Article

pubs.acs.org/journal/apchd5

Electron Energy-Loss Spectroscopy Calculation in Finite-Dierence

Time-Domain Package
Yang Cao,, Alejandro Manjavacas,*,, Nicolas Large,*,,, and Peter Nordlander*,,

Department of Physics and Astronomy, Department of Electrical and Computer Engineering, and Laboratory for Nanophotonics,
Rice University, 6100 Main Street, Houston, Texas 77005, United States
S Supporting Information
*

ABSTRACT: Electron energy-loss spectroscopy (EELS) is a unique tool that is

extensively used to investigate the plasmonic response of metallic nanostructures. We
present here a novel approach for EELS calculations using the nite-dierence timedomain (FDTD) method (EELS-FDTD). We benchmark our approach by direct
comparison with results from the well-established boundary element method (BEM)
and published experimental results. In particular, we compute EELS spectra for
spherical nanoparticles, nanoparticle dimers, nanodisks supported by various
substrates, and a gold bowtie antenna on a silicon nitride substrate. Our EELSFDTD method can be easily extended to more complex geometries and congurations. This implementation can also be directly
exported beyond the FDTD framework and implemented in other Maxwells equation solvers.
KEYWORDS: numerical recipe, electron energy-loss spectroscopy (EELS), nite-dierence time-domain (FDTD),
plasmonic nanostructures, gold nanoparticle, silver dimer, gold nanodisk, bowtie antennas

simulations with optical excitations.11,2325 A recently published review article by Kociak and Stephan explicitly states
that, despite an implementation for cathodoluminescence (CL)
spectroscopy in FDTD2628 and the EELS implementation in
DGTD,22 there is a clear lack of time-domain numerical
methods for electron-beam spectroscopy calculations.6
In this paper, we propose a novel numerical procedure for
EELS simulations employing a reliable and widely used
commercial package: Lumerical FDTD Solutions.29 This
package implements a high performance 3D solver that oers
a user-friendly environment for solving Maxwells equations
using the FDTD method.30 We benchmark our method by
direct comparison with results from the well-established BEM
method15 for three representative systems: (i) isolated
nanospheres, (ii) nanosphere dimers, and (iii) nanodisks
supported by a substrate. In the particular case of nonpenetrating trajectories for nanospheres, we also compare our
method with analytical results from Mie theory.9,31,32 Finally, in
order to demonstrate the power of this implementation, we
perform EELS calculations for a more complex geometry
consisting of a supported bowtie antenna where experimental
results are available.

nergy-loss spectroscopy using fast electrons was used in

the rst experimental detection of surface plasmons in
metals.14 Since these pioneering studies, electron energy-loss
spectroscopy (EELS) has become a unique tool for probing
surface plasmons of metallic nanostructures with unprecedented spatial (<1 nm) and energy (<100 meV) resolution.510
With the ability of electrons to probe dark surface
plasmons,1013 to provide accurate information on the
nanostructure morphology and local environment,5 quantitative
information on surface plasmon kinetics, damping, and
dispersion,4,14 EELS has become an irreplaceable tool for the
experimental study of surface plasmons. Consequently, EELS is
widely used to investigate optical properties of complex metallic
nanostructures (i.e., complicated geometries, strongly coupled
nanosystems).9 The particular nature of the electromagnetic
eld of an electron (i.e., fast moving point charge, Figure 1a)
makes the theoretical modeling of EELS data more involved
than the modeling of optical experiments. For such simulations
a myriad of dierent numerical techniques have been developed
in the past years, including boundary element method
(BEM),15,16 discrete dipole approximation (DDA),1719
nite-element method (FEM),10,12,20 nite-dierence timedomain (FDTD) method,21 and discontinuous Galerkin timedomain (DGTD) method.22 Although these methods are able
to predict and interpret experimental EELS spectra, they
possess some of the following signicant drawbacks: (i) the
need for large computational resources,1719 (ii) limitations to
nonpenetrating electron trajectories,1820,22 (iii) requiring
highly symmetrical geometries,1519 and (iv) complexity (i.e.,
programming skills required, absence of user-friendly interface).1522 These drawbacks are the main reason why EELS
experimental results are generally compared to numerical
2015 American Chemical Society

METHOD
The theoretical formalism for EELS simulations has been
extensively discussed in the literature (cf. refs 1, 9, and
references therein). The physics of EELS can be understood as
follows: the electric eld produced by an electron moving with
constant velocity v polarizes the nanostructure placed in the
Received: October 30, 2014
Published: February 16, 2015
369

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

ACS Photonics

Article

Figure 1. Description of the geometry employed to calculate the electron energy-loss spectrum. (a) Schematics showing the electric eld of an
electron moving with velocity v along a straight-line trajectory separated from a metallic nanostructure by the impact parameter b. (b, c) Geometry
0
0
used to calculate the induced electric eld Eind
z (z,) = Ez(z,) Ez (z,) for a nanostructure. Here Ez and Ez represent the z-component of the
electric eld generated by an electric dipole p(z,) at position z in presence and in absence of the metallic nano-object, respectively. (d) FDTD
simulation setup showing (i) the FDTD simulation domain with PML boundary conditions, (ii) the nanostructure geometry, (iii) the electric dipole
source, (iv) the 1D monitor used to record the electric eld along the electron path, and (v) the override meshes used to improve the discretization
for the nanostructure and the electron trajectory.

2
1
+ (r, ) 2 G(r, r , ) = 2 (r r)I

c
c

vicinity of its trajectory (cf. Figure 1a). This, in turn, induces an

electric eld Eind that acts back on the electron, exerting a force
that produces the energy loss. This energy loss can be written
as9
E EELS = e

vEind[re(t ), t ]dt = 0

where (r,) is the permittivity of the medium and I is the unit

tensor. Using the electron current density relevant for EELS
j(r,) = e(x b)(y)eiz/z and eq 2 we can rewrite the
loss probability (eq 1) as9

EELS()d

where e is the elementary charge, re(t) represents the electron

trajectory, and
EELS() =

EELS() =

Re{eit vEind[re(t ), ]}dt

+ bulk()

(3)

Here,
= z [G(z,z,) G (z,z,)]z , with
G0(r,r,) being the Green tensor for vacuum (to simplify
the notation we have omitted the lateral spatial coordinates x =
b and y = 0 in the arguments of the Green functions and related
quantities). In the derivation of this expression, we have also
used the reciprocity property of the Green tensor G(r,r,) =
GT(r,r,). Interestingly, Gind
zz (z,z,) can be obtained from the
z-component of the electric eld induced at position z, by an
electric dipole of amplitude p(z,) placed at z and oriented
along z-axis

(1)

where L is the length of the electron trajectory inside the

medium, is the medium dielectric permittivity, and qc
1[(meout)2 + (/)2]1/2 is the cuto momentum
determined by the electron mass me, the electron velocity ,
and the collection angle out of the microscope. This expression
is valid within the local response approximation, in which only
low enough momentum transfers below qc are collected. In the
remainder of this paper, we do not consider the bulk
contribution. In addition, and without loss of generality, we
assume the electron trajectory to be in the (x,z) plane, parallel
to z-axis, and separated from the origin by the impact
parameter b, so v = z and re(t) = (b,0,t).
The electric eld created by an arbitrary current distribution
(such as a beam of electrons) can be written as33

G(r, r, )j(r, )dr

ind
Gzz
(z,z,)

2 2 2

2

1 c qc /
e 2L

Im
ln

c 2/2
2
c

E[r(t ), ] = 4i

cos (z z)

ind
(z , z , )]dz dz
Im[ Gzz

is the energy loss probability per unit of frequency .9 The

second term of this expression, bulk, represents the bulk loss
probability. This contribution can be calculated using the
following analytical expression9
bulk =

4e 2

ind
(z , z, ) =
Gzz

ind
1 Ez (z , )
4 2 p(z, )

Using this expression, we can rewrite eq 3 as

EELS() =

(z z)
e2
cos

E ind(z , )
dz dz
Im z
p(z, )

