Download as pdf or txt
Download as pdf or txt
You are on page 1of 382

Design and Metrology of

Freeform Gradient-Index Optics


for Imaging and Illumination

by
David Henry Lippman

Submitted in Partial Fulfillment of the


Requirements for the Degree
Doctor of Philosophy

Supervised by Professor Duncan T. Moore, Professor Greg R. Schmidt,


and Professor Julie L. Bentley

The Institute of Optics


Arts, Sciences and Engineering
Edmund A. Hajim School of Engineering and Applied Sciences

University of Rochester
Rochester, New York

2023
To my family and friends,
and above all, my lovely Lindsey
Table of Contents

Biographical Sketch viii

Acknowledgments x

Abstract xii

Contributors and Funding Sources xiv

List of Acronyms xvi

List of Tables xviii

List of Figures xx

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Gradient-index (GRIN) optics . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Freeform GRIN (F-GRIN) . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.2 Application to optical design . . . . . . . . . . . . . . . . . . . 6
1.3.3 Current fabrication & metrology techniques . . . . . . . . . . 9
1.4 Dissertation objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

iii
Table of Contents iv

2 Freeform Gradient-Index Annular Folded Lenses 12


2.1 Annular folded lenses (AFLs) . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.1 Design form advantages and disadvantages . . . . . . . . . . 15
2.2 Design scope, specifications, and process . . . . . . . . . . . . . . . . 19
2.3 Global search of the homogeneous AFL solution space . . . . . . . . 21
2.4 GRIN representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Linear, two-material composition model . . . . . . . . . . . . 30
2.4.2 Quantifying GRIN dispersion . . . . . . . . . . . . . . . . . . . 33
2.4.3 Spatial refractive index variation . . . . . . . . . . . . . . . . . 34
2.4.4 GRIN optimization constraints . . . . . . . . . . . . . . . . . . 36
2.5 Monochromatic GRIN AFL designs . . . . . . . . . . . . . . . . . . . . 38
2.6 Color-corrected GRIN AFL designs . . . . . . . . . . . . . . . . . . . . 42
2.6.1 Two-material designs . . . . . . . . . . . . . . . . . . . . . . . 42
2.6.2 Multi-material designs . . . . . . . . . . . . . . . . . . . . . . . 48
2.7 GRIN AFL sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . 51

3 Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 56


3.1 Surface Alvarez lenses (SALs) . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 Gradient-index Alvarez lenses (GALs) . . . . . . . . . . . . . . . . . . 59
3.2.1 Refractive index profile . . . . . . . . . . . . . . . . . . . . . . 60
3.2.2 Tilt and ∆n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.3 Power variation . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2.4 Imaging performance . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.5 Fabrication, alignment, and metrology . . . . . . . . . . . . . . 89
3.3 Optimized GAL designs . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3.1 Higher-order polynomial designs . . . . . . . . . . . . . . . . 95
Table of Contents v

3.3.2 Hybrid polynomial-discrete designs . . . . . . . . . . . . . . . 103


3.3.3 Discontinuous plate designs . . . . . . . . . . . . . . . . . . . 112
3.4 Zoom riflescope designs with GALs . . . . . . . . . . . . . . . . . . . 114
3.4.1 Design goal and specifications . . . . . . . . . . . . . . . . . . 116
3.4.2 Design process . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4.3 Final designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

4 Gradient-Index Cover Glass for Field Curvature Compensation 132


4.1 Field curvature (FC) aberrations . . . . . . . . . . . . . . . . . . . . . . 132
4.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.1.2 Impact on imaging optical design . . . . . . . . . . . . . . . . 135
4.1.3 Impact on imaging performance . . . . . . . . . . . . . . . . . 137
4.1.4 Compensation with curved image detectors . . . . . . . . . . 139
4.2 Homogeneous detector cover glass . . . . . . . . . . . . . . . . . . . . 141
4.2.1 Longitudinal image displacement . . . . . . . . . . . . . . . . 143
4.3 Inhomogeneous detector cover glass . . . . . . . . . . . . . . . . . . . 145
4.3.1 FC compensation . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.3.2 Petzval field curvature . . . . . . . . . . . . . . . . . . . . . . . 153
4.3.3 Schmidt telescope design study . . . . . . . . . . . . . . . . . . 155
4.3.4 Connection to conventional field flatteners . . . . . . . . . . . 162
4.3.5 Future work: higher-order application spaces . . . . . . . . . 163

5 F-GRIN Prescribed Illumination Optics 168


5.1 Prescribed illumination design . . . . . . . . . . . . . . . . . . . . . . 168
5.2 F-GRIN for prescribed illumination . . . . . . . . . . . . . . . . . . . . 173
5.2.1 Design scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.3 Analytical linear GRIN ray tracing . . . . . . . . . . . . . . . . . . . . 177
Table of Contents vi

5.3.1 Skew ray tracing . . . . . . . . . . . . . . . . . . . . . . . . . . 184


5.4 Design process for F-GRIN prescribed illumination optics . . . . . . . 191
5.4.1 Optic-target mapping . . . . . . . . . . . . . . . . . . . . . . . 193
5.4.2 Linear GRIN array . . . . . . . . . . . . . . . . . . . . . . . . . 200
5.4.3 Integrated and interpolated piecewise-continuous F-GRIN . . 205
5.4.4 Monte Carlo ray trace evaluation . . . . . . . . . . . . . . . . . 213
5.5 Example designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
5.6 Fabricated and measured designs . . . . . . . . . . . . . . . . . . . . . 221

6 Three-Dimensional Gradient-Index Modal Reconstruction 230


6.1 Reconstruction overview . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.1.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.1.2 Application and scope . . . . . . . . . . . . . . . . . . . . . . . 235
6.1.3 OP D measurement modalities . . . . . . . . . . . . . . . . . . 237
6.1.4 Comparing ray and wave propagation . . . . . . . . . . . . . 238
6.1.5 Comparing modal and zonal reconstruction . . . . . . . . . . 242
6.1.6 Modal reconstruction and mid-spatial frequency errors . . . . 244
6.2 Modal reconstruction process . . . . . . . . . . . . . . . . . . . . . . . 245
6.2.1 Starting point generation . . . . . . . . . . . . . . . . . . . . . 246
6.2.2 Iterative optimization . . . . . . . . . . . . . . . . . . . . . . . 251
6.2.3 Effect of measurement noise . . . . . . . . . . . . . . . . . . . . 261

7 Conclusion and Future Research 264


7.1 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
7.2 Suggestions for future research . . . . . . . . . . . . . . . . . . . . . . 267

Bibliography 271
Table of Contents vii

Appendix A CODE V User-Defined Gradient-Index Source Code 289


A.1 “University of Rochester” linear, two-material composition . . . . . . 289
A.2 XYZ polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293

Appendix B Annular Folded Lenses Grid Search Code 304

Appendix C Analytical ∆n of Gradient-Index Alvarez Lens Elements 314

Appendix D Derivation of the Linear Gradient-Index Ray Path 321

Appendix E Algorithms used for Gradient-Index Modal Reconstruction 331


Biographical Sketch

David Henry Lippman was born in Boca Raton, Florida in 1996. He graduated
from high school in 2014 from Saint Andrew’s School in Boca Raton, Florida. After-
wards, he attended the University of Rochester and in 2018 received a Bachelor of
Science in optical engineering with a minor in mathematics, graduating magna cum
laude and with highest distinction. David began graduate studies at the University
of Rochester in 2018, pursuing a Doctor of Philosophy in optics focused on optical
design and metrology under the supervision of Professors Duncan T. Moore, Greg
R. Schmidt, and Julie L. Bentley. Concurrently, he received a Master of Science in
technical entrepreneurship and management from Simon Business School in 2021.
David is the 2020 recipient of the SPIE Optical Design and Engineering Scholarship
and the 2022 recipient of the Michael Kidger Memorial Scholarship.

List of Publications
N. S. Kochan, D. Xu, S. Iqbal, B. Moon, J. Hrdina, D. Lippman, S. A. Choudhury, M.
T. Banet, K. J. Dunn, A. A. G. Dewage, R. Draham, N. Takaki, J. D. T. Kruschwitz,
“Light and lilacs: an interactive exploration of colorimetry,” Proc. SPIE 11143, Fif-
teenth Conference on Education and Training in Optics and Photonics (2019).

D. H. Lippman and G. R. Schmidt, “Prescribed irradiance distributions with freeform


gradient- index optics,” Opt. Express 28(20), 29132-29147 (2020).

D. H. Lippman, D. S. Teverovsky, J. L. Bentley, “Monte Carlo first-order design


method for anamorphic cinema zoom lenses,” Opt. Eng. 60(5), 051203 (2021).

viii
Biographical Sketch ix

D. H. Lippman, R. Chou, A. X. Desai, N. S. Kochan, T. Yang, G. R. Schmidt, J. L.


Bentley, D. T. Moore, “Design of annular folded lenses using freeform gradient-
index optics,” Proc. SPIE 12078, International Optical Design Conference, 120781S
(2021).

D. H. Lippman, J. L. Bentley, D. T. Moore, “Learning lens design from Rudolf


Kingslake,” Proc. SPIE 12078, International Optical Design Conference, 1207809 (2021).

T. Yang, D. H. Lippman, R. Y. Chou, N. S. Kochan, A. X. Desai, G. R. Schmidt, J. L.


Bentley, D. T. Moore, “Material optimization in the design of broadband gradient-
index optics,” Proc. SPIE 12078, International Optical Design Conference, 120780Z
(2021).

D. H. Lippman, N. S. Kochan, T. Yang, G. R. Schmidt, J. L. Bentley, D. T. Moore,


“Freeform gradient-index media: a new frontier in freeform optics,” Opt. Express
29(22), 36997-37012 (2021).

D. H. Lippman, R. Chou, A. X. Desai, N. S. Kochan, T. Yang, G. R. Schmidt,


J. L. Bentley, D. T. Moore, “Polychromatic annular folded lenses using freeform
gradient-index optics,” Appl. Opt. 61(3), A1-A9 (2022).

D. H. Lippman, R. Xu, and G. R. Schmidt, “Freeform gradient-index optics for


prescribed illumination,” Proc. SPIE 12220, Nonimaging Optics: Efficient Design for
Illumination and Solar Concentration XVIII, 1222005 (2022).

D. H. Lippman, G. R. Schmidt, J. L. Bentley, D. T. Moore, H. Akhavan, J. P. Har-


mon, G. M. Williams, “Gradient-index Alvarez lenses,” Appl. Opt. 62(13), 3485-
3495 (2023).

D. H. Lippman, G. R. Schmidt, J. L. Bentley, D. T. Moore, H. Akhavan, J. P. Har-


mon, “Compact zoom visual scope using gradient-index Alvarez lenses,” Interna-
tional Optical Design Conference (2023).

D. H. Lippman and D. R. Shafer, “Unusual imaging between equi-curvature mir-


rors,” International Optical Design Conference (2023).

J. A. Sacks, D. H. Lippman, J. L. Bentley, “Generating Zoom Lens Starting Designs


With Particle Swarm Optimization and K-Means Clustering,” International Optical
Design Conference (2023).
Acknowledgments

There are many people I would like to thank for their support and guidance through-
out my academic career at the Institute. Without them, this work would not have
been possible.
First and foremost, I would like to thank my “triumvirate” of advisors: Duncan
Moore, Greg Schmidt, and Julie Bentley. In a synergistic way, each of them sup-
ported, engaged, and encouraged me in forming the researcher and engineer that I
am today. I am grateful for the research diversity they offered, and the stimulating
discussion we’ve had. It is an honor to have been their student.
I am also grateful to Jannick Rolland and Yuhao Zhu, the members of my dis-
sertation committee, for their insight and feedback during my graduate studies.
Thank you as well to Mark Bocko for serving as the chair of my committee.
Thank you to the Institute administrative staff who logistically enabled my re-
search and teaching. Thanks to Lori Russell, Tal Haring, Kari Kafka, Meir Brea,
and Dustin Newman. I am grateful for the help and guardrails Kai Davies offered
in managing the requirements of the PhD program. Special thanks to Lynn Reiner
for the advice, conference room keys, and birthday cake throughout the years. I
am also grateful to Per Adamson and Ed Herger, the lab managers during my time
here, for their assistance and lending of equipment.

x
Acknowledgments xi

I would like to thank Mike Pomerantz and Jim Alkins who taught me how to
build things, an invaluable skill for an engineer to have.
I am grateful to my graduate student colleagues whose discussion helped shape
the direction of my work. Thanks to the members of the GRIN research group dur-
ing my time here: Eryn Fennig, Tianyi Yang, Nick Kochan, Robert Chou, Ankur
Desai, Evan James, and Lyse Mugeni. I also appreciate master’s students Rongze
Xu and Junfu Zheng for their assistance. Last, I offer thanks to the members of the
five Advanced Lens Design classes I helped teach; you taught me even more.
Thank you to the city of Rochester, New York – a hidden gem – for having a
perfectly intricate logo to serve as the prescribed illumination target in this work.
Thanks to my family and friends for their unwavering support and for, at times,
listening to optics jargon with a smile on their faces. There are too many of you to
name, but I appreciate you all. Thanks also to my two muses, Peanut and Butter-
cup, for always greeting me with a lick and a wag. I am grateful to my parents,
Steven and Beth, and my sister, Suzanne, for their love, support, and encourage-
ment throughout the years.
And finally, a big thank you to my wife, Lindsey, my guide star and resolution
target, for her endless love and compassion.
Abstract

Freeform gradient-index (F-GRIN) optics present new opportunities for high per-
formance optical systems with compact geometries. As a type of freeform optic,
F-GRIN is similar in many ways to freeform surfaces but with different function-
ality that can be leveraged for unique optical influence. F-GRIN has only recently
been introduced due to advancements in the fabrication of arbitrary refractive in-
dex distributions such as via additive manufacturing. As a result, countless design
spaces are now accessible that require investigation to determine the strengths and
weakness of F-GRIN in optical design for both imaging and illumination. Of par-
ticular interest are the ways in which F-GRIN can reduce the size and weight of
conventional optical systems.
In this work, F-GRIN is studied in four different design spaces, three for imag-
ing and one for illumination, with the focus of achieving new system geometries
and novel optical functionality. First, annular folded lenses are designed using F-
GRIN for its unique color correcting properties in a compact, monolithic catadiop-
tric system. Second, GRIN Alvarez lenses (GALs) are introduced and analyzed
for their power variation and performance. GALs are then applied to the design
of a zoom riflescope where it is demonstrated that a reduction in system length is
achievable. Third, the unique field flattening capability of F-GRIN cover glass is
studied, offering potential reduction in imaging system complexity. Fourth, the il-

xii
Abstract xiii

lumination optical design considers the inverse problem of generating a prescribed


radiance distribution using a piecewise-continuous F-GRIN with plane-parallel
surfaces.
For any application of F-GRIN, adequate volumetric refractive index metrol-
ogy is required for practical and producible designs. Currently, non-destructive
F-GRIN metrology is in a premature state, and significant obstacles remain to be
solved before an accurate and precise technique is developed. A central challenge
is the algorithm needed to reconstruct a volume of three-dimensional (3D) refrac-
tive index variation from a series of independent measurements. In this work, a 3D
GRIN modal reconstruction technique is described for non-destructively obtain-
ing volumetric refractive index information. Formatted as a tomography problem,
multi-angular transmitted phase data are used to infer the GRIN volume. The re-
construction process is applicable to media with strong gradients and does not rely
on the presence of scattering or diffraction.
Contributors and Funding Sources

This dissertation has been supervised by a committee consisting of Professors Dun-


can Moore (advisor), Greg Schmidt (advisor), Julie Bentley (advisor), and Jannick
Rolland of the Institute of Optics, Professor Yuhao Zhu of the Computer Science
department, and Professor Mark Bocko of the Electrical and Computer Engineer-
ing department (chair).
The design study in Chapter 2 was a group effort lead by the author and with
contributions from Tianyi Yang, Ankur Desai, Nick Kochan, and Robert Chou at
the University of Rochester.
The CODE V user-defined GRIN source code in Appendix A.1 for the design
study in Chapter 2 is written by Tianyi Yang. Tianyi and the author collaborated
on the chromatic optimization tools used in Chapter 2.
The two grayscale illumination designs in Chapter 5 were executed by Rongze
Xu using software written by the author.
The reconstruction process in Chapter 6 benefited from conversations with Pro-
fessor Nebojsa Duric and Rehman Ali of the University of Rochester Medical Cen-
ter as well as Professor Nick Antipa of the University of California, San Diego.
The optical design and modeling in Chapters 2, 3, and 4 was performed with a
student license of CODE V® provided by Synopsys.
Fabrication of the F-GRIN elements presented in Chapters 3 and 5 was per-

xiv
Contributors and Funding Sources xv

formed by Nanovox LLC.


The work in Chapter 3 was funded by Nanovox LLC. The work in Chapter 5
was funded by the Technology Development Fund from the University of Rochester.
The author was also partially supported during his graduate studies by the SPIE
Optical Design and Engineering Scholarship and the Michael Kidger Memorial
Scholarship.
List of Acronyms

AFL annular folded lens


BFGS Broyden–Fletcher–Goldfarb–Shanno
BPM beam propagation method
BSE boresight error
CCD charge-coupled device
CG cover glass
CMOS complementary metal-oxide semiconductor
COTS commercial off-the-shelf
CPU central processing unit
CT computed tomography
DMD digital micromirror device
DOE diffractive optical element
DOF degree of freedom
ER eye relief
FC field curvature
FFP first focal plane
F-GRIN freeform gradient-index
GAL gradient-index Alvarez lens
GRIN gradient-index
LAP linear assignment problem
MA Monge-Ampère
MSF mid-spatial frequency
MTF modulation transfer function
NURBS non-uniform rational B-splines

xvi
List of Acronyms xvii

OCT optical coherence tomography


ODT optical diffraction tomography
OPD optical path difference
OPL optical path length
PPP plane-parallel plate
PV peak-to-valley
ROI ray of interest
RMS root mean square
SAL surface Alvarez lens
SELFOC self-focusing
SFP second focal plane
SHWFS Shack-Hartmann wavefront sensor
SPDT single-point diamond turning
SQM supporting quadratic method
SRA straight-ray approximation
SWAP-C size, weight, power, and cost
TEA thin element approximation
TIR total internal reflection
TTL total track length
USCT ultrasound computed tomography
WFE wavefront error
WPM wave propagation method
2D two-dimensional
3D three-dimensional
List of Tables

2.1 F-GRIN AFL exploratory design specifications. MTF specifications


are based on data listed in Table 2.2 for the Sigma 135 mm f /1.8 DG
HSM, a high-performing conventional telephoto lens. . . . . . . . . . 20
2.2 As-built MTF data for the Sigma 135 mm f /1.8 DG HSM measured
across the field-of-view [71]. MTF is recorded at different spatial
frequencies and for tangential (T) and sagittal (S) pupil sections. . . . 20
2.3 Different design parameters and values considered in the grid search
of the homogeneous AFL solution space. . . . . . . . . . . . . . . . . 22
2.4 System specifications for five homogeneous AFL design starting points
(see Fig. 2.4). The focal length for all designs is fixed at f 0 = −100 mm,
and all designs are made of PMMA. . . . . . . . . . . . . . . . . . . . 26
2.5 Specifications for the homogeneous base materials and GRIN for the
polychromatic GRIN AFL design in Fig. 2.11. . . . . . . . . . . . . . . 47
2.6 Tolerances for homogeneous base materials constituting the GRIN.
Material tolerances are based on standard values for “commercial”
and “precision” applications [86]. Errors in the base materials result
in errors in the GRIN dispersion. . . . . . . . . . . . . . . . . . . . . . 52

xviii
List of Tables xix

2.7 Tolerances for GRIN spatial coefficients governing radial and axial
variation in Eq. (2.8). Errors in the coefficients result in low spa-
tial frequency errors in the GRIN refractive index variation. Since
the GRIN spatial coordinates are normalized, both types contribute
approximately equal refractive index errors. . . . . . . . . . . . . . . . 54

3.1 Standardized GAL design parameters analyzed for imaging perfor-


mance in Sec. 3.2.4, except where noted otherwise. . . . . . . . . . . . 76
3.2 Fabricated base GAL design specifications. . . . . . . . . . . . . . . . 90
3.3 XYZ polynomial terms up to 7th order. In total, 120 terms are con-
sidered for higher-order polynomial GAL design. 50 terms (bold)
are GAL-compatible as even-order in x and odd-order in y. 13 select
terms (green) are found to be effective in improving GAL perfor-
mance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.4 Specifications of the f /4 Wood lens starting point for the optimized
design using the hybrid polynomial-discrete model. . . . . . . . . . . 108
3.5 Specifications for the 6x zoom riflescope design study. Low mag-
nification, M = 1.0, enables “eyes open” operation at the widest
field-of-view zoom position. The eye relief and total track length
are constant across zoom. Specifications are drawn from a conven-
tional zoom riflescope (Leupold Mark 6 1-6x20). . . . . . . . . . . . . 118
3.6 GRIN fabrication constraints in the zoom riflescope design study.
Constraints derive from current boundaries on Nanovox F-GRIN
additive manufacturing. . . . . . . . . . . . . . . . . . . . . . . . . . . 119
List of Tables xx

3.7 COTS elements used in the 6x zoom riflescope design. Lens element
numbers correspond to those noted in Fig. 3.38. Together, all COTS
elements can be purchased for $1,097. . . . . . . . . . . . . . . . . . . 129

4.1 Cover glass (CG) center thicknesses for different image detector man-
ufacturers [153]. Notice there is significant variation between man-
ufacturers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.2 Specifications for the Schmidt telescope design study using GRIN
CG to perform FC compensation. . . . . . . . . . . . . . . . . . . . . . 156

5.1 Fabrication constraints for F-GRIN prescribed illumination optics


made by additive manufacturing. . . . . . . . . . . . . . . . . . . . . . 221
List of Figures

1.1 Three standard GRIN geometries of refractive index variation: (a)


axial, (b) radial, and (c) spherical. . . . . . . . . . . . . . . . . . . . . . 4
1.2 Examples of different freeform GRIN (F-GRIN) with two or three in-
dependent spatial coordinates for different coordinate systems. Col-
ormap represents refractive index variation. . . . . . . . . . . . . . . . 7

2.1 Annular folded lenses (AFLs) with different numbers of reflections


R. For R > 2, multiple reflections are required per surface where
annular regions possess potentially different surface figure. Only
the on-axis field is shown for each design. . . . . . . . . . . . . . . . . 13
2.2 Modulation transfer function (MTF) for diffraction-limited imaging
systems with different obscuration ratios Robs . The spatial frequency
cutoff fcutof f = 1000 lp/mm set according to Eq. (2.1) for λ = 500 nm
and f /1.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

xxi
List of Figures xxii

2.3 Grid search results for the homogeneous AFL solution space. The
distribution of locally optimized designs is depicted for all nine pa-
rameters listed in Table 2.3. Vertical axes show the field-averaged
geometrical RMS spot size on a log scale. Each scatter point (col-
ored) represents an individual design, the boxes denote the 25th
through 75th percentile solutions, the bold horizontal lines show
the median performance, and the whiskers show the 2nd through
98th percentile. Solutions with unrecoverable performance (e.g., ray
trace failure) are filtered out of this analysis. . . . . . . . . . . . . . . . 24
2.4 Five homogeneous AFL design starting points for later incorpora-
tion of F-GRIN. Each design exhibits a different set of system specifi-
cations to explore a diverse solution space (see Table 2.4). Limited to
spherical surfaces, these starting points do not perform adequately,
as can be seen by the visibly aberrated image spots. . . . . . . . . . . 26
2.5 Homogeneous, monochromatic AFL design based on starting point
Design 3 (f /1.5, 10◦ full field-of-view). All four surfaces are aspheres
on base conics. The homogeneous material in use is PMMA. . . . . . 27
2.6 Homogeneous, polychromatic AFL design based on starting point
Design 3 (f /1.5, 10◦ full field-of-view). The design is a reoptimiza-
tion of the monochromatic design in Fig. 2.5 for the visible spec-
trum. All four surfaces are aspheres on base conics. The homoge-
neous material in use is PMMA. . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Transverse ray aberration plots for the homogeneous, polychromatic
AFL design in Fig. 2.6. Rays blocked by the central obscuration are
omitted. Chromatic aberrations limit performance and cannot be
corrected using a single homogeneous material. . . . . . . . . . . . . 28
List of Figures xxiii

2.8 Homogeneous, polychromatic AFL design based on starting point


Design 5 (f /2, 5◦ full field-of-view). Chromatic aberrations are bet-
ter controlled in this design with a front refractive surface that is
near-planar and a rear refractive surface that is near-concentric with
the image. All four surfaces are aspheres on base conics. The homo-
geneous material in use is PMMA. . . . . . . . . . . . . . . . . . . . . 30
2.9 Linear two-material composition model. GRIN dispersion is shown
for different volume fractions, C1 and C2 , of base homogeneous ma-
terials of differing dispersions, n1 (λ) and n2 (λ). . . . . . . . . . . . . 32
2.10 Monochromatic GRIN AFL design based on starting point Design 3
(f /1.5, 10◦ full field-of-view). All four surfaces are aspheres on base
conics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.11 Polychromatic GRIN AFL design based on starting point Design 3
(f /1.5, 10◦ full field-of-view). All four surfaces are aspheres on base
conics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.12 Transverse ray aberration plots across the field-of-view for the poly-
chromatic GRIN AFL design in Fig. 2.11. (a) Rays blocked by the
central obscuration (Robs = 0.7) are included (gray region). (b) Rays
blocked by the central obscuration are omitted (only rays transmit-
ted by the system are shown), allowing for a 100x reduction in aber-
ration scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.13 RMS wavefront error across the field-of-view for the polychromatic
two-material GRIN AFL design in Fig. 2.11. . . . . . . . . . . . . . . . 46
List of Figures xxiv

