June 15, 2004 / Vol. 29, No.

12 / OPTICS LETTERS 1393

Exact solution of Maxwell’s equations for optical interactions

with a macroscopic random medium
Snow H. Tseng, Jethro H. Greene, and Allen Taflove
Department of Electrical and Computer Engineering, Northwestern University, Evanston, Illinois 60208

Duncan Maitland
Medical Physics and Biophysics Division, Lawrence Livermore National Laboratory, Livermore, California 94550

Vadim Backman and Joseph T. Walsh, Jr.

Department of Biomedical Engineering, Northwestern University, Evanston, Illinois 60208

Received December 11, 2003

We report what we believe to be the first rigorous numerical solution of the two-dimensional Maxwell equations
for optical propagation within, and scattering by, a random medium of macroscopic dimensions. Our solution
is based on the pseudospectral time-domain technique, which provides essentially exact results for electro-
magnetic field spatial modes sampled at the Nyquist rate or better. The results point toward the emerging
feasibility of direct, exact Maxwell equations modeling of light propagation through many millimeters of bio-
logical tissues. More generally, our results have a wider implication: Namely, the study of electromagnetic
wave propagation within random media is moving toward exact rather than approximate solutions of Maxwell’s
equations. © 2004 Optical Society of America
OCIS codes: 240.6680, 060.0060, 060.2320, 240.6690, 170.3880, 290.7090.

Tissue optics deals with light scattering by biological
structures, on which noninvasive optical imaging
techniques such as optical coherence tomography are
based. Most studies of tissue optics have utilized
heuristic approximations in the categories of radiative
transfer theory1 and Mie theory,2 including Beer’s
law, the Kubelka– Monk theory, the adding–doubling
method, the diffusion approximation, and the Monte
Carlo method. To various degrees, these methods
neglect the full-vector electromagnetic wave nature of
light based on Maxwell’s equations, especially with
regard to near-f ield interactions of closely spaced
particles.
Recently, finite-difference time-domain numeri-
cal solutions of Maxwell’s equations3 were applied
to model optical interactions with models of single
biological cells.4 In principle, the finite-difference
time-domain method (FDTD) could be used to model
cell collections spanning macroscopic dimensions (mil-
limeters) and thus to attack the tissue-optics problem
on the most fundamental basis. However, using the
FDTD may not be feasible for many years because
computers lack capabilities to deal with the enormous
database of electromagnetic f ield vector components
mandated for the FDTD by its mesh-density require-
ment of 20 or more samples per optical wavelength in
each spatial dimension.
In this Letter we report the initial application to the
tissue-optics problem of an emerging advanced vari-
ant of the FDTD: the pseudospectral time-domain
technique (PSTD).5,6 For large electromagnetic wave
interaction models in D dimensions that do not have
geometric details or material inhomogeneities smaller
than one-half wavelength the PSTD reduces computer
storage and running time by approximately 8D :1 rela-
0146-9592/04/121393-03$15.00/0
tive to standard the FDTD while it achieves com-
parable accuracy.5,6 This advantage is sufficient to
permit, for the first time to the authors’ knowledge
rigorous numerical solution of the full-vector Maxwell
equations for optical propagation within, and scatter-
ing by, a random medium of macroscopic dimensions.
The most basic version of the PSTD is implemented
on an unstaggered, collocated Cartesian space grid.
Let 兵Vi 其 denote the values of field component V at all
points along an x-directed cut through the grid, and let
兵共≠V 兾≠x兲i 其 denote the x derivatives of V at the same
points needed in Maxwell’s equations. Using the dif-
ferentiation theorem for Fourier transforms, we can
write
Ω Çæ
≠V
苷 2F 21 共j k̃x F 兵Vi 其兲 , (1)
≠x i
where F and F 21 denote, respectively, the forward
and inverse discrete Fourier transform and k̃x is the
Fourier transform variable that represents the x com-
ponent of the numerical wave vector. In this way,
兵共≠V 兾≠x兲i 其 can be calculated in one step. In multiple
dimensions, this process is repeated for each cut par-
allel to one of the major axes of the space lattice.
According to the Nyquist sampling theorem, the
representation in Eq. (1) is exact (i.e., of infinite order)
for electromagnetic field spatial modes sampled at
the Nyquist rate or better. This permits the PSTD’s
meshing density to approach two samples per wave-
length in each spatial dimension. The wraparound
caused by the periodicity in the discrete Fourier trans-
form is eliminated by use of the anisotropic perfectly
matched layer absorbing boundary condition.7
We introduce an arbitrary incident wave by im-
plementing the scattered-f ield formulation.3 Here,
© 2004 Optical Society of America
1394 OPTICS LETTERS / Vol. 29, No. 12 / June 15, 2004

