U.; Quidant, R.; Käll, M. Lab Chip 2009,
9 (2), 193–195.
(46) Sepulveda, B.; Alegret, J.; Käll, M. Opt. Exp. 2007, 15 (22), 14914–
Acknowledgment. We thank Timur Shegai, and Lianming 14920.
Tong for stimulating discussions. This work was financially (47) Johansson, P. Unpublished.
(48) Barton, J. P.; Alexander, D. R.; Schaub, S. A. J. Appl. Phys. 1989,
supported by the Swedish Research Council. 66 (10).
(49) Draine, B. T. Astrophys. J. 1988, 333, 848–872.
References and Notes (50) Draine, B. T.; Flatau, P. J. J. Opt. Soc. Am. A 1994, 11, 1491–
1499.
(1) Moskovits, M. J. Chem. Phys. 1978, 69 (9), 4159–4161. (51) Meier, M.; Wokaun, A. Opt. Lett. 1983, 8 (11).
(2) Moskovits, M. ReV. Mod. Phys. 1985, 57, 783–826. (52) Novotny, L. Top. Appl. Phys. 2001, 81, 123.
(3) Lambe, J.; McCarthy, S. L. Phys. ReV. Lett. 1976, 37, 923. (53) Gordon, J. P. Phys. ReV. A 1973, 8, 14–21.
(4) Berndt, R.; Gimzewski, J. K.; Johansson, P. Phys. ReV. Lett. 1991, (54) Draine, B. T.; Weingartner, J. C. Astrophys. J. 1996, 470, 551–565.
67, 3796. (55) Hoekstra, A. G.; Frijlink, M.; Waters, L. B. F. M.; Sloot, P. M. A.
(5) Fleischmann, M.; Henra, P. J.; McQuillan, A. J. Chem. Phys. Lett. J. Opt. Soc. Am. A 2001, 18 (8).
1974, 26 (2), 163. (56) Draine, B. T.; Flatau, P. J. http://arxiv.org/abs/0809.0337v4.
(6) Jeanmaire, D. L.; Van Duyne, R. P. J. Electroanal. Chem. 1977, (57) Novotny, L.; Hecht B. Principles of nano-optics; Cambridge
84, 1–20. University Press: 2006.
(7) Albrecht, M. G.; Creighton, J. A. J. Am. Chem. Soc. 1977, 99 (15), (58) Novotny, L.; Henkel C. Opt. Lett. 2008, 33 (9).
5215–5217. (59) Jackson, J. D. Classical Electrodynamics, 3rd ed.; Wiley: New
(8) Anker, J. N.; Hall, W. P.; Lyandres, O.; Shah, N. C.; Zhao, J.; van York, 1998.
Duyne, R. P. Nat. Mater. 2008, 7, 442. (60) Johnson, P. B.; Christy, R. W. Phys. ReV. B 1972, 6, 4370–4379.
(9) Sepúlveda, B.; Angelomé, P. C.; Lechuga, L. M.; Liz-Marzán, L. M.
Nano Today 2009, 4, 244–251. JP911371R