2.14 Homogeneous base materials used in the polychromatic GRIN AFL


design in Fig. 2.11. The design’s two base materials (red ‘X’s) are
shown alongside available materials from Nanovox (points). The
design process constrains the base materials within the region of
available materials (white) but does not fit to specific materials. The
red curves show the achievable values from a linear composition of
the two base materials. . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.15 GRIN dispersion for the polychromatic GRIN AFL design in Fig.
2.11. The dispersion is quantified by the spatially varying (a) Abbe
number ν and (b) relative partial dispersion PF,d . (c) Spatial varying
values for nd , ν, and PF,d are shown in three dimensions. With two
base materials, allowable values for nd , ν, and PF,d are confined to a
curve in three-dimensional space. . . . . . . . . . . . . . . . . . . . . . 48
2.16 Polychromatic multi-material GRIN AFL design based on starting
point Design 3 (f /1.5, 10◦ full field-of-view). All four surfaces are
aspheres on base conics. . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.17 RMS wavefront error across the field-of-view for the polychromatic
multi-material GRIN AFL design in Fig. 2.16. For comparison, the
same scale is used as in Fig. 2.13 for the two-material design. . . . . . 50
2.18 GRIN dispersion for the polychromatic GRIN AFL design in Fig.
2.16. The dispersion is quantified by the spatially varying (a) Abbe
number ν and (b) relative partial dispersion PF,d . (c) Spatial varying
values for nd , ν, and PF,d are shown in three dimensions. With mul-
tiple base materials, allowable values for nd , ν, and PF,d can occupy
a volume in three-dimensional space. . . . . . . . . . . . . . . . . . . 51
List of Figures xxv

2.19 Sensitivity analysis of GRIN dispersion for the polychromatic GRIN


AFL design in Fig. 2.11. Spectral error is randomly introduced by
deviations in homogeneous base materials with (a) “commercial”
tolerances (δn = ±0.001, δν = ±0.8%) and (b) “precision” tolerances
(δn = ±0.0005, δν = ±0.5%) [86]. 100 Monte Carlo trials are per-
formed. The perturbed MTF is shown on-axis and at full field. . . . . 53
2.20 Sensitivity analysis of low-spatial frequency refractive index error
for the polychromatic GRIN AFL design in Fig. 2.11. Spatial error
is randomly introduced via all eight spatial coefficients (four radial,
four axial) with maximum coefficient errors of (a) δC = ±0.5% and
(b) δC = ±0.25%. 100 Monte Carlo trials are performed. The per-
turbed MTF is shown on-axis and at full field. . . . . . . . . . . . . . 55

3.1 Surface Alvarez lenses (SALs) offer a variable amount of optical


power Φ based on laterally translating elements orthogonal to the
optical axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2 GRIN Alvarez lenses (GALs) offer a variable amount of optical power
Φ based on laterally translating elements orthogonal to the optical
axis. Unlike for SALs, GALs use plane-parallel surfaces and require
no internal airspace. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 GAL element shift δ along y imparts optical power. For the configu-
ration shown, the GAL produces positive power. . . . . . . . . . . . . 61
List of Figures xxvi

3.4 GRIN ∆n for base GAL elements with different values of tilt coef-
ficient B. Two specific cases of interest are denoted. (i) Zero tilt,
B = 0, is the conventional Alvarez form in Eq. (3.8). (ii) ∆n is at
a minimum when B = B0 , requiring 2.4 times less refractive in-
dex change than for zero tilt. ∆n increases approximately linearly
about B = B0 . This example is evaluated for Alvarez coefficient
A = −8.485 × 10−4 mm−3 and element diameter D = 10 mm. . . . . . 64
3.5 GAL element refractive index profiles for different values of tilt co-
efficient B. In each profile, maximum refractive index values are
marked with black points while minimum values are marked with
white points. When B/B0 = 1, the GRIN ∆n is minimized, as shown
in Fig. 3.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.6 GAL average refractive index. In the overlapping region, navg is a
radially quadratic GRIN power distribution. For element diameters
D and clear aperture CA, ±δmax is the maximum element shift before
vignetting occurs. When used in an optical system, all rays outside
CA are blocked. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.7 GAL power range ∆Φ versus element diameter D for a fixed clear
aperture CA. The power range is maximized for D/CA = 1.5. This
example is evaluated using Eq. (3.23) for no tilt with CA = 10 mm,
t = 3 mm, and ∆n = 0.2 Power range is shown in diopters [D = m−1 ]. 71
3.8 GAL element refractive index profiles with different coordinate shifts
y0 and bias powers Φbias in units of diopters. Each element produces
the same power range ∆Φ, but greater ∆n is needed for larger bias
powers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
List of Figures xxvii

3.9 Comparison of first-order and real ray power variation in GALs.


First-order power is calculated using the expressions derived in Sec.
3.2.3. Real ray power is evaluated from a wavefront fit to real ray
trace data. The three cases consider the GAL configuration in Table
3.1 with (a) zero tilt, (b) tilt minimizing ∆n, and (c) bias power of 5
diopters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.10 RMS WFE across a circular 10◦ full field-of-view for three differ-
ent power configurations. The evaluated GAL is defined in Table
3.1 with zero tilt. Diffraction-limited performance according to the
Maréchal criterion is limited to a small region in the center of the
field-of-view while performance significantly deteriorates with in-
creasing field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.11 Geometrical spot diagram across a circular 10◦ full field-of-view in
the 5 diopter power configuration. The evaluated GAL is defined in
Table 3.1 with zero tilt. The spots reflect the freeform aberration con-
tent of GALs with on-axis coma and trefoil and off-axis aberrations
that are not symmetric with field. Symmetry about the shift axis y is
maintained, same as for the field performance in Fig. 3.10. . . . . . . 80
3.12 Field curvature and astigmatism full field displays across GAL power
variation. The evaluated GAL is defined in Table 3.1 with zero tilt.
Field curvature is calculated with best-fit power removed according
to the on-axis field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.13 GAL power range ∆Φ versus on-axis RMS WFE, shown across ele-
ment shift δ. The evaluated GAL is defined in Table 3.1 with zero tilt.
The black dashed line shows the Maréchal criterion for the diffrac-
tion limit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
List of Figures xxviii

3.14 GAL bias power Φbias versus on-axis RMS WFE, shown across ele-
ment shift δ. The evaluated GAL is defined in Table 3.1 with zero tilt.
The black dashed line shows the Maréchal criterion for the diffrac-
tion limit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.15 GAL element thickness t versus on-axis RMS WFE, averaged across
power variation. For the change in t, the corresponding change in
GRIN ∆n is also shown that maintains the GAL power variation
defined in Table 3.1. The black dashed line shows the Maréchal cri-
terion for the diffraction limit. Performance deteriorates monotoni-
cally with increasing t and decreasing ∆n. . . . . . . . . . . . . . . . . 84
3.16 GAL diameter to clear aperture ratio D/CA versus on-axis RMS
WFE, averaged across power variation. For the change in D/CA,
the corresponding change in GRIN ∆n is also shown that maintains
the GAL power variation defined in Table 3.1. The black dashed line
shows the Maréchal criterion for the diffraction limit. Performance
improves monotonically with increasing D/CA while the ∆n is min-
imized for D/CA = 1.5 and increases beyond that. . . . . . . . . . . . 85
3.17 GAL tilt coefficient versus on-axis RMS WFE, averaged across power
variation. The tilt is quantified by its ratio with the tilt value mini-
mizing the refractive index change, B/B0 . For the change in B/B0 ,
the corresponding change in GRIN ∆n is also shown. The black
dashed line shows the Maréchal criterion for the diffraction limit.
WFE is at a minimum for zero tilt and only moderately degraded
with B/B0 = 1 for minimum ∆n. . . . . . . . . . . . . . . . . . . . . . 86
List of Figures xxix

3.18 GAL tilt coefficient versus (a) angular and (b) spatial boresight error
(BSE), shown across element shift δ. The GRIN tilt is quantified by
its ratio with the tilt value minimizing the refractive index change,
B/B0 . BSE is calculated from the ray passing through the centroid
of the geometrical on-axis spot. For spatial BSE, the ray height is
calculated in the best-focus image plane, which is nonexistent for
the zero power configuration, δ = 0. The evaluated GAL is defined
in Table 3.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.19 GAL fabricated via additive manufacturing by Nanovox LLC. (a) A
single GAL element is shown with a grid in the background from
which freeform refractive index change can be noticed. (b) Custom
machined housing is used for precision GAL element translation. . . 91
3.20 Mach-Zehnder interferograms of the fabricated GAL elements after
a clocking adjustment is performed to align elements with the trans-
lation axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.21 Mach-Zehnder interferograms of the fabricated GAL for different
element shifts δ. The red circle indicates the 3 mm clear aperture. . . 91
3.22 GAL power variation qualitatively demonstrated by change in im-
age magnification with element shift δ. In this example, the letter
“G” serves as the object and is of fixed size. The magnified “G”
image for different δ is the same modality used in the quantitative
power measurement shown in Fig. 3.23 for the USAF bar target. . . . 92
3.23 Fabricated GAL power variation measured by change in image mag-
nification. The first-order power according to Eq. (3.20) for this GAL
is shown for comparison. There is close agreement between mea-
sured and first-order power. . . . . . . . . . . . . . . . . . . . . . . . . 94
List of Figures xxx

3.24 10 select higher-order terms identified as providing significant GAL


performance improvement. RMS WFE improvement from the base
design is averaged over both power variation and field-of-view. The
percent increase in ∆n is shown for the same terms. . . . . . . . . . . 100
3.25 Optimized higher-order polynomial GAL design. Refractive index
profiles are shown in y-z and x-y cross-sections, all on the same col-
orbar scale. RMS WFE performance for this design is shown in Fig.
3.26. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.26 RMS WFE across a circular 10◦ full field-of-view for three different
power configurations for the higher-order polynomial optimized GAL
design in Fig. 3.25. Diffraction-limited performance is found in an
enlarged region in the center of the field-of-view that changes in po-
sition with power variation. RMS WFE is shown on the same scale
as in Fig. 3.10 for the base GAL design. . . . . . . . . . . . . . . . . . 102
3.27 Hybrid polynomial-discrete GRIN representation. In this example,
the refractive index is defined in N = 21 discrete planes separated
by ∆z. In each plane, the transverse refractive index ni (x, y) is
defined continuously by a two-dimensional polynomial. Between
planes, the axial refractive index is defined continuously by inter-
polation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
List of Figures xxxi

3.28 Refractive index profile and transverse ray aberration plots for (a)
the Wood lens defined in Table 3.4 and (b) the optimized design rep-
resented by the hybrid polynomial-discrete model. Starting from
the Wood lens design, the optimized design incorporates axial re-
fractive index change and higher-order transverse variation to cor-
rect spherical aberration and coma. The optimized design is ulti-
mately limited by astigmatism. The ray aberrations for both designs
are shown on the same scale. The optimized design has an increase
of 0.039 in total refractive index change. . . . . . . . . . . . . . . . . . 109
3.29 Optimized GAL elements represented by the hybrid polynomial-
discrete model. The optimized design achieves a 12.8x reduction
in angular ray aberration compared with the starting point base GAL.111
3.30 Optimized GAL using a series of transversely varying plates with
discontinuous axial refractive index change. The optimization is
performed in CODE V. The design performance shown in Fig. 3.31. . 113
3.31 RMS WFE across a circular 10◦ full field-of-view for three different
power configurations for the discontinuous plate GAL design in Fig.
3.30. Diffraction-limited performance is found across much of the
field-of-view. RMS WFE is shown on the same scale as in Fig. 3.10
for the base GAL design in Table 3.1. . . . . . . . . . . . . . . . . . . . 113
3.32 Conventional zoom riflescope design consisting of objective, zoom
relay, and eyepiece. The X-ray image is of the Leupold Mark 6 1-
6x20 riflescope, which is used to devise the specifications in Table
3.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
List of Figures xxxii

3.33 6x zoom riflescope design using GALs and custom optics. The total
track length T T L = 237 mm is a 25 mm reduction compared with
the conventional design listed in Table 3.5. . . . . . . . . . . . . . . . 124
3.34 Angular ray aberration plots through zoom for the custom 6x zoom
riflescope design. Note, both x and y fields are present, and the
aberration scale is in units of arcminutes. . . . . . . . . . . . . . . . . 125
3.35 Diffraction MTF performance for the custom 6x zoom riflescope de-
sign. MTF is evaluated across a 2.5 mm eye pupil, calculated out
to the eye’s spatial frequency cutoff, and considered monochromat-
ically at the design wavelength. . . . . . . . . . . . . . . . . . . . . . . 127
3.36 GAL element shift δ through zoom for the custom 6x zoom rifle-
scope design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.37 Angular boresight error through zoom for the custom 6x zoom rifle-
scope design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.38 6x zoom riflescope design using GALs and the COTS elements listed
in Table 3.7. The total track length T T L = 300 mm is longer than
specified, and ER is changing through zoom. . . . . . . . . . . . . . . 130
List of Figures xxxiii

4.1 Field curvature (FC) aberrations yield a non-planar image surface


(bold black lines) that can be either (a) stigmatic or (b) astigmatic.
(a) Stigmatic FC results in all rays in the pupil intersecting at an
infinitesimal point for a given field, introducing no “blur” in the
image spot. (b) Astigmatic FC produces spatially extended image
spots for a field, regardless of the longitudinal image position, due
to differing focal distances for different rays in the pupil. Tangential
(T) and sagittal (S) rays in the pupil focus at different longitudinal
positions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.2 Field curvature degrades image performance across the field-of-view
when restricted to a planar image detector. Transverse spot dia-
grams are shown across the field for a telecentric imaging system
suffering only from Petzval field curvature, a stigmatic FC aberra-
tion. The dashed curve shows the stigmatic Petzval image surface.
A planar detector is longitudinally positioned at the ideal focus for
the (a) on-axis, (b) mid-field, and (c) full field image heights. (a)-(c)
The image spots are extended for the out of focus fields, reducing
image resolution. (d) A curved image detector captures the stig-
matic curved image in focus across the full field-of-view. . . . . . . . 138
4.3 Longitudinal image displacement d for a beam converging through
a PPP of center thickness t and homogeneous refractive index n. For
each the upper and lower marginal rays, the entering and exiting
ray segments are parallel but offset from one another. . . . . . . . . . 143
List of Figures xxxiv

4.4 Longitudinal image displacement d calculated according to Eq. (4.2)


for increasing refractive index n and for different numerical aper-
tures N A. The image displacement increases monotonically with re-
fractive index. For this example, the PPP center thickness is t = 2 mm.144
4.5 Transversely varying image displacement d (x, y) for a beam con-
verging through a GRIN CG of center thickness t and spatially vary-
ing refractive index n (x, y). . . . . . . . . . . . . . . . . . . . . . . . . 146
4.6 FC introduction and compensation using GRIN CG. The image sur-
face for the imaging system preceding the CG is shown with a dot-
ted line and the image surface upon transmitting through the CG
is shown with a solid line. (a) For an imaging system producing
a planar image, the addition of a GRIN CG yields a curved image
surface, similar to in Fig. 4.5. (b) For an imaging system produc-
ing a curved image surface with sag z (x, y), the incorporation of
the proper GRIN CG compensates for FC aberrations and yields a
planar image surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.7 CAD depiction of GRIN CG directly incorporated into a planar de-
tector, yielding an artificially curved image detector. . . . . . . . . . . 149
4.8 Refractive index n required to flatten an image surface of sag z cal-
culated according to Eq. (4.13). (a) The full theoretical range for n
and z is shown within the achievable bounds (red). (b) A sub-range
is shown centered around z = 0 and n = n0 where the relationship is
closer to linear. (a)-(b) The required index increases monotonically
with increasing sag, either positive or negative. For this example,
the PPP center thickness is t = 2 mm and the base refractive index is
n0 = 1.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
List of Figures xxxv

4.9 Comparison of exact and approximate GRIN CG refractive index


for correcting Petzval field curvature. In this example, GRIN CG
thickness is t = 2 mm, and the Petzval surface curvature is cP =
0.25 mm−1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.10 Schmidt camera design study using GRIN CG for FC compensation.
System specifications are listed in Table 4.2. Ray trace, astigmatic
field curves, and MTF plot are shown for each design. Different
levels of FC compensation are used including (a) planar detector at
best focus, (b) parabolic curved detector, (c) GRIN CG with analyt-
ical refractive index according to Eq. (4.13), and (d) GRIN CG with
subsequently optimized refractive index profile. (c)-(d) Refractive
index profiles are shown separately in Fig. 4.11. . . . . . . . . . . . . 158
4.11 GRIN CG in Schmidt camera designs. (i) The GRIN profile used in
the design in Fig. 4.10(c). The refractive index is obtained analyt-
ically to flatten the Petzval surface, as specified in Eqs. (4.17) and
(4.18) to flatten the Petzval surface. (ii) The GRIN profile used in the
design in Fig. 4.10(d). The refractive index in this design is initial-
ized with the profile in (i) and then optimized for further correction
with the form in Eq. (4.19). . . . . . . . . . . . . . . . . . . . . . . . . . 160
List of Figures xxxvi

4.12 Concept illustration of applying GRIN CG to reduce track length in


a mobile device imager by performing FC compensation. (a) A con-
ventional mobile device imager applies six highly aspheric elements
followed by a homogeneous CG [156]. (b) A reduction from six to
four elements decreases system track length while sacrificing field
flatness with curved image surface z (x, y). (c) The incorporation of
GRIN CG compensates for FC while maintaining the reduced sys-
tem length. In this example, the GRIN CG three-dimensional refrac-
tive index variation accommodates the highly non-telecentric system. 165
4.13 Concept illustration of GRIN CG used to compensate freeform FC
aberrations. (a) Conventional three-mirror freeform telescope has a
sufficiently planar image surface [11]. (b) System volume is reduced
by increasing mirror power and surface tilt. As a result, the image
surface z (x, y) suffers from freeform FC. (c) The addition of GRIN
CG is used to compensate for freeform FC aberrations while keeping
system volume reduced. . . . . . . . . . . . . . . . . . . . . . . . . . . 166

5.1 The inverse problem central to illumination optics, including pre-


scribed illumination design. . . . . . . . . . . . . . . . . . . . . . . . . 169
5.2 Prescribed illumination generated using a freeform surface with sag
z (x, y). The black contour on the surface is a slope discontinuity
needed to form the hole in the illumination target. . . . . . . . . . . . 170
List of Figures xxxvii

5.3 Prescribed illumination generated using a freeform gradient-index


(F-GRIN) optic with refractive index n (x, y, z). The black contour in
the refractive index is a gradient discontinuity needed to form the
hole in the illumination target. Unlike for the freeform surface in
Fig. 5.2, the F-GRIN illumination optic requires only plane-parallel
surfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.4 Three different examples of linear GRIN with gradient directions ĝ
(a) in three dimensions, (b) parallel to z (i.e., axial gradient), and (c)
orthogonal to z in the transverse x-y plane. (c) Linear GRIN defined
by its gradient in the x-y plane is what is used in the prescribed
illumination design process. . . . . . . . . . . . . . . . . . . . . . . . . 178
5.5 Coordinates and geometry of linear GRIN used in the design pro-
cess, shown in cross-section. (a) The transverse x-y cross-section
shows the gradient direction along the g-axis, defined by the an-
gle θg measured from the y-axis. (b) The longitudinal cross-section
shows the g-z plane containing the gradient and optical axis. The
coordinate origin is located at the front surface vertex. . . . . . . . . . 179
5.6 Surface refraction at a linear GRIN. The ray is traced in the g-z plane
containing the gradient and the optical axis. The ray path g (z) is
a section of the hyperbolic cosine function determined by the ray
initial conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.7 Beam deflecting property of linear GRIN, demonstrated for (a) col-
limated light, (b) diverging light, and (c) converging light. No sig-
nificant power or aberration is introduced by the linear GRIN. . . . . 184
List of Figures xxxviii

5.8 Ray tracing in a linear GRIN for (a) a ray incident in the g-z plane
containing the gradient and the optical axis and (b) a skew ray in-
cident outside the g-z plane. The planar cross-section for the skew
ray obliquely dissects the volume but continues to experience a lin-
ear gradient. (a)-(b) In these cases, the gradient is aligned along the
y-axis, although this not a requirement, and the same skew planar
cross-section can be obtained regardless of θg . . . . . . . . . . . . . . . 185
5.9 Coordinate transformations performed in tracing a linear GRIN skew
ray incident outside the plane of the gradient. Unlike the gradient,
the initial ray direction may have components in all three dimen-
sions and is not confined to the transverse x-y plane. (a) The co-
ordinate system is rolled around the z-axis by −θg . (b) Next, the
coordinate system undergoes a yaw about the ĝ = ŷ direction by θs
such that the ray again falls in the g-z plane. (c) The longitudinal
cross-section depicts the effective propagation thickness tef f found
by the obliquity of the planar cross-section. . . . . . . . . . . . . . . . 187
5.10 Flowchart of the four-step process used in F-GRIN prescribed illu-
mination design. The four steps are described in detail in Sec. 5.4.1
– 5.4.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.11 Number of points N in a discretized illumination target. (a) For a
binary target, N is simply the number of white pixels in the target.
(b) For a grayscale target, N is the sum of bit values for all pixels. In
this grayscale example, there are three gray levels with black (value
0), gray (value 1), and white (value 2) regions. . . . . . . . . . . . . . 193
List of Figures xxxix

5.12 Different combinations of source distance zs and optic array element


size De for maintaining a fixed target pixel size Dpx and distance zt ,
i.e., throw ratio Rpx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.13 Mapping between points in the discretized GRIN array and illumi-
nation target. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.14 The linear assignment problem (LAP) demonstrated for N = 4 where
the objective is to minimize the Euclidean distance between optic
(black) and target (colored) points. (a) All pair combinations are
shown. (b) The pairs that minimize the Euclidean separation among
all points is solved. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.15 Regions of attainable ray coordinates at the target plane after travers-
ing linear GRIN of different parameters. Radial contours (blue gra-
dient) show ray coordinates for constant values of the refractive in-
dex slope α. Azimuthal contours (black) show ray coordinates for
constant values of gradient angle θg . The center point shows the ray
coordinate for homogeneous media, α = 0. Rays are traced from an
on-axis source with position zs = −100 mm and evaluated in a tar-
get plane with position zt = 1000 mm. (a) The linear GRIN element
is centered on the optical axis with the ray normally incident on the
front planar surface. (b) The linear GRIN element is offset from the
optical axis with center point (30, 30, 0) mm, which yields a 23.0◦ ray
angle of incidence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
List of Figures xl

5.16 Design example of a linear GRIN array that generates prescribed


illumination. The transverse refractive index distribution n (x, y) is
shown, including an exploded view of individual array elements.
Each element has a unique combination of n0 , α, and θg . In this
design, there are N = 104 elements in the array. . . . . . . . . . . . . . 204
5.17 Three steps in F-GRIN integration from (a) a linear GRIN array to
(b) discrete refractive index values from Southwell reconstruction to
(c) continuous refractive index from bicubic interpolation. . . . . . . 206
5.18 Bicubic interpolation scheme for (a) a continuous 4×4 enclosing grid
and (b) a grid including predefined gradient discontinuities. The
red ‘X’ shows the coordinate at which to interpolate the refractive
index. The yellows points are the innermost 2 × 2 grid of discrete
refractive index values used directly in interpolating index values.
The blue points are the two outermost rows and columns of discrete
refractive index values used in calculating index derivatives. The
gray points lying on the opposite side of a discontinuity are not used
in any capacity. The coordinates (xn , yn ) are normalized according
to the innermost 2 × 2 grid. . . . . . . . . . . . . . . . . . . . . . . . . 209
5.19 Design example of an integrated piecewise-continuous F-GRIN that
generates prescribed illumination. The transverse refractive index
distribution n (x, y) is shown, including an exploded view of a gra-
dient discontinuity (black lines). The linear GRIN array form of this
design is shown in Fig. 5.16. . . . . . . . . . . . . . . . . . . . . . . . . 212
List of Figures xli

5.20 F-GRIN prescribed illumination optics generating binary illumina-


tion targets containing (a) the Rochester “Flower City” logo and (b)
the letters “U of R.” For each optic, the illumination target, design re-
fractive index n (x, y), and Monte Carlo irradiance are shown. Both
optics are designed by the process outlined in Sec. 5.4 with zs ∼ −25
mm, t = 2 mm, zt = 1, 000 mm, Rt = 0.5, and N ∼ 104 . The designs
are both shifted and scaled to self-imposed refractive index fabrica-
tion limits of nmid = 1.5 and ∆n = 0.1. . . . . . . . . . . . . . . . . . . 215
5.21 F-GRIN prescribed illumination optics generating grayscale illumi-
nation targets containing (a) the Mona Lisa and (b) a simple auto-
mobile headlamp distribution. For each optic, the illumination tar-
get, design refractive index n (x, y), and Monte Carlo irradiance are
shown. The automobile headlamp irradiance, both target and ray
trace, are shown on a log intensity scale to make dim regions more
visible. Both designs were executed by Rongze Xu using the pro-
vided design program. . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.22 Histogram of relative irradiance for the Monte Carlo ray trace in Fig.
5.20(b) for the “U of R” design. Due to statistical noise inherent to
Monte Carlo ray tracing, there is a negligible yet non-zero number of
points with relative irradiance greater than 60%. The denoted 90th
percentile for non-zero relative irradiance (i.e., points with at least
one ray hit) is used to scale the target for the cross-sectional analysis
in Fig. 5.23. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
List of Figures xlii