the total electric 共E兲 and magnetic 共H 兲 fields within cylinders of individual diameter d 苷 5 mm. Here, five
the grid are decomposed into a sum of unknown cases are shown 共N 苷 80, 200, 320, 400, 480兲. We see
scattered f ields 共Escat , Hscat 兲 and known incident that, as N exceeds approximately 200, the TSCS above
fields 共Einc , Hinc 兲 propagating in a lossless vacuum. 60 THz saturates at the level indicated in Fig. 1.
Following Ref. 3, we write Maxwell’s curl equations Figure 3 shows results analogous to those of Fig. 2
for this case as for a bundle of N dielectric cylinders for individ-
ual diameter d 苷 10 mm. Five cases are shown
≠Hscat
m 1 s ⴱ Hscat 苷 2 = 3 Escat 2 s ⴱ Hinc 共N 苷 20, 50, 80, 100, 120兲. As N exceeds approxi-
≠t mately 50, the TSCS above 60 THz saturates at the
≠Hinc , same level as in Figs. 1 and 2.
2 共m 2 m0 兲 (2) Together, Figs. 1 – 3 show that a saturation effect for
≠t
the TSCS of a fixed-overall-sized bundle of randomly
≠Escat
e 1 sEscat 苷 = 3 Hscat 2 sEinc positioned, noncontacting cylinders can be achieved in
≠t different ways. In Fig. 1 the position and number of
≠Einc the cylinders within the bundle are constant, while the
2 共e 2 e0 兲 (3) diameter of each cylinder increases. In Figs. 2 and 3
≠t
the diameter of each cylinder is constant, while the to-
in a source-free region where material parameters tal number of cylinders within the bundle increases.
m, e, s ⴱ , and s are mapped into the grid to specify As the average dielectric coverage of each bundle in-
the geometry. Space derivatives of 共Escat , Hscat 兲 are creases beyond a certain threshold, the TSCS of the
implemented by means of Eq. (1), and time stepping bundle becomes independent of its internal geometric
is implemented by the Yee leapfrog.3,5,6 The total E details such as the size, position, and number of its con-
and H fields are obtained in a postprocessing step by stituent cylinders.
addition of the PSTD-computed scattered f ields to the
known incident f ields.
The PSTD has been extensively validated.5,6,8 For
electromagnetic wave interaction structures whose
primary geometrical or material feature sizes exceed
one half of the dielectric wavelength 共ld 兲, the PSTD
has been shown to exhibit the same computational
accuracy and dynamic range as FDTDs that have
approximately eight-times-finer resolution.5,6 That
is, a PSTD grid with coarse ld 兾4 resolution provides
approximately the same accuracy as a FDTD grid
with f ine ld 兾32 resolution. Much experience with
FDTD modeling has shown that this level of spatial
resolution yields an accuracy of better than 1 dB over
dynamic ranges that exceed 50 dB for the scattering
Fig. 1. PSTD-computed TSCS of a 160-mm overall-
intensity observed at all possible angles, including diameter cylindrical bundle of 34 randomly positioned,
backscatter.3 For the millimeter-scale optical paths noncontacting n 苷 1.2 dielectric cylinders of individual
of great current interest in biophotonics, we expect diameter d. Four cases [(a) –(c)] are shown, with the po-
PSTD grids with ld 兾4 resolution to provide compa- sition of each cylinder fixed. As d exceeds approximately
rable accuracy and dynamic range. 10 mm, the TSCS above 60 THz saturates.
In this Letter we report the initial application
of the PSTD to model two-dimensional transverse-
magnetic scattering of light by large bundles of in-
finitely long dielectric cylinders in free space. We
use a PSTD grid that has a uniform spatial resolu-
tion of 0.25 mm, equivalent to 0.3ld at 300 THz for
cylinder refractive index n 苷 1.2. Figure 1 shows the
PSTD-computed total scattering cross section (TSCS)
of a 160-mm overall-diameter cylindrical bundle of
34 randomly positioned, noncontacting, dielectric
cylinders of diameter d. For each of the cases studied
共d 苷 5, 10, 15, 20 mm兲, a single PSTD run provides a
frequency response of 0.5– 300 THz 共l0 苷 60 1 mm兲
with a resolution of 0.5 THz. Note that the position
of the center of each cylinder is fixed for each case. Fig. 2. PSTD-computed TSCS of a 160-mm overall-
We can see that, as d exceeds approximately 10 mm, diameter cylindrical bundle of N randomly positioned,
the bundle’s TSCS above 60 THz saturates. noncontacting n 苷 1.2 dielectric cylinders of fixed individ-