(4)

Therefore, all we need to do in order to obtain the loss

probability is to compute the induced electric eld generated by
an electric dipole along the electron trajectory. Notice that eq 4
involves only the imaginary part of the induced eld, which
remains nite even at the position of the dipole. The use of the
induced eld, in place of the total eld, for nonpenetrating
trajectories is not required by the theoretical formalism, since
an electron cannot produce energy loss in absence of material
structures. However, due to the nite accuracy of the FDTD

(2)

in terms of the Green tensor of Maxwells equations G(r,r,)

(note that we use Gaussian units). This quantity is dened as
the solution of
370

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

ACS Photonics

calculations we choose to work with the induced eld in order

to minimize any numerical instability originating from the
calculation of the elds. In order to calculate the eld, we can
employ any Maxwells equation solver. Here, we choose to
work with the commercial software package Lumerical FDTD
Solutions29 due to its convenient user environment. The
computation procedure starts by setting a 3D FDTD simulation
domain with perfectly matched layers (PMLs) to prevent
spurious reections from outer boundaries (Figure 1b,d).34 We
then insert the nanostructure and dene an override mesh that
allows us to manually adjust the mesh grid size in a particular
region. This allows us to optimize the discretization of the
physical object and to improve the convergence. After that, we
place a 1D (linear) monitor along the electron trajectory that
allows us to calculate the electric eld at specic points. The
monitor is extended across the entire simulation domain
through the PMLs. Again, to improve the convergence and to
ensure a proper spatial discretization along the electron
trajectory, we place a second override mesh on top of the
monitor. Next, we position an electric point dipole p(z,) on
the mesh grid points, aligned with the electron path, and
successively displace it from mesh point to mesh point along
the electron trajectory from z = zmax (upper PML) to z = zmin
(lower PML). This electric dipole acts as a source in Maxwells
equations. For each position, we record the z-component of the
total electric eld Ez(z,) along the entire monitor (Figure 1b).
To obtain the induced electric eld Eind
z (z,), we subtract the
background electric eld E0z (z,) from Ez(z,). The former
quantity is calculated using the previously described protocol
but removing all the physical objects (e.g., nano-object and
substrate) from the simulation domain (Figure 1c). For
situations where the electron trajectory penetrates into the
absorbing medium (e.g., the metal), one needs to calculate the
corresponding background electric eld generated by an electric
dipole placed in an innite space lled with the corresponding
material. In practice, this can be accomplished by placing the
dipole at the center of a sphere composed of this material,
whose diameter must be chosen large enough to minimize the
eld spill out into the surrounding medium. Here we choose
this diameter equal to the monitor length. In metals, as the eld
of the electric dipole decays to zero after a few tens of
nanometers, no eld exits the micron-sized metallic domain
which, thus, can be considered as innite from the dipoles
point of view. Incidentally, since FDTD simulations are not
stable when an absorbing medium is extended through the
PMLs, this forces us to enlarge the simulation domain and the
monitor to prevent the absorbing medium to reach into the
PMLs. Once the induced electric eld is calculated, the EELS
spectrum is readily computed using eq 4. Lumerical and Matlab
scripts used for the postprocessing (i.e., calculation of the
induced electric eld and calculation of the integral using eq 4)
are presented in the Supporting Information, S1. Finally, we
note that the velocity of the electron only enters in eq 4
through the cosine function. This allows us to compute EELS
spectra for any electron velocity from a single FDTD
calculation. It is important to notice that one can either
calculate all the dipole positions in one single FDTD
calculation or split each dipole position into smaller
subcalculations to increase the parallelization and optimization.29 The results presented in this paper are performed using
the later procedure.

Article

ISOLATED NANOSTRUCTURE

We rst illustrate the EELS-FDTD implementation for the case

of an isolated gold nanosphere of diameter a = 160 nm placed
in vacuum. To benchmark our method, BEM calculations are
performed using an electron source implemented in the axialsymmetry version of this semianalytical method following the
formalism established in ref 15. In addition, the simple spherical
geometry allows us to perform analytical (Mie theory).31 We
use the dielectric function of gold tabulated by Johnson and
Christy.35 While the experimental values are used as is in the
BEM and Mie calculations, analytical multicoecient models
(MCMs) are used in FDTD to t these experimental data and
overcome the diculty of adapting spectrally tabulated
dielectric permittivities into time-domain methods (cf.
Supporting Information, S2).29 The nanoparticle center is
placed at the origin of the coordinate system. To ensure a good
convergence (cf. Supporting Information, S3), we set a monitor
length of 1500 nm (i.e., zmax(min) = 750 nm). The other
parameters used in the Lumerical FDTD Solutions simulation
are set as follows: a simulation time of 100 fs with an autoshuto parameter of 105, a mesh accuracy of 5 (i.e., 22 mesh
points per wavelength), and mesh renement algorithm set to
conformal variant 0 allowing for a nonuniform mesh over the
FDTD domain. The physical meaning of these proprietary
parameters is provided in Supporting Information, S3. An initial
simulation is performed to calculate the total electric eld Ez at
each point along the electron trajectory in the frequency range
13 eV. A second calculation is then performed for the same
electric dipole positions in absence of the nanoparticle to
calculate the background electric eld generated by the dipole
in vacuum E0z . Schematics of these two congurations is shown
in Figure 1b,c. Then, the induced electric eld is computed as
0
Eind
z (z) = Ez(z) Ez (z) and inserted into eq 4 to obtain the
EELS spectra. The results of this calculation are shown with
blue lines in Figure 2 for three dierent impact parameters: b =
120 nm (away from the nanoparticle, bottom), b = 82 nm (in
close proximity to the nanoparticle, center), and b = 0 nm
(through the center of the nanoparticle, top). In all the cases,
we use an electron velocity equal to half of the speed of light in
vacuum c (i.e., = 0.5c), which corresponds to a kinetic energy
of 80 keV.
The results obtained with the EELS-FDTD implementation
(blue lines) show a strong peak at 2.4 eV, in very good
agreement with BEM (red lines) and Mie theory (black
triangles) calculations. This peak corresponds to the quadrupolar mode of the nanoparticle. The dipolar mode of the
nanoparticle only appears as a shoulder in the spectrum at
smaller energies. Interestingly, the position of this peak
depends on the impact parameter due to retardation eects,
originating from the frequency dependence of the eld from
the electron. In contrast to several published methods,1820,22
the EELS-FDTD implementation can also handle penetrating
trajectories (Figure 2, top). However, for such cases one has to
be careful when performing the FDTD simulation.
The relative position of the electric dipoles with respect to
the nanoparticle surface can articially introduce numerical
errors. We show in Supporting Information, S5, that when an
electric dipole is placed exactly at the nanoparticle surface, it
produces an overestimation of the EELS signal. This is easily
solved by slightly displacing the entire nanostructure along the
z-axis with respect to the monitor mesh grid (typically 1/3
mesh step). The discrepancies observed between FDTD and
371

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

ACS Photonics

Article

Figure 2. EELS spectra for a gold nanosphere of diameter a = 160 nm

calculated with the EELS-FDTD implementation (blue lines) for three
impact parameters. From top to bottom: b = 0 nm, b = 82 nm, and b =
120 nm. The EELS-FDTD calculations are compared with BEM
calculations (red lines), and with Mie theory (black triangles, only for
b > a/2). The spectra for b = 120 nm are multiplied by 10 to
improve the clarity of the gure. The electron velocity is taken equal to
0.5c (i.e., 80 keV) in all cases.

Figure 3. (a) EELS spectra for a dimer of silver nanospheres of

diameter a = 160 nm separated by a gap g = 5 nm. Results obtained
with the EELS-FDTD implementation (blue lines) are compared with
BEM calculations (red lines) for three dierent impact parameters,
from top to bottom: b = 164.5 nm, b = 82.5 nm, and b = 0 nm. In all
cases, the electron velocity is xed to 0.5c (i.e., 80 keV). Inset: Zoomin view of the low energy part of the upper spectrum. (b) EELS maps
calculated at (i) 1.5, (ii) 2.3, (iii) 3.25, and (iv) 3.5 eV.

