APPLIED PHYSICS LETTERS 89, 221116 共2006兲
Nonlinear inverse scattering and three-dimensional near-field optical
imaging
George Y. Panasyuk
Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania 19104
Vadim A. Markel
Department of Radiology, University of Pennsylvania, Philadelphia, Pennsylvania 19104
P. Scott Carney
Beckman Institute and Department of Electrical & Computer Engineering, University of Illinois, Urbana,
Illinois 61801
John C. Schotlanda兲
Department of Bioengineering, University of Pennsylvania, Philadelphia, Pennsylvania 19104
共Received 18 September 2006; accepted 16 October 2006; published online 29 November 2006兲
The nonlinear inverse scattering problem for electromagnetic fields with evanescent components is
considered. A solution to this problem is obtained in the form of a functional series expansion. The
first term in the expansion corresponds to the pseudoinverse solution to the linearized inverse
problem. The higher order terms represent nonlinear corrections to this result. Applications to the
problem of three-dimensional optical imaging with subwavelength resolution are described and
illustrated with numerical simulations. © 2006 American Institute of Physics.
关DOI: 10.1063/1.2396921兴
Optical imaging at the nanoscale presents a formidable letter is to extend these results to account for the effects of
challenge with correspondingly great scientific rewards.1 Of multiple scattering. This situation, in which the ISP is non-
particular current interest is the possibility of three- linear, is of interest when variations in the susceptibility are
dimensional nanoscale imaging.2–4 Potential applications not small or when the sample is sufficiently large—
range across multiple fields within both the physical and bio- conditions which characterize a large class of physical
logical sciences. In micro- and nanofabrication, for example, systems.
manufacturing processes are transitioning from planar to We begin by considering the scattering of a monochro-
stacked platforms. In cell biology, the study of subcellular matic electromagnetic field from a nonmagnetic medium
nanostructures is recognized as increasingly important. In characterized by a position-dependent dielectric susceptibil-
these and numerous other examples, near-field microscopy ity 共r兲. The electric field E is taken to consist of an incident
realizes nanometer scale resolution but with a significant part Ei and a scattered part Es according to E = Ei + Es. The
limitation, namely, only surface features of the sample are scattered field obeys the integral equation
visualized.3 Imaging of subsurface structures may be per-
formed by a destructive method in which the sample is ab-
lated and imaged layer by layer by scanning probe
E␣s 共r兲 = k20 冕 d3r⬘G␣共r,r⬘兲E共r⬘兲共r⬘兲, 共1兲
microscopy.5
where k0 = 2 / is the free-space wave number, G denotes
Near-field tomography 共NFT兲 is a recently proposed
Green’s tensor, and summation over repeated indices is im-
computed imaging modality for three-dimensional optical
plied. Evaluation of the integral in Eq. 共1兲 is nontrivial since
imaging with subwavelength resolution.2,6–8 NFT is nonde- the scattered field appears on both the left- and right-hand
structive and achieves nanoscale resolution by combining the sides of the equation. An explicit formula for Es, involving
experimental methods of near-field optics with the math- only the incident field, can be written in the form of the
ematical tools of inverse scattering theory. The diffraction perturbation expansion,
limit to resolution is overcome by exploiting the ability of
evanescent waves to carry information on subwavelength
scales. In NFT the sample is illuminated sequentially by a
series of incident waves while the scattered field is measured
E␣s 共r兲 = k20 冕 ⬘ d3r G␣共r,r⬘兲共r⬘兲Ei 共r⬘兲
in the near zone. Using the data gathered from such an ex-
periment, it is possible to reconstruct the three-dimensional + k40 冕 ⬘ d3r d3r⬙G␣共r,r⬘兲共r⬘兲G␥共r⬘,r⬙兲共r⬙兲
sample structure as encoded by the spatial dependence of the
susceptibility. This inverse scattering problem 共ISP兲 has been ⫻E␥i 共r⬙兲 + ¯ . 共2兲
studied within the accuracy of the single-scattering or first The first term corresponds to single scattering of the incident
Born approximation.6 The question of existence and unique- field, the second term corresponds to double scattering, and
ness of solutions has been addressed, and computationally so on. If the scattered field is much smaller than the incident
efficient inversion formulas are known.2 The purpose of this field, then only the first term in Eq. 共2兲 need be retained. This
result, which is known as the first Born approximation, lin-
a兲
Electronic mail: schotland@[Link] earizes the integral equation 共1兲 with respect to . Note that
0003-6951/2006/89共22兲/221116/3/$23.00 89, 221116-1 © 2006 American Institute of Physics
Downloaded 23 Jun 2011 to [Link]. Redistribution subject to AIP license or copyright; see [Link]
221116-2 Panasyuk et al. Appl. Phys. Lett. 89, 221116 共2006兲
multiple scattering leads to a nonlinear relationship between for every internal solid line. Applying these diagrammatic
E and . rules it can be seen that the first and second diagrams in Eq.