Figure 2 shows the PSTD-computed TSCS of a ual diameter d 苷 5 mm. Five cases [(a) –(e)] are shown.
160-mm overall-diameter cylindrical bundle of N As N exceeds approximately 200, the TSCS above 60 THz
randomly positioned, noncontacting n 苷 1.2 dielectric saturates at the level indicated in Fig. 1.
 June 15, 2004 / Vol. 29, No. 12 / OPTICS LETTERS
Fig. 3. PSTD-computed TSCS of a 160-mm overall-
diameter cylindrical bundle of N randomly positioned,
noncontacting n 苷 1.2 dielectric cylinders of fixed individ-
ual diameter d 苷 10 mm. Five cases [(a) – (e)] are shown.
As N exceeds approximately 50, the TSCS above 60 THz
saturates at the same level as in Figs. 1 and 2.
Fig. 4. PSTD-computed TSCSs of (a) a 160-mm overall-
diameter cylindrical bundle of 120 randomly positioned,
noncontacting n 苷 1.2 dielectric cylinders of individual di-
ameter d 苷 10 mm; (b) as in (a) but for 480 cylinders of
individual diameter d 苷 5 mm; (c) a single cylinder of re-
fractive index n 苷 1.0938, the average refractive index for
(a) and (b).
This conclusion is further supported by Fig. 4, which
illustrates the PSTD-computed TSCSs of (a) a 160-mm
overall-diameter cylindrical bundle of 120 randomly
positioned, noncontacting n 苷 1.2 dielectric cylinders
of individual diameter d 苷 10 mm; (b) as in (a) but
for 480 cylinders of individual diameter 5 mm; and
(c) a single cylinder of refractive index n 苷 1.0938, the
volume-averaged refractive index in (a) and (b). We
can see that the frequency dependence of the TSCSs
of the bundles of (a) and (b) represents essentially the
average behavior of the TSCS of the volume-averaged
homogeneous cylinder of (c). The primary difference
1395
is that the homogeneous cylinder exhibits ripples of its
TSCS versus frequency as a result of coherent internal
wave-interference effects that are suppressed by scat-
tering events within the random clusters.
We have reported what we believe to be the f irst
rigorous numerical solution of the two-dimensional
Maxwell equations for optical propagation within, and
scattering by, a random medium of macroscopic dimen-
sions. Our solution is based on the pseudo-spectral
time-domain technique, which provides essentially
exact results for electromagnetic field spatial modes
sampled at the Nyquist rate or better. In ongoing
research we have found that it is straightforward to
extend the PSTD to the full-vector Maxwell equations
in three dimensions. We have validated PSTD models
of scattering by isolated spheres and are commencing
studies of optical interactions with millimeter-scale
three-dimensional volumes of biological tissues.
The results reported in this Letter point toward the
emerging feasibility of direct, exact Maxwell-equations
modeling of light propagation through, and scatter-
ing by, millimeters of biological tissues. More gener-
ally, our results have a wider implication. Namely,
the study of electromagnetic wave propagation within
random media is moving toward exact rather than ap-
proximate solutions of Maxwell’s equations.
The authors thank Zhigang Chen for helpful tech-
nical discussions. Also, the authors thank the Na-
tional Institutes of Health National Cancer Institute
(contract grant 5R01CA085991-03) and the Pittsburgh
Supercomputing Center (grant ECS020006P) for their
support of this research. This study was performed
under the auspices of the U.S. Department of Energy
under contract W-7405-ENG-48. S. H. Tseng’s e-mail
address is snow@ece.northwestern.edu.
References
1. A. J. Welch and M. J. C. van Gemert, in Lasers, Pho-
tonics, and Electro-Optics, H. Kogelnik, ed. (Plenum,
New York, 1995).
2. G. Mie, Ann. Phys. (Leipzig) 25, 377 (1908).
3. A. Taf love and S. C. Hagness, Computational Electro-
dynamics: the Finite-Difference Time-Domain Method
(Artech House, Boston, Mass., 2000).
4. R. Drezek, A. Dunn, and R. Richards-Kortum, Opt. Ex-
press 6, 147 (2000), http://www.opticsexpress.org.
5. Q. H. Liu, Microwave Opt. Technol. Lett. 15, 158 (1997).
6. Q. H. Liu, IEEE Trans. Geosci. Remote Sens. 37, 917
(1999).
7. S. D. Gedney, IEEE Trans. Antennas Propag. 44, 1630
(1996).
8. G. Zhao and Q. H. Liu, IEEE Trans. Antennas Propag.
51, 619 (2003).
56 OPTICS LETTERS / Vol. 30, No. 1 / January 1, 2005