BEM results (Figure 2, top) can be minimized by increasing the

number of dipoles used in the calculation, in particular in the
part of the trajectory close to the nanostructure, as shown in
Supporting Information, S6.
Although our EELS-FDTD implementation requires performing of a large number of short subcalculations (i.e., one per
dipole position), it allows for reaching a better convergence
level at lower computational cost than DDA18 in specic
congurations (cf. Figure S4 and Table S2 in the Supporting
Information). BEM, due to the axial-symmetry nature of this
particular problem, allows the user to perform the same
calculation much faster. A comparison of the computational
resource used by FDTD, DDA, and BEM to calculate the EELS
spectrum for an impact parameter b = 82 nm (Figure 2, center)
is provided in Supporting Information, S4. A more general and
detailed comparison between the three methods can be found
in ref 36.

164.5 nm (dimer end, top), b = 82.5 nm (through the center of

one of the NP, center), and b = 0 nm (in the gap, bottom). In
all the cases, we use an electron velocity of = 0.5c (80 keV). It
is well-known that for large gaps the nanoparticles interact only
weakly and the resulting dimer plasmons are essentially
bonding and antibonding combinations of the nanoparticle
plasmons of the same multipole order (e.g., l = 1, dipole).40
Here, due to the very small gap-to-diameter ratio (g/a = 0.03),
the plasmon modes of the dimer contain contributions from all
multipole orders. When the electrons pass in close proximity to
the dimer end (upper spectra), both FDTD and BEM show the
appearance of two weak localized surface plasmon resonances
(LSPRs) at 1.5 and 2.3 eV and stronger features at 3.25 and 3.5
eV (i, ii, iii, and iv, respectively). Though mixed with higher
values of l, the 3.5 eV is dominated by the antibonding dipole,
while the 3.25 eV feature results from the strong hybridization
of high order modes. The low energy and weak features
observed at 1.5 and 2.3 eV are the longitudinal and transverse
bonding dipole modes, respectively. The EELS maps associated
with the four LSPRs (iiv) are shown in Figure 3b with a 4 nm
spatial resolution. For the sake of simplicity and for
computational considerations, we here excluded penetrating
trajectories. Although the present dimer is much larger (i.e.,
more retardation eects) and has a smaller gap-to-diameter
ratio (i.e., more hybridization), the results are in good
agreement with EELS measurements reported in literature.11,23,24,37,38 The large energy splitting between the bonding
and antibonding dipolar LSPRs is the signature of a strong
coupling.37,38,40 When the electron beam follows a penetrating
trajectory (center spectra) the high energy antibonding dipole

INTERACTING NANO-OBJECTS
Nanoparticle dimers have been extensively studied using
EELS.11,13,2224,37,38 For this reason, they constitute another
ideal system to benchmark our EELS-FDTD method. Here, we
study a dimer of closely spaced (i.e., strongly interacting) silver
nanospheres of diameter a = 160 nm placed in vacuum. The
gap size is xed to g = 5 nm, and we use the dielectric function
for silver tabulated by Palik39 in the BEM calculations and
MCMs t of the later in the FDTD calculations (Supporting
Information, S2). The other simulation parameters are the
same as for the gold nanosphere (cf. Supporting Information,
S3).
Figure 3 shows the results obtained with our EELS-FDTD
method (blue lines) for three dierent impact parameters: b =
372

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

ACS Photonics

Article

very thin substrates can have a signicant impact on the EELS

spectrum and may need to be included in the simulations.
Finally, to highlight the power and the exibility of our
EELS-FDTD implementation, we perform EELS calculations
for a supported bowtie antenna. This complex structure is
composed of two gold equilateral triangles with a lateral length
a = 80 nm, a height h = 15 nm, separated by a gap g = 4 nm.
The gold bowtie structure is placed on top of a 30 nm thick SiN
substrate (calculations for other substrates are shown in
Supporting Information, S7). We also include in the simulation
a 2.5 nm chromium (Cr) adhesion layer. The dielectric
permittivity of SiN is assumed to be constant and equal to
5.5,45 while the corresponding one for Cr is described by
MCMs t of tabulated data (Supporting Information, S2).39
The FDTD simulation parameters are given in Supporting
Information, S3.
The results of this simulation are shown in Figure 5. There,
we observe that for edge excitations (Figure 5a) the spectra

(3.5 eV) remains strong and the transverse bonding dipole (2.3
eV) is strengthened. For a penetrating electron trajectory, also
the bulk plasmon mode at 3.8 eV can be excited.11,24 However,
as we mentioned in the Method section we have chosen not to
include the bulk contribution. For the gap center trajectory
(lower spectra), the bonding dipolar dimer mode cannot be
excited and the spectrum is dominated by the antibonding
dipolar LSPR.11,23,24 Incidentally, due to retardation eects the
position of this mode is slightly dierent for the dierent
excitation congurations. This spectral shift has been observed
experimentally for dimers of large nanoparticles.24

SUPPORTED NANOSTRUCTURES
EELS experiments require very thin nonabsorbing substrates to
minimize energy losses. Typical EELS substrates are made of
mica, silica (SiO2), silicon nitride (Si3N4, SiNx), or carbon (C),
and their thickness generally ranges from 5 to 50 nm. Even
though such thin substrates introduce a negligible EELS
background and are often considered to have a negligible eect
on the optical properties of the supported nanostructure (i.e.,
small spectral shift with respect to the free-standing
nanostructure), they can be crucial in some situations.41 For
this reason, we calculate the EELS spectra for gold nanodisks of
diameter a = 50 nm and height h = 15 nm placed on 30 nm
thick substrates of dierent dielectric permittivities: = 1 (freestanding); = 2 (SiO2);42,43 and = 4 (Si3N4).42,44 As in the
case of the isolated sphere, we use the gold dielectric function
tabulated by Johnson and Christy35 in the BEM calculations
and MCMs t of the later in the FDTD calculations
(Supporting Information, S2). The FDTD simulation parameters are the same as in previous cases (cf. Supporting
Information, S3), and the electron velocity is set to = 0.5c
(i.e., 80 keV). The results are shown in Figure 4 for FDTD
(blue lines) and BEM (red lines) for an impact parameter b =
27 nm (in close proximity to the nanodisk). As expected, the
dipolar LSPR (2.3 eV for = 1) red-shifts with increasing
permittivity of the substrate (1.97 eV for = 4). Interestingly, it
has to be noticed a change in the LSPR line shape for = 4.
This eect, along with the spectral shift clearly shows that even

Figure 5. EELS spectra for a gold bowtie antenna calculated with the
EELS-FDTD implementation. Each triangle has a lateral length a = 80
nm, height h = 15 nm, and gap g = 4 nm. The bowtie antenna is
supported by a 30 nm thick SiN substrate. A 2.5 nm chromium
adhesion layer is included. The electron velocity is taken equal to 0.5c
(i.e., 80 keV) in all cases. (a) EELS spectra for edge excitation with
four dierent impact parameters: 2 (blue), 5 (purple), 10 (red), and
15 nm (black; cf. inset). (b) EELS spectra for gap excitation with ve
dierent impact parameters: 0 (blue), 2 (purple), 5 (red), 10 (brown),
and 15 nm (black; cf. inset). (c) EELS maps calculated at (i) 1.27, (ii)
1.68, (iii) 2.17, and (iv) 2.39 eV.