5.23 Cross-section of the relative irradiance for the “U of R” design in Fig.


5.20(b). The cross-sectional plots compare the irradiance with the
target (black dashed line). The target is scaled to have its maximum
irradiance occur at the 90th percentile of ray traced values in order
to remove the contrast reduction from statistical noise (see Fig. 5.22).
Artifacts are noticeable from rays crossing gradient discontinuities,
yielding bands in the irradiance. . . . . . . . . . . . . . . . . . . . . . 217
5.24 Change in Monte Carlo irradiance with target evaluation distance zt
for the “U of R” design shown in Fig. 5.20(b). The “U of R” design is
performed for evaluation distance zt = 1 m, although the ray trace
shows preservation of the target over larger distances due to the
far-field design process. Evaluation distances closer than the design
value show how the irradiance transforms from the square F-GRIN
optic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
5.25 Change in Monte Carlo irradiance with circular disk source diame-
ter Ds for the “U of R” design shown in Fig. 5.20(b). The design is
performed for an infinitesimal point source, although illumination
fidelity adequate for many applications is maintained for extended
sources used at a distance zs = −25 mm. . . . . . . . . . . . . . . . . . 220
5.26 Fabricated F-GRIN prescribed illumination optic with Rochester “Flower
City” design. Top: illumination target and design refractive index.
Bottom: illumination optic fabricated by additive manufacturing. . . 222
5.27 Fabricated F-GRIN prescribed illumination optic with Nanovox logo
design. Top: illumination target and design refractive index. Bot-
tom: illumination optic fabricated by additive manufacturing. . . . . 222
List of Figures xliii

5.28 Experimental setup used in measuring generated illumination. For


the shown “Flower City” configuration, the source distance is zs =
−115 mm, and the illumination is viewed on a screen with distance
zt = 700 mm. In this configuration, the illumination profile on the
screen has dimensions of 150 mm × 150 mm. . . . . . . . . . . . . . . 223
5.29 Measured irradiance from fabricated F-GRIN prescribed illumina-
tion optics. (a) The illumination generated by the Rochester “Flower
City” design in Fig. 5.26. (b) The illumination generated by the
Nanovox logo design in Fig. 5.27. Qualitatively, both designs gener-
ate the target illumination with adequate fidelity for many applica-
tions. A skewing effect is found in both irradiances due to the line of
sight of the measuring camera differing from the source-optic-target
axis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
5.30 Cross-section of the measured relative irradiance for the fabricated
Nanovox logo design in Fig. 5.27. The cross-sectional plots compare
the irradiance with the target (black dashed line). A skewing effect
is found in the irradiance due to the line of sight of the measuring
camera differing from the source-optic-target axis. . . . . . . . . . . . 225
5.31 Mach-Zehnder interferogram of the fabricated Rochester “Flower
City” design in Fig. 5.26. The exploded view shows a rounded-off
gradient discontinuity. . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
List of Figures xliv

5.32 Change in measured relative irradiance with increasing circular disk


source diameter Ds for the fabricated Rochester “Flower City” de-
sign in Fig. 5.26. The increasing source extent rounds the edges of
the illumination target, and the increasing source area increases the
absolute irradiance. The changing source diameter is achieved by
adjustment of an iris in contact with the white light LED source. . . . 229
5.33 Measured irradiance using a collimated source for the fabricated
Rochester “Flower City” design in Fig. 5.26. The sun serves as
the light source and presents some angular extent that reduces edge
contrast similar to a spatially extended source. . . . . . . . . . . . . . 229

6.1 Illustration of the 3D GRIN tomography problem. Multi-angular


OP D data are needed to gain volumetric information about the re-
fractive index variation. . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.2 Comparison of wave propagation through Maxwell’s fisheye for (a)
BPM and (b) WPM. WPM succeeds in capturing ray curvature where
BPM fails. Adapted with permission from [220] © The Optical Society.240
6.3 Starting point for 3D GRIN reconstruction generated using the straight-
ray approximation (SRA). The ground truth, SRA generated starting
point, and residual are shown for cross-sections in x-y (z = 0.5 mm)
and y-z. The starting point is obtained from simulated OP D ray
trace data and a regularization parameter Λ = 50. The measure-
ment angle and pupil sampling used in the SRA is shown in Fig.
6.4. The SRA result does not exactly produce the axial refractive in-
dex change but rather is meant to provide an advantageous starting
point for subsequent iterative optimization. . . . . . . . . . . . . . . . 250
List of Figures xlv

6.4 OP D ray sampling used for the cost function summation in Eq.
(6.15). (a) Field angle sampling is performed on a polar grid where
each point indicates a plane wave angle of incidence across which
OP D is measured. (b) Pupil sampling is determined by Gaussian
quadrature [225, 226] where, for a given field angle, each point indi-
cates the entrance pupil coordinate of a traced ray for which OP D
is measured. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.5 The missing cone problem in 3D GRIN reconstruction, illustrated
for three plane wave measurement angles. The GRIN cylindrical
volume under test is embedded within a larger element with a ho-
mogeneous outer annulus. Circular apertures mask the clear aper-
ture on front and back surfaces. (a) The normally incident measure-
ment captures OP D information for the entire cylindrical GRIN vol-
ume, and the entrance pupil is filled. (b)-(c) Non-normally incident
measurements capture OP D information for only a subset of the
volume, and the entrance pupil is underfilled due to vignetting. . . . 255
6.6 3D GRIN reconstruction obtained using iterative optimization. The
ground truth, final reconstruction, and residual are shown for cross-
sections in x-y (z = 0.5 mm) and y-z. Reconstruction is performed
using simulated OP D ray trace data and a regularization parameter
Λ = 6.11 × 10−2 . The optimization starting point is generated by
the SRA result shown in Fig. 6.3. The measurement angle and pupil
sampling used in the reconstruction is shown in Fig. 6.4. . . . . . . . 259
6.7 Visualization of 3D GRIN coefficients Θ for the ground truth and
reconstructed result. Dominant terms in the ground truth are suc-
cessfully identified by the reconstruction. . . . . . . . . . . . . . . . . 259
List of Figures xlvi

6.8 Change with optimization cycle of (a) cost function and (b) RMS
error of reconstructed refractive index. The reconstruction result
shown in Fig. 6.6 is plotted for the final optimization cycle 73. The
RMS error is calculated by 3D sampling on a 10 µm grid. For all
cycles, the iterative optimization successfully reduces the cost func-
tion, which from Eq. (6.14) is based on both OP D and regulariza-
tion. Although, the RMS error of reconstructed refractive index does
not decrease for all cycles while ultimately reaching a minimum
upon completion of optimization. . . . . . . . . . . . . . . . . . . . . . 260
6.9 Distribution of random noise introduced to OP D data used in the
noisy reconstruction. The noise bandwidth is 3σ = 1/4 λ at λ =
632.8 nm, meaning 99.7% of OP D data points experience noise within
plus-or-minus a quarter-wave . . . . . . . . . . . . . . . . . . . . . . . 262
6.10 3D GRIN reconstruction obtained using iterative optimization for
OP D data containing Gaussian noise (3σ = ±1/4 λ). The ground
truth, final reconstruction, and residual are shown for cross-sections
in x-y (z = 0.5 mm) and y-z. Reconstruction is performed using
simulated, noisy OP D ray trace data and a regularization parameter
Λ = 5.50 × 10−1 . The optimization starting point is generated by the
SRA result shown in Fig. 6.3. The measurement angle and pupil
sampling used in the reconstruction is shown in Fig. 6.4. . . . . . . . 262
6.11 Visualization of 3D GRIN coefficients Θ for the ground truth and
the result reconstructed with noise. Dominant terms in the ground
truth are successfully identified by the reconstruction. . . . . . . . . . 263
List of Figures xlvii

C.1 GAL element total refractive index change ∆n for varying refractive
index tilt B. The Alvarez coefficient A and element diameter D are
held constant. The function ∆n (B) is piecewise with four different
subdomains. Example refractive index profiles are shown for each
subdomain with index minima and maxima marked with white and
black points, respectively. This example is evaluated for Alvarez
coefficient A = 8.485 × 10−4 mm−3 and element diameter D = 10 mm. 315
Chapter 1

Introduction

1.1 Motivation

The frontier of optical design is ever-expanding. As new technologies emerge and


new techniques are developed, the designer incessantly strives for improvements
in optical properties such as aperture, field-of-view, waveband, and image qual-
ity. These improvements must be made while balancing real-world constraints im-
posed on size, weight, power, and cost (SWAP-C) of the optical system. Included in
cost is the manufacturability and tolerances required to achieve the desired optical
performance when fabricated. Other design factors of note include environmental
conditions such as temperature and external mechanical forces.
The scope of traditional optical design is limited to refractive and reflective
surfaces of spherical shape, of which planar is a special case [1]. These spherical
surfaces are formed from optical media that is homogeneous, isotropic, and linear.
Although, advancements in modeling, fabrication, and testing have recently intro-
duced new optical design capabilities with additional degrees of freedom (DOFs).
These new DOFs allow access to solution spaces with higher optical specification
and reduced SWAP-C. Examples of these new technologies include advanced op-

1
Chapter 1. Introduction 2

tical surface types such as aspheres [2], diffractives [3], freeforms [4], and metasur-
faces [5]. Advancements have also been made in the domain of optical materials.
A notable example is gradient-index (GRIN) optics, which offer DOFs by a spatial
variation in refractive index [6].
In particular, freeform optics have enabled a new realm of optical design that
is no longer restricted to a single shared optical axis. Instead, complex system ge-
ometries can be obtained by tilting and/or decentering optical elements. In doing
so, freeform aberrations are introduced according to nodal aberration theory [7–9],
yet freeform optics have the capacity to correct these aberrations [10, 11]. Freeform
optics may also possess unique optical functionality that is unattainable with con-
ventional optics. One such example is the Alvarez lens, which provides a tunable
amount of optical power by use of freeform optics [12]. Within the class of freeform
optics, freeform surfaces are the most common type, possessing rotationally vari-
ant surface sag z (x, y) for use in refraction or reflection [4].
Recently, a new type of freeform optic has been introduced: freeform gradient-
index (F-GRIN) media [13–15]. This work explores how the application of F-GRIN
can further expand the optical design frontier. Both imaging and illumination
design forms are studied where theoretical and practical findings are collated.
The motivation for applying F-GRIN optics is similar to that of freeform surfaces:
achieving new system geometries and imparting new optical functionality.

1.2 Gradient-index (GRIN) optics

GRIN optics possess a spatial variation in refractive index, which is unlike conven-
tional homogeneous optical media where the refractive index is spatially uniform.
According to Fermat’s principle, the ray path between two points in a refractive
Chapter 1. Introduction 3

index distribution – homogeneous and inhomogeneous, alike – must be station-


ary with respect to variations in ray path. For homogeneous materials, Fermat’s
principle yields rays that travel linearly within the medium. On the other hand,
spatially varying refractive index entails that, in general, rays follow curved tra-
jectories within GRIN [16, 17]. From the calculus of variations, the Euler-Lagrange
equation applied to Fermat’s principle yields a differential equation for the ray
path,
0
(n~r 0 ) = ∇n (1.1)

where n is the refractive index distribution, ~r is the ray position vector, and (· · · )0
indicates d/ds.
GRIN’s capability to introduce ray curvature allows for unique optical func-
tionality that is unavailable in homogeneous optics. Perhaps the most classical
example of this is the Wood lens, which introduces optical power but with plane-
parallel surfaces [18]. The GRIN distribution in a Wood lens allows for a planar,
powered optic, a feat unattainable with conventional optics. Another example is
self-focusing (SELFOC) GRIN rods, which by oscillatory ray refraction serves as a
series of image relays within a single monolithic element [17].
There are three canonical geometries for refractive index variation within GRIN
media. All three options derive from fabrication techniques that rely on symmetry
of the underlying optical blank. The first is axial GRIN which has refractive index
variation along the optical axis direction, n (z), producing planar isoindicial sur-
faces, as shown in Fig. 1.1(a). The second is radial GRIN which has radial index
variation, n (r), with cylindrical isoindicial surfaces, as shown in Fig. 1.1(b). The
Wood lens and SELFOC rods described previously are both examples of radial
GRIN. The third type is spherical GRIN which has spherically symmetric index
Chapter 1. Introduction 4

Refractive index
(a) (b) (c)

Figure 1.1: Three standard GRIN geometries of refractive index variation: (a) axial, (b)
radial, and (c) spherical.

change, n (ρ), yielding spherical isoindicial surfaces, as shown in Fig. 1.1(c). Ex-
amples of spherical GRIN include the Maxwell fisheye and the Luneberg lens [17].
For the three canonical GRIN geometries, equations for the ray path can be
solved directly from the ray differential equation in Eq. (1.1). These ray solutions
are enabled by the symmetry of the refractive index, which reduces the complexity
of the differential equation [17]. For example, in a spherical GRIN, all rays lies in
a plane containing the origin of the spherical coordinate system. As a result, the
three-dimensional (3D) ray trajectory in a spherical GRIN can instead be consid-
ered more simply in a plane in polar coordinates.
More generally, an arbitrary refractive index distribution must be ray traced nu-
merically. One common technique which is used throughout this work is solving
the ray differential equation in Eq. (1.1) by the fourth-order Runge-Kutta method,
using a format presented by Sharma et al. [19–21]. The ray tracing scheme is it-
erative where the ray path is segmented into multiple smaller steps. Thus, the
generality of numerical ray tracing comes at the cost of computational efficiency
when compared to evaluation of the ray equations derived for symmetrical GRIN.
Chapter 1. Introduction 5

1.3 Freeform GRIN (F-GRIN)

1.3.1 Definition

Freeform GRIN (F-GRIN) is a type of GRIN media as well as a type of freeform


optic. The “freeform” qualifier in F-GRIN refers to what degree of symmetry, if
any, is present in the refractive index variation. The shape of accompanying optical
surfaces is irrelevant when it comes to whether an optic is F-GRIN or not. For
example, F-GRIN may possess plane-parallel surfaces while freeform surfaces, a
different type of freeform optic, may not.
The definition of freeform optics is not universally agreed upon. For example,
the definition of freeform surfaces has evolved over the past two decades [4]. The
definition of freeform surfaces cannot be directly applied to F-GRIN due to a dif-
ference in dimensionality. Freeform surfaces described a surface sag as a function
of two spatial coordinates, z (x, y). Meanwhile, F-GRIN obeys a volumetric func-
tion with the refractive index as a function of three spatial coordinates, n (x, y, z).
As a result, some degrees of symmetry not allowed in freeform surfaces may be
found in F-GRIN. For example, F-GRIN may possess an axis of rotational symme-
try, yet vary independently in the two remaining spatial coordinates, still impart-
ing freeform aberrations [15].
A formal definition for F-GRIN has been proposed [15]:

Any GRIN medium that must be specified in two or more independent spatial
coordinates is a freeform GRIN (F-GRIN).

Notice, “coordinates” is used in the definition rather than “dimensions” to reflect


comparable DOFs compared with other freeform optics. For instance, the spheri-
cal GRIN shown in Fig. 1.1(c) has 3D refractive index variation but only a single
Chapter 1. Introduction 6

coordinate of index change, meaning it is not a type of F-GRIN, as expected. The


same is true for axial and radial GRIN, neither of which are considered F-GRIN.
F-GRIN can ultimately have an arbitrary 3D refractive index profile. With no
assumed degree of symmetry, there is no way of simplifying the ray differential
equation in Eq. (1.1). As a result, there is no general analytical solution for the ray
path within F-GRIN. Instead, ray tracing must be performed numerically, such as
with the Runge-Kutta method [19–21] mentioned in Sec. 1.2. This numerical ray
tracing procedure is necessary for obtaining optical designs using F-GRIN.
The classification of different types of F-GRIN can be performed in terms of
the number of independent spatial coordinates as well as the coordinate system
in which the F-GRIN is defined. For example, F-GRIN with two and three co-
ordinate dependence are shown in Fig. 1.2 for Cartesian, cylindrical, and spher-
ical coordinate systems. F-GRIN with exclusively transverse variation, n (x, y),
presents largely the same optical effect as a freeform surface of sag z (x, y) [14].
However, F-GRIN also has the capacity for longitudinal refractive index change,
n (x, y, z), which allows for new volumetric DOFs not available in a single freeform
surface [15].

1.3.2 Application to optical design

F-GRIN offers new DOFs to the design process by its multiple spatial coordinates
of refractive index variation. The added DOFs allow for further improvement in
system specification as well as performance. Also, F-GRIN enables new system
geometries and novel optical functionality, similar to other freeform optics, as de-
scribed in Sec. 1.1.
F-GRIN presents freeform aberration contributions that prove valuable in op-
Chapter 1. Introduction 7

2 Coordinates 3 Coordinates
Cartesian xy F-GRIN xz F-GRIN yz F-GRIN F-GRIN
coordinates
y
z
x

Cylindrical ρθ F-GRIN ρz F-GRIN θz F-GRIN F-GRIN


coordinates
ρ
θ
z

Spherical rθ F-GRIN rφ F-GRIN θφ F-GRIN F-GRIN


coordinates
r
θ
φ

Figure 1.2: Examples of different freeform GRIN (F-GRIN) with two or three independent
spatial coordinates for different coordinate systems. Colormap represents refractive index
variation.

tical design [14, 15]. F-GRIN that is transversely varying but axially constant can
impart the same wavefront aberration as a freeform surface, at least within the
thin-element approximation (TEA) [22]. The magnitude of this aberration contri-
bution is proportional to the F-GRIN thickness t and total refractive index change
∆n whereas for freeform surfaces the aberration magnitude is established by the
extent of the surface sag. Field dependent aberrations are also introduced for
F-GRIN located away from the aperture stop, similar to freeform surfaces [10].
Where F-GRIN and freeform surfaces differ is when the F-GRIN also incorporates
axial refractive index change. Axial variation affects the field dependence of intro-
duced freeform aberrations. For example, linear axial change removes field con-
stant aberrations. As the order of axial variation increases, so does the aberration
Chapter 1. Introduction 8

field dependence that is removed [14]. In fact, these different axial terms can be
combined to target the introduction or removal of specific freeform field depen-
dent aberrations [15]. In this way, F-GRIN varying both transversely and axially
has a similar effect to multiple freeform surfaces.
Furthermore, the mathematical representation of the F-GRIN refractive index
profile greatly influences a resultant design. Similar to freeform surfaces, F-GRIN
can be defined with a global representation (e.g., polynomial basis in three-dimensions
[14]) or a local representation (e.g., interpolated point-cloud [23], NURBS [24], etc.).
Local representations typically present more DOFs but at the cost of increased com-
putational complexity. This work applies both global and local F-GRIN represen-
tations where this trade-off becomes apparent.
As for all optical media, F-GRIN is dispersive which means its refractive index
varies both spatially and spectrally, n (x, y, z, λ). Upon refraction, dispersion gives
rise to chromatic aberrations that reduce image quality, assuming polychromatic
imaging. Conventionally, optical design relies on multiple homogeneous mate-
rials of differing dispersions for controlling chromatic effects. Meanwhile, GRIN
dispersion is dependent on the constituent base materials used in fabrication [25].
As a result, the GRIN dispersion profile can itself be engineered by tailoring the
base materials’ dispersion. This GRIN dispersion offers valuable chromatic DOFs
absent in homogeneous optical materials, allowing, for example, achromatization
with a single GRIN medium [26]. The chromatic properties of F-GRIN can also
be leveraged for controlling chromatic aberrations in more complex freeform de-
signs [27, 28].
Several designs incorporating F-GRIN have been presented in the literature
thus far for both imaging [13–15, 27–32] and illumination [33–36]. Nevertheless,
countless new opportunities in optical design remain to be investigated using F-
Chapter 1. Introduction 9

GRIN and its vast DOFs. Of particular interest is the impact F-GRIN can have in
reducing the SWAP-C of conventional optical systems.

1.3.3 Current fabrication & metrology techniques

Traditionally, GRIN distributions were limited by available fabrication methods to


single coordinates of index variation. This restriction is what gave rise to the three
canonical GRIN profiles depicted in Fig. 1.1 where the form of the optical blank af-
fected the index change [6]. More recently, F-GRIN index profiles have been made
possible by advancements in fabrication techniques that break free from imposed
symmetry. These new techniques are the primary impetus behind recent interest
in applying F-GRIN to optical design.
One promising fabrication option for F-GRIN media is additive manufacturing.
F-GRIN additive manufacturing is a sub-aperture fabrication technique where in-
dividual voxels (i.e., 3D pixels) of material are deposited, and the spatial index
variation is implemented by changing the deposited voxel index. Additive man-
ufacturing of GRIN media has been demonstrated both in polymer [37–39] and in
glass [40]. By the nature of additive manufacturing, any refractive index distribu-
tion can be fabricated so long as the constituent base materials are available. This
capability includes fabricating discontinuities in either the refractive index or the
gradient of F-GRIN.
In addition to additive manufacturing, there are other fabrication options ca-
pable of producing F-GRIN, although not all with the same generality as addi-
tive manufacturing. Perhaps the earliest technique consolidates sheets of polymer
nanolayers into a tailored axial GRIN and then mechanical deforms the substrate
to introduce further spatial index change [41]. Another fabrication option applies
Chapter 1. Introduction 10

photopolymerization, although the total refractive index change ∆n is limited for


these techniques [42]. A third method uses direct laser writing in a porous sili-
con wafer to produce 3D varying index distributions, albeit for micron-scale op-
tics [43]. Lastly, some of these techniques, such as direct laser writing, have also
been applied to materials for use in the infrared [44].
While fabrication methods have begun to emerge, the state of metrology for
F-GRIN remains in a premature state. Current non-destructive GRIN measure-
ments primarily rely on transmitted relative phase measurements, such as with
a Mach-Zehnder interferometer [45]. These techniques are well-suited for mea-
suring GRIN that is varying transverse to the transmitted wavefront; however,
information about the index gradient along the propagation direction is masked.
As a result, new metrological tools are required for non-destructively measuring
F-GRIN with 3D index change, strong gradients, and no measurable scattering or
diffraction. There are significant challenges that must be overcome for achieving a
general 3D GRIN metrology method.

1.4 Dissertation objectives

There are three primary objectives of this dissertation, all of which relate to the
design and metrology of F-GRIN for application in optical design. The objectives
are achieved through five chapters, some of which accomplish multiple objectives.
The first objective is to apply F-GRIN in designing systems with desirable ge-
ometries that are unattainable with conventional optics. Novel system configura-
tions is a chief feature of freeform optics, and the F-GRIN design space remains
largely unexplored. Unlike freeform surfaces which are more often implemented
as reflective optics, F-GRIN may only exert its optical influence in refraction, giv-
Chapter 1. Introduction 11

ing rise to different types of systems. Unique system geometries by aid of F-GRIN
are demonstrated for monolithic annular folded lenses (Chapter 2), compact zoom
riflescopes using GRIN Alvarez lenses (Chapter 3), and GRIN cover glass for field
curvature compensation (Chapter 4).
The second objective is to obtain novel optical functionality using F-GRIN, an-
other hallmark of freeform optics. One example that is considered are GRIN Al-
varez lenses, which offer a tunable amount of optical power without longitudinal
translation (Chapter 3). Like other types of variable power components, the effect
of GRIN Alvarez lenses is not achievable using conventional optics. Furthermore,
a significant difference of F-GRIN compared with conventional optics as well as
other freeform optics is its optical influence can be achieved as a “flat” optic with
plane-parallel surfaces. The planar form factor of F-GRIN is leveraged in three dif-
ferent chapters for GRIN Alvarez lenses elements (Chapter 3), GRIN cover glass
acting analogously to a field flattener (Chapter 4), and a flat prescribed illumi-
nation optic (Chapter 5). The F-GRIN illumination optic also takes advantage of
discontinuities in the refractive index gradient, which can be fabricated with ad-
ditive manufacturing. Moreover, the unique chromatic properties of F-GRIN are
applied for polychromatic correction of monolithic annular folded lenses, a feat
unattainable with homogeneous media (Chapter 2).
The third objective is to tackle one of the challenges facing non-destructive
3D GRIN metrology. The reconstruction problem is examined where the goal is
to infer 3D GRIN information from a series of independent measurements. In
this work, an approach is demonstrated that applies a modal reconstruction algo-
rithm to recover a 3D GRIN volume from multi-angular relative phase information
(Chapter 6). This reconstruction process can be used in metrology for obtaining
any smooth F-GRIN distribution.
Chapter 2

Freeform Gradient-Index Annular


Folded Lenses

2.1 Annular folded lenses (AFLs)

Advanced optical technologies like aspheric and freeform surfaces, diffractive op-
tical elements (DOEs), and gradient-index (GRIN) media are typically applied to
optical design for two main reasons. The first is in offering new optical functional-
ity. Alvarez lenses, for example, use freeform optics to provide a variable amount
of optical power, an effect unattainable with conventional optics [12]. A second op-
portunity granted by advanced optical components is in obtaining new system ge-
ometries. Conventional optical systems rely on a series of elements aligned to the
optical axis. Axially symmetric systems present convenient options in fabrication
and alignment due to their rotational symmetry. Deviating from this conventional
mold by tilting or decentering elements introduces a new class of aberrations as
described by nodal aberration theory [7–9].
Moreover, many unconventional system geometries rely on reflective compo-
nents to fold the optical path. Some folded systems maintain an axis of rotational
symmetry with the introduction of an obscuration and/or by use of an off-axis por-

12
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 13

tion of the pupil or field-of-view. For example, classical reflective telescopes such
as the Cassegrain design possess rotational symmetry with reflective elements and
subsequently require a central obscuration in the pupil [46]. Many of these clas-
sical reflective designs also apply non-spherical surfaces, typically conics used at
their stigmatic conjugates as Cartesian reflectors.
Meanwhile, annular folded lenses (AFLs) are an unusual design form that rely
on many of the described techniques to achieve a system geometry that is unattain-
able with conventional optics [47]. AFLs use reflective surfaces to axially fold the
system geometry while maintaining rotational symmetry. Consequently, a central
obscuration is present with multiple reflective surfaces aligned to the optical axis.
The key feature of AFLs is that the reflective surfaces are closely positioned to one
another, yielding a very axially compact geometry (see Fig. 2.1). The proximity of
reflective surfaces requires multiple reflections per surface, meaning unique annu-
lar regions of a surface are used for each reflection, hence “annular” in AFL.