Exact solution of Maxwell’s equations for optical interactions

with a macroscopic random medium: addendum
Snow H. Tseng, Jethro H. Greene, and Allen Taflove
Department of Electrical and Computer Engineering, Northwestern University, Evanston, Illinois 60208

Duncan Maitland
Medical Physics and Biophysics Division, Lawrence Livermore National Laboratory, Livermore, California 94550

Vadim Backman and Joseph T. Walsh, Jr.

Department of Biomedical Engineering, Northwestern University, Evanston, Illinois 60208

Received August 17, 2004

This Addendum provides a revised set of figures containing converged numerical data for total scattering
cross section (TSCS), replacing the figures in our recent publication [Opt. Lett. 29, 1393 (2004)]. Due to the
use of an overly large time step, our original TSCS data exhibited a systematic, nonphysical diminution above
150 THz for all cases studied. We have determined that numerical convergence in the temporal sense for the
pseudospectral time-domain (PSTD) algorithm employed previously requires limiting the time step to no more
than 1兾60th of the sinusoidal period at the maximum frequency of interest, which in the previous case was
300 THz. This is an important point that we hereby report to future users of PSTD simulations in electro-
dynamics and optics. Note that all our original conclusions remain valid. © 2005 Optical Society of America
OCIS codes: 240.6680, 060.0060, 060.2320, 240.6690, 170.3880, 290.7090.

In our recent publication,1 we employed the pseudo-

spectral time-domain (PSTD) method for the full-
vector Maxwell’s equations to compute the total
scattering cross section (TSCS) of a bundle of
randomly positioned dielectric cylinders in two dimen-
sions. While all the conclusions reported in Ref. 1
remain valid, in this Addendum we provide a set of
modif ied figures that contain converged numerical
data for the TSCS above 150 THz, which in Ref. 1
exhibited a systematic diminution of TSCS for all cases
studied. For a 160-mm-diameter bundle of dielectric
cylinders, we have determined that numerical conver-
gence is achieved by reducing the time step used in
the PSTD algorithm to 1兾60th of the sinusoidal period Fig. 1. PSTD-computed TSCS of a 160-mm overall-
at 300 THz. Figures 1 –4 in this Addendum show the diameter cylindrical bundle of 34 randomly positioned,
new results. noncontacting, n 苷 1.2 dielectric cylinders of individual
We conclude that, in all future applications of PSTD diameter d. Four cases are shown (d 苷 5, 10, 15, 20 mm)
with the position of each cylinder f ixed. As d exceeds
for large-scale models of optical interactions, it is
approximately 10 mm, the TSCS above 60 THz saturates.
mandatory to maintain a temporal resolution of at
least 60 samples per sinusoidal period at the high-
est frequency of interest. This temporal resolution
exceeds that previously reported in the literature.2,3
S. H. Tseng’s e-mail address is s-tseng@northwest-
ern.edu.

References
1. S. H. Tseng, J. H. Greene, V. Backman, D. Maitland,
and J. T. Walsh, Opt. Lett. 29, 1393 (2004).
2. Q. H. Liu, in Proceedings of the 1997 IEEE Antennas
and Propagation Society International Symposium,
Vol. 1 (Institute of Electrical and Electronics Engineers,
New York, 1997), pp. 122– 125.
3. A. Taf love and S. C. Hagness, Computational Electro- Fig. 2. PSTD-computed TSCS of a 160-mm overall-
dynamics: the Finite-Difference Time-Domain Method diameter cylindrical bundle of N randomly positioned,
(Artech House, Norwood, Mass., 2000), p. 852. noncontacting, n 苷 1.2 dielectric cylinders of f ixed in-
dividual diameter d 苷 5 mm. Five cases are shown
(N 苷 80, 200, 320, 400, 480). As N exceeds approximately
200, the TSCS above 60 THz saturates at the level
indicated in Fig. 1.
0146-9592/05/010056-02$15.00/0 © 2005 Optical Society of America
 January 1, 2005 / Vol. 30, No. 1 / OPTICS LETTERS 57

Fig. 3. PSTD-computed TSCS of a 160-mm overall-
diameter cylindrical bundle of N randomly positioned,
noncontacting, n 苷 1.2 dielectric cylinders of f ixed in-
dividual diameter d 苷 10 mm. Five cases are shown
(N 苷 20, 50, 80, 100, 120). As N exceeds approximately
50, the TSCS above 60 THz saturates at the same level as
in Figs. 1 and 2.
Fig. 4. PSTD-computed TSCS of (a) 160-mm overall-
diameter cylindrical bundle of 120 randomly positioned,
noncontacting, n 苷 1.2 dielectric cylinders of individual di-
ameter d 苷 10 mm; (b) same as in (a) but for 480 cylinders
of individual diameter d 苷 5 mm; and (c) single cylinder of
refractive index n 苷 1.0938, the average refractive index
for cases (a) and (b).