display two LSPRs (i, iv) located at 1.27 and 2.39 eV,
respectively. In this case, dierent impact parameters produce
very dierent intensity ratios for these LSPRs. On the other
hand when electron trajectory crosses the center of the gap
(Figure 5b) the EELS spectrum displays two distinct LSPRs (ii,
iii) at 1.68 and 2.17 eV, respectively, with relative intensities
that are much less dependent on the impact parameter. When
displacing the electron trajectory o-axis, the LSPR at 2.17 eV
(iii) becomes weaker progressively, while the one at 1.68 eV
(ii) remains unchanged. All these results, along with the results
for a single triangular prism (Supporting Information, S8), are
in good agreement with the experimental observations by Yang
and co-workers.10,12 The small discrepancies, mainly in the line
widths, are related to the presence of a thicker Cr adhesion
layer in our calculations, which is known to introduce a

Figure 4. EELS spectra for a gold nanodisk of diameter a = 50 nm and

height h = 15 nm placed on top of a 30 nm thick substrate with
dielectric permittivity = 1, 2, and 4 calculated with the EELS-FDTD
implementation (blue lines) and BEM (red lines). The impact
parameter is xed to b = 27 nm and the electron velocity is 0.5c (i.e.,
80 keV).
373

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

ACS Photonics

Article

Present Address

broadening and a red-shift of the LSPRs (cf. Supporting

Information, S8).46,47
Figure 5c shows the EELS maps corresponding to the four
LSPRs (iiv) calculated using our EELS-FDTD implementation with a 2 nm spatial resolution. Similarly to the dimer, we
here choose to exclude penetrating trajectories. The nature of
the LSP modes can straightforwardly be determined from these
maps. Modes (i) and (ii) correspond to the dipolar bonding
(bright) and antibonding (dark) modes, respectively. The map
for 2.39 eV shows strong EELS signal from each bowtie edge
underlining the high-order nature of mode (iv). These results
are also in excellent quantitative agreement with the recent
studies by Yang and co-workers.10,12 Finally, mode (iii) displays
a EELS signal which is strongly localized at the bowtie gap. This
spatial connement directly correlates with the rapid vanishing
of mode (iii) when the electron trajectory is displaced o-gap
(Figure 5b). Interestingly, this mode was not imaged by Duan
et al. because their excitation geometry was o-gap excitation.10

Department of Chemistry, Northwestern University, Evanston, IL, United States (N.L.).

Notes

The authors declare no competing nancial interest.

ACKNOWLEDGMENTS
The authors thank F. J. Garc a de Abajo for his insight and
stimulating discussions. This work was supported by the Robert
A. Welch Foundation under Grant C-1222, the Cyberinfrastructure for Computational Research funded by NSF under
Grant CNS-0821727, and by the Data Analysis and Visualization Cyberinfrastructure under NSF Grant OCI-0959097.
A.M. acknowledges support from the Welch Foundation under
the J. Evans Attwell-Welch Fellowship for Nanoscale Research,
administrated by the Richard E. Smalley Institute for Nanoscale
Science and Technology (Grant L-C-004).

CONCLUSIONS
We have presented a simple procedure to calculate the energy
loss probability of fast electrons interacting with metallic
nanostructures. Although this method can be implemented
with any Maxwells equation solver we have chosen here to
work with the commercial package Lumerical FDTD
Solutions29 due to the exibility of the FDTD method and
its user-friendly environment. Contrary to most of the wellestablished methods, we have shown that this implementation
can deal with both penetrating and nonpenetrating trajectories
and nanostructures of arbitrary geometries and morphologies,
including substrates and adhesion layers. We have benchmarked our EELS-FDTD implementation by comparing the
results with the well-established BEM method for dierent
representative nanostructures, such as nanospheres, nanoparticle dimers, and a nanodisk supported by a substrate.
Furthermore, we have applied this method to study the EELS
spectrum of a complex system consisting of a supported bowtie
antenna, and we have mapped the LSPR modes of the latter.
Our EELS-FDTD method provides a simple and convenient
approach for the calculation of EELS spectra and maps from
complex nanostructures of arbitrary shape and composition.

(1) Ritchie, R. H. Plasma losses by fast electrons in thin films. Phys.

Rev. 1957, 106, 874.
(2) Powell, C. J.; Swan, J. B. Origin of the characteristic electron
energy losses in aluminum. Phys. Rev. 1959, 115, 869.
(3) Watanabe, H. Experimental evidence for the collective nature of
the characteristic energy loss of electrons in solids: studies on the
dispersion relation of plasma frequency. J. Phys. Soc. Jpn. 1956, 11, 112.
(4) Pettit, R. B.; Silcox, J.; Vincent, R. Measurement of surfaceplasmon dispersion in oxidized aluminum films. Phys. Rev. B 1975, 11,
31163123.
(5) Botton, G. Probing bonding and electronic structure at atomic
resolution with spectroscopic imaging. MRS Bull. 2012, 37, 21.
(6) Kociak, M.; Stephan, O. Mapping plasmons at the nanometer
scale in an electron microscope. Chem. Soc. Rev. 2014, 43, 38653883.
(7) Garca de Abajo, F. J.; Kociak, M. Probing the photonic local
density of states with electron energy loss spectroscopy. Phys. Rev. Lett.
2008, 100, 106804.
(8) Garca de Abajo, F. J.; Kociak, M. Electron energy-gain
spectroscopy. New J. Phys. 2008, 10, 073035.
(9) Garca de Abajo, F. J. Optical excitations in electron microscopy.
Rev. Mod. Phys. 2010, 82, 209.
(10) Duan, H.; Fernandez-Domnguez, A. I.; Bosman, M.; Maier, S.
A.; Yang, J. K. W. Nanoplasmonics: Classical down to the nanometer
scale. Nano Lett. 2012, 12, 1683.
(11) Koh, A. L.; Bao, K.; Kahn, I.; Smith, W. E.; Kothleitner, P.;
Nordlander, G.; Maier, S. A.; McComb, D. W. Electron energy-loss
spectroscopy (EELS) of surface plasmons in single silver particle and
dimers: Influence of beam damage and mapping of dark modes. ACS
Nano 2009, 3, 3015.
(12) Koh, A. L.; Fernandez-Domnguez, A. I.; McComb, D. W.;
Maier, S. A.; Yang, J. K. W. High-resolution mapping of electron-beamexcited plasmon modes in lithographically defined gold nanostructures. Nano Lett. 2011, 11, 13231330.
(13) Barrow, S. J.; Rossouw, D.; Funston, A. M.; Botton, G. A.;
Mulvaney, P. Mapping bright and dark modes in gold nanoparticle
chains using electron energy loss spectroscopy. Nano Lett. 2014, 14,
37993808.
(14) Bosman, M.; Ye, E.; Tan, S. F.; Nijhuis, C. A.; Yang, J. K. W.;
Marty, R.; Mlayah, A.; Arbouet, A.; Girard, C.; Han, M.-Y. Surface
plasmon damping quantified with an electron nanoprobe. Sci. Rep.
2013, 3, 1312.
(15) Garca de Abajo, F. J.; Howie, A. Retarded field calculation of
electron energy loss in inhomogeneous dielectrics. Phys. Rev. B 2002,
65, 115418.
(16) Hohenester, U. Simulating electron energy loss spectroscopy
with the MNPBEM toolbox. Comput. Phys. Commun. 2014, 185, 1177.

ASSOCIATED CONTENT

S Supporting Information
*

(S1) Lumerical and Matlab scripts for postprocessing and EELS

spectrum computation. (S2) Dielectric function of metals:
Multicoecient models in FDTD. (S3) Convergence of the
EELS spectra: FDTD Lumerical parameters. (S4) Computational resource: EELS-FDTD versus BEM versus e-DDA. (S5)
Penetrating trajectory: Numerical error and workaround. (S6)
Convergence of the EELS spectra: Electron path meshing. (S7)
EELS spectra of a bowtie antenna: Substrate eect. (S8) EELS
spectra of a single gold triangular prism: Eect of the Cr layer.
This material is available free of charge via the Internet at
http://pubs.acs.org.