number of reflections

2 4 6 8

Smaller obscuration Smaller volume


Larger field-of-view More degrees of freedom

Figure 2.1: Annular folded lenses (AFLs) with different numbers of reflections R. For R >
2, multiple reflections are required per surface where annular regions possess potentially
different surface figure. Only the on-axis field is shown for each design.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 14

The number of reflections in an AFL is a free variable in the design process


and strongly influences a design’s characteristics [47, 48]. For example, Fig. 2.1
shows AFLs designed for different numbers of reflections. The number of reflec-
tions is typically restricted to an even number in order for the object and image to
be positioned on opposite sides of the optic. Also, with fewer reflections, a smaller
obscuration and larger field-of-view can be achieved. The reason for this is, with
fewer reflections, there is more available clearance between annular regions before
larger fields “skip” a surface region. On the other hand, with a greater number of
reflections, a smaller system volume can be attained and more degrees of freedom
(DOFs) are available for aberration correction (see Fig. 2.1). The smaller volume
can be attributed to weaker individual surfaces requiring a smaller center thickness
to, again, ensure clearance between annular surface regions while still capturing
the full field-of-view.
AFLs have been demonstrated in different forms and for use in different ap-
plication spaces. These various designs differ primarily based on the number of
reflections and the specifications required for each application. One of the ear-
liest AFL designs is the well-known all-reflective Schwarzschild microscope ob-
jective [49]. The Schwarzschild objective employs a two reflection AFL design,
as have several other microscope-style objectives including for solid-immersion
microscopy [50] and deep-ultraviolet microlithography [51]. AFLs have also gar-
nered appeal as imagers, particularly when restricted to axially compact form fac-
tors [47, 48, 52–55]. For example, Tremblay et al. studied ultra-thin AFLs need-
ing four or more reflections [47, 48]. AFLs have been applied for an assortment of
other applications including scanning [56], telescopic contact lenses [57,58], zooms
systems employing liquid lenses [59], and even as nonimaging concentration op-
tics [60, 61].
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 15

2.1.1 Design form advantages and disadvantages

There are two key advantages to the AFL design form. The first has already been
mentioned: their folded geometry yields a very axially compact package compared
with conventional optical systems (see Fig. 2.1). For most imaging applications,
the system volume and weight are limiting factors. In many cases, it is desired to
limit the axial system extent. For example, imagers in mobile devices are severely
limited in total track length (TTL) due to the thickness of the device. AFLs easily
meet stringent TTL requirements when they are necessary.
The second advantage of AFLs is that they can support very fast apertures with
adequate image quality. As a result, very high imaging resolution is attainable in
AFLs. The spatial frequency cutoff of an incoherent imaging system is [62]

1
fcutof f = (2.1)
λ · f /#

where λ is the imaging wavelength and f /# is the system f-number. AFLs can
typically have f-numbers faster than f /2, which at the visible wavelength λ = 500
nm offers a cutoff frequency fcutof f > 1000 line pairs per millimeter (lp/mm). Some
AFLs even immerse the image at the rear surface of the optic to achieve even faster
f-numbers and higher diffraction-limited resolution [47].
From both of these advantages, AFLs are primed for use as telephoto imaging
systems. First, the T T L of a telephoto lens is often quantified in terms of its focal
length f 0 by the telephoto ratio k,

TTL
k= , (2.2)
f0

where smaller values for k yield more axially compact systems. Achieving tele-
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 16

photo ratios . 0.7 proves difficult for high-performing conventional telephoto lens
designs [63]. Meanwhile, AFLs can reach far smaller telephoto ratios due to their
folded geometry [47].
The fast aperture but limited volume of AFLs is a second reason why they are
well-positioned as alternatives to conventional telephoto lenses. Fast telephoto
lenses are very desirable due to their high resolution imaging as well as greater
image irradiance and resultant signal-to-noise ratio. Although, conventional tele-
photo lenses require very large system diameters at fast f-numbers. This require-
ment is intrinsic to the positive-negative power separation of the classical tele-
photo design form. On the other hand, fast telephoto AFLs have approximately
the same diameter as conventional designs (for the same focal length) but require
a much smaller system volume due to their axial compactness. As such, telephoto
AFLs can also dramatically reduce system weight.
While telephoto AFLs possess several attractive features, they also present some
disadvantages that must be carefully considered. First and foremost, AFLs re-
quire a central obscuration, which reduces modulation transfer function (MTF)
mid-spatial frequency contrast (see Fig. 2.2) [64,65]. The contrast loss can be quan-
tified in terms of the linear obscuration ratio Robs ,

Dobs
Robs = , (2.3)
DEP

where Dobs is the diameter of a circular central obscuration and DEP is the entrance
pupil diameter. The change in MTF with Robs is depicted in Fig. 2.2, from which
there are two important takeaways. First, the cutoff frequency fcutof f defined in
Eq. (2.1) is unaffected by the obscuration ratio. Second, large obscuration ratios
can still achieve high contrast imaging at high spatial frequencies given fcutof f is
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 17

Figure 2.2: Modulation transfer function (MTF) for diffraction-limited imaging systems with
different obscuration ratios Robs . The spatial frequency cutoff fcutof f = 1000 lp/mm set
according to Eq. (2.1) for λ = 500 nm and f /1.5.

sufficiently large. The explanation for this is a narrowing of the central peak of
the diffraction-limited point spread function for an annular aperture [66]. As an
example, even with an 80% obscuration, an f /1.5 system can resolve > 20% con-
trast at 200 lp/mm, which is sufficiently high for many imaging applications. This
demonstrates how high-resolution imaging can still be achieved with an AFL, even
in the presence of a large obscuration, due to its fast f-number. As a result, the cen-
tral obscuration in AFLs does not significantly hinder its capability to serve as a
high-performing telephoto objective.
From a more practical standpoint, a second downside is that AFLs require far
more complex fabrication, alignment, metrology, and optomechanics than conven-
tional telephoto designs. The AFL’s annular regions are often designed with as-
pheric surfaces, which means sub-aperture figuring such as single-point diamond
turning (SPDT) is required. Moreover, the standard AFL consists of two reflective
optics facing each other, separated in air. As a result, struts are required to hold the
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 18

object-side reflective optic, which yields additional diffraction effects in the point-
spread function. The added mechanical complication of holding one optic is very
similar to the “spiders” resultant from positioning the secondary mirror in con-
ventional reflective telescopes. Lastly, baffling may be required to suppress stray
light entering from angles outside the field-of-view [48].
One way of easing the requirements on AFL alignment and optomechanics is
by making the design monolithic. A monolithic AFL incorporates a bulk optical ma-
terial in the space between the two reflective optics. As a result, the monolithic AFL
is a catadioptric system where two refractive surfaces are now introduced, one pre-
ceding all reflections and the other following them. The main benefit of monolithic
AFLs is that the alignment of the reflective surfaces is established when fabricat-
ing the singular bulk optic. As a result, the alignment of the reflective surfaces
is also impervious to any vibration- or shock-induced misalignment. Monolithic
AFLs also avoid the need for optomechanical struts to hold the object-side reflec-
tive surfaces. The two added refractive surfaces offer valuable DOFs to the design
process as well. Due to these advantages, the majority of AFLs demonstrated in
the literature have monolithic designs.
As with all refractive optical elements, the disadvantage of monolithic AFL de-
signs is the introduction of chromatic aberrations at its two refractive surfaces due
to the bulk material’s dispersion. Chromatic aberrations have been shown to se-
riously degrade imaging performance in monolithic AFLs [47], and with a single
homogeneous material, the DOFs are not fully available to correct these aberra-
tions. One approach that has been taken to correct chromatic aberrations is to split
the AFL into two elements consisting of different homogeneous materials, simi-
lar to a cemented doublet [67]. Other approaches have relied on DOEs to provide
chromatic control [68,69]. Both DOE designs use four reflections where one design
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 19

incorporates the DOE at the third reflective surface [68] while the other design uses
a multi-layer DOE following the exiting refractive surface [69].

2.2 Design scope, specifications, and process

The primary goal of this work is to achieve chromatic correction in a monolithic


AFL using the unique dispersive properties of GRIN materials [27,70]. It is known
from prior work that GRIN can function analogously to a cemented achromatic
doublet [26]. Thus, GRIN offers an alternate route for chromatic control in AFLs
beyond the previously pursued homogeneous doublet [67] and DOE designs [68,
69].
A secondary goal is to obtain higher specification AFL designs while maintain-
ing adequate imaging performance, even if only monochromatically. The DOFs
offered by F-GRIN’s spatial refractive index variation allow for further monochro-
matic aberration correction than what is available in homogeneous AFLs. Conse-
quently, more ambitious AFL designs can be performed with the hope of main-
taining chromatic correction as well.
The F-GRIN AFL design study is intended to be exploratory, so no hard speci-
fications are established from the start. Rather, what is sought are the boundaries
of the design space that allow the two outlined goals to be met. As guidance, a set
of exploratory specifications are devised, as outlined in Table 2.1. MTF data for a
high-performing conventional telephoto lens are also listed in Table 2.2 for com-
parison. Note, these specifications serve only as targets. Later, different designs
are analyzed that surpass as well as fail these targets. For example, in Sec. 2.3 a
search of the solution space considers designs that are slower than f /2 while the
final designs in Sec. 2.6 are faster than f /2. Similarly, the final designs in Sec. 2.6
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 20

Parameter Value
Focal length [mm] 100.0
F-number f /2 or faster
Full field-of-view [deg] ≥5
Spectrum visible (d, F, C)
Telephoto ratio, k ≤ 0.6
MTF @ 200 lp/mm, on-axis ≥ 0.3
MTF @ 150 lp/mm, all fields ≥ 0.3
Table 2.1: F-GRIN AFL exploratory design specifications. MTF specifications are based
on data listed in Table 2.2 for the Sigma 135 mm f /1.8 DG HSM, a high-performing con-
ventional telephoto lens.

Half MTF
field-of-view 48 lp/mm 96 lp/mm 144 lp/mm 192 lp/mm 240 lp/mm
[deg] T S T S T S T S T S
0.0 0.79 0.79 0.58 0.58 0.43 0.43 0.34 0.34 0.28 0.28
3.5 0.76 0.78 0.57 0.54 0.42 0.39 0.31 0.28 0.22 0.20
5.0 0.72 0.76 0.52 0.46 0.36 0.30 0.25 0.21 0.16 0.13
Table 2.2: As-built MTF data for the Sigma 135 mm f /1.8 DG HSM measured across the
field-of-view [71]. MTF is recorded at different spatial frequencies and for tangential (T)
and sagittal (S) pupil sections.

have a 10◦ full field-of-view, which is twice the target value listed in Table 2.1.
As depicted in Fig. 2.1, the number of reflections in the design strongly influ-
ences the achievable system specifications. A greater number of reflections can
reach smaller telephoto ratios but limits the field-of-view and increases the cen-
tral obscuration. Consequently, the F-GRIN AFL design scope is limited to two-
reflection configurations, which promote more ambitious system specifications yet
also present fewer DOFs. As monolithic designs, two-reflections AFLs have four
optical surfaces total, two refractive and two reflective. The decision to only exam-
ine two-reflection designs is partially based on a feasibility design study where it
is found that homogeneous four-reflections designs can readily achieve diffraction-
limited monochromatic performance using only aspheric surfaces at the specifi-
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 21

cations listed in Table 2.1. Restricting the number of reflections also narrows the
breadth of the global solution space that will be explored.
The design scope is also limited in two other ways. First, the center thickness
separating surface vertices is held constant for all subsequent ray passes. Doing
so does not guarantee a continuous surface sag across annular regions, although it
does serve to limit potential sag discontinuities between adjacent regions. A sec-
ond constraint is that the central obscuration aperture is restricted to be coincident
with the aperture stop. As a result, the rays in the pupil that are obscured remains
constant across the field.
The design process is planned in three steps. First, the global solution space for
homogeneous AFLs is surveyed. Upon doing so, different starting points of inter-
est can be identified. Then, a monochromatic design is pursued from the identified
starting points with the incorporation of F-GRIN for further performance improve-
ment. Finally, the visible spectrum is introduced, imparting chromatic aberrations.
The dispersive properties of the F-GRIN are then leveraged for attempting a final
color-corrected design.

2.3 Global search of the homogeneous AFL solution space

The goal of the design study is to search the AFL solution space for what designs
lie at its periphery, achieving high specification while still allowing for monochro-
matic and polychromatic correction via F-GRIN. A logical first step in reaching this
goal is to perform a global search of the monolithic AFL solution space for a homo-
geneous material. Then, the results and designs can be used to inform on the later
design steps where F-GRIN is incorporated.
One methodical way of exploring a global solution space is using a grid search.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 22

A grid search explores a multi-dimensional space by evaluating all possible combi-


nations for a set of predefined values in each dimension. In lens design, a global so-
lution space can be explored using a grid search by predefining values for different
design parameters and then assessing the resultant designs from all combinations.
As such, a grid search is a brute-force approach to surveying a higher-dimensional
space, offering a thorough and stable yet computationally inefficient option.
For the homogeneous AFL solution space, a nine-dimensional grid search is
performed across nine variable design parameters (see Appendix B for CODE V
MacroPLUS code). The values assessed for each design parameter are listed in Ta-
ble 2.3. The choice of parameter values within the grid search is in consultation
with the exploratory design specifications in Table 2.1. In total, 23,040 different de-
sign combinations are evaluated in the search. Upon initializing each combination,
the AFL design is then locally optimized for surface curvatures (currently limited
to spherical surfaces) and center thickness while constraining the nine imposed
parameter values from Table 2.1. Finally, data are evaluated and saved for each
design, which provides insight into the solution space and which specific designs
may be advantageous starting points for incorporating F-GRIN.

Design parameter Values


F-number f /1.5, f /2.0, f /4.0
Full field-of-view [deg] 5, 10, 15
Obscuration ratio, Robs 0.5, 0.6, 0.7, 0.8
Telephoto ratio, k 0.5, 0.6, 0.7, 0.8
Spectrum monochromatic, polychromatic
Aperture stop surface 1, 2, 3, 4
Refractive surface type planar, spherical
Surface clearance constraint on, off
Starting point design A, B, C, D, E
Table 2.3: Different design parameters and values considered in the grid search of the
homogeneous AFL solution space.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 23

Some of the design parameters explored in the grid search must be explained,
namely the bottom four rows of Table 2.3. First, the aperture stop is evaluated on
all four optical surfaces, two refractive and two reflective. This is done to capture
any stop-shift effects. In conventional telephoto lenses, the stop is located toward
the object-space side of the system to reduce ray angles at the image and limit the
front group diameter [63]. Second, the two refractive surfaces are evaluated as ei-
ther planar or spherical in shape. For spherical refractive surfaces, the curvatures
are subsequently obtained by local optimization. Third, a constraint maintaining
clearance between adjacent annular surface regions is either turned on or off. A
final AFL design requires clearance, although at this initial design stage, it shrinks
the available solution space. A clearance constraint is imposed using the “JMRCC”
function within CODE V, modified for proper use inside an immersed refractive in-
dex [72,73]. This constraint function calculates the perpendicular distance between
a defined ray segment and a point of ray-surface intersection. Lastly, the starting
point designs, A through E, are five unique homogeneous AFLs that are loaded ini-
tially in the grid search. The eight other design parameters in the search are then
imposed on the starting point design before being locally optimized. The reason
for the multiple starting point designs is to seek diversity in the explored solution
space. All starting point designs as well as all evaluated homogeneous AFLs use
PMMA (nd = 1.4918, ν = 57.47, PF,d = 0.7018) as the bulk optical material.
The end product of the grid search is 23,040 unique homogeneous AFL de-
signs along with their evaluated data. With this data, analysis can be performed
to identify starting point designs of interest for incorporating F-GRIN. First, de-
signs with unrecoverable performance, such as those with ray trace failures, are
filtered out. 4,246 solutions are filtered out, meaning the grid search has an over-
all success rate of 81.6%. From the remaining 18,794 solutions, the distribution of
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 24

Figure 2.3: Grid search results for the homogeneous AFL solution space. The distribu-
tion of locally optimized designs is depicted for all nine parameters listed in Table 2.3.
Vertical axes show the field-averaged geometrical RMS spot size on a log scale. Each
scatter point (colored) represents an individual design, the boxes denote the 25th through
75th percentile solutions, the bold horizontal lines show the median performance, and the
whiskers show the 2nd through 98th percentile. Solutions with unrecoverable performance
(e.g., ray trace failure) are filtered out of this analysis.

homogeneous AFL designs is shown in Fig. 2.3 with performance as a function of


the nine parameters listed in Table 2.3. Imaging performance is quantified by the
geometrical (i.e., ray-based) RMS spot size. From this global solution data, several
trends can be noticed. First, as expected, performance improves with slower f-
numbers, smaller fields-of-view, larger obscuration ratios, and larger telephoto ra-
tios. The performance distribution is only slightly improved for a monochromatic
spectrum, showing that polychromatic aberrations are not limiting the homoge-
neous AFL designs with spherical surfaces. Designs also perform progressively
worse as the stop surface moves away from object space. Moreover, designs with
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 25

spherical refractive surfaces perform better than those with planar surfaces due to
their curvatures offering additional DOFs. On the other hand, designs constrained
for surface clearance perform worse. Finally, the design performance is largely in-
dependent of starting point design A through E, as intended so that the search is
not biased toward any one starting point.
Using the results of the grid search, homogeneous AFL designs can now be
identified for use as starting points when incorporating F-GRIN. Specifically, de-
signs of interest include those with high specifications yet decent enough perfor-
mance that may later be fully recovered with optimized F-GRIN. From the various
niches fitting this description, five starting point AFL designs are selected, each
with a unique set of specifications. All five designs are shown in Fig. 2.4, and
the specifications for each design are outlined in Table 2.4. (Note, throughout this
chapter, designs are ray traced with three field angles, plus-or-minus, although
denser field sampling is used in the design process.) Each design has its own
unique qualifications. Design 1 has the fastest aperture and largest field as well as
the aperture stop located on the third surface. Design 2 also has the largest field
but a slower aperture than Design 1. Design 3 also has the fastest aperture but has
a smaller field than Design 1. Design 4 is slower and has the smallest field but
has almost half the telephoto ratio of other designs. Design 5 is also slow and has
the smallest field but uses only planar refractive surfaces. Among the five designs,
Design 3 is the primary focus of this work.1
At the current stage, all homogeneous designs use only spherical surfaces.
The logical next step is to incorporate additional DOFs for further performance
1
Throughout the design process, all five starting points are pursued concurrently. Each design
is performed by a different individual in a “design group” setting where design techniques are
collaboratively devised. I focused on Design 3, and as a result, the findings presented here stem
from that starting point. For other contributing authors, see the Contributors and Funding Sources
section as well as references [27, 70].
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 26

Design 1 Design 2 Design 3 Design 4 Design 5

30 mm

Figure 2.4: Five homogeneous AFL design starting points for later incorporation of F-
GRIN. Each design exhibits a different set of system specifications to explore a diverse
solution space (see Table 2.4). Limited to spherical surfaces, these starting points do not
perform adequately, as can be seen by the visibly aberrated image spots.

Design parameter Design 1 Design 2 Design 3 Design 4 Design 5


F-number f /1.5 f /2.0 f /1.5 f /2.0 f /2.0
Full field-of-view [deg] 15 15 10 5 5
Obscuration ratio, Robs 0.7 0.7 0.7 0.5 0.7
Telephoto ratio, k 0.49 0.59 0.60 0.35 0.65
Spectrum mono. mono. mono. mono. mono.
Aperture stop surface 3 1 1 2 1
Refractive surface type sph. sph. sph. sph. planar
Surface clearance constraint on on on on on
Table 2.4: System specifications for five homogeneous AFL design starting points (see
Fig. 2.4). The focal length for all designs is fixed at f 0 = −100 mm, and all designs are
made of PMMA.

improvement. By the folded and monolithic nature of AFLs, surface generation


would likely be achieved in fabrication using SPDT. Consequently, surface defini-
tions with greater DOFs can be leveraged without a significant reduction in man-
ufacturability. In order to maintain system rotational symmetry, each surface in
all five starting point designs in Fig. 2.4 is converted to an Q-con asphere [2] de-
fined on top of a base conic surface (the only exceptions are the two intentionally
planar refractive surfaces of Design 5). Six Q-con terms are used per surface, and
the aspheric coefficients are then obtained by optimization in concert with the base
surface curvatures, AFL center thickness, and image distance.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 27

15 mm

Figure 2.5: Homogeneous, monochromatic AFL design based on starting point Design 3
(f /1.5, 10◦ full field-of-view). All four surfaces are aspheres on base conics. The homoge-
neous material in use is PMMA.

One optimized homogeneous AFL design is shown in Fig. 2.5. The design
originates from starting point Design 3. Incorporation of the four aspheric sur-
faces, two reflective and two refractive, has allowed for near diffraction-limited
monochromatic performance across the field-of-view, as shown by the MTF plot in
Fig. 2.5. For this design and all subsequent designs, the MTF plot is evaluated out
to 200 lp/mm, even though the cutoff frequency is much higher (> 1000 lp/mm).
The low diffraction-limited MTF at 200 lp/mm in each MTF plot is due to the cen-
tral obscuration’s reduction in mid-spatial frequency contrast.
Previously it is claimed that monolithic AFL designs suffer severely from chro-
matic aberrations when used across a polychromatic spectrum. To evaluate this
assumption, the homogeneous AFL design in Fig. 2.5 is redesigned for the visible
spectrum (d, F, C Fraunhofer line wavelengths) while maintaining the same system
specifications. The resultant polychromatic design can be seen in Fig. 2.6. By com-
paring the MTF plots in Figs 2.5 and 2.6, the addition of the visible spectrum results
in approximately a 4x reduction in image resolution. Notably, the front refractive
surface changes dramatically in form between the monochromatic and polychro-
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 28

15 mm

Figure 2.6: Homogeneous, polychromatic AFL design based on starting point Design 3
(f /1.5, 10◦ full field-of-view). The design is a reoptimization of the monochromatic design
in Fig. 2.5 for the visible spectrum. All four surfaces are aspheres on base conics. The
homogeneous material in use is PMMA.

Figure 2.7: Transverse ray aberration plots for the homogeneous, polychromatic AFL de-
sign in Fig. 2.6. Rays blocked by the central obscuration are omitted. Chromatic aberra-
tions limit performance and cannot be corrected using a single homogeneous material.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 29

matic designs, indicating this change is needed to attain a better balance of both
monochromatic and polychromatic aberrations. In the polychromatic design, the
limiting aberration is spherochromatism introduced by the aspheric refractive sur-
faces, followed by secondary chromatic aberrations based on the relative partial
dispersion of PMMA. The transverse ray aberrations for this design are shown
in Fig. 2.7 where rays blocked by the central obscuration are omitted. The poly-
chromatic performance is lacking such that further improvement is needed. As
described in Sec. 2.6, the further chromatic correction can be accomplished using
F-GRIN rather than homogeneous media.
While chromatic aberrations cannot be fully eliminated using a single homoge-
neous material, their effect can largely be controlled with the proper configuration
of refractive surfaces. By reducing the system specifications (e.g., slower aperture
and smaller field), the influence of monochromatic aberrations can be reduced. In
doing so, the refractive surface degrees of freedom can be devoted to correcting
chromatic aberrations. For example, if designing the refractive surfaces to con-
trol exclusively chromatic aberrations, axial color can be fully corrected by making
the front surface planar and the rear surface concentric with the image. Then, the
marginal ray goes unrefracted at both refractive surfaces, introducing no on-axis
chromatic aberration. Meanwhile, the reflective surfaces can be used to mitigate
remaining monochromatic aberrations. However, lateral color is still present in off-
axis fields, although it is limited by the reduced field-of-view. An example of such
a design is shown in Fig. 2.8 based on starting point Design 5 and its less ambitious
system specifications with half the field-of-view compared with Design 3. For the
design in Fig. 2.8, the front refractive surface is near-planar while the rear surface is
near-concentric with the image, yielding diffraction-limited polychromatic perfor-
mance on-axis. The full field performance is not diffraction-limited but improved
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 30

15 mm

Figure 2.8: Homogeneous, polychromatic AFL design based on starting point Design 5
(f /2, 5◦ full field-of-view). Chromatic aberrations are better controlled in this design with
a front refractive surface that is near-planar and a rear refractive surface that is near-
concentric with the image. All four surfaces are aspheres on base conics. The homoge-
neous material in use is PMMA.

compare with the polychromatic design in Fig. 2.6. Having said that, the off-axis
field performance is still not close to meeting the MTF specification at 150 lp/mm
outlined in Table 2.1.
Even with the more modest system specifications of Design 5, adequate poly-
chromatic performance is not achievable in homogeneous AFLs. Rather, the unique
chromatic properties of GRIN may be leveraged for further chromatic control, as
is done in Sec. 2.6. First, the mathematical representation of the GRIN must be
explained.

2.4 GRIN representation

2.4.1 Linear, two-material composition model

In GRIN materials, the refractive index varies both spatially and spectrally, n (x, y, z, λ).
Both features must be accurately accounted for when modeling GRIN for use in
optical design. The spectral description is of particular interest since the design
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 31

goal is to correct chromatic aberrations in monolithic AFLs.