The Green’s tensor may be expanded into plane-wave 共3兲 correspond to the integrals
modes of the form G␣共r , r⬘兲 = 1 / 共2兲2 兰 d2q exp关iq · 共
− ⬘兲兴g␣共z , z⬘ ; q兲, where we have used the notation r
= 共 , z兲 and the form of g␣共z , z⬘ ; q兲 is given in Ref. 2. Note
k20 冕
0
L
dz 兺 g␣共zd,z;q̇1 − q2兲˜共q2 − Q,z兲e共q1,z兲
Q
共4兲
that this decomposition facilitates the treatment of systems and
冕冕 冕
which are translationally invariant in the transverse direction L L
d 2q
and that the plane-wave modes are labeled by the transverse
wave vector q. The modes for which 兩q兩 艋 k0 correspond to
k40
0 0
dz1dz2 兺 g␣共zd,z1 ;q1 − q2兲˜共q − q1
共2兲2 Q
propagating waves while the modes with 兩q兩 ⬎ k0 correspond
to evanescent waves. The contribution of the evanescent + q2 − Q,z1兲g␥共z1,z2 ;q兲˜共q1 − q,z2兲e␥共q1,z2兲, 共5兲
modes is exponentially small in the far zone of the scatterer where we have performed a shift in the argument of ⌽ so
which leads to loss of high spatial-frequency information that q2 → q1 − q2 and Q is a reciprocal lattice vector.
upon propagation. The inverse problem in NFT is to reconstruct ˜ from ⌽.
We will assume that the sample occupies the region 0 We begin by considering the linearized ISP. This corresponds
艋 z 艋 L and that the scattered field is measured on the plane to solving the integral equation K␣˜ = ⌽␣. Here the linear
z = zd with zd ⬎ L. The sample is illuminated by a plane wave operator K␣ is obtained by retaining only the first diagram in
of the form Ei = E共0兲 exp共iq1 · + ikz共q1兲z兲 with polarization Eq. 共2兲 and the action of K␣ on ˜ is defined by the integral
E共0兲, transverse wave vector q1, and kz共q兲 = 冑k20 − q2. The 共4兲. The linearized inverse problem has the solution ˜
scattered light is collected in the near zone of the sample by = K␣+ ⌽␣, where K␣+ denotes the pseudoinverse of K␣ which
a small aperture in a probe tip which is scanned over a square has been obtained elsewhere.2,6 To derive the solution to the
lattice with lattice spacing h. nonlinear ISP, we act on Eq. 共2兲 with K+ and thereby obtain
It will prove useful to define a data function by the
lattice Fourier transform of the sampled scattered field,
namely, ⌽␣共q1 , q2兲 = 兺e−iq2·E␣s 共 , zd ; q1兲, where the sum is 共6兲
carried out over all lattice vectors, q2 belongs to the first where ˜共1兲 = K␣+ ⌽␣. Next, we recognize that since Eq. 共6兲
Brillouin zone of the lattice, and the dependence on the determines ˜ self-consistently, we can iterate this result and
incident transverse wave vector q1 has been made explicit. obtain a diagrammatic series for ˜ of the form
Making use of Eq. 共2兲, we find that ⌽ can be
represented in a convenient diagrammatic form:
共3兲
Diagrams through the third order of multiple scattering are 共7兲
shown and are to be understood as follows. A solid line cor- Here we have introduced the notation that = 쎲, a double ˜ 共1兲
responds to a factor of k20g␣, a wavy line to a factor of solid line represents K+, and the effect of the overall appli-
共1 / h2兲exp共±ikz共q兲z兲E␣共0兲 ⬅ e␣共q , z兲, and a circle to a factor of cation of K+ is to join the ends of the diagrams. Applying the
˜共q , z兲 = 兰d2 exp共iq · 兲共 , z兲. In addition, a z integration is diagrammatic rules, it can be seen that the second diagram in
performed over each vertex and a q integration is carried out Eq. 共7兲 is given by the integral
k40 冕 d2q1K␣+ 共z;q1,q2兲 冕冕
L
0
L
0
dz1dz2 冕 d 2q
兺 g␣共zd,z1 ;q1 − q2兲˜共1兲共q − q1 + q2 − Q,z1兲g␥共z1,z2 ;q兲˜共1兲共q1 − q,z2兲e␥共q1,z2兲.