REFERENCES

AUTHOR INFORMATION

Corresponding Authors

*E-mail: alejandro.manjavacas@rice.edu.
*E-mail: nicolas.large@northwestern.edu.
*E-mail: nordland@rice.edu.
374

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

ACS Photonics

Article

gold and silver dimers probed by EELS. J. Phys. Chem. C 2014, 118,
54785485.
(39) Palik, E. D., Ed. Handbook of Optical Constants of Solids;
Academic Press: New York, 1985; Vol. 1.
(40) Nordlander, P.; Oubre, C.; Prodan, E.; Li, K.; Stockman, M. I.
Plasmon hybridization in nanoparticle dimers. Nano Lett. 2004, 4,
899903.
(41) Knight, M. W.; Wu, Y. P.; Lassiter, J. B.; Nordlander, P.; Halas,
N. J. Substrates matter: Influence of an adjacent dielectric on an
individual plasmonic nanoparticle. Nano Lett. 2009, 9, 2188.
(42) Baak , T. Silicon oxynitride; a material for GRIN optics. Appl.
Opt. 1982, 21, 10691072.
(43) Gao, L.; Lemarchand, F.; Lequime, M. Refractive index
determination of SiO2 layer in the UV/vis/NIR range: Spectrophotometric reverse engineering on single and bilayer designs. J. Eur. Opt.
Soc. 2013, 8, 13010.
(44) Philipp, H. R. Optical properties of silicon nitride. J. Electrochem.
Soc. 1973, 120, 295300.
(45) Tanaka, M.; Saida, S.; Tsunashima, Y. Film properties of lowk
silicon nitride films formed by hexachlorodisilane and ammonia. J.
Electrochem. Soc. 2000, 147, 22842289.
(46) Vial, A.; Laroche, T. Description of dispersion properties of
metals by means of the critical points model and application to the
study of resonant structures using the FDTD method. J. Phys. D: Appl.
Phys. 2007, 40, 71527158.
(47) Zheng, Y. B.; Juluri, B. K.; Mao, X.; Walker, T. R.; Huang, T. J.
Systematic investigation of localized surface plasmon resonance of
long-range ordered Au nanodisk arrays. J. Appl. Phys. 2008, 103,
014308.

(17) Geuquet, N.; Henrard, L. EELS and optical response of a noble

metal nanoparticle in the frame of a discrete dipole approximation.
Ultramicroscopy 2010, 110, 1075.
(18) Bigelow, N.; Vaschillo, A.; Iberi, V.; Camden, J. P.; Masiello, D.
Characterization of the electron- and photon-driven plasmonic
excitations of metal nanorods. ACS Nano 2012, 6, 7497.
(19) Bigelow, N.; Vaschillo, A.; Camden, J. P.; Masiello, D.
Signatures of Fano interferences in the electron energy loss
spectroscopy and cathodoluminescence of symmetry-broken nanorod
dimers. ACS Nano 2013, 7, 4511.
(20) Reed, N. W.; Chen, J. M.; MacDonald, N. C.; Silox, J.; Bertsch,
G. F. Fabrication and STEM/EELS measurements of nanometer-scale
silicon tips and filaments. Phys. Rev. B 1999, 60, 5641.
(21) Talebi, N.; Sigle, W.; Vogelsang, R.; van Aken, P. Numerical
simulations of interference effects in photon-assisted electron energyloss spectroscopy. New J. Phys. 2013, 15, 053013.
(22) Matyssek, C.; Niegemann, J.; Hergert, W.; Busch, K. Computing
electron energy loss spectra with the discontinuous Galerkin timedomain method. Photon. Nanostruct.: Fundam. Appl. 2011, 9, 367.
(23) Song, F.; Wang, T.; Wang, X.; Xu, C.; He, L.; Wan, J.; van
Haesendonck, C.; Ringer, S. P.; Han, M.; Liu, Z.; Wang, G. Visualizing
plasmon coupling in closely spaced chains of Ag nanoparticles by
electron energy-loss spectroscopy. Small 2010, 6, 446.
(24) Kadkhodazadeh, S.; Wagner, J. B.; Joseph, V.; Kneipp, J.;
Kneipp, H.; Kneipp, K. Electron energy loss and one- and two-photon
excited SERS probing of hot plasmonic silver nanoaggregates.
Plasmonics 2013, 2, 763.
(25) Scholl, J. A.; Koh, A. L.; Dionne, J. A. Quantum plasmon
resonances of individual metallic nanoparticles. Nature 2012, 483, 421.
(26) Chaturvedi, P.; Hsu, K. H.; Kumar, A.; Fung, K. H.; Mabon, J.
C.; Fang, N. X. Imaging of plasmonic modes of silver nanoparticles
using high-resolution cathodoluminescence spectroscopy. ACS Nano
2009, 3, 29652974.
(27) Das, P.; Chini, T. K.; Pond, J. Probing higher order surface
plasmon modes on individual truncated tetrahedral gold nanoparticle
using cathodoluminescence imaging and spectroscopy combined with
FDTD simulations. J. Phys. Chem. C 2012, 116, 15610.
(28) Das, P.; Kedia, A.; Kumar, P. S.; Large, N.; Chini, T. K. Local
electron beam excitation and substrate effect on the plasmonic
response of single gold nanostars. Nanotechnology 2013, 24, 405704.
(29) Lumerical Solutions, Inc.; http://www.lumerical.com/tcadproducts/fdtd/.
(30) Yee, K. Numerical solution of initial boundary value problems
involving Maxwells equations in isotropic media. IEEE Trans.
Antennas Propag. 1966, 14, 302.
(31) Garca de Abajo, F. J.; Howie, A. Relativistic electron energy loss
and electron-induced photon emission in inhomogeneous dielectrics.
Phys. Rev. Lett. 1998, 80, 51805183.
(32) Garca de Abajo, F. J. Relativistic energy loss and induced
photon emission in the interaction of a dielectric sphere with an
external electron beam. Phys. Rev. B 1999, 59, 30953107.
(33) Novotny, L., Hecht, B. Principles of Nano-Optics, 1st ed.;
Cambridge University Press: New York, 2006.
(34) Berenger, J.-P. Perfectly Matched Layer (PML) for Computational Electromagnetics. Synthesis Lectures on Computational Electromagnetics 2007, 2, 1117 DOI: 10.2200/S00030ED1V01Y200605CEM008.
(35) Johnson, P. B.; Christy, R. Optical constants of the noble metals.
Phys. Rev. B 1972, 6, 4370.
(36) Myroshnychenko, V.; Rodrguez-Fernandez, J.; Pastoriza-Santos,
I.; Funston, A. M.; Novo, C.; Mulvaney, P.; Liz-Marzan, L. M.; Garca
de Abajo, F. J. Modelling the optical response of gold nanoparticles.
Chem. Soc. Rev. 2008, 37, 17921805.
(37) Kadkhodazadeh, S.; Wagner, J. B.; Kneipp, H.; Kneipp, K.
Coexistence of classical and quantum plasmonics in large plasmonic
structures with subnanometer gaps. Appl. Phys. Lett. 2013, 103,
083103.
(38) Kadkhodazadeh, S.; de Lasson, J. R.; Beleggia, M.; Kneipp, H.;
Wagner, J. B.; Kneipp, K. Scaling of the surface plasmon resonance in
375

DOI: 10.1021/ph500408e
ACS Photonics 2015, 2, 369375

Supporting information for:

Electron Energy-Loss Spectroscopy Calculation in
Finite-Difference Time-Domain Package
Yang Cao,, Alejandro Manjavacas,,, Nicolas Large,,,, and Peter
Nordlander,,,
Department of Physics and Astronomy, Department of Electrical and Computer
Engineering, and Laboratory for Nanophotonics, Rice University, 6100 Main Street,
Houston, Texas 77005, United States
E-mail: alejandro.manjavacas@rice.edu; nicolas.large@northwestern.edu; nordland@rice.edu

S1. Lumerical and Matlab scripts for postprocessing and EELS spectrum
computation
S2. Dielectric function of metals: Multicoefficient models in FDTD
S3. Convergence of the EELS spectra: FDTD Lumerical parameters
S4. Computational resource: EELS-FDTD vs BEM vs e-DDA
S5. Penetrating trajectory: Numerical error and workaround
S6. Convergence of the EELS spectra: Electron path meshing
S7. EELS spectra of a bowtie antenna: Substrate effect
S8. EELS spectra of a single gold triangular prism: Effect of the Cr layer

To whom correspondence should be addressed

Department of Physics and Astronomy

ECE Department

LANP, Rice University

Current address: Department of Chemistry, Northwestern University, Evanston, IL, United States

S1

S1. Lumerical and Matlab scripts for postprocessing and EELS

spectrum computation
(a) Lumerical script (.lsf)
1
2
3
4

#
#
#
#

Run th e s c r i p t i n th e f o l d e r c o n t a i n i n g a l l th e s i m u l a t i o n f i l e s named
' s w e e p i i . f s p ' where i i i s th e d i p o l e p o s i t i o n l a b e l .
NB: 'E ' has to be r e p l a c e d by ' E 0 ' when p r o c e s s i n g th e background f i e l d
i n vacuum o r by ' E 0m ' f o r th e background f i e l d i n th e m e ta l .

