Professional Documents
Culture Documents
Improved Analytical Fit of Gold Dispersion: Application To The Modeling of Extinction Spectra With A Finite-Difference Time-Domain Method
Improved Analytical Fit of Gold Dispersion: Application To The Modeling of Extinction Spectra With A Finite-Difference Time-Domain Method
Improved analytical fit of gold dispersion: Application to the modeling of extinction spectra
with a finite-difference time-domain method
Alexandre Vial,* Anne-Sophie Grimault, Demetrio Macas, Dominique Barchiesi, and Marc Lamy de la Chapelle
Laboratoire de Nanotechnologie et dInstrumentation Optique - CNRS FRE 2671, Universit de Technologie de Troyes,
12 rue Marie Curie, BP-2060 F-10010 Troyes Cedex, France
sReceived 13 July 2004; revised manuscript received 14 October 2004; published 23 February 2005d
We propose an accurate description for the dispersion of gold in the range of 1.242.48 eV. We implement
this improved model in an FDTD algorithm and evaluate its efciency by comparison with an analytical
method. Extinction spectra of gold nanoparticle arrays are then calculated.
I. INTRODUCTION vD2
e Ds v d = e ` , s1d
The employment of the nite-difference time-domain v s v + i g Dd
sFDTDd method in the study of different electromagnetic
where vD is the plasma frequency and gD is the damping
phenomena has raised constantly increasing interest over the
coefcient. Nevertheless, if frequencies within the range of
past 15 years. Since then, an extensive number of references
the visible spectrum are required for a specic study, the
describing the principles of the method have been published;
model in Eq. s1d may not be complete enough to provide
see, e.g., Refs. 1 and 2. Also, a wide variety of software
accurate results. To illustrate this fact, we try to t the rela-
based on this technique has been developed and is commer-
tive permittivity of gold eJC, tabulated by Johnson and
cially, and noncommercially, available elsewhere.
Christy12 through the optimization of e`, vD, and gD for
Due to the fact that accurate results for a full spectrum
energies between 1.24 and 2.48 eV swavelengths between
can be obtained in a single run of the program, the FDTD has
500 and 1000 nmd. In order to determine the best set of
proven to be well adapted for different kinds of spectro-
parameters, we dene a tness function ssometimes called an
scopic studies.3 Nevertheless, a strong limitation is the re-
objective functiond F as
quirement of an analytical model of dispersion. Typical
laws used for FDTD simulations are the Debye, Lorentz, or
Drude dispersion models.36 Also, a modied Debye law can
F= o
v
hRefeJCsv jd eDsv jdgj2 + hImfeJCsv jd eDsv jdgj2 ,
j
be used.7 At least in principle, any dispersion law could be
s2d
described in terms of a linear combination of Debye and
Lorentz laws.2 Surprisingly, this property has not been used where v j are the discrete values of the frequency v
in studies using the FDTD method. Rather, it has been suc- = 2pc / l for which the permittivity is calculated. The real
cessfully applied to the description of optical functions for 11 and imaginary parts of a complex value z are, respectively,
metals over a wide spectrum,8 or to a calculation of the re- Reszd and Imszd. The minimization of F is performed em-
ectance of single wall nanotubes.9 In this paper, we will ploying the simulated annealing procedure described in Ref.
employ a scheme similar to the one that appears in this last 13, and results are presented in the rst row of Table I.
reference for the study of the optical response of gold nano- The real and imaginary parts of the permittivity eDsvd,
structures. calculated with the Drude model, are, respectively, plotted
The structure of this work is as follows. In Sec. II, we with a dotted line in Figs. 1 and 2.
show the results obtained from the implementation of two It can be seen that neither ReseJCd nor ImseJCd are well
classic dispersion models employing the FDTD method. In described for energies above 2.2 and 1.9 eV, respectively. To
order to validate the numerical approaches, we apply them to emphasize this difference, we also plot in Figs. 1 and 2 the
the case of a simple structure and compare the results with relative errors on ReseDd and ImseDd. The existence of a
those obtained using the analytical method described in Ref. strong discrepancy for energies above 1.9 eV when using the
10. In Sec. III, we employ our FDTD-based implementation single Drude model is evident.
of the Drude-Lorentz model to the calculation of extinction Due to the inability of the Drude model to describe the
spectra. In Sec. IV, we present our main conclusions and permittivity of metals over a wide range of frequencies,
nal remarks. some of the authors working with the FDTD restrict their
studies to a zone of the spectrum where the Drude model is
II. MODELS OF DISPERSION valid sRef. 7 for a gold tip, Refs. 14 and 15 for silver and
aluminum structures; it should be noted that for these two
A. The Drude model
metals, the Drude model alone works well for the optical
It is well known that in the near infrared, the relative wavelengthsd. Others try to t the permittivity in the range of
permittivity of several metals can be described by means of interest by modifying the values of the parameters e`, vD,
the Drude model,11 and gD of the model, as shown in Ref. 5 for the case of silver
TABLE I. Values of the parameters used for the optimization of the Drude and the Drude-Lorentz models.