For GRIN media that are fabricated from two or more base homogeneous ma-
terials, the GRIN dispersion must be modeled in accordance with the dispersion of
the constituent base materials. For example, the simplest case of forming a GRIN
from two base materials requires the GRIN refractive index to be some intermedi-
ate of the two base materials’ indices at all wavelengths. This relationship of how
GRIN dispersion is determined from different proportions of homogeneous mate-
rials is a question of light-matter interaction on the molecular scale. One way of
empirically determining this physical behavior is by measuring the refractive in-
dex for media with different proportions of constituent base materials. Although,
the complexity of analytically describing the physical GRIN dispersion imposes
an obstacle for optical design. Instead, a simplified model can be adopted to ap-
proximately describe the GRIN dispersion with, ideally, adequate accuracy. By
these dispersion models, the chromatic GRIN properties can then be leveraged by
proper adjustment of base material dispersion.
In this work, the GRIN dispersion is approximated by a linear model, also
known as Arago-Biot theory [25, 74–77]. The primary design focus is GRIN AFLs
consisting of two base homogeneous materials, n1 (λ) and n2 (λ). For this case,
a linear two-material composition of the GRIN dispersion n (λ) is found by sim-
ply the linear combination of the base materials according to their relative volume
fractions, C1 and C2 ,
n (λ) = C1 n1 (λ) + C2 n2 (λ) , (2.4)

where C1 ∈ [0, 1], C2 ∈ [0, 1], and C1 + C2 = 1. Example GRIN dispersions for
different combinations of C1 and C2 can be seen by this model in Fig. 2.9.
The advantage of the linear model is its mathematical simplicity, which adapts
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 32

C1 =
0.0, C
C1 = 2 = 1.0
0.1, C
C1 = 2 = 0.9
0.2, C
2 = 0.8
C1 =
0.3, C
2 = 0.7
C1 = 0
.4, C =
C1 = 0 2 0.6
.5, C =
C1 = 0 2 0.5
.6, C2 =
C1 = 0 0.4
.7, C2 =
0.3
C1 = 0.
8, C2 = 0.
2
C1 = 0.9
, C2 = 0.1
C1 = 1.0
, C2 = 0.0

Figure 2.9: Linear two-material composition model. GRIN dispersion is shown for differ-
ent volume fractions, C1 and C2 , of base homogeneous materials of differing dispersions,
n1 (λ) and n2 (λ).

well for use in optical design. Although, it is important that the linear model ac-
curately describes the physical GRIN dispersion upon fabrication. For many ma-
terials, the linear approximation agrees closely with more accurate models such as
the quadratic or Lorentz-Lorenz models [25, 75]. The linear model errs the greatest
at the extremes of the volume fraction range where either C1 or C2 becomes small
and the other becomes large [74]. The model accuracy may also differ depending
on whether the base materials are glasses or polymers, the latter of which is the
focus of this work. Ultimately, prior to design fabrication, it is important to exper-
imentally verify the accuracy of the linear model for the base materials in use. For
this work, it is assumed that the linear model provides adequate accuracy.
Since the volume fractions must sum to unity, C1 +C2 = 1, substituting C1 = 1−
C2 into Eq. (2.4) yields a different format for the linear two-material composition
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 33

model,

n (λ) = n1 (λ) + [n2 (λ) − n1 (λ)] C2


(2.5)
= n1 (λ) + ∆n (λ) C2 ,

where ∆n (λ) = n2 (λ) − n1 (λ) is the total refractive index range between the two
base materials. By reformulating the dispersion model as in Eq. (2.5), only a single
volume fraction C2 is needed to describe the gradient refractive index.

2.4.2 Quantifying GRIN dispersion

The GRIN dispersion from the linear composition model can be quantified in mul-
tiple ways. First, the dispersion of the individual homogeneous base materials can
be expressed by the classical Abbe number ν and relative partial dispersion PF,d ,

nd − 1 nF − nd
ν= , PF,d = , (2.6)
nF − nC nF − nC

where ni = n (λi ) and d, F , C are Fraunhofer line wavelengths. V- and P-numbers


can also be defined for the GRIN medium as a whole [25],

∆nd ∆nF − ∆nd


νGRIN = , PF,d,GRIN = , (2.7)
∆nF − ∆nC ∆nF − ∆nC

where ∆ni = n2 (λi ) − n1 (λi ). The GRIN V- and P-numbers were originally in-
troduced as analogs to the homogeneous V- and P-numbers in the context of cor-
recting primary and secondary chromatic aberrations. To this end, their derivation
relies on the assumption that the GRIN only possesses the radial quadratic power
term. The polychromatic GRIN AFL designs considered in Sec. 2.6 are not re-
stricted to this GRIN form, so the values for νGRIN and PF,d,GRIN no longer strictly
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 34

apply in their original context of chromatic aberration contributions. Nevertheless,


the values for νGRIN and PF,d,GRIN remain convenient quantities when characteriz-
ing the dispersion of two-material GRIN media.
A final way of quantifying the GRIN dispersion is, again, using the conven-
tional Abbe number and partial dispersion in Eq. (2.6), although now calculated
at different points throughout the GRIN volume rather than for the base materials.
Within a GRIN medium, ν and PF,d are spatially varying, yet remain within the
bounds of the base materials. Evaluation of spatially varying ν and PF,d provides
insight into a design’s chromatic correction as well as what GRIN dispersion is
accessible [77].

2.4.3 Spatial refractive index variation

The linear two-material composition model described in Eq. (2.5) governs the re-
fractive index as a function of wavelength n (λ) when composed from two base
materials. For a GRIN medium where the refractive index variation occurs volu-
metrically, the spatial dependence of n is notably absent in Eq. (2.5). Consequently,
this model can also be thought of as defining the refractive index at a single spa-
tial coordinate according to the base material volume fraction C2 . To impart the
spatial index variation found in GRIN media, the model can be expanded by im-
plementing a spatial variation in volume fraction, C2 (x, y, z), while making sure
C2 (x, y, z) ∈ [0, 1] for all points in space. By defining the GRIN representation
this way, the terms containing the material composition and dispersion, n1 (λ) and
∆n (λ), are separate from the term governing spatial variation, C2 (x, y, z). This
proves advantageous when performing the GRIN design since each term can be
varied independently without fear of violating the dispersion model or the form
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 35

of the spatial variation.


For AFLs whose surfaces maintain an axis of rotational symmetry along the
optical axis, the form of the GRIN spatial index change is selected to preserve a
system axis of symmetry. In doing so, the complexity of the design problem is
reduced since field symmetry is maintained for rigid rotation of the entire system
about the optical axis (i.e., only a one-dimensional field-of-view must be consid-
ered) [78]. One GRIN representation that preserves an axis of rotational symmetry
while providing ample DOFs is the “University of Rochester” notation consist-
ing of even orders of radial terms, both even and odd orders of axial terms, and
radial-axial cross terms [79]. In this work, no cross terms are considered and four
terms are included each for radial variation and axial variation. Embodied in the
volume fraction C2 (r, z), this spatial refractive index variation is mathematically
represented by
4
X 4
X
0 2i j
C2 (r, z) = Ci0 (r ) + C0j (z 0 ) , (2.8)
i=1 j=1

r z
r0 = , z0 = , (2.9)
rnorm znorm

where Ci0 and C0j are spectrally-independent scalar coefficients, r2 = x2 + y 2 is


the radial coordinate, rnorm is a normalization radius, and znorm is a normalization
thickness. By having two independent coordinates of refractive index change (r
and z), the profiles defined by Eq. (2.8) qualify as F-GRIN according to the defini-
tion laid out in [15].
All together, both the spatial and spectral behavior of the GRIN is governed by

n (r, z, λ) = n1 (λ) + ∆n (λ) C2 (r, z) . (2.10)

In design, the base materials can be optimized to achieve a certain GRIN disper-
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 36

sion using n1 (λ) and ∆n (λ). These dispersions can also be constrained to remain
within the bounds of available base materials. Likewise, the spatial variation en-
compassed by C2 (r, z) can also be obtained via optimization through its coeffi-
cients Ci0 and C0j . Both options are used in concert for the design of GRIN AFLs.
The normalization dimensions, rnorm and znorm , may also be obtained by optimiza-
tion. This may be necessary if the clear aperture and center thickness of the AFL de-
sign is changing considerably. Otherwise, optimization convergence is improved
by freezing these values at or slightly larger than the optic dimensions.
The source code for the CODE V user-defined GRIN is provided in Appendix
A.1 for the “University of Rochester” linear, two-material composition model.

2.4.4 GRIN optimization constraints

A central requirement in GRIN design is the ability to constrain the refractive index
in optimization within bounds that are achievable in fabrication. For the canonical
types of GRIN, these constraints include the total refractive index change as well
as the spatial index geometry. For F-GRIN, including the profiles described by
Eq. (2.10), new challenges arise due to the increased mathematical complexity and
added DOFs. The primary obstacle derives from the non-monotonic nature of the
spatial distribution where extrema may be located internal to the volume. For the
distribution in Eq. (2.10), the constraints on spatial variation fall on C2 (r, z) where
it is required that for all points in space

min {C2 (r, z)} ≥ 0,


(2.11)
max {C2 (r, z)} ≤ 1.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 37

This requirement is needed for producing valid combinations of the base materials
(e.g, at a single coordinate, C2 > 1 would require a nonphysical negative volume
fraction C1 < 0 in order to maintain C1 +C2 = 1). In optimization, it proves difficult
to constrain the minimum and maximum values for C2 (r, z) throughout a volume
due to the non-monotonic nature of the summations in Eq. (2.8). Rather, an alter-
nate design approach that is found to yield improved results is to leave the volume
fraction unconstrained in optimization, allowing C2 (r, z) ∈ (−∞, ∞). By doing so,
the base materials described by n1 (λ) and ∆n (λ) no longer accurately embody the
actual base materials needed to fabricated the GRIN. However, the actual base ma-
terials can subsequently be back-calculated from n1 (λ) and ∆n (λ) via any instan-
taneous pair of min {C2 (r, z)} and max {C2 (r, z)}. These actual base materials are
what are constrained within optimization, controlling their refractive index, Abbe
number, and relative partial dispersion to remain within manufacturable bounds.
For computational efficiency, it is also important to have a robust means of find-
ing the extrema of C2 (r, z) throughout a volume. For optical design in commercial
software CODE V, it is also necessary for the function to be compatible within the
available optimization framework [72]. One brute-force approach calculates the
refractive index along a three-dimensional sampling grid to identify the minimum
and maximum values for C2 (r, z). Adequate accuracy can be achieved with this
method if using a dense enough grid but at significant computational expense.
Instead, a different approach is used to more efficiently find the extrema of
C2 . Due to the absence of r, z cross terms in Eq. (2.8), the extrema of C2 can be
identified by solving for the roots of the derivatives of C2 with respect to r and
z [80]. Differentiating Eq. (2.8) and equating to zero provides the polynomials of
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 38

interest,

4
∂C2 (r, z) X 2i 2i−1
= Ci0 (r0 ) = 0,
∂r r
i=1 norm
4
(2.12)
∂C2 (r, z) X j j−1
= C0j (z 0 ) = 0.
∂z z
j=1 norm

Since the radial derivative is a 7th order polynomial and the axial derivative is a
3rd order polynomial, the roots must be solved numerically within the optimiza-
tion routine. A numerical root finding technique is used that combines concepts
from the bisection method and Newton’s method while leveraging the fact that the
number of possible roots is capped by the polynomial order. The result is a series
of coordinates at which to evaluate C2 where extrema occur,


C2 (rmin , 0) , C2 (rmax , 0) , C2 rroot(s) , 0 ,
 (2.13)
C2 (0, zmin ) , C2 (0, zmax ) , C2 0, zroot(s) .

(In this instance, rmin = 0 along the optical axis, rmax equals the clear semi-aperture,
and any roots rroot(s) , zroot(s) are not guaranteed to exist depending on the design).
Lastly, from these handful of evaluated coordinates the minimum and maximum
values of C2 can be identified throughout the entire volume far more efficiently
than the brute-force approach where thousands of coordinates must be evaluated.

2.5 Monochromatic GRIN AFL designs

GRIN AFL designs can now be performed using the optimized homogeneous
starting points identified in Sec. 2.3 combined with the F-GRIN representation out-
lined in Sec. 2.4. While the ultimate goal is to obtain a color-corrected design, the
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 39

first step in the design process is to perform monochromatic AFL designs. Then,
the GRIN dispersion can subsequently be incorporated to address present chro-
matic aberrations, as described next in Sec. 2.6.
The homogeneous, monochromatic AFL design shown in Fig. 2.5 uses opti-
mized aspheric surfaces to correct monochromatic aberrations. The accompanying
MTF plot in Fig. 2.5 shows that residual aberrations in the optimized design are
significant enough to drop performance below the specification outlined in Table
2.1, both on-axis as well as across the field-of-view. In fact, at f /1.5 with a 0.7 ob-
scuration ratio, nearly diffraction-limited performance is required to meet the MTF
target.
Further monochromatic aberration correction can be performed using the newly
incorporated F-GRIN DOFs. Several examples in the literature highlight GRIN’s
monochromatic aberration control. An axial gradient located at a spherical surface
acts equivalently to an aspheric surface made from a homogeneous medium [79].
Field-dependent aberrations such as Petzval field curvature, astigmatism, and dis-
tortion have also be corrected using radial and spherical gradients [81, 82].
Prior to optimizing a full-fledged GRIN AFL design, analysis is performed to
identify the effect of different F-GRIN spatial terms in Eq. (2.8). First, the identi-
fied starting point AFLs possessing spherical surfaces (see Fig. 2.4) are examined
in order to isolate the effects of the GRIN and neglect any effects of aspheric sur-
faces. This process is done in three steps using exclusively radial terms (C0j = 0),
exclusively axial terms (Ci0 = 0), and lastly a combination of both radial and axial
terms. For each case, the design is initialized with a different GRIN profile. In this
procedure, it is found that the initial GRIN profile strongly influences the resul-
tant optimized design. For example, starting from homogeneous versus starting
from a certain initial amount of radial or axial GRIN yields significantly different
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 40

outcomes upon optimization. This is an important finding where much of the de-
sign time is devoted not to optimizing but to manually exploring different solution
spaces.
Three primary findings are discovered when evaluating the effect GRIN terms
have on monochromatic AFLs. First, the GRIN power via its radial quadratic term
optimally takes the opposite sign of the net surface powers. The explanation for
this is GRIN power is being used for balancing field curvature by contributing
an opposite contribution than the surfaces [81]. The amount of field curvature
correction needed for adequate performance also depends on the field-of-view.
Since the different AFL designs in Fig. 2.4 vary in full field-of-view from 5 to 15
degrees, the optimal radial GRIN profile differs between designs based on field
size.
Second, the optimized GRIN axial variation often presents a single dominant
polynomial term. Which axial term is dominant depends on the convexity or con-
cavity of the accompanying refractive surfaces. Whether an even or odd order
polynomial term is needed depends on whether the index change approaching
the surface must be increasing or decreasing for spherical aberration compensa-
tion. Specifically, the axial index change must impart a GRIN surface contribution
to spherical aberration that balances that of the homogeneous surface contribu-
tion [79]. Since the intermediate reflective aspheric surfaces also possess spherical
contributions, the axial term is best obtained via optimization.
Third, it is found for all designs that performance improves with increasing
AFL center thickness. The reason for this is a relaxation of the previously described
surface clearance constraint as well as an increase in the optical path difference that
the GRIN may impart. Unfortunately, there is typically a limit on achievable GRIN
center thickness due to both fabrication constraints as well as system volume and
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 41

1.5549

1.5369

Refractive index
1.5188

1.5007

15 mm
1.4827

Figure 2.10: Monochromatic GRIN AFL design based on starting point Design 3 (f /1.5,
10◦ full field-of-view). All four surfaces are aspheres on base conics.

weight targets.
With these various findings in mind, final monochromatic GRIN AFL designs
are sought. The GRIN and aspheric surfaces are optimized in concert, striving for
improved monochromatic aberration correction compared with the homogeneous
designs in Sec. 2.3. The GRIN is optimized for both its spatial variation as well
as its base material refractive indices. The base materials are constrained as de-
scribed in Sec. 2.4.4, although for these monochromatic GRIN, dispersion is yet to
be introduced.
Again considering the results stemming from starting point Design 3, a final
monochromatic GRIN AFL design is shown in Fig. 2.10 along with its MTF per-
formance. Comparing the homogeneous design MTF in Fig. 2.5 with the GRIN
design MTF in Fig. 2.10, the addition of the F-GRIN is clearly responsible for
providing further monochromatic performance improvement. The GRIN design
is diffraction-limited across the field-of-view and successfully meets the perfor-
mance target laid out in Table 2.1. The additional DOFs of F-GRIN allow for further
monochromatic aberration correction for the three reasons outlined above, helping
with field curvature, spherical aberration, and relaxing the surface clearance.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 42

2.6 Color-corrected GRIN AFL designs

2.6.1 Two-material designs

Given the successful monochromatic GRIN AFL designs from Sec. 2.5, the next
step is to attempt color correction in the visible spectrum. The standard d, F , C
Fraunhofer line wavelengths are considered as adequately spanning the visible.
GRIN has been shown to possess unique chromatic properties that are bene-
ficial for correcting chromatic aberrations. Most notably, GRIN dispersion can be
used for achromatization by a single optical material, unlike conventional homo-
geneous media which require a doublet for color correction [26, 77]. Axial GRIN
also offers control over spherochromatism introduced at a surface [83].
The simplest option for attempting color correction is to introduce the visible
spectrum to the monochromatic design in Fig. 2.10. Then, the GRIN dispersion
can be optimized via the two homogeneous base materials’ dispersion described
by n1 (λ) and ∆n (λ) = n2 (λ) − n1 (λ). In this optimization process, the base ma-
terials’ refractive index nd , Abbe number ν, and relative partial dispersion PF,d are
constrained using the process described in Sec. 2.4.4. Unfortunately, the design
resulting from simple GRIN dispersion optimization does not achieve adequate
color correction. The explanation for this is the GRIN spatial distribution also in-
fluences the chromatic aberration contributions, not simply the material disper-
sion. The same is also true for conventional homogeneous optics where chromatic
aberrations depend on both the material properties as well as the optical surfaces.
Instead, the spatial index variation for the design in Fig. 2.10 must be modi-
fied to promote simultaneous monochromatic and polychromatic aberration cor-
rection. An updated color-corrected design is obtained, again, by optimization
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 43

of the base materials, n1 (λ) and ∆n (λ), although now in concert with the spatial
variation governed by Ci0 and C0j . The homogeneous base refractive indices, Abbe
numbers, and partial dispersions continue to be constrained using the process de-
scribed in Sec. 2.4.4. These quantities are constrained based on 64 polymer base
materials available from Nanovox LLC. With no immediate plans for fabrication,
designs are not fitted to specific materials but rather are kept within the region of
available materials,
1.4 ≤ nd ≤ 1.8
15 ≤ ν ≤ 70 (2.14)

0.63 ≤ PF,d ≤ 0.75.

With the same spatial DOFs available as for the monochromatic design, im-
proved color correction entails a loss in monochromatic aberration correction. Find-
ing the proper balance of these two regimes proves to be challenging. The pri-
mary reason for this is the refractive index variation and aspheric surface forms
best suited for monochromatic correction differs from those that provide the best
polychromatic correction (e.g., recall the specific color-correcting surface configu-
ration for the design in Fig. 2.8). As a result, starting from the highly-corrected
monochromatic design in Fig. 2.10 poses a challenging optimization problem that
leads to stagnation when attempting to add chromatic correction. One approach
that is found to work well is to start with an early monochromatic GRIN AFL de-
sign that is not high-performing and, therefore, not deeply entrenched in a local
minima. Then, the visible spectrum is introduced gradually by slowly increasing
the optimization weight on the two extreme F and C wavelengths. This careful
process also requires close examination of the optimizer’s derivative calculation,
the accuracy of which is essential for efficient convergence. One accurate although
computationally expensive option for derivative calculation is finite differences.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 44

A final polychromatic GRIN AFL design is shown in Fig. 2.11. The four as-
pheric surfaces are similar in form to the starting point Design 3 in Fig. 2.4, except
for the exiting refractive surface which has changed from concave to convex. The
system f-number, full field-of-view, obscuration ratio, and telephoto ratio remain
unchanged from those listed in Table 2.4. The MTF performance is also shown in
Fig. 2.11 where it can be seen that diffraction-limited performance is now achieved,
although only for the on-axis field. Performance decreases with increasing field
due to the presence of lateral color, as can be seen in the transverse ray aberration
plots in Fig. 2.12. The ray aberration plots also show that rays transmitted through
the system are well-corrected while rays blocked by the central obscuration result
in a geometrical spot that is approximately 100 times increased in size. This ef-
fect is to be expected since rays blocked by the obscuration are not included in the
optimization merit function.
The full field performance fails the MTF target listed in Table 2.1. The field per-
formance is also depicted in Fig. 2.13 in terms of RMS wavefront error. The full
field display shows how performance is rotationally symmetric with field since
the system and GRIN profile is rotationally symmetric about the optical axis. Nev-
ertheless, the design’s polychromatic performance is far improved over the pre-
viously described option of simply optimizing the GRIN dispersion in isolation.
Also, the overall system resolution remains very high given the f-number and tele-
photo ratio.
The GRIN profile for the design in Fig. 2.11 is also notably different from that
of the monochromatic GRIN design in Fig. 2.10 since, as described previously,
the profile must strike the proper balance between monochromatic and polychro-
matic aberration correction. The total refractive index change for this design is
∆nd = 0.1359, which remains within the allowed bounds for the base materials
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 45

1.5761

1.5421

Refractive index
1.5082

1.4742

15 mm
1.4402

Figure 2.11: Polychromatic GRIN AFL design based on starting point Design 3 (f /1.5, 10◦
full field-of-view). All four surfaces are aspheres on base conics.

(a) (b)

Figure 2.12: Transverse ray aberration plots across the field-of-view for the polychromatic
GRIN AFL design in Fig. 2.11. (a) Rays blocked by the central obscuration (Robs = 0.7) are
included (gray region). (b) Rays blocked by the central obscuration are omitted (only rays
transmitted by the system are shown), allowing for a 100x reduction in aberration scale.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 46

Y Field Angle in Object Space - degrees


4

-2

-4

-6

-6 -4 -2 0 2 4 6
X Field Angle in Object Space - degrees

RMS WAVEFRONT ERROR


vs
Polychromatic GRIN FIELD ANGLE IN OBJECT SPACE
AFL Design #3 Minimum = 0.024256
Maximum = 0.38578
Average = 0.19182
Std Dev = 0.10897
DHL 31-Jul-23 1 wave (546.1 nm)

Figure 2.13: RMS wavefront error across the field-of-view for the polychromatic two-
material GRIN AFL design in Fig. 2.11.

defined in Eq. (2.14). The base materials’ dispersion also satisfies the ν and PF,d
constraints (see Fig. 2.14). By these two homogeneous materials, the GRIN V- and
P-numbers can also be calculated. Table 2.5 lists the complete dispersion informa-
tion for this design. As described in Sec. 2.4.2, the conventional Abbe number and
partial dispersion can also be calculated spatially throughout the GRIN volume.
The spatial variation in ν and PF,d is shown in Fig. 2.15(a)-(b) for the polychro-
matic GRIN design.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 47

Figure 2.14: Homogeneous base materials used in the polychromatic GRIN AFL design
in Fig. 2.11. The design’s two base materials (red ‘X’s) are shown alongside available
materials from Nanovox (points). The design process constrains the base materials within
the region of available materials (white) but does not fit to specific materials. The red
curves show the achievable values from a linear composition of the two base materials.

Base material 1 Base material 2


Refractive index, nd 1.4402 1.5761
Abbe number, ν 68.06 47.95
Relative partial dispersion, PF,d 0.6466 0.6467

GRIN
Total refractive index range, ∆nd 0.1359
Abbe number, νGRIN 24.50
Relative partial dispersion, PGRIN,F,d 0.6468
Table 2.5: Specifications for the homogeneous base materials and GRIN for the polychro-
matic GRIN AFL design in Fig. 2.11.

The DOFs are not available to achieve diffraction-limited performance across


the field-of-view for the design shown in Fig. 2.11. Unlike the spatial DOFs which
can be increased by adding higher-order coefficients Ci0 and C0j , the GRIN dis-
persion is currently restricted by the linear two-material composition model in Eq.
(2.5). For a two-material GRIN, the spatial variation in nd , ν, and PF,d is confined
to a curve in this three-dimensional space [77]. For the design in Fig. 2.11, this nd ,
ν, PF,d curve is shown in Fig. 2.15(c).
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 48

68.06 0.6467

Rel. partial dispersion, PF,d


0.6467
63.03

Abbe number, ν
PF,d
58.01
0.6466
47.95
52.98
υ 1.4402
nd
47.95 0.6466 68.06 1.5761

(a) (b) (c)

Figure 2.15: GRIN dispersion for the polychromatic GRIN AFL design in Fig. 2.11. The
dispersion is quantified by the spatially varying (a) Abbe number ν and (b) relative partial
dispersion PF,d . (c) Spatial varying values for nd , ν, and PF,d are shown in three dimen-
sions. With two base materials, allowable values for nd , ν, and PF,d are confined to a curve
in three-dimensional space.