5
6
7
8
9

## NUMERICAL PARAMETERS ##
N f r e q =100;
# T o ta l number o f f r e q u e n c y / e n e r g y p o i n t s
N mon=501;
# T o ta l number o f m o n i to r p o i n t s ( where E i s r e c o r d e d )
N dip =501;
# T o ta l number o f e l e c t r i c d i p o l e s

10
11

E = m a tr i x ( N dip , N mon , N f r e q ) ; # I n i t i n i a l i z e th e Im{E( z , z ' , w) } m a tr i x

12
13
14
15
16
17
18
19
20
21
22
23

## POSTPROCESS THE FDTD SIMULATION AT EACH DIPOLE POSITION ##

f o r ( i i =1: N dip ) {
f i l e n a m e = s w e e p +num2str ( i i ) ;
# Simulation f i l e f o r di pol e i i
load ( filename ) ;
# Load th e f s p f i l e
Ez = g e t d a t a ( m o n i to r , Ez ) ;
# E x t r a c t Ez a l o n g th e m o n i to r [V/m]
f = g e t d a t a ( m o n i to r , f ) ;
# E x t r a c t th e f r e q u e n c y v e c t o r [ Hz ]
z = g e t d a t a ( m o n i to r , z ) ;
# E x t r a c t th e v e c t o r p o s i t i o n z ' [m]
F = pinch ( f ) ;
# Remove s i n g l e t o n d i m e n s i o n
w = 2 p i F ;
# Convert f to w ( omega )
[ rad / s ]
E( i i , 1 : N mon , 1 : N f r e q )=p i n c h ( imag ( Ez ) ) ;# 3D m a tr i x Im{E( z , z ' , w) }
}

24
25

m a tl a b s a v e ( E ,w, E) ; # Save w and Im{E( z , z ' , w) } to a Matlab w o r k s p a c e E . mat

(b) Matlab script (.m)

1
2
3
4
5
6
7
8
9

% This s c r i p t computes th e l o s s p r o b a b i l i t y from e q u a t i o n 4 .

%
% In p u t : i m a g i n a r y p a r t o f t o t a l and background e l e c t r i c f i e l d s E and E 0 ( and / o r E 0m ) ,
%
r e s p e c t i v e l y . The e l e c t r i c f i e l d s a r e o f th e form E( z , z ' , w) where z , z ' , and w
%
a r e th e d i p o l e and m o n i to r p o i n t p o s i t i o n s and th e a n g u l a r f r e q u e n c y , r e s p e c t i v e l y .
%
w, E , and E 0 ( and / o r E 0m ) have to be l o a d i n t o th e Matlab w o r k s p a c e ( l o a d E.mat ) .
%
% Output : e n e r g y l o s s p r o b a b i l i t y p e r u n i t o f e n e r g y (%gamma ( : , 3 ) ) and e n e r g y l o s s
%
p r o b a b i l i t y p e r u n i t o f f r e q u e n c y (%gamma ( : , 2 ) ) a s f u n c t i o n o f e n e r g y (%gamma ( : , 1 ) )

10
11
12
13
14
15
16
17

%% PHYSICAL CONSTANTS AND PARAMETERS %%

e=1. 6 0 2 1 7 6 5 7 e 19;
% Elementary c h a r g e [ C ]
hbar=1. 0 5 4 5 7 1 7 2 6 e 34;
% Reduced Planck c o n s t a n t [ J . s ]
c =299732484;
% Speed o f l i g h t i n vacuum [m/ s ]
v=0 . 5 c ;
% E l e c t r o n v e l o c i t y [m/ s ]
p0=1. 9 7 2 7 6 e 31;
% D i p o l e a m p l i tu d e [ C.m ]
R=50e 9;
% Radius / s i z e / h e i g h t o f th e n a n o p a r t i c l e [m]

18
19
20
21
22
23
24
25

%% NUMERICAL PARAMETERS %%
N f r e q =100;
N mon=501;
N dip =501;
z max=500e 9;
dz=2z max / ( N dip 1) ;
dz2 =2z max / ( N mon1) ;

%
%
%
%
%
%

T o ta l number o f f r e q u e n c y / e n e r g y p o i n t s
T o ta l number o f m o n i to r p o i n t s
T o ta l number o f e l e c t r i c d i p o l e s
2 z max=2z min i s th e m o n i to r l e n g t h [m]
D i s t a n c e between d i p o l e p o s i t i o n s [m]
D i s t a n c e between m o n i to r p o i n t s [m]

26
27
28
29
30

%% VECTOR INITIALIZATION %%
gamma=z e r o s ( N f r e q , 3 ) ;
% L o s s p r o b a b i l i t y o u tp u t v e c t o r [ eV , s , 1/eV ]
I=z e r o s ( N dip , N mon , N f r e q ) ;
% I n t e g r a n d : c o s [ w( zz ' ) / v ] Im [ E i n d /p ]
E i n d=z e r o s ( N dip , N mon , N f r e q ) ; % Im a g i n a r y p a r t o f th e i n d u c e d e l e c t r i c f i e l d [V/m]

S2

31
32
33
34
35

DD=1: N dip ;
MM=1:N mon ;
z d i p=z max+(DD1) dz ;
z mon=z max+(MM1) dz2 ;

% P o s i t i o n s o f th e e l e c t r i c d i p o l e s [m]
% P o s i t i o n s where th e e l e c t r i c f i e l d i s c a l c u l a t e d

[m]

36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52

%% ENERGY LOSS PROBABILITY CALCULATION %%

% C a l c u l a t i o n o f th e i m a g i n a r y p a r t o f th e i n d u c e d e l e c t r i c f i e l d [V/m]
% Comment/Uncomment 'OPTION 1 ( 2 ) ' a c c o r d i n g to th e c o n s i d e r e d t r a j e c t o r y .
%
% OPTION 1 : P e n e t r a t i n g t r a j e c t o r y ( to be adapted to th e c o n s i d e r e d geometry )
f o r kk = 1 : 1 : N dip
i f abs ( z d i p ( kk ) ) R; % i f th e d i p o l e i s i n s i d e th e m e ta l then s u b t r a c t E 0m o f th e
m e ta l
E i n d ( kk , : , : ) =E( kk , : , : ) E 0m ( kk , : , : ) ;
e l s e % i f th e d i p o l e i s o u t s i d e m e ta l then s u b s t r a c t E 0 o f vacuum
E i n d ( kk , : , : ) =E( kk , : , : ) E 0 ( kk , : , : ) ;
end
end
%
% OPTION 2 : Nonp e n e t r a t i n g t r a j e c t o r y
E i n d=EE 0 ; % s u b s t r a c t E 0 o f vacuum
%

...