The value of the tness function is given in the last column.
FIG. 1. Real part of the permittivity of gold as published in Ref. FIG. 2. Imaginary part of the permittivity of gold as published
12, calculated with the single Drude model and calculated with the in Ref. 12, calculated with the single Drude model and calculated
Drude-Lorentz model. with the Drude-Lorentz model.
085416-2
IMPROVED ANALYTICAL FIT OF GOLD PHYSICAL REVIEW B 71, 085416 s2005d
analytical method. For simplicity, the structure we will con- drawn. In this case, the maximal error achieved when using
sider for our tests consists of a thin layer of gold surrounded the Drude-Lorentz model is 5.2%, and the Drude model per-
by air. We will calculate the coefcient of transmission in forms worse only for energies above 2.34 eV swavelengths
amplitude for the electric eld using the formulas that appear below 530 nmd. If now we still consider the threshold previ-
in Ref. 10 for an absorbing lm on a transparent substrate. ously dened smaximal error of 2.7%d, then the Drude-
In Fig. 3, we present the results obtained considering a Lorentz model can be accepted for energies below 2.39 eV
thin lm of thickness e = 20 nm. The absolute values of the swavelengths greater than 520 nmd, whereas the Drude
transmission coefcients utu, computed through the FDTD model is only acceptable for energies below 1.78 eV swave-
implementations of Drude and Drude-Lorentz models, are, lengths above 697 nmd and in a narrow zone between 2.08
respectively, depicted with a dashed and a thick solid curve. and 2.24 eV s555 and 596 nmd.
The thin solid curve corresponds to the analytical result. The As a last test, we compute the intensity 5 nm above an
values of the permittivity used for the analytical calculation innite gold cylinder of radius 15 nm illuminated with a
are the ones published in Ref. 12 and not the values tted by plane wave, the incident eld being polarized perpendicu-
the Drude-Lorentz model. larly with the axis of the cylinder. Results are compared with
It can be observed in Fig. 3 that the agreement between the Mie theory and presented in Fig. 5. It is conrmed that
the Drude model and the analytical one decreases for ener- the Drude-Lorentz models perform better than the single
gies above 1.8 eV swavelength below 700 nmd. On the other Drude model, which is unable to predict the peak of absorp-
hand, the agreement between the Drude-Lorentz model and tion around 2.4 eV. Nevertheless, the error calculated in this
the analytical one is quite good. This behavior can be visu- case is higher than previously observed, mainly due to the
alized further by plotting the relative error on the values of utu difculty to accurately describe a circle using an orthogonal
for the two models used with the FDTD method. It is obvi- mesh. For this calculation, in order to rene the description
ous from Fig. 3 that the Drude-Lorentz model not only per- of the shape of the cylinder, a spatial discretization of 0.5 nm
forms better than the Drude model, but also that it leads to a was used, instead of 5 nm for the thin-layer cases.
very small error on a wide spectrum. The maximal relative Once we have veried that an improvement can be ob-
error obtained with the Drude-Lorentz model is 2.7%, and if served not only on the description of the permittivity e, but
we consider this value as a threshold for the validity of the also on the calculation of transmission coefcients, we feel
model, the Drude model is only valid for energies below 1.97 condent to apply our new dispersion model to a more com-
eV swavelengths greater than 630 nmd. plicated case, namely the calculation of extinction spectra
According to Figs. 1 and 2, it is clear that the imaginary above gold nanostructures arrays.