2.6.2 Multi-material designs

Additional chromatic freedom can be gained using an alternate GRIN composition


model that applies greater than two base homogeneous materials. As shown by
Desai et al., the incorporation of a third base material expands the achievable GRIN
dispersion to a convex surface in nd , ν, and PF,d space [77]. Four or more bases
expand the dispersion space further to a non-convex surface or volume in nd , ν,
and PF,d space. In both of these cases, the expanded dispersion space yields a
polygonal area [84] of achievable values in the Abbe diagram and ν, PF,d plots in
Fig. 2.14.
Accessing an unrestricted continuum of dispersion values, multi-material com-
positions significantly increase the GRIN chromatic DOFs. These properties can
be leveraged for higher-performing polychromatic GRIN AFLs at the cost of more
complex GRIN media fabrication. By this chromatic control, the spatial refrac-
tive index distribution can be prioritized for correcting monochromatic aberra-
tions while the dispersion can independently affect polychromatic aberrations. As
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 49

a result, the spatial index change can revert to the optimal form found for the
monochromatic GRIN AFL design in Fig. 2.10. Simultaneously, the GRIN disper-
sion according to a multi-material composition can be optimized to correct chro-
matic aberrations. (One effective way of incorporating additional base materials is
by allowing variation of Ci0 and C0j with wavelength, a capability that was previ-
ously restricted for all previous two-material designs.)
The resultant polychromatic GRIN AFL design applying a multi-material com-
position can be seen in Fig. 2.16. The corresponding design MTF is once again
diffraction-limited across the field-of-view, surpassing the MTF target in Table 2.1
and the Sigma 135 mm f /1.8 DG HSM performance in Table 2.2. The spatial refrac-
tive index change closely resembles that of the monochromatic GRIN AFL design
in Fig. 2.10, as expected. Now, the multi-material GRIN composition is being used
to independently correct chromatic aberrations, yielding the improved MTF per-
formance. Interestingly, the design refractive index range ∆n = 0.0779 is nearly
half that of the two-material polychromatic design in Fig. 2.11.
With more than two base materials, the GRIN V- and P-numbers from Eq. (2.7)
become no longer applicable. Instead, multi-material GRIN dispersion can again
be visualized by the spatial variation in ν and PF,d , as shown in Fig. 2.18(a)-(b). The
spatial variation in Abbe number and partial dispersion resembles a homogeneous
cemented doublet where the entering and exiting refractive surfaces possess differ-
ent material dispersions. Recall, Galan et al. demonstrated such a homogeneous
cemented doublet AFL design for improved color correction [67]. Interestingly,
the quasi-interface within the ν and PF,d distributions do not correspond to the
same surface curvature, meaning this design could not be readily substituted by
two homogeneous media. Furthermore, the GRIN dispersion represented in three-
dimensional n, ν, PF,d space in Fig. 2.18(c) is no longer confined to a curve as for
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 50

1.5553

1.5358

Refractive index
1.5163

1.4969

15 mm
1.4774

Figure 2.16: Polychromatic multi-material GRIN AFL design based on starting point De-
sign 3 (f /1.5, 10◦ full field-of-view). All four surfaces are aspheres on base conics.

6
Y Field Angle in Object Space - degrees

-2

-4

-6

-6 -4 -2 0 2 4 6
X Field Angle in Object Space - degrees

RMS WAVEFRONT ERROR


vs
Polychromatic GRIN FIELD ANGLE IN OBJECT SPACE
AFL Design #3 Minimum = 0.0096227
Maximum = 0.17858
Average = 0.094794
Std Dev = 0.036629
DHL 31-Jul-23 1 wave (546.1 nm)

Figure 2.17: RMS wavefront error across the field-of-view for the polychromatic multi-
material GRIN AFL design in Fig. 2.16. For comparison, the same scale is used as in Fig.
2.13 for the two-material design.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 51

78.86 0.6369

Rel. partial dispersion, PF,d


0.6369
63.03 0.6376

Abbe number, ν
PF,d
47.20 0.6357
0.6317
15.54
31.37 0.6337
υ 1.4774
nd
15.54 0.6317 78.86 1.5553

(a) (b) (c)

Figure 2.18: GRIN dispersion for the polychromatic GRIN AFL design in Fig. 2.16. The
dispersion is quantified by the spatially varying (a) Abbe number ν and (b) relative partial
dispersion PF,d . (c) Spatial varying values for nd , ν, and PF,d are shown in three dimen-
sions. With multiple base materials, allowable values for nd , ν, and PF,d can occupy a
volume in three-dimensional space.

a two-material composition. Instead, the GRIN dispersion occupies a non-convex


surface. As a result, at least four base materials are required to fabricate this GRIN.
The four base materials must form a volume in n, ν, PF,d space that encloses the
non-convex surface in Fig. 2.18(c). As a result, there is no unique set of base ma-
terials needed to generate this optic, unlike for the two-material design where the
two bases are required to lie at the ends of (or beyond) the curve in Fig. 2.15(c).

2.7 GRIN AFL sensitivity analysis

When designing any optical system intended for fabrication, it is wise to perform
a sensitivity analysis of how the design responds to fabrication errors. This re-
sponse to error almost always yields a decrease in performance. From this sensi-
tivity analysis, a set of manufacturing tolerances can be specified that, when met
in fabrication, yield designs with a known range of as-built performance.
The act of tolerancing freeform GRIN designs has received little attention, largely
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 52

due to the enormity and complexity of the task. While tolerance analysis has been
performed for conventional GRIN profiles [85], there are many additional sources
of errors when fabricating F-GRIN elements including deviations from the nom-
inal freeform refractive index profile as well as registration of the F-GRIN to the
accompanying optical surfaces. In this work, the former type of error is the sole
focus while any surfaces errors are neglected, including their registration to the
GRIN.
Errors in refractive index variation can occur both spectrally and spatially. Spec-
tral errors derive from differences in chromatic behavior from the nominal model,
including differences in base material dispersion as well as differences in refractive
index composition (i.e., linear versus nonlinear GRIN compositions). The spatial
index variation may also deviate from the prescribed distribution. These spatial er-
rors occur across a range of different scales or, inversely, spatial frequencies. Both
types of fabrication errors are analyzed here for their impact on performance.
For polychromatic GRIN AFL designs, the sensitivity with regard to achromati-
zation is of particular interest. Errors in GRIN dispersion are studied for their effect
on MTF performance for the linear two-material design in Fig. 2.11. The change
in MTF on-axis and at full field is evaluated for tolerances in the homogeneous
base materials’ refractive indices δn and Abbe numbers δν (partial dispersion goes
unchanged). Two sets of material tolerances are evaluated, as listed in Table 2.6.
These tolerances are based on standard index and Abbe number tolerances for

Tolerance Type “Commercial” “Precision”


Base material refractive index, δn ±0.001 ±0.0005
Base material Abbe number, δν ±0.8% ±0.5%
Table 2.6: Tolerances for homogeneous base materials constituting the GRIN. Material
tolerances are based on standard values for “commercial” and “precision” applications [86].
Errors in the base materials result in errors in the GRIN dispersion.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 53

On-axis On-axis
1.0 1.0
Diffraction limit Diffraction limit
0.8 Nominal design 0.8 Nominal design
Modulation

Modulation
Monte Carlo Monte Carlo
0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 50 100 150 200 0 50 100 150 200
Spatial Frequency (cycles/mm) Spatial Frequency (cycles/mm)

Full field Full field


1.0 1.0
Diffraction limit Diffraction limit
0.8 Nominal design 0.8 Nominal design
Modulation

Modulation
Monte Carlo Monte Carlo
0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 50 100 150 200 0 50 100 150 200
Spatial Frequency (cycles/mm) Spatial Frequency (cycles/mm)

(a) (b)

Figure 2.19: Sensitivity analysis of GRIN dispersion for the polychromatic GRIN AFL de-
sign in Fig. 2.11. Spectral error is randomly introduced by deviations in homogeneous
base materials with (a) “commercial” tolerances (δn = ±0.001, δν = ±0.8%) and (b) “pre-
cision” tolerances (δn = ±0.0005, δν = ±0.5%) [86]. 100 Monte Carlo trials are performed.
The perturbed MTF is shown on-axis and at full field.

“commercial” and “precision” applications [86]. A Monte Carlo sensitivity anal-


ysis is performed with 100 trials where both base materials are perturbed by a
random amount within δn and δν (uniform probability distribution), the system is
refocused, and the as-built MTF is calculated. The results are shown in Fig. 2.19.
For commercial tolerances, the on-axis MTF at 200 lp/mm drops for most cases by
< 10% and in the worst case by ≈ 15% in Fig. 2.19(a). For precision tolerances, the
on-axis MTF typically drops by < 5% and in the worst case ≈ 10% in Fig. 2.19(b).
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 54

Similar results are found at full field, although in some cases the full field perfor-
mance of the perturbed design is improved at the expense of on-axis performance.
It is also of interest how performance changes for GRIN AFL designs with error
in the spatial refractive index variation. A similar sensitivity analysis is performed
where low-spatial frequency error is introduced via random perturbations of the
four radial coefficients (C10 , C20 , C30 , C40 ) and the four axial coefficients (C01 , C02 ,
C03 , C04 ). Since the radial and axial coordinates are normalized, random coeffi-
cient errors translate to a comparable magnitude refractive index error across the
element. In the sensitivity analysis, different sets of “commercial” and “precision”
coefficient tolerances are evaluated, as listed in Table 2.7. In a Monte Carlo process,
each coefficient is randomly changed by some percent error, δCi0 or δC0j , with uni-
form probability distribution. The perturbed design is then refocused, and the
on-axis and full field MTF data are recorded.
The findings from the Monte Carlo sensitivity analysis are depicted in Fig. 2.20
for 100 trials. For δC = ±0.5%, the on-axis MTF at 200 lp/mm drops for most cases
by < 15% and in the worst case by ≈ 25% in Fig. 2.20(a). For δC = ±0.25%, the
on-axis MTF typically drops by < 10% and in the worst case ≈ 15% in Fig. 2.20(b).
As before, similar conclusions are drawn for full field where some cases experience
a performance improvement at the cost of on-axis performance.
While the performed sensitivity analysis adequately captures low-spatial fre-

Tolerance Type “Commercial” “Precision”


GRIN radial coefficients, δCi0 ±0.5% ±0.25%
GRIN axial coefficients, δC0j ±0.5% ±0.25%
Table 2.7: Tolerances for GRIN spatial coefficients governing radial and axial variation
in Eq. (2.8). Errors in the coefficients result in low spatial frequency errors in the GRIN
refractive index variation. Since the GRIN spatial coordinates are normalized, both types
contribute approximately equal refractive index errors.
Chapter 2. Freeform Gradient-Index Annular Folded Lenses 55

On-axis On-axis
1.0 1.0
Diffraction limit Diffraction limit
0.8 Nominal design 0.8 Nominal design
Modulation

Modulation
Monte Carlo Monte Carlo
0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 50 100 150 200 0 50 100 150 200
Spatial Frequency (cycles/mm) Spatial Frequency (cycles/mm)

Full field Full field


1.0 1.0
Diffraction limit Diffraction limit
0.8 Nominal design 0.8 Nominal design
Modulation

Modulation
Monte Carlo Monte Carlo
0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 50 100 150 200 0 50 100 150 200
Spatial Frequency (cycles/mm) Spatial Frequency (cycles/mm)

(a) (b)

Figure 2.20: Sensitivity analysis of low-spatial frequency refractive index error for the
polychromatic GRIN AFL design in Fig. 2.11. Spatial error is randomly introduced via
all eight spatial coefficients (four radial, four axial) with maximum coefficient errors of (a)
δC = ±0.5% and (b) δC = ±0.25%. 100 Monte Carlo trials are performed. The perturbed
MTF is shown on-axis and at full field.

quency index error, other spatial defects such as mid-spatial frequency (MSF) er-
rors may also appear in GRIN parts. MSF errors may be especially prevalent in
GRIN fabricated by sub-aperture techniques like additive manufacturing. To ac-
count for MSF errors, a different tolerancing analysis is needed where higher spa-
tial frequency refractive index perturbations must be introduced.
Chapter 3

Compact Zoom Riflescope using


Gradient-Index Alvarez Lenses

3.1 Surface Alvarez lenses (SALs)

The Alvarez lens is an optical component that provides a tunable amount of opti-
cal power [12]. The atypical component consists of two freeform optical elements
where power variation is achieved by translating the elements in equal and op-
posite amounts along an axis orthogonal to the optical axis, as shown in Fig. 3.1.
Classically, the two Alvarez elements are identical in form and are oriented 180◦
rotated with respect to one another.

Φ=0 Φ>0 Φ<0

Figure 3.1: Surface Alvarez lenses (SALs) offer a variable amount of optical power Φ
based on laterally translating elements orthogonal to the optical axis.

56
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 57

Historically, Alvarez lenses have used freeform surfaces for the component’s
two freeform elements. The recent introduction of new types of freeform optics
including freeform GRIN (F-GRIN) [15] requires the distinction of surface Alvarez
lenses (SALs). SALs serve as one of the earliest examples of freeform optics of any
type. Introduced in the late 1960s, SALs were originally devised concurrently by
Nobel laureate Luis Alvarez [12, 87, 88] and Adolf Lohmann [89, 90].
SALs have found use in a variety of different fields. Their primary application
is in ophthalmics [87, 91, 92] where SALs can provide tunable corrective power for
myopia or hyperopia in the human eye. Ophthalmic applications were the original
motivation for Alvarez’s invention, including a similar concept that provides a
tunable amount of cylindrical power for astigmatism compensation in the eye [87].
More recent applications have explored zoom and varifocal configurations us-
ing SALs. New zoom kernels can be obtained with variable power components
such as SALs that remain longitudinally fixed in position [93, 94]. Zoom sys-
tems using SALs have been demonstrated in a variety of different imaging system
designs including cameras [95–98], telescopes [99], microscopes [100, 101], endo-
scopes [102, 103], and head-mounted displays [104].
Moreover, SALs have primarily been applied in the visible spectrum; although,
a germanium SAL has been demonstrated for use in the mid-wave infrared spec-
trum [105]. The variable power generation of SALs has also been applied to the
design of nonimaging optics [106,107]. While no longer used in forming an image,
SALs offer the ability of tunable illumination distributions.
The first-order power variation and imaging performance of SALs has been
analyzed by several [91, 92, 95, 105, 108–110]. The origin of SAL power variation
is related to the slope of the freeform surface along the shift axis [108]. In this
way, the SAL concept can also be expanded to generate a tunable amount of any
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 58

continuous wavefront aberration by laterally shifting the proper surface pair [108].
In terms of performance, SALs suffer from significant induced aberrations [100,
108,110]. These aberrations are induced by the spacing of the two freeform surfaces
in SALs. Space between surfaces is necessary for clearance when translating the el-
ements for power variation. The spacing is ideally kept to the minimum allowable
separation in order to minimize the aberration contribution. For the same reason,
the SAL freeform surfaces are typically positioned adjacent to one another, internal
to the component rather than on the exterior surfaces. Induced aberrations can be
corrected to a degree with higher-order terms in the surface figure [100, 108, 110].
With these higher-order terms, SALs may take advantage of the two Alvarez ele-
ments having different surface prescriptions. Nevertheless, the aberration correc-
tion in SALs is limited for the full power range. As a result of the induced aberra-
tions, adequate performance is also limited to a small field-of-view [95, 100, 110],
a feature that is acceptable in ophthalmic applications, for example. On the other
hand, SAL performance that is limited in power range and/or field-of-view makes
their integration in complete zoom imaging systems challenging.
A second unwanted effect of SALs is their introduction of a variable bore-
sight error (BSE) that changes in magnitude with element shift and power vari-
ation [91, 95]. Also known as prismatic error, BSE results in a deflection of the ray
traveling along the optical axis [111], an effect that is absent in nominal rotationally
symmetric systems. BSE is detrimental to imaging applications where line-of-sight
accuracy is critical. One such example is the design of riflescopes, which is the fo-
cus of Sec. 3.4. While BSE can be compensated with a properly oriented wedged
optic, the change in BSE with power variation means a static corrector is no longer
sufficient. Rather, a dynamic means of BSE correction is required, a feature that is
returned to in Sec. 3.4.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 59

3.2 Gradient-index Alvarez lenses (GALs)

GRIN offers a new avenue for constructing Alvarez lenses, using F-GRIN for the
freeform elements rather than freeform surfaces. GRIN Alvarez lenses (GALs) em-
ploy two F-GRIN elements with plane-parallel surfaces that are shifted in equal
and opposite amounts orthogonal to the optical axis, as shown in Fig. 3.2.
As has been done for SALs [91, 92, 95, 105, 108–110], GALs can be better under-
stood by developing an analytical framework governing their power variation and
performance. The refractive index profile for the two GAL elements can be derived
analytically, as shown in Sec. 3.2.1. Next, the incorporation of a tilt refractive index
term and its effect on ∆n is considered, as done in Sec. 3.2.2. Then, the first-order
power variation can be derived based on these profiles and the element shift, as
demonstrated in Sec. 3.2.3. Finally, for these “base” (i.e., unoptimized) GALs, the
overall imaging performance can be assessed, as shown in Sec. 3.2.4.
There are several interesting opportunities available with GALs that are not
achievable with SALs. Due to their planar surfaces, the GAL elements can be po-
sitioned with zero internal airspace, as shown in Fig. 3.2. As a result, the source of

Φ=0 Φ>0 Φ<0


Refractive index

Figure 3.2: GRIN Alvarez lenses (GALs) offer a variable amount of optical power Φ based
on laterally translating elements orthogonal to the optical axis. Unlike for SALs, GALs use
plane-parallel surfaces and require no internal airspace.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 60

induced aberrations in SALs is eliminated. Although, aberrations in GALs are still


induced, now due to their element center thicknesses, as described in Sec. 3.2.4.
A second advantage of GALs is the volumetric control of F-GRIN that is unavail-
able with freeform surfaces. F-GRIN functions analogous to a series of freeform
surfaces in terms of aberration contributions [14, 15]. Consequently, F-GRIN of-
fers new degrees of freedom for correcting induced aberrations as well as BSE in
GALs. These volumetric degrees of freedom are studied for optimized GAL de-
signs in Secs. 3.3 and 3.4.

3.2.1 Refractive index profile

The refractive index profile for the two GAL elements can be derived in a similar
manor to that presented by Palusinski et al. for SAL surface sag [108]. The optical
path difference (OPD) imparted by a GRIN element with plane-parallel surfaces
for a normally incident wavefront can be related to the transverse refractive index
variation n (x, y) and center thickness t,

W (x, y) ≈ t · [n (x, y) − n0 ] = t · nr (x, y) . (3.1)

where nr (x, y) = n (x, y) − n0 is the refractive index variation relative to the value
at the origin, n0 = n (0, 0). As a result, it is arbitrarily established that W (0, 0) = 0.
The OPD expression in Eq. (3.1) relies on the thin element approximation (TEA) for
GRIN media [22] where any potential ray curvature within the GRIN is neglected.
In this case, the GRIN contribution is approximated as confined to a plane rather
than a volume.
Two GRIN elements with plane-parallel surfaces can also be jointly considered.
With no internal airspace between the elements, the resultant OPD according to
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 61

-nr (x, y - δ)

Refractive index

y
nr (x, y + δ)
z
t t
Figure 3.3: GAL element shift δ along y imparts optical power. For the configuration
shown, the GAL produces positive power.

the TEA is simply the sum of the individual contributions of the elements,

W (x, y) = W1 (x, y) + W2 (x, y) = t1 nr1 (x, y) + t2 nr2 (x, y) . (3.2)

For the case of base GALs, the two elements are identical, meaning t1 = t2 = t
and nr1 = nr2 = nr . Assuming the GAL implements power variation by lateral
element translation along y, the second element is flipped along y, as depicted in
Fig. 3.3. As a result of the element flip, there is a sign inversion in the relative
refractive index of the second element. For the equal and opposite element shifts
±δ shown in Fig. 3.3, the GAL imparted wavefront OPD according to Eq. (3.2) is

W (x, y) = t [nr (x, y + δ) − nr (x, y − δ)] . (3.3)

Notice that for no element shift, δ = 0, the GAL imparts no wavefront aberration,
W (x, y) = 0.
Then, multiplication of Eq. (3.3) by unity, 1 = (2δ) / (2δ), yields an interesting
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 62

result,
nr (x, y + δ) − nr (x, y − δ)
W (x, y) = 2δt . (3.4)

The fraction in Eq. (3.4) can be recognized as the difference quotient of single-
variable calculus. Consequently, for small element shifts δ, a differential relation-
ship can be formed between the imparted wavefront aberration W and the relative
refractive index nr ,
∂nr
W (x, y) = 2δt . (3.5)
∂y

Thus, the imparted wavefront is dependent on the derivative of the relative refrac-
tive index along the axis of element translation, which in this framework is y. Eq.
(3.5) also shows that the magnitude of W is directly proportional to the shift extent
δ and each element’s center thickness t.
Rearranging and integrating Eq. (3.5) gives the inverse scenario of the requisite
relative refractive index in terms of the wavefront,

Z
1
nr (x, y) = W (x, y) dy. (3.6)
2δt

By Eq. (3.6), the relative index nr can be solved that produces a tunable amount of
any arbitrary wavefront aberration W . For GALs, the desired wavefront aberration
is power, W = x2 + y 2 , which inserted into Eq. (3.6) yields

y3
Z  
1 2 2 1
+ x2 y

nr (x, y) = x +y dy = (3.7)
2δt 2δt 3

where the constant of integration is set to zero to maintain nr (0, 0) = 0.


From the established definition of the relative refractive index, nr = n − n0 , as
well as the flipped second element’s sign inversion, the absolute refractive index
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 63

profiles for the two GAL elements, n1 (x, y) and n2 (x, y), can be obtained directly
from Eq. (3.7),

y3
 
2
n1 (x, y) = n0 + A +x y ,
3
 3  (3.8)
y 2
n2 (x, y) = n0 − A +x y
3

where A is a scalar coefficient of units mm−3 describing the magnitude of the


power variation with element shift. As seen next in Sec. 3.2.2, it is advantageous
to incorporate an additional term in these refractive index profiles.

3.2.2 Tilt and ∆n

The first-order power variation in SALs is ultimately limited by the sag of its
freeform surfaces. As with all freeform surfaces, there are limitations on the sur-
face sag and slope departures that are achievable in fabrication and metrology. The
same limitation is imposed in GALs due to a restriction of the total refractive index
change ∆n in fabrication and metrology. In fact, the ∆n is far more restricted in
GALs than the sag is in SALs when considering an equivalent element OPD. As a
result, the ∆n is a driving factor in GAL design that must be carefully understood.
One way of alleviating pressure on the GAL ∆n constraint is by incorporating
an additional tilt refractive index term along the shift axis y,

y3
 
2
n1 (x, y) = n0 + A + x y + By,
3
 3  (3.9)
y 2
n2 (x, y) = n0 − A + x y − By
3

where B is a scalar coefficient of units mm−1 describing the magnitude of the GRIN
tilt. According to Eq. (3.5), this tilt term introduces a variable amount of piston to
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 64

Figure 3.4: GRIN ∆n for base GAL elements with different values of tilt coefficient B. Two
specific cases of interest are denoted. (i) Zero tilt, B = 0, is the conventional Alvarez form
in Eq. (3.8). (ii) ∆n is at a minimum when B = B0 , requiring 2.4 times less refractive index
change than for zero tilt. ∆n increases approximately linearly about B = B0 . This example
is evaluated for Alvarez coefficient A = −8.485 × 10−4 mm−3 and element diameter D =
10 mm.

the wavefront, which does not impact image quality and does not alter power
variation.
The tilt term proves extremely useful in the design of GALs because it can be
used to reduce the GRIN ∆n while maintaining the same power variation. Said
another way, tilt allows for greater power variation given a restriction on the max-
imum value for ∆n. For example, Fig. 3.4 shows the relationship between B and
∆n. The resultant curve is piecewise, and importantly, there is a certain tilt value
denoted B = B0 for which the ∆n is at a minimum. Notice that the sign of B0 is
opposite that of A.
The refractive index profile for GAL elements with different values for B is
shown in Fig. 3.5. The tilt is quantified in terms of the ratio B/B0 where B/B0 = 0
with no tilt and B/B0 = 1 when ∆n is minimized. It can be seen by the black
and white points marking the refractive index maxima and minima, respectively,
why it is that B/B0 = 1 minimizes the ∆n. With no tilt, B/B0 = 0, and increasing
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 65

Figure 3.5: GAL element refractive index profiles for different values of tilt coefficient B. In
each profile, maximum refractive index values are marked with black points while minimum
values are marked with white points. When B/B0 = 1, the GRIN ∆n is minimized, as
shown in Fig. 3.4.

amounts of tilt, B/B0 < 1, the index extrema are located along the clear aperture
edge. Only when reaching B/B0 = 1 do the internal nodes equal the outer extrema
in refractive index. Then, even larger tilts B/B0 > 1 yield index extrema only at
these internal nodes, including at the clear aperture edge.
In Eq. (3.9), the refractive index profile for the base GAL configuration is de-
fined in terms of coefficients A and B. Although, the GRIN ∆n is typically the
physical quantity of interest when designing, fabricating, and measuring GRIN
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 66

optics. Using the findings of Figs. 3.4 and 3.5, analytical expressions can be de-
rived that forge a relationship between ∆n, the coefficients A and B, as well as the
diameter of the GAL element D.
The derivation of the multi-dimensional ∆n function is performed in Appendix
C with the result listed in Eq. (C.17). The expression is continuous in A and D but
piecewise-continuous in B with four different subdomains, as expected given Fig.
3.4.
In this chapter, two specific values for B are studied due to their relevance
in GAL design. The first case is a GAL with zero tilt, which corresponds to the
more conventional Alvarez form listed in Eq. (3.8). With no tilt, B = 0, the total
refractive index change ∆n according to Eq. (C.17) is


2 3
∆n (B = 0) = D |A|. (3.10)
12

The second case of interest is a GAL with tilt that minimizes the element ∆n, a
tilt value denoted B = B0 . The value of B0 is also derived in Appendix C where it
is found to be
B0 = −κD2 A (3.11)

where
1 + 22/3 − 21/3
κ= ≈ 0.1106. (3.12)
12

Then, the GAL element ∆n for this case of B = B0 is

4κ3/2 3
∆n (B = B0 ) = D |A|. (3.13)
3

There are several interesting features for the ∆n expression in Eqs. (3.10) and
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 67

(3.13) that are common to both cases. First, both expressions increase linearly with
the magnitude of A and cubically with the element diameter D. As proven next
in Sec. 3.2.3, the GAL power variation is directly proportional to A and, there-
fore, also directly proportional to ∆n. Moreover, the expressions are scaled by two
different constant factors. Taking the ratio of Eqs. (3.10) and (3.13) finds

∆n (B = B0 ) √
= 8 2κ3 ≈ 0.4163. (3.14)
∆n (B = 0)

This means that with B = B0 only 42% the ∆n is required compared with the
conventional no tilt case. As found in Sec. 3.2.4, the value of B has minor impact
on the imaging performance of GALs, so to maximize power variation, the optimal
value for the tilt coefficient in most cases is B = B0 .