53
54
55
56
57
58
59
60

f o r i i =1:1: N freq
% C a l c u l a t i o n o f th e i n t e g r a n d I=c o s [ w( zz ' ) / v ] Im [ E i n d /p ]
f o r j j = 1 : 1 : N mon
f o r kk = 1 : 1 : N dip
I ( kk , j j , i i )=c o s (w( i i ) ( z d i p ( kk )z mon ( j j ) ) /v ) E i n d ( kk , j j , i i ) / p0 ;
end
end

61
62
63
64
65
66
67

% C a l c u l a t i o n o f th e i n t e g r a l u s i n g th e t r a p e z o i d a l method
gamma ( i i , 1 )=hbar w( i i ) / e ;
% Energy [ eV ]
gamma ( i i , 2 )=e 2 / ( p i hbar w( i i ) 2 ) . . .
t r a p z ( z mon , t r a p z ( z d i p , I ( : , : , i i ) , 1 ) , 2 ) ; % L o s s p r o b a b i l i t y [ s ]
gamma ( i i , 3 )=gamma( i i , 2 ) e / hbar ;
% L o s s p r o b a b i l i t y [ 1 / eV ]
end

68
69
70
71
72
73
74
75
76
77
78
79

%% PLOT OF THE EELS SPECTRUM %%

f i g u r e ; h o l d on ; g r i d on ;
[ f f , l l 1 , l l 2 ]= p l o t y y (gamma ( : , 1 ) ,gamma ( : , 3 ) ,gamma ( : , 1 ) ,gamma ( : , 2 ) ) ;
s e t ( f f ( 2 ) , ' y l i m ' , [ g e t ( f f ( 1 ) , ' y l i m ' ) ] hbar / e )
linkaxes ( ff , 'x ' ) ;
set ( ff , ' f o n t s i z e ' ,16) ;
set ( ll1 , ' color ' , ' r ' ) ;
set ( l l 2 , ' color ' , 'b ' , ' linewidth ' ,3) ;
x l a b e l ( ' \ b f E n e r g y ( eV ) ' , ' f o n t s i z e ' , 1 6 ) ;
y l a b e l ( f f ( 1 ) , ' \ b f L o s s p r o b a b i l i t y ( eV{ 1}) ' , ' f o n t s i z e ' , 1 6 ) ;
yl abel ( f f (2) , ' \ bfLoss p r o b a b i l i t y ( s ) ' , ' f o n t s i z e ' ,16) ;

Figure S1: (a) Lumerical script used for postprocessing the Lumerical FDTD simulations.
The script successively loads the .fsp files for each dipole position, extracts the z-component
of the electric field, and built a 3D matrix containing the imaginary part of the electric field
Im{Ez (z, z , )} along the monitor (z ), for each dipole position (z) and as function of the
frequency (). (b) Matlab script used for computing the energy loss probability EELS ()
from the induced electric field Ezind (z, ) extracted from Lumerical FDTD simulations. It
uses the 3D matrices built with the Lumerical script and eq 4.

S3

S2. Dielectric function of metals: Multicoefficient models in FDTD

To deal with the dispersive nature of optical materials, time-domain based methods typically employ analytical models (e.g.,, Drude, Debye, and Lorentz models) to approximate
the dielectric permittivity. Here, we use multicoefficient models (MCMs), implemented in
Lumerical FDTD solutions, that rely on a more extensive set of basis functions to better fit
dispersion profiles that are not easily described by Drude, Debye, and Lorentz models.

Figure S2: Real and imaginary parts of the dielectric permittivity of (a) gold, (b) silver, and
(c) chromium. The squares are the experimental data tabulated by Johnson and Christy S1
and by Palik S2 and the full lines are the MCM fits used in the FDTD simulations.

S4

S3. Convergence of the EELS spectra: FDTD Lumerical parameters

The override mesh is an object that enables the user to indicate where the interfaces are
physically significant to the problem. It allows for manually setting a finer mesh on particular
regions than the automatic meshing would have generated.
The mesh accuracy is an integer from 1 to 8 defining the fineness of the mesh in the
FDTD domain. The number of mesh points per wavelength (ppw), where the wavelength
is the shortest wavelength of the simulation bandwidth, is a major consideration for the
meshing algorithm. Accuracy 1 to 8 corresponds to a target of 6 to 34 ppw by increment of
4 ppw.
The auto-shutoff parameter is a built-in convergence criterion of Lumerical FDTD Solutions, associated to the total amount of energy remaining in the simulation domain. The
lower the auto-shutoff, the less energy remains, the better the convergence.
More detailed information about the Lumerical FDTD specific parameters (shown in
Figure S3 and Table S1) can be found in ref S3.

Figure S3: Convergence of the EELS-FDTD spectrum of a gold S1 nanosphere of diameter

a = 160 nm. The impact parameter is fixed to b = 82 nm and the electron velocity is taken
equal to 0.5c (i.e., 80 keV) in all the cases. (a) FDTD domain size and monitor length; the
simulation domain is taken as cubic. (b) FDTD Lumerical mesh accuracy, ma. (c) FDTD
Lumerical auto-shutoff parameter. (d) Nanostructure override mesh (dx = dy = dz).
S5

Table S1: FDTD parameters for: (i) isolated gold S1 nanosphere of diameter a = 160 nm,
(ii) dimer of silver S2 nanoparticles of diameter a = 160 nm and gap g = 5 nm, (iii) gold S1
nanodisk of diameter a = 50 nm, and thickness h = 15 nm on a 30 nm thick Si3 N4 S4,S5
substrate, and (iv) gold S1 bowtie antenna of edge length a = 80 nm, thickness h = 15 nm,
and gap g = 4 nm on a 30 nm thick SiN S6 substrate and 2.5 nm chromium S2 adhesion layer.
FDTD simulation domain
Sim. time
[fs]
Sphere
100
Dimer
100
Disk
100
Bowtie
100

Size
[m]
1
1
1
1

auto
auto
auto
auto

Mesh type
(accuracy)
non-uniform
non-uniform
non-uniform
non-uniform

S6

(5)
(7)
(5)
(5)

Mesh
refinement
conformal 0
conformal 0
conformal 0
conformal 0

Autoshutoff
105
106
105
106

Override mesh
Structure e-path
dx, dy, dz
dz
[nm]
[nm]
2,2,2
2
1.5,2,2
2
2,2,2
2
1.5,2,2
2

S4. Computational resource: EELS-FDTD vs BEM vs e-DDA

We calculate the EELS spectrum of an isolated gold S1 nanosphere of diameter a = 160 nm
with an impact parameter b = 82 nm, for an energy range from 1 to 4 eV, and an energy
resolution of 15 meV. The electron velocity is taken equal to 0.5c (i.e., 80 keV). The EELSFDTD calculation is performed for a cubic simulation domain size (and monitor length)
of 1.5 m, a mesh size dx = dy = dz = 1 nm for the nanoparticle with a mesh accuracy
of 5 (i.e., 22 ppw), and an auto-shutoff parameter set to 106 . The dipole position step
is fixed at dz = 2 nm. Figure S4a shows the computational time for each dipole position
in 0 z 750 nm (red dots and blue line). The average calculation time obtained for
this calculation is 8.61 min per dipole position (green dashed line), and the median value
is 8.80 min (orange dashed line). However, considerable time reduction can be achieved by
reducing the mesh size, the mesh accuracy, and the auto-shutoff parameter while preserving
the convergence of the results (cf. Figure S3). Figure S4b shows the computational time for
each dipole position, for the same calculation, when the FDTD simulation domain size is
set to 1 m, the mesh accuracy set to 2 (i.e., 10 ppw), and the auto-shutoff parameter set to
105 . The average time is decreased to 4.62 min.

Figure S4: (a) EELS-FDTD computational time for each dipole position in 0 z 750 nm
(red dots and blue line). The mean value is 8.61 min (green dashed line), and the median
value is 8.80 min (orange dashed line). (b) Computational time in 0 z 500 nm (red
dots and blue line). The mean value is 4.62 min (green dashed line), and the median value
is 4.53 min (orange dashed line). (c) FDTD, BEM, DDA, and Mie EELS spectra of an
isolated gold nanosphere of diameter a = 160 nm with an impact parameter b = 82 nm and
an electron velocity of 0.5c (i.e., 80 keV). The spectra are calculated with the parameters
given in Table S2.