part of the permittivity is less well described than the real
one even with the Drude-Lorentz model, thus we can check
results obtained for utu when considering a layer of thickness III. APPLICATION TO THE CALCULATION
e = 50 nm sFig. 4d in order to verify that we still get satisfy- OF EXTINCTION SPECTRA
ing results. It can be seen that both the transmission coef-
cients utu and the relative errors present a behavior analogous Recently, several experimental results obtained with gold
to the one shown in Fig. 3, and similar conclusions can be nanoparticle arrays have been published.1621 Various sizes,
085416-3
VIAL et al. PHYSICAL REVIEW B 71, 085416 s2005d
085416-4
IMPROVED ANALYTICAL FIT OF GOLD PHYSICAL REVIEW B 71, 085416 s2005d
IV. CONCLUSION
For a three-dimensional computation window, with a spatial discretization Dx and a temporal discretization Dt, we represent
the value of any eld U at position x = iDx, y = jDx, z = kDx and for the instant t = nDt by uUuni,j,k, and the standard recursion
equations in a nondispersive and nonmagnetic medium for the electric and magnetic elds are
Dt n
n+1/2
uHxui1/2,j+1,k+1 n1/2
= uHxui1/2,j+1,k+1 + usE u n
uEyui1/2,j+1,k+1/2 n
+ uEzui1/2,j+1/2,k+1 n
uEzui1/2,j+3/2,k+1 d, sA1d
m0Dx y i1/2,j+1,k+3/2
Dt n
n+1/2
uHyui,j+1/2,k+1 n1/2
= uHyui,j+1/2,k+1 + usE u n
uEzui1/2,j+1/2,k+1 n
+ uExui,j+1/2,k+1/2 n
uExui,j+1/2,k+3/2 d, sA2d
m0Dx z i+1/2,j+1/2,k+1
Dt n
n+1/2
uHzui,j+1,k+1/2 n1/2
= uHzui,j+1,k+1/2 + usE u n
uExui,j+1/2,k+1/2 n
+ uEyui1/2,j+1,k+1/2 n
uEyui+1/2,j+1,k+1/2 d, sA3d
m0Dx x i,j+3/2,k+1/2
Dt n+1/2
n+1
uExui,j+1/2,k+1/2 n
= uExui,j+1/2,k+1/2 + usHzui,j+1,k+1/2 uHzui,j,k+1/2
n+1/2 n+1/2
+ uHyui,j+1/2,k n+1/2
uHyui,j+1/2,k+1 d, sA4d
e0ei,j+1/2,k+1/2Dx
085416-5
VIAL et al. PHYSICAL REVIEW B 71, 085416 s2005d
Dt n+1/2
n+1
uEyui1/2,j+1,k+1/2 n
= uEyui1/2,j+1,k+1/2 + usHxui1/2,j+1,k+1 uHxui1/2,j+1,k
n+1/2 n+1/2
+ uHzui1,+1j,k+1/2 n+1/2
uHzui,j+1,k+1/2 d,
e0ei1/2,j+1,k+1/2Dx
sA5d
Dt n+1/2
n+1
uEzui1/2,j+1/2,k+1 n
= uEzui1/2,j+1/2,k+1 + usHyui,j+1/2,k+1 uHyui1,j+1/2,k+1
n+1/2 n+1/2
+ uHxui1/2,j,k+1 n+1/2
uHxui1/2,j+1,k+1 d.
e0ei1/2,j+1/2,k+1Dx
sA6d
uCLuni = uCLr CLun1 + uCLd Eun , sA8d Coefcients CLr and CLd can now be written as
CLr = esa+ibdDt , sA14d
uEun+1 = uCaEun + Cb 3 Hn+1/2 + Cg ResuCDun + uCLund,
sA9d CLd = DxL0 . sA15d
where the coefcients Ca, Cb, Cg, Cd, and Cr depend on the Finally, we dene x0 = xD0 + ResxL0 d, and the last three co-
coefcients of Eq. s3d. efcients needed are
0
For the Drude term, we dene de = svD / gDd2, xD
0 2 e`
= des1 egDDtd, DxD = des1 egDDtd2, and sD = vD / gD. Then Ca = , sA16d
the coefcients CD r d
and CD are dened as e` + x0 + sDDt
r
CD = egDDt , sA10d Dt
Cb = , sA17d
d 0
Dxe0se` + x0 + sDDtd
CD = DxD . sA11d
For the Lorentz term, we rst dene a = GL / 2, b 1
Cg = . sA18d
= VL2 a2, and g = DeVL2 / b. Then we dene e` + x0 + sDDt
IEEE Trans. Geosci. Remote Sens. 36, 1928 s1998d. 241 s2003d.
5 S. K. Gray and T. Kupka, Phys. Rev. B 68, 045415 s2003d. 16 S. Linden, J. Kuhl, and H. Giessen, Phys. Rev. Lett. 86, 4688
6 s2001d.
M. Futamata, Y. Maruyama, and M. Ishikawa, J. Phys. Chem. B
107, 7607 s2003d. 17 N. Flidj, J. Aubard, G. Lvi, J. Krenn, M. Salerno, G. Schider,
7 J. T. K. II, E. J. Sanchez, and X. S. Xie, J. Chem. Phys. 116, B. Lamprecht, A. Leitner, and F. Aussenegg, Phys. Rev. B 65,
10895 s2002d. 075419 s2002d.
8 A. Raki, A. Djurisi, J. Elazar, and M. Majewski, Appl. Opt. 37, 18 N. Flidj, J. Aubard, G. Lvi, J. Krenn, G. Schider, A. Leitner,
085416-6
IMPROVED ANALYTICAL FIT OF GOLD PHYSICAL REVIEW B 71, 085416 s2005d
A. Leitner, and F. Aussenegg, Appl. Phys. Lett. 82, 3095 21 N. Flidj, S. Truong, G. Lvi, J. Krenn, A. Hohenau, A. Leitner,
s2003d. and F. Aussenegg, J. Chem. Phys. 120, 7141 s2004d.
20 22
J. Grand, S. Kostcheev, J.-L. Bijeon, M. L. D. L. Chapelle, P.-M. K. Kelly, E. Coronado, L. L. Zhao, and G. C. Schatz, J. Phys.
Adam, A. Rumyantseva, G. Lrondel, and P. Royer, Synth. Met. Chem. B 107, 668 s2003d.
139, 621 s2003d.
085416-7