3.2.3 Power variation

The optical power imparted by a GAL for a given element shift δ can be understood
analytically, to first-order, from its refractive index distribution. GRIN paraxial op-
tical power Φ derives from radially quadratic refractive index change. The Wood
lens embodies this index profile [18],

nw (x, y) = n0 + N10 x2 + y 2

(3.15)

where N10 is the scalar coefficient governing the radially quadratic contribution to
the index. For this profile, the Wood lens power is directly proportional to N10 and
the lens center thickness t [17],

Φw = −2N10 t. (3.16)
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 68

Meanwhile, the base GAL refractive index is derived in Eq. (3.9) to produce op-
tical power from the overlapping region of the two elements. Within the TEA, the
refractive index in the overlap can be considered as the average of the two elements’
profiles in this region,

n1 (x, y + δ) + n2 (x, y − δ)
navg (x, y) = . (3.17)
2

Applying the GAL definitions of n1 and n2 in Eq. (3.9) and reducing gives

navg (x, y) = n∗0 + Aδ x2 + y 2



(3.18)

where n∗0 is a spatially constant refractive index, dependent on δ, A, and B,

Aδ 3
n∗0 = n0 + + Bδ. (3.19)
3

By comparing Eq. (3.18) with the Wood lens profile in Eq. (3.15), the same radially
quadratic spatial term is recognized where now N10 = Aδ. The average refractive
index is also plotted in Fig. 3.6 where in the overlap the familiar radially quadratic
profile can be seen.
Thus, by Eq. (3.16), the GAL optical power for shift configuration δ is

Φ = −4Atδ (3.20)

where center thickness 2t is needed for the two GAL elements in series.
The GAL power expression in Eq. (3.20) shows that first-order power changes
linearly with transverse element shift δ as well as Alvarez coefficient A and GAL
element thickness t. While A and t are fixed in sign, δ may be either positive or neg-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 69

Refractive index
+δmax
CA
- δmax

D
y
x
Figure 3.6: GAL average refractive index. In the overlapping region, navg is a radially
quadratic GRIN power distribution. For element diameters D and clear aperture CA, ±δmax
is the maximum element shift before vignetting occurs. When used in an optical system,
all rays outside CA are blocked.

ative depending on the configuration, meaning both positive and negative power
Φ can be imparted.
It is established in Sec. 3.2.2 that each circular GAL element has diameter D.
Although, as evident from Fig. 3.6, the used clear aperture of a GAL is necessar-
ily smaller than D when elements are shifted and power is introduced from navg
in the overlapping region. For a circular clear aperture CA centered on the opti-
cal axis, there is a maximum allowable element shift δmax before vignetting of the
GAL clear aperture occurs. The maximum shift configuration occurs when CA is
inscribed in the overlapping region, as shown in Fig. 3.6. From this configuration,
a relationship between δmax , D, and CA can be noticed,

D − CA
δmax = . (3.21)
2
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 70

The range of allowable element shift without vignetting is then δ = [−δmax , δmax ].
Since the GAL power in Eq (3.20) is proportional to δ, this entails that there is a cor-
responding restriction on the GAL total power range, ∆Φ = Φmax − Φmin ,

∆Φ = |−4Atδmax + 4Atδmin |
(3.22)
= 4|A|t (D − CA) .

Rather than expressing ∆Φ in terms of Alvarez coefficient |A|, the power range
can be described in terms of ∆n using the relationships presented in Sec. 3.2.2 for
the two specific cases of tilt. For no tilt, B = 0, solving for |A| in Eq. (3.10) and
inserting into Eq. (3.22) gives

√ D − CA
∆Φ (B = 0) = 24 2 t ∆n . (3.23)
D3

For tilt minimizing the refractive index change, B = B0 , the power range in terms
of ∆n is found via Eqs. (3.13) and (3.22),

D − CA
∆Φ (B = B0 ) = 3κ−3/2 t ∆n (3.24)
D3

where κ is defined in Eq. (3.12).


Notably, both Eqs. (3.23) and (3.24) have the same functional form but different
scalar coefficients. Similar to Eq. (3.14), taking the ratio of these two expressions
finds √
∆Φ (B = B0 ) κ−3/2 2
= ≈ 2.4023, (3.25)
∆Φ (B = 0) 16

meaning, for the same GAL ∆n, a 2.4 times larger power range is achievable with
B = B0 than B = 0. Eqs. (3.23) and (3.24) also show that the power range increases
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 71

linearly with ∆n as well as thickness t, as expected within the TEA.


A surprising result of Eqs. (3.23) and (3.24) is that the GAL power range is
proportional to the term (D − CA) /D3 . The power range for B = 0 according to
Eq. (3.23) is plotted in Fig. 3.7 versus element diameter for a fixed clear aperture.
The curve in Fig. 3.7 shows the behavior of (D − CA) /D3 . When D = CA, the
power range is zero since no element shift is possible without vignetting of the
clear aperture. Meanwhile, the power range does not increase monotonically with
D, but instead presents a local maximum when D/CA = 1.5. This finding is also
given by differentiating with respect to D and equating to zero,

 
d D − CA 3CA − 2D D 3
= =0 ⇒ = . (3.26)
dD D3 D4 CA 2

This shows that for a fixed clear aperture, as is typically the case when using GALs
in optical design, the power range is maximized when the element diameter is 1.5
times the clear aperture. The same finding is discovered for SALs, although it has
not been reported. While this ratio provides the greatest power range, other factors

Figure 3.7: GAL power range ∆Φ versus element diameter D for a fixed clear aperture
CA. The power range is maximized for D/CA = 1.5. This example is evaluated using
Eq. (3.23)
 for no tilt with CA = 10 mm, t = 3 mm, and ∆n = 0.2 Power range is shown in
diopters D = m−1 .
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 72

such as imaging performance may influence the optimal value for D/CA, as found
in Sec. 3.2.4.
In addition to the range of power variation, an interesting aspect of Alvarez
lenses that has been largely overlooked in SALs is the capability to bias the power
variation [112–114]. The base GAL described thus far can achieve power variation
with equal magnitudes of positive and negative power,

 
∆Φ ∆Φ
Φ∈ − ,+ . (3.27)
2 2

This power variation can also be thought of as having zero bias power, Φbias = 0,
yet in both SALs and GALs, one can obtain bias power Φbias 6= 0 for greater control
over the power variation,

 
∆Φ ∆Φ
Φ ∈ Φbias − , Φbias + . (3.28)
2 2

Bias power proves very advantageous when incorporating Alvarez lenses in zoom
imaging systems where zoom kernels may require exclusively positive or negative
power [93, 94].
Bias power shifts the center of the power range that occurs for δ = 0. One way
of achieving this effect is using GAL elements with coordinate centers offset from
the optical axis for the base refractive index in Eq. (3.9). As a result, bias power
is implemented using a coordinate shift along the axis of element translation. For
GALs translating along y, bias power is introduced in the refractive index profiles
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 73

by a coordinate shift y0 ,

" #
(y − y0 )3
n1 (x, y) = n0 + A + x2 (y − y0 ) + B (y − y0 ) ,
3
" # (3.29)
(y + y0 )3
n2 (x, y) = n0 − A + x2 (y + y0 ) − B (y + y0 ) .
3

Notably, even with the incorporation of y0 , the two GAL elements remain identical
to one another upon flipping of the second element about y.
From these profiles, the average refractive index in the overlapping region ac-
cording to Eq. (3.17) is now

navg (x, y) = n∗0 + A (δ − y0 ) x2 + y 2



(3.30)

where
A (δ − y0 )3
n∗0 = n0 + + B (δ − y0 ) . (3.31)
3

As before, the average index consists of a spatially constant term as well as a radi-
ally quadratic term, scaled by coefficient N10 = A (δ − y0 ), as found by comparing
with Eq. (3.15) for the Wood lens. The revised expression for N10 yields an altered
form of the GAL power,
Φ = −4At (δ − y0 ) . (3.32)

GAL bias power can now be recognized in Eq. (3.32) for the case of no element
shift, δ = 0,
Φbias = 4Aty0 . (3.33)

From Eq. (3.33), Φbias increases linearly with coordinate shift y0 .


GAL elements that introduce different amounts of bias power can be seen in
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 74

Φbias = -7.54 D Φbias = -3.77 D Φbias = 0.00 D Φbias = 3.77 D Φbias = 7.54 D
Δn = 0.117 Δn = 0.104 Δn = 0.100 Δn = 0.104 Δn = 0.117
7.5 7.5 7.5 7.5 7.5 1.559

Refractive index
y [mm]

y [mm]

y [mm]

y [mm]

y [mm]
0.0 0.0 0.0 0.0 0.0 1.500

-7.5 -7.5 -7.5 -7.5 -7.5 1.441


-7.5 0.0 7.5 -7.5 0.0 7.5 -7.5 0.0 7.5 -7.5 0.0 7.5 -7.5 0.0 7.5
x [mm] x [mm] x [mm] x [mm] x [mm]

Figure 3.8: GAL element refractive index profiles with different coordinate shifts y0 and
bias powers Φbias in units of diopters. Each element produces the same power range ∆Φ,
but greater ∆n is needed for larger bias powers.

Fig. 3.8. While the refractive index profiles differ between elements, an important
takeaway is that the refractive index range ∆n increases with the magnitude of
Φbias . Similar to Appendix C, a piecewise function can be constructed to determine
∆n from the different parameters, now including the coordinate shift y0 in the
refractive index defined in Eq. (3.29). Although, the form of this ∆n function is
more complex and requires additional subdomains. For the simplest case of zero
tilt, the ∆n in terms of y0 is found by

   √
3/2
2 3 y0 2
∆n (B = 0) = D |A| 4 +1 (3.34)
12 D

where now an extra term in hard brackets is included compared with Eq. (3.10).
Bias power changes linearly with y0 while in Eq. (3.34) the ∆n changes nonlinearly,
as is also apparent by the ∆n values in Fig. 3.8.
As before, there is also a tilt coefficient that minimizes the ∆n for y0 6= 0. Al-
though, the expression for this tilt value is also piecewise and not analytically sim-
ple. Instead, the minimizing tilt can be solved numerically where it is found that
∆n increases approximately linearly with y0 , a lower-order dependence than in Eq.
(3.34) for zero tilt.
In addition to its effect on ∆n, bias power also affects the imaging performance,
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 75

as examined next in Sec. 3.2.4. Overall, the capability of introducing bias power
is useful for certain cases where power variation is sought centered around a non-
zero value.

3.2.4 Imaging performance

The imaging performance of GALs is studied next. It is described in Sec. 3.1 how
SALs deviate from their first-order formulation due to the finite extent of their
surface sag. Aberrations are induced due to the finite airspace between the two
freeform surfaces that is needed for surface clearance while translating.
On the other hand, GALs applying planar surfaces avoid the need for airspace
between elements. Although, aberrations are still induced, now due to the finite
center thickness of each element. Since the GAL formulation in Sec. 3.2 is based
on the thin element approximation (TEA), only when t approaches zero does the
GAL performance converge on paraxial. This realization provides guidance that
the thinnest possible elements provide optimal imaging performance in GALs. The
connection between performance and t is explored further in this section.
GAL imaging performance is studied in this section in three different ways: (1)
power variation, (2) wavefront error, and (3) boresight error. For all analyses, the
same base GAL design is considered with the parameters listed in Table 3.1.

Power variation

The GAL first-order power variation derived in Sec. 3.2.3 relies on the TEA by
assuming each element to be infinitesimally thin and coincident with one another.
As a result, the power imparted by a GAL according to real ray tracing through
elements of finite thickness differs from the analytical result. It is important to un-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 76

Parameter Value
Element diameter, D [mm] 15.0
Clear aperture, CA [mm] 10.0
Maximum element shift, δmax [mm] 2.5
Element thickness, t [mm] 3.0

Power range, ∆Φ [D] 10.0


Alvarez coefficient, A [mm−3 ] 7.03125 × 10−2
Bias power, Φbias [D] 0.0
Coordinate offset, y0 [mm] 0.0

Base refractive index, n0 1.5


GRIN ∆n
Zero tilt, B = 0 0.066
Tilt minimizing, B = B0 0.028

Circular full field-of-view [deg] 10.0


monochromatic,
Spectrum
λd = 587.56 nm
Table 3.1: Standardized GAL design parameters analyzed for imaging performance in
Sec. 3.2.4, except where noted otherwise.

derstand the actual imparted power when designing systems incorporating GALs.
For example, if the difference in power were significant, a correction factor in the
power expression would be needed for compensation.
For comparison with the analytical results of Sec. 3.2.3, the power imparted
by a GAL is calculated using real ray tracing in the optical design software CODE
V [72]. The effective focal length of a GAL is not a clearly defined quantity (e.g.,
where are the principal planes?) and is infinite for the zero power configuration.
As a result, GAL power is evaluated using a wavefront consisting of the first 37
fringe Zernike polynomial terms fit to real ray trace data. The power Zernike term
Z4 is then subsequently converted to units of diopters via the entrance pupil diam-
eter. The fit power term encompasses deviations from the TEA, averages over any
potential wavefront aberrations, and ignores the effects of boresight error.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 77

The first-order and real ray power variations are compared for three different
cases of interest, as shown in Fig. 3.9. First, the power is calculated by both meth-
ods across the full element shift range for the GAL configuration in Table 3.1 with
no tilt in the GRIN, B = 0. The results are shown in Fig. 3.9(a) where the real ray
power changes approximately linearly with element shift, agreeing with the first-
order result in Eq. (3.20). The difference between first-order and real ray power
is also plotted where there is at most one-twentieth diopter of departure. Second,
the same GAL configuration is evaluated except with B = B0 for tilt minimiz-
ing the ∆n. For this case, the power comparison is shown in Fig. 3.9(b) where
now the error in the TEA analysis is at most one-thirtieth diopter. Third, the zero

Figure 3.9: Comparison of first-order and real ray power variation in GALs. First-order
power is calculated using the expressions derived in Sec. 3.2.3. Real ray power is evalu-
ated from a wavefront fit to real ray trace data. The three cases consider the GAL configu-
ration in Table 3.1 with (a) zero tilt, (b) tilt minimizing ∆n, and (c) bias power of 5 diopters.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 78

tilt GAL configuration in Table 3.1 is adopted with the addition of a bias power,
Φbias = 5 D. The bias power shifts the first-order power range from Φ ∈ [−5, 5] D
to Φ ∈ [0, 10] D, now imparting only positive power. The results shown in Fig.
3.9(c) confirm the addition of the bias power, and the difference from first-order is
at most one-sixth diopter.
The GAL power variation based on real ray tracing confirms the validity of
the analytical expressions derived in Sec. 3.2.3. The difference between the two
calculations is significantly sub-diopter and likely can be attributed to other (non-
power) wavefront aberrations influencing the wavefront fit used in determining
ray-based power. As seen next, wavefront error in GALs increases with power
magnitude, a trend that is also reflected in Fig. 3.9. Overall, the first-order power
formulation can be confidently applied to GALs.

Wavefront error

Next, GAL image performance is assessed in terms of wavefront error (WFE).


GALs present a vast parameter space, and it is of interest how WFE changes with
a variety of different factors. In total, six different parameters are considered for
their effect on WFE: (1) field-of-view, (2) power range ∆Φ, (3) bias power Φbias , (4)
element thickness t, (5) diameter to clear aperture ratio D/CA, and (6) GRIN tilt B.
Wavefront performance is evaluated in CODE V via the RMS WFE, calculated rel-
ative to the best-fit power for the on-axis field. The RMS WFE is quantified in units
of waves at the monochromatic wavelength λd = 587.56 nm and in the following is
abbreviated as [wvs].
First, WFE is assessed across the field-of-view for the standard GAL configura-
tion with zero tilt. The results can be seen in Fig. 3.10 where performance is evalu-
ated across a 10◦ circular full field-of-view. The WFE versus field is also shown for
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 79

Figure 3.10: RMS WFE across a circular 10◦ full field-of-view for three different power
configurations. The evaluated GAL is defined in Table 3.1 with zero tilt. Diffraction-limited
performance according to the Maréchal criterion is limited to a small region in the center of
the field-of-view while performance significantly deteriorates with increasing field.

three distinct power configurations that span the full power range. The results in
Fig. 3.10 show that diffraction-limited performance is achieved only over a small
central region in the field-of-view. Beyond this on-axis region, performance dete-
riorates rapidly with more than 1.5 waves of error at the edge of the field for all
three power configurations, a 50x increase in error compared with on-axis. This
inadequate field performance makes incorporating GALs in full imaging systems
difficult, especially when large fields-of-view are required. It can also be noticed
that the performance is not rotationally symmetric with field as it would be in a
rotationally symmetric system. Instead, GALs being freeform optical components
means upper and lower field points may perform differently. Although, there is
field symmetry about the shift axis y. Consequently, two-dimensional fields must
be modeled when designing with GALs, and non-rotationally symmetric perfor-
mance can be expected.
The freeform aberration composition of GALs is very similar to that of SALs.
The geometrical spot diagrams across the field-of-view are shown in Fig. 3.11 at
the 5 diopter power configuration. For the on-axis field, the dominant aberrations
are coma and trefoil, the same as for SALs [95, 105]. The freeform coma and trefoil
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 80

Φ=5D

Field angle, Y [deg]


0

-5
6 arcmin

-5 0 5
Field angle, X [deg]

Figure 3.11: Geometrical spot diagram across a circular 10◦ full field-of-view in the 5
diopter power configuration. The evaluated GAL is defined in Table 3.1 with zero tilt. The
spots reflect the freeform aberration content of GALs with on-axis coma and trefoil and
off-axis aberrations that are not symmetric with field. Symmetry about the shift axis y is
maintained, same as for the field performance in Fig. 3.10.

are largely constant with field, change in magnitude with power variation, and
are oriented along the shift axis y. Meanwhile, in both GALs and SALs, off-axis
fields suffer from the same freeform aberrations, namely field conjugate, field lin-
ear astigmatism followed by field linear, medial field curvature (see Fig. 3.12).
The two aberrations have linear nodes that are orthogonal to one another and ori-
ented along the Cartesian axes. For the most part, the astigmatism is constant with
power variation while the field curvature changes with variation. Due to the lack-
ing off-axis performance, these freeform aberrations are the key to performance
improvement in optimized GALs, as is the focus of Sec. 3.3.
Second, the change in wavefront performance with power range is considered.
For the GAL in Table 3.1 with no tilt, eight different values for ∆Φ are evaluated.
With mechanical dimensions t, D, and CA fixed, the increase in ∆Φ is achieved by
a corresponding increase in ∆n according to Eq. (3.23). For the different values
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 81

Field Curvature (Z4) Astigmatism (Z5/Z6)


5 5

Y field angle [deg]

Y field angle [deg]


Φ = -5.0 D 0 0

-5 -5
-5 0 5 -5 0 5
X field angle [deg] X field angle [deg]
Y field angle [deg] 5 5

Y field angle [deg]


Φ = -2.5 D 0 0

-5 -5
-5 0 5 -5 0 5
X field angle [deg] X field angle [deg]
5 5
Y field angle [deg]

Φ = 0.0 D 0 Y field angle [deg] 0

-5 -5
-5 0 5 -5 0 5
X field angle [deg] X field angle [deg]
5 5
Y field angle [deg]

Y field angle [deg]

Φ = 2.5 D 0 0

-5 -5
-5 0 5 -5 0 5
X field angle [deg] X field angle [deg]
5 5
Y field angle [deg]

Y field angle [deg]

Φ = 5.0 D 0 0

-5 -5
-5 0 5 -5 0 5
X field angle [deg] X field angle [deg]

5 waves 9 waves

Figure 3.12: Field curvature and astigmatism full field displays across GAL power varia-
tion. The evaluated GAL is defined in Table 3.1 with zero tilt. Field curvature is calculated
with best-fit power removed according to the on-axis field.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 82

Figure 3.13: GAL power range ∆Φ versus on-axis RMS WFE, shown across element shift
δ. The evaluated GAL is defined in Table 3.1 with zero tilt. The black dashed line shows
the Maréchal criterion for the diffraction limit.

of ∆Φ, the RMS WFE for the on-axis field is plotted versus element shift δ in Fig.
3.13. The results show a clear trend of wavefront aberration increasing with ∆Φ
and in equal amounts across δ. All designs perform best at δ = 0 with zero power.
RMS WFE also increases in equal amounts with ±δ for either positive or negative
power. Performance is also compared against the diffraction limit according to the
Maréchal criterion (black dashed line in Fig. 3.13). Only ∆Φ = 5 D is diffraction-
limited across the full power range while the standard GAL listed in Table 3.1 with
∆Φ = 10 D is only diffraction-limited over the central region of the power range.
Third, the same process is repeated for bias power to assess the change in wave-
front performance for Φbias 6= 0 but a fixed power range ∆Φ. The results are shown
in Fig. 3.14 where the bias power is shown to change the position of the minimum
wavefront aberration without significantly changing the error magnitude. The ele-
ment shift δ corresponding to this minimum error occurs at the position with zero
power, Φ = 0. Consequently, non-zero bias powers yield greater peak-to-valley
(PV) WFE across the power range than the standard GAL configuration with zero
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 83

Figure 3.14: GAL bias power Φbias versus on-axis RMS WFE, shown across element shift
δ. The evaluated GAL is defined in Table 3.1 with zero tilt. The black dashed line shows
the Maréchal criterion for the diffraction limit.

bias power. Φbias 6= 0 also requires a larger ∆n, as found by Eq. (3.34).
For both power range ∆Φ and bias power Φbias , it is identified that WFE in-
creases with GAL power Φ. Since all examples consider the same clear aperture
CA = 10 mm defined in Table 3.1, the change in performance is ultimately linked
to the GAL f-number rather than any specific power range or bias. As power is
varied with element shift δ, the GAL f-number changes according to

1 1
f /# = = (3.35)
Φ (δ) · CA 4At (δ − y0 ) CA

where Eq. (3.32) is inserted for Φ (δ). Faster systems with smaller values for f /#
are found to suffer from greater aberrations while slower systems with larger f /#
perform better. This is expected as the same trend appears in classical lens design.
Fourth, the GAL design in Table 3.1 is interrogated for performance changes
with element thickness t. It is found in Sec. 3.2.3 that the generated power is
proportional to the product of t and ∆n. As a result, a decrease in t also requires
a proportionate increase in ∆n, and vice versa. This inverse relationship is shown
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 84

Figure 3.15: GAL element thickness t versus on-axis RMS WFE, averaged across power
variation. For the change in t, the corresponding change in GRIN ∆n is also shown that
maintains the GAL power variation defined in Table 3.1. The black dashed line shows
the Maréchal criterion for the diffraction limit. Performance deteriorates monotonically with
increasing t and decreasing ∆n.

in Fig. 3.15 where ∆n approaches infinite as t approaches zero. The change in


RMS WFE is also shown in Fig. 3.15 for changing element thickness. Wavefront
aberration increases monotonically with element thickness t. This agrees with the
prior conclusion that aberrations are induced by the finite GAL element thickness
as they are induced in SALs by the finite separation between freeform surfaces.
By confirming the connection between t and imaging performance, the important
conclusion to be drawn is that GALs should ideally use the maximum allowable
∆n in order to reduce t. Doing so maintains the same power variation but with
vastly improved performance.
Fifth, the ratio of the GAL element diameter D to the component clear aperture
CA is analyzed for its effect on WFE. Previously, Fig. 3.7 shows that, for a fixed
GRIN ∆n, the GAL power range is maximized when D/CA = 1.5. Likewise, for
a fixed power range, the GRIN ∆n is minimized when D/CA = 1.5, as shown in
Fig. 3.16. Nevertheless, this finding does not address how imaging performance
changes with D/CA. In the following analysis, CA is held constant while D is
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 85

Figure 3.16: GAL diameter to clear aperture ratio D/CA versus on-axis RMS WFE, aver-
aged across power variation. For the change in D/CA, the corresponding change in GRIN
∆n is also shown that maintains the GAL power variation defined in Table 3.1. The black
dashed line shows the Maréchal criterion for the diffraction limit. Performance improves
monotonically with increasing D/CA while the ∆n is minimized for D/CA = 1.5 and in-
creases beyond that.

changed, leaving the f-number unaffected. In order to maintain a constant power


range, the ∆n is also appropriately adjusted. The resultant imaging performance
in terms of D/CA is also shown in Fig. 3.16. It is clear that WFE improves as the
D/CA ratio increases. For a fixed power range ∆Φ, a change in D/CA imparts
a change in the rate of power variation per element shift δ. With small values of
D/CA, small element shifts result in rapid changes in power, necessitating stronger
refractive index gradients and yielding worse performance. The GRIN ∆n shown
in Fig. 3.16 finds that beyond its minimum at D/CA = 1.5, ∆n increases slowly
with D/CA in order to maintain the same power variation. From these results,
the key takeaway is that for improved performance, it is best to maximize D/CA
while remaining within fabrication constraints on D and ∆n.
Sixth, the change in WFE is considered for different values of the refractive
index tilt coefficient B. The tilt coefficient is quantified by its ratio with the tilt
value minimizing the refractive index change, B/B0 . The resultant performance
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 86

Figure 3.17: GAL tilt coefficient versus on-axis RMS WFE, averaged across power vari-
ation. The tilt is quantified by its ratio with the tilt value minimizing the refractive index
change, B/B0 . For the change in B/B0 , the corresponding change in GRIN ∆n is also
shown. The black dashed line shows the Maréchal criterion for the diffraction limit. WFE is
at a minimum for zero tilt and only moderately degraded with B/B0 = 1 for minimum ∆n.