S7

Table S2 summarizes the statistics of the two FDTD calculations and compares them
to the results obtained using BEM S7 and DDA. S8 BEM uses 100 boundary elements (i.e.,
mesh size of about 2.5 nm) to descretize the profile of the nanoparticle. In DDA, we set
an interdipole distance of d = 0.8 nm, corresponding to about 4.2 million dipoles in the
nanoparticle. The FDTD, BEM, and DDA calculations are performed on 10 Intel Sandybridge E2670 processors (64-bit, 20 MB Cache, 2.6 Ghz) with 40 GB of available memory
(DDR3 1600 MHz).
Table S2: Comparison of the computational resource (time and memory) between FDTD
(two calculations), DDA, and BEM. The EELS spectra are calculated for an isolated gold
nanosphere of diameter a = 160 nm with an impact parameter b = 82 nm, for an energy
range from 1 to 4 eV, and an energy resolution of 15 meV (Figure S4c). The discretization
parameter d corresponds to the minimum volume mesh size (FDTD), the boundary element
size (BEM), and the interdipole distance (DDA). The total time is calculated for each EELS
calculation when the sequences are run in serial. FDTD (a) and (b) refers to the panels (a)
and (b) in Figure S4.
d
[nm]
FDTD (a)

FDTD (b)

DDA

0.8

BEM

2.5

Number of
elements

Number of
sequences

6, 011, 499
750 dipole
Yees cells
positions
1, 642, 560
500 dipole
Yees cells
positions
4, 188, 896
200 energy
dipoles
values
100 boundary 200 energy
elements
values

Average time
per sequence
[min]

Total
comput. time
[h]

Average
memory
[MB]

8.6 2.1

105

2, 874

4.6 0.6

35

1, 659

62 20

> 210

15, 135

6.8 s

25 min

14.50

Due to its partitioning in subcalculations, the total computational time of the EELSFDTD simulation can be drastically reduced by running the subcalculations in parallel (e.g.,
using job arrays). Also, as the nanosphere possesses a symmetry with respect to the z-axis,
the number of calculations, and hence the total computational time, can be reduced by a
factor of 2 (i.e., 375 dipole positions) by using the reciprocity property of the Green tensor.
As it can be seen from Figure S4c and Table S2, EELS-FDTD allows for reaching a better
convergence level at lower computational cost than DDA. Furthermore, it is important to
notice that, in the current conditions, a DDA calculation with a smaller interdipole distance
results in errors and cannot be performed due to the extremely large number of dipoles (over
17 millions for d = 0.5 nm). Also, the number of iterations increases with the magnitude of
the permittivity in DDA, and becomes extremely large for gold in the NIR. This results in
computation time of several hours for each frequency when reaching energies below 2.25 eV.

S8

S5. Penetrating trajectory: Numerical error and workaround

Figure S5: EELS spectra for a spherical gold S1 nanoparticle of diameter a = 160 nm calculated with the EELS-FDTD implementation at b = 0 nm (i.e., penetrating trajectory). The
electron velocity is taken equal to 0.5c (i.e., 80 keV). The override mesh placed along the
electron trajectory discretizes space for the placement of the electric dipoles. Each electric
dipole are placed at one of these mesh points. (i) When the physical surface of the nanostructure coincides with the override mesh (i.e., when an electric dipole is placed at the NP
surface) the EELS signal is overestimated. This overestimation, introduced by numerical/
computational errors, can be corrected by introducing a slight offset between the surface and
the mesh grid as shown in panels (ii) and (iii).

S9

S6. Convergence of the EELS spectra: Electron path meshing

Figure S6: EELS-FDTD spectra for a gold S1 nanodisk of diameter a = 50 nm and thickness
h = 15 nm placed over a 30 nm thick SiO2 substrate with dielectric permittivity = 2. S5,S9
The impact parameter is fixed to b = 27 nm and the electron velocity is 0.5c (i.e., 80 keV).
Convergence of the calculated EELS spectrum can be achieved by increasing the number of
electric dipoles. This is done by changing the step dz of the override mesh used along the
electron trajectory. (i) Uniform distribution of electric dipoles along the z-axis (dz = 2 nm).
(ii) Uniform distribution of electric dipoles along the z-axis (dz = 1 nm).

S10

S7. EELS spectra of a bowtie antenna: Substrate effect

Figure S7: EELS spectra for a gold S1 bowtie antenna of edge length a = 80 nm, thickness
h = 15 nm, and gap g = 5 nm on top of a 30 nm thick substrate. We vary the substrate
dielectric permittivity from 1 to 4. The electron velocity is fixed to 0.5c (i.e., 80 keV) in
all the cases. The impact parameter is b = 0 nm (gap center). The bowtie nanostructure
appears to be more sensitive to the presence of the substrate than a single triangular prism
(cf. Figure 4). In addition to a significant red-shift of the LSPRs, an increase of also gives
rise to the appearance of high order modes.

S11

S8. EELS spectra of a single gold triangular prism: Effect of the

Cr layer

Figure S8: EELS spectra for a single gold S1 triangular prism of edge length a = 80 nm and
thickness h = 15 nm, on top of a 30 nm thick SiN S6 substrate. The nanostructures include a
2.5 nm (blue lines) and 1 nm (red lines) chromium S2 adhesion layer. The electron velocity is
fixed to 0.5c (i.e., 80 keV) in all the cases. (a) Edge excitation b = 15 nm. (b) Apex excitation
b = 15 nm. The impact parameter b is taken as the distance from the prism end in panel
(a) and prism apex in panel (b). The results are in good agreement with the experimental
observations by Yang and co-workers. S10,S11

S12

References
(S1) Johnson, P. B.; Christy, R. Optical constants of the noble metals. Phys. Rev. B 1972,
6, 4370.
(S2) Palik, E. D., Ed. Handbook of Optical Constants of Solids; Academic Press: New York,
1985; Vol. 1.
(S3) Lumerical Solutions, Inc. http://www.lumerical.com/tcad-products/fdtd/.
(S4) Philipp, H. R. Optical properties of silicon nitride. J. Electrochem. Soc. 1973, 120,
295300.
(S5) B
a
ak, T. Silicon oxynitride; a material for GRIN optics. Appl. Opt. 1982, 21, 1069
1072.
(S6) Tanaka, M.; Saida, S.; Tsunashima, Y. Film properties of lowk silicon nitride films
formed by hexachlorodisilane and ammonia. J. Electrochem. Soc. 2000, 147, 2284
2289.
(S7) Garca de Abajo, F. J.; Howie, A. Retarded field calculation of electron energy loss in
inhomogeneous dielectrics. Phys. Rev. B 2002, 65, 115418.
(S8) Bigelow, N.; Vaschillo, A.; Iberi, V.; Camden, J. P.; Masiello, D. Characterization of
the electron- and photon-driven plasmonic excitations of metal nanorods. ACS Nano
2012, 6, 7497.
(S9) Gao, L.; Lemarchand, F.; Lequime, M. Refractive index determination of SiO2 layer in
the UV/Vis/NIR range: Spectrophotometric reverse engineering on single and bi-layer
designs. J. Europ. Opt. Soc. Rap. Public. 2013, 8, 13010.
(S10) Koh, A. L.; Fernandez-Domnguez, A. I.; McComb, D. W.; Maier, S. A.; Yang, J. K. W.
High-resolution mapping of electron-beam-excited plasmon modes in lithographically
defined gold nanostructures. Nano Lett. 2011, 11, 13231330.
(S11) Duan, H.; Fernandez-Domnguez, A. I.; Bosman, M.; Maier, S. A.; Yang, J. K. W.
Nanoplasmonics: Classical down to the nanometer scale. Nano Lett. 2012, 12, 1683.

S13