共2兲2 Q
共8兲
We note that there are three diagrams of third order, seven of Series solutions to the inverse problems of quantum me-
fourth order, and that the number of diagrams asymptotically chanical backscattering, acoustic scattering, and optical to-
doubles at each order. mography with diffuse light have been previously
Equation 共7兲 provides a formally exact solution to the reported.9–11 Though the algebraic structure of series solu-
inverse problem of NFT. It may be viewed as a nonlinear tions to different ISPs is similar, their analytic structure and
inversion formula whose first term coincides with the convergence properties are quite different. This is a reflection
pseudoinverse solution to the linearized ISP. The higher- of the underlying physical difference between the short-
order terms represent nonlinear corrections, which, in prin- range propagation of evanescent waves in NFT and the long-
ciple, can be computed to arbitrarily high order. We note that range propagation of waves in quantum mechanics or acous-
Eq. 共7兲 implies that it is necessary only to solve the linear tics. A detailed analysis of the convergence of the inverse
ISP in order to formally solve the nonlinear ISP. series is beyond the scope of this letter. Finally, we stress that
Downloaded 23 Jun 2011 to [Link]. Redistribution subject to AIP license or copyright; see [Link]
221116-3 Panasyuk et al. Appl. Phys. Lett. 89, 221116 共2006兲
FIG. 2. 共Color online兲 One-dimensional profiles of the reconstructed
susceptibility.
second, third, and fourth rows show the second, third, and
fourth order nonlinear reconstructions, respectively. Figure 2
shows the one-dimensional profiles of the reconstructed sus-
ceptibility along the line y = 0 in the equatorial plane. All
curves are normalized by the susceptibility 0 of the sphere.
It can be seen that the effect of the nonlinear corrections is to
systematically improve the reconstructions order by order.
In conclusion, we have described and illustrated in nu-
merical simulations a nonlinear inverse scattering method for
tomographic near-field optical imaging. We emphasize
that our results provide a direct rather than a iterative
numerical solution to the inverse problem. Applications to
three-dimensional optical imaging of nanostructures are
anticipated.
This work was supported by the NSF Grant No. DMR
FIG. 1. 共Color online兲 Tomographic images of a spherical scatterer. The 0425780 to one of the authors 共J.C.S.兲 and by the Air Force
field of view in each image is 18 ⫻ 18. MURI Grant No. F49620-03-1-0379 to another author
共P.S.C.兲.
the inverse series is a direct nonlinear inversion method, in 1
contrast to iterative Newton-type methods,12 which require L. Novotny and B. Hecht, Principles of Nano-Optics 共Cambridge Univer-
sity Press, New York, 2006兲.
the forward scattering problem to be solved for each 2
P. S. Carney and J. C. Schotland, MSRI Publications in Mathematics 47,
iteration. 133 共2003兲.
We now illustrate the use of the inverse series 共7兲 with a 3
T. Taubner, F. Keilmann, and R. Hillenbrand, Opt. Express 13, 8893
numerical example. The scatterer consists of a sphere of ra- 共2005兲.
4
dius R = 2 with index of refraction n = 1.1 related to the N. Anderson, P. Anger, A. Hartschuh, and L. Novotny, Nano Lett. 6, 744
共2006兲.
susceptibility by 0 = 共n2 − 1兲 / 4. The data function is com- 5
R. Magerle, Phys. Rev. Lett. 85, 2749 共2000兲.
puted from the Mie solution.13 The incident field was polar- 6
P. S. Carney and J. C. Schotland, Appl. Phys. Lett. 77, 2798 共2000兲.
7
ized in the x̂ direction, and 31 incident plane waves were P. S. Carney and J. C. Schotland, J. Opt. A, Pure Appl. Opt. 4, S140
employed with transverse wave vectors pointing in the ŷ 共2002兲.
8
direction and ranging uniformly from −0.9k0ŷ to 0.9k0ŷ. The P. S. Carney, R. Frazin, S. Bozhevolnyi, V. Volkov, A. Boltasseva, and J.
C. Schotland, Phys. Rev. Lett. 92, 163903 共2004兲.
plane of detection was located at a distance zd = / 3 from the 9
R. T. Prosser, J. Math. Phys. 10, 1819 共1969兲.
top of the sphere. The field was computed on a 600⫻ 600 10
A. B. Weglein, F. V. Arajo, P. M. Carvalho, R. H. Stolt, K. H. Matson, R.
grid with a lattice spacing h = / 6. Figure 1 presents the re- T. Coates, D. Corrigan, D. J. Foster, S. A. Shaw, and H. Zhang, Inverse
constructions obtained. The central column shows the results Probl. 19, R27 共2003兲.
11
of reconstructions in the equatorial plane of the sphere. The V. A. Markel, J. A. O’Sullivan, and J. C. Schotland, J. Opt. Soc. Am. A
20, 903 共2003兲.
left and right columns are the results of reconstructions in the 12
W. C. Chew and Y. M. Wang, IEEE Trans. Med. Imaging 9, 2318 共1990兲.
planes 0.7R above and below the equatorial plane. The first 13
C. F. Bohren and D. R. Huffman, Absorption and Scattering of Light by
row illustrates the results of linear reconstructions while the Small Particles 共Wiley, New York, 1983兲.
Downloaded 23 Jun 2011 to [Link]. Redistribution subject to AIP license or copyright; see [Link]