along with the change in ∆n is shown in Fig. 3.17. WFE is minimized with zero
tilt, B/B0 = 0, but increases in equal amounts with either positive or negative tilt
values. The change in performance with non-zero tilt such as for B/B0 = 1 is
not as significant compared with the prior analyses (note Figs. 3.13 – 3.17 are all
presented on the same scale). As a result, the 2.4x decrease in ∆n for B/B0 = 1
compared with B/B0 = 0 typically outweighs the moderate loss in performance,
especially when power range is sought to be maximized.
The conclusions drawn from the explored relationships between different pa-
rameters and WFE are beneficial when designing with GALs, such as in Sec. 3.4.
The findings are performed for a base GAL defined by the refractive index in Eq.
(3.29). In Sec. 3.3, GALs are optimized with higher-order refractive index terms to
improve aberration correction.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 87

Boresight error

In addition to WFE, GAL imaging performance is also affected by boresight error


(BSE). BSE describes a change in position and direction of the ray traveling along
the optical axis (i.e., normally incident on the GAL front surface) [111]. As a re-
sult, BSE alters the line-of-sight of an instrument compared with its mechanical
“pointing” axis. This effect is detrimental to certain imaging applications such as
riflescopes, the focus of Sec. 3.4.
For finite and infinite conjugate systems, BSE results in a transverse image shift
away from the optical axis due to a transverse shift in ray coordinates. On the
other hand, in afocal systems like riflescopes, BSE alters the angular position of
the observed image due to a shift in ray angle. When assessing visual systems like
riflescopes, a helpful target is for angular BSE to be less than one arcminute, which
falls beyond the visual acuity of the eye [115]. Regardless of conjugate, BSE does
not result in a loss in image resolution, unlike WFE.
BSE is nominally absent in rotationally symmetric systems. On the other hand,
by breaking rotational symmetry, both GALs and SALs introduce BSE. Moreover,
the imparted BSE also changes in magnitude with element shift and power varia-
tion, making its correction difficult. Variable BSE correction requires an additional
dynamic optical component to provide an equal and opposite BSE contribution.
In this work, BSE is calculated using the ray passing through the centroid of
the geometrical spot for the on-axis field, as determined by CODE V. The centroid
ray is used rather than the chief ray due to the presence of on-axis coma (see Fig.
3.11). Due to GAL symmetry about y, this centroid ray is required to lie in the y-z
plane. Then, BSE can be quantified either angularly or spatially using this centroid
ray’s trajectory. Angular BSE is calculated by the signed angle of the centroid ray
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 88

relative to the optical axis. Spatial BSE is found from the y-offset of the centroid ray
from the optical axis in the best-focus image plane (either real or virtual depending
on the power sign). For this reason, spatial BSE is undefined for the zero power
configuration where no image is formed.
The primary factor influencing the magnitude of BSE is the GAL tilt coefficient
B. Both angular and spatial BSE are shown in Fig. 3.18 across element shift δ for
different tilts coefficients, quantified by B/B0 . For any tilt value, BSE is seen to
change with element shift, entailing that BSE changes in magnitude with power
variation.
For angular BSE, zero GRIN tilt, B/B0 = 0, introduces BSE that increases ap-
proximately linearly and with positive slope, as found in Fig. 3.18(a). Changing to
B/B0 6= 0 affects this slope where positive values introduce even greater PV BSE.
(Recall, B/B0 = +1 yields the minimum GRIN ∆n.) On the other hand, B/B0 < 0

Figure 3.18: GAL tilt coefficient versus (a) angular and (b) spatial boresight error (BSE),
shown across element shift δ. The GRIN tilt is quantified by its ratio with the tilt value mini-
mizing the refractive index change, B/B0 . BSE is calculated from the ray passing through
the centroid of the geometrical on-axis spot. For spatial BSE, the ray height is calculated
in the best-focus image plane, which is nonexistent for the zero power configuration, δ = 0.
The evaluated GAL is defined in Table 3.1.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 89

tilts the curve toward zero BSE. From the five demonstrated cases, B/B0 = −1
yields the least PV angular BSE. Unfortunately, this tilt also results in greater GRIN
∆n according to Fig. 3.5.
Meanwhile, spatial BSE changes approximately quadratically with element shift,
as can be seen in Fig. 3.18(b). This spatial BSE variation occurs with about equal
PV variation across δ, regardless of GRIN tilt. On the other hand, the baseline
offset of BSE variation does depend on the GRIN tilt. For B/B0 = +2, the on-axis
image spots across power variation lie closest to the optical axis. For the worst case
scenario of B/B0 = −2, the on-axis image spots are offset ≈ 45 µm from the opti-
cal axis. Recall, this 45 µm offset is in the best-focus image plane, which is located
100’s of millimeters from the GAL pair.
The angular and spatial BSE evaluated in this section is for a GAL pair in iso-
lation. Although, when incorporated in an optical design with other components,
the total system BSE depends on the GAL BSE as well as the effects of surround-
ing elements. For example, a zoom riflescope design using GALs is demonstrated
in Sec. 3.4 where the total system BSE depends on the effects of two GAL pairs
as well as several fixed power elements. As a result, for BSE specifications on an
entire system, there is no exact requirement on the GAL BSE.

3.2.5 Fabrication, alignment, and metrology

A base GAL design is fabricated to verify the analytical power variation derived
in Sec. 3.2.3. The two GAL elements with their freeform refractive index change
are fabricated via polymer additive manufacturing by Nanovox LLC. The specifi-
cations for the fabricated GAL are listed in Table 3.2. One of the 3D printed GAL
elements can be seen in Fig. 3.19(a) with a grid placed in the background, demon-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 90

Parameter Value
Element diameter, D [mm] 12.0
Clear aperture, CA [mm] 3.0
Maximum element shift, δmax [mm] 4.5
Element thickness, t [mm] 3.0

Power range, ∆Φ [D] 62.0


Alvarez coefficient, A [mm−3 ] 5.74524 × 10−4
Tilt coefficient, B [mm−1 ] 0.0
Bias power, Φbias [D] 0.0
Coordinate offset, y0 [mm] 0.0

Base refractive index, n0 1.4765


GRIN ∆n 0.117

monochromatic,
Spectrum
λd = 587.56 nm
Table 3.2: Fabricated base GAL design specifications.

strating the freeform nature of the GRIN profile. A translation stage custom ma-
chined by Nanovox houses the elements and allows for precise element shift δ, as
shown in Fig. 3.19(b).
Upon receiving the GAL pair, the orientation of each element is examined with
the aid of interferometry. A clocking adjustment is made for each elements to
ensure the F-GRIN and translation axes are aligned. The Mach-Zehnder intefer-
ograms for the two individual GAL elements can be seen in Fig. 3.20 after the
clocking adjustment is performed.
After performing the clocking alignment, the two elements are returned to the
translation stage. The joint effect of the two elements upon translation is then ob-
served in the Mach-Zehnder interferometer. The results are shown in Fig. 3.21
for three element shifts, yielding three different power configurations. The in-
troduction of power fringes is apparent with δ 6= 0. Although, it is difficult to
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 91

(a) (b)

2 mm

Figure 3.19: GAL fabricated via additive manufacturing by Nanovox LLC. (a) A single
GAL element is shown with a grid in the background from which freeform refractive index
change can be noticed. (b) Custom machined housing is used for precision GAL element
translation.

1 mm 1 mm

Figure 3.20: Mach-Zehnder interferograms of the fabricated GAL elements after a clocking
adjustment is performed to align elements with the translation axis.
δ = -3 mm δ = 0 mm δ = +3 mm

1 mm 1 mm 1 mm

Figure 3.21: Mach-Zehnder interferograms of the fabricated GAL for different element
shifts δ. The red circle indicates the 3 mm clear aperture.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 92

quantitatively assess the power due to aberrations, mid-spatial frequency errors in


fabrication, and high fringe density.
Instead, the power variation is measured indirectly by determining the mag-
nification for an image formed by a GAL. From first-order optics, the power of a
component Φ can be related to the transverse magnification m by

1−m
Φ= (3.36)
ml

where l is the object distance from the front principal plane of the component. Since
the GAL principal planes are undefined, it is assumed that l is approximately equal
to the object distance from the GAL front surface. The object distance is measured
with a ruler, l = −55.0 ± 1.0 mm. An example of change in image magnification
with GAL power variation is shown in Fig. 3.22.
The image magnification is measured with a backlit USAF bar target serving as
the object. Group 0, element 6 is used when measuring positive power. Group 0,
element 1 is used for negative power. Different target elements are used to ensure
their entirety remains enlarged but within the field-of-view for either positive or

δ = -2 mm δ = -1 mm δ = 0 mm δ = +1 mm δ = +2 mm

10 mm

Figure 3.22: GAL power variation qualitatively demonstrated by change in image magni-
fication with element shift δ. In this example, the letter “G” serves as the object and is of
fixed size. The magnified “G” image for different δ is the same modality used in the quanti-
tative power measurement shown in Fig. 3.23 for the USAF bar target.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 93

negative GAL power. By doing so, measurement precision is improved by increas-


ing the number of pixels spanning the magnified images. For example, Fig. 3.22
shows the change in magnification with both positive and negative GAL power
variation for a “G” objective of fixed size. The “G” object spans significantly fewer
pixels when δ = −2 mm than it does when δ = +2 mm. The use of two USAF target
elements adjusts for this effect.
In performing the measurement, a camera is focused on a target element with
no GAL present to determine the object size h, measured in pixels. Then, the GAL
is inserted, and the camera observes the target element with an image size h0 re-
sulting from the GAL power, again measured in pixels. From these quantities, the
transverse magnification is determined by their ratio m = h0 /h. The GAL forms
a virtual image of the target which changes in both size as well as position with
power variation. However, the camera is not refocused with GAL power variation
due to the non-parfocality (i.e., “focus breathing”) of the imaging lens. Without
parfocality, refocusing the camera lens would erroneously alter the magnification
imparted by the GAL power.
The GAL power measurement is repeated three times for performing error
analysis. The final results can be seen in Fig. 3.23. Shown along side the first-order
result of Eq. (3.20), the measured GAL power agrees closely with the derived the-
ory. The change in power with element shift δ proves to be approximately linear.

3.3 Optimized GAL designs

In SALs, performance improvement has been demonstrated by incorporating higher-


order spatial terms in the two freeform surfaces [100, 108, 110]. To access greater
degrees of freedom, the two freeform surfaces also become decoupled, allowing
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 94

Figure 3.23: Fabricated GAL power variation measured by change in image magnification.
The first-order power according to Eq. (3.20) for this GAL is shown for comparison. There
is close agreement between measured and first-order power.

for different contributions of higher-order terms. As a result, the two surfaces are
no longer equivalent in form. Although, with only two surfaces, SAL performance
improvement is limited in power range and/or field-of-view. One reason for this
is the freeform influence of the surfaces is limited to two discrete longitudinal po-
sitions.
F-GRIN’s volumetric degrees of freedom offer further control over aberrations
[14, 15]. The correction of induced aberrations in GALs in studied next. The GALs
considered in Sec. 3.2 consists of “base” (i.e., unoptimized) designs where the
refractive index is governed by three transverse spatial terms in Eq. (3.9). By in-
corporating higher-order transverse and longitudinal refractive index terms, per-
formance can be significantly improved.
In this section, optimized GAL designs are examined that apply higher-order
terms via three different methods. First, in Sec. 3.3.1, a more conventional ap-
proach is taken where the refractive index is described by a three-dimensional
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 95

polynomial with coefficients that are obtained by optimization. Then, in Sec. 3.3.2,
a novel approach is presented that relies on a hybrid polynomial-discrete method
of F-GRIN optimization. Lastly, a design consisting of a series of transversely vary-
ing thin GRIN plates is demonstrated.

3.3.1 Higher-order polynomial designs

The three transverse terms in the base GAL refractive index in Eq. (3.9) are y 3 , x2 y,
and y. In devising a polynomial basis with additional higher-order terms, these
three base terms must be readily included. Thus far, Zernike polynomials have
primarily been relied on for implementing transverse refractive index change in F-
GRIN [14, 15, 116]. While the three base GAL terms can be obtained by the proper
combination of Zernike terms in certain proportions, it is not ideal for multiple
Zernike terms to be coupled. Also, since GAL elements are translating along y, it
is beneficial to have direct control over terms varying parallel or perpendicular to
the shift axis.
For these various reasons, the polynomial basis applied in this work is the
three-dimensional XYZ polynomial [15]. The mathematical representation for the
XYZ polynomial is

X i j k
n (x0 , y 0 , z 0 , λ) = n0 (λ) + ∆n (λ) cijk (x0 ) (y 0 ) (z 0 ) . (3.37)
i,j,k

Assuming a two material composition consisting of n1 (λ) and n2 (λ), the spectral
components are defined the same as by Yang et al. [14],

|n1 (λ) + n2 (λ)| |n1 (λ) − n2 (λ)|


n0 (λ) = , ∆n (λ) = . (3.38)
2 2
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 96

As in [14], ∆n (λ) describes the allowable material index range, not necessarily the
index range used in any given design.
Meanwhile, the spatial refractive index variation within the summation in Eq.
(3.37) is independent of wavelength. Within this summation, cijk are scalar coef-
ficients, and the spatial coordinates are defined in terms of Cartesian coordinates
by
x0 = x − x 0 , y 0 = y − y0 , z 0 = z − z0 (3.39)

where (x0 , y0 , z0 ) is the polynomial’s coordinate origin. The coordinate shift y0 is


also the same coefficient that provides power bias, as described in Sec. 3.2.3. (Note,
the spatial coordinates x0 ,y 0 ,z 0 may also be normalized over the unit cylinder as
done in [14]. The following design work implemented in CODE V does apply
normalized coordinates, but the mathematical description presented here is unnor-
malized in order to remain consistent with the analysis performed in Sec. 3.2.)
For an mth order XYZ polynomial, it is required that

i + j + k ≤ m, (3.40)

which allows for a finite number of spatial terms N ,

3
1 Y
N (m) = (m + i) . (3.41)
3! i=1

The XYZ polynomial is implemented in CODE V as a user-defined gradient, which


currently is limited to 150 refractive index coefficients (see Appendix A.2 for source
code) [72]. Accordingly, up to 7th order XYZ polynomial terms can be accommo-
dated, allowing for 120 terms by Eq. (3.41), as listed in Table 3.3. From these 120
terms, the three base GAL terms can be recognized, providing a relationship be-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 97

tween the coefficients A and B from Sec. 3.2 and the XYZ coefficients cijk ,

A
c030 = , c210 = A, c010 = B. (3.42)
3

For higher-order polynomial GAL design, coefficient values for all 120 terms
may be obtained by optimization. Although, optimization convergence can be im-
proved by reducing the problem size via the exclusions ineffective terms [117].
There are different approaches one can take to determine which terms do not pro-
vide significant performance improvement and can be excluded in optimization.
First, base GALs possess y-z symmetry. It is unlikely that breaking symmetry
will provide performance improvement since additional field dependent aberra-
tions would require correction. To maintain y-z symmetry in higher-order designs,
all terms that are odd-order in x are excluded. For the 120 terms listed in Table 3.3,
50 are odd-order in x. The x coordinate shift is also set to zero, x0 = 0, to maintain
y-z symmetry.
Next, it is found in Sec. 3.2.1 that the wavefront imparted with element shift is
related to the refractive index slope along y. In order to impart an even-order (i.e.,
symmetric) wavefront such as power, the refractive index variation along y must
be odd-order. For example, the y 2 term imparts a tunable amount of wavefront tilt
in y, yielding unwanted BSE. As a result, all terms even-order in y are excluded.
For the 120 terms listed in Table 3.3, 34 are even-order in y, not counting factors of
y0.
After removing all terms that are odd-order in x or even-order in y, 50 poly-
nomial terms remain, as indicated in bold in Table 3.3. From these 50 terms, an
analysis is performed to identify which provide significant performance improve-
ment in GALs. The standardized base GAL with specifications listed in Table 3.1
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 98

Table 3.3: XYZ polynomial terms up to 7th order. In total, 120 terms are considered for
higher-order polynomial GAL design. 50 terms (bold) are GAL-compatible as even-order
in x and odd-order in y. 13 select terms (green) are found to be effective in improving GAL
performance.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 99

is again considered. Going one term at a time, the base GAL is loaded and a sin-
gle higher-order term is optimized in CODE V to minimize ray aberrations across
both power variation and the 10◦ circular full field-of-view. In each case, the term’s
coefficient cijk is free to differ between the two GAL elements. The GRIN refractive
index range is unconstrained in optimization for the sake of identifying effective
terms. For each optimized design, the RMS WFE is recorded across power varia-
tion and field. The change in GRIN ∆n is also monitored.
One caveat in this analysis is the assumption that each term’s influence is inde-
pendent from that of other terms. A second assumption is that the standardized
GAL in Table 3.1 is representative of other GALs when it comes to higher-order
terms and performance improvement. For example, some terms may be more
helpful in thicker GAL elements or across a larger field-of-view, effects that are
overlooked in this analysis. Finally, four terms are excluded from this optimiza-
tion analysis: the three base Alvarez terms (y 3 , x2 y, y) and the piston index term
(1), which is a redundant variable with n0 (λ) and ∆n (λ).
From the results of this analysis, select higher-order terms are identified that
provide GAL performance improvement. In particular, 10 higher-order terms are
recognized as having a significant effect on improving RMS WFE averaged across
the power variation and field. The percent RMS WFE improvement compared with
the base GAL design is shown in Fig. 3.24 for each term, sorted by improvement.
Notably, all 10 terms possess axial refractive index variation, clearly demonstrating
its value in correcting field dependent aberrations [14, 15]. The axial variation is
also seen to be odd-order for all 10 terms. The percent increase in GRIN ∆n is also
shown in Fig. 3.24 for the 10 select terms. The terms with exclusively longitudinal
variation (z, z 3 , z 5 ) are found to have significantly larger increases in ∆n. Due to
fabrication constraints, these terms may not be as effective in GAL design.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 100

Figure 3.24: 10 select higher-order terms identified as providing significant GAL perfor-
mance improvement. RMS WFE improvement from the base design is averaged over both
power variation and field-of-view. The percent increase in ∆n is shown for the same terms.

From the 13 XYZ polynomial terms of interest – 3 base and 10 higher-order


– a higher-order polynomial GAL design is optimized in CODE V. Starting from
the base design in Table 3.1, the 13 terms are varied independently between the
two GAL elements. The term coefficients are obtained by optimization on mini-
mizing ray aberrations across power variation and field-of-view. The optimization
requires careful inspection to ensure favorable convergence. Due to the highly
nonlinear design space, derivative increments for each coefficient are calculated
by finite difference for each optimization cycle. The only constraint in optimiza-
tion is that the ∆n of each element is held less than twice the base design’s value,
∆n < 0.132. The ∆n constraint is imposed in CODE V using a penalty cost
(PTC) where the total refractive index change is calculated from a dense three-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 101

Element 1 Element 2

Figure 3.25: Optimized higher-order polynomial GAL design. Refractive index profiles are
shown in y-z and x-y cross-sections, all on the same colorbar scale. RMS WFE perfor-
mance for this design is shown in Fig. 3.26.

dimensional sampling of index values.


The refractive index profile for the optimized GAL design is shown in Fig. 3.25.
While the two elements are allowed to differ within optimization, comparing their
cross-sections finds quasi-symmetry remains, an indicator that the optimization
has converged on a favorable design. From inspecting individual higher-order co-
efficients, the two most impactful terms in both elements are x2 yz and x2 y 3 z, both
of which change linearly in z. Meanwhile, the terms z, z 3 , z 5 are absent from both
elements, even though included in optimization. This is likely due to the ∆n con-
straint and these terms’ large contributions according to Fig. 3.24. The optimized
design’s total refractive index change ∆n = 0.122 confirms the optimization con-
straint is met.
The optimized design’s RMS WFE across the field-of-view is depicted in Fig.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 102

Figure 3.26: RMS WFE across a circular 10◦ full field-of-view for three different power con-
figurations for the higher-order polynomial optimized GAL design in Fig. 3.25. Diffraction-
limited performance is found in an enlarged region in the center of the field-of-view that
changes in position with power variation. RMS WFE is shown on the same scale as in Fig.
3.10 for the base GAL design.

3.26 through power variation. Shown on the same scale as in Fig. 3.10 for the base
design, significant improvement is found with a 2.3x reduction in RMS WFE at
the edge of the field. A mere 1.1x improvement in WFE is found in the center of
the field; although, this region of best performance is enlarged. In the optimized
design, the center of the high-performance region is now offset from the on-axis
field when Φ 6= 0.
The higher-order polynomial GAL design obtained by optimization provides
significant improvement in performance. Since designed in CODE V, higher-order
polynomial GALs can be readily integrated into larger optical systems and jointly
optimized with other components. In this way, F-GRIN polynomial optimization
is utilized in Sec. 3.4 for the design of a zoom riflescope; however, other techniques
can also be applied for GAL optimization in custom software, as described next in
Sec. 3.3.2.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 103

3.3.2 Hybrid polynomial-discrete designs

Optimization via a polynomial basis is the current standard for GRIN design. The-
oretically, orthogonal three-dimensional polynomials such as the Zernike-Legendre
polynomial [14] can produce any continuous refractive index distribution, although
an infinite number of terms are required. On the other hand, optimization con-
verges most readily when the fewest number of variables tolerable are considered.
For this reason, a finite number of polynomial terms (e.g., 150 maximum in CODE
V) cannot capture all orders of refractive index change in optimization. A second
limiting factor is that the coordinate origin about which the polynomial basis is de-
fined dramatically affects what refractive index distributions are attainable given a
finite number of terms. Consequently, many optimal refractive index distributions
are overlooked when designing GRIN via polynomials.
An alternative for GRIN design that has not been explored is to use local refrac-
tive index representations. Unlike a polynomial defined continuously through-
out a volume, local definitions consider discrete refractive index values in three-
dimensional space. The refractive index distribution is then obtained continuously
from these discrete values by some assumption of how the index changes between
points.
Previously, freeform surface designs have leveraged local surface sag defini-
tions such as non-uniform rational B-splines (NURBS) [118] and basis function in-
terpolation [119]. Both techniques could be adapted for use in GRIN design where
an expansion to four dimensions (three in space and one in refractive index) is
needed.
The value of local GRIN representations is that there is no restriction on what
refractive index can be defined. The downside is that their optimization poses a
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 104

significantly more challenging numerical problem. Polynomial GRIN design in


CODE V optimizes at most ∼ 102 coefficients. Meanwhile, optimizing a relatively
small GRIN with dimensions 1 mm × 1 mm × 1 mm defined on a uniform grid with
10 µm pitch requires optimization of 106 coefficients, a four order of magnitude in-
crease in complexity! This poses a daunting problem of how locally represented
GRIN can be efficiently designed via optimization. This problem is also exacer-
bated in GRIN compared with optical surfaces since an extra spatial dimension
must be described. For example, in Chrisp’s work [118] freeform surfaces are de-
signed with ∼ 103 NURBS control points.
Besides the numerical size of this design problem, local GRIN definitions must
also satisfy requirements imposed by fabrication techniques. For example, the
smoothness of the refractive index distribution is often desired for fabrication as
well as metrology. As in polynomial designs, constraining the GRIN material com-
position and allowable refractive index range is also a requirement. On top of
that, GRIN dispersion adds another dimension that further increases the numeri-
cal complexity.
Faced with these challenges, a simple yet effective initial approach is to con-
sider a hybrid GRIN representation, consisting of both polynomial and local defi-
nitions. In this work, a polynomial basis is used to define the transverse refractive
index change while a local definition is used axially. The decision to use the local
definition along z rather than x or y is due to the influence of axial polynomial
terms that is discovered for GALs in Sec. 3.3.1. More generally, the axial index
change is often the least well-posed in polynomial optimization since it has a sub-
tler effect with rays traveling close to parallel to the axial gradient. Many GRIN
designs also have smaller axial dimensions than transverse ones.
The hybrid GRIN representation defines the refractive index in a series of N
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 105

y
x

y Δz
z
t
Figure 3.27: Hybrid polynomial-discrete GRIN representation. In this example, the re-
fractive index is defined in N = 21 discrete planes separated by ∆z. In each plane, the
transverse refractive index ni (x, y) is defined continuously by a two-dimensional polyno-
mial. Between planes, the axial refractive index is defined continuously by interpolation.

planes that are equally-spaced along z, as shown in Fig. 3.27. The spacing of these
planes has pitch ∆z = t/ (N − 1) where t is the GRIN center thickness between the
first and last plane, which are assumed to be the exterior surfaces. The refractive
index in each plane, ni (x, y), is governed by a two-dimensional polynomial with
M coefficients. The polynomial coefficients are free to differ between planes, which
allows for axial refractive index change. In total, this hybrid GRIN definition is
fully contained in a two-dimensional matrix Θ with dimensions N × M .
While this hybrid GRIN is defined discretely in z, the refractive index is gov-
erned volumetrically where it is assumed that the index changes continuously be-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 106

tween planes. In other words, this hybrid approach is not equivalent to a discrete
series of stacked plates. Instead, the refractive index is interpolated between dis-
crete planes. For ray tracing, the refractive index n and its gradient ∇n must be
known throughout a volume [19–21]. To find their values, the refractive index is
linearly interpolated, making it axially continuous but not necessarily smooth. The
ray tracing step size ∆t can also be dynamically adjusted for each ray step in or-
der to take axially constant steps ∆z between the defined planes. This step size
adjustment is implemented by
∆z
∆t = (3.43)
Γ

where Γ is the optical direction cosine along z for the current ray position.
As explained previously, numerical complexity is a concern with local refrac-
tive index definitions. The outlined hybrid GRIN model and ray tracing procedure
is devised with computational efficiency in mind. Ray tracing and optimization
can be formulated via a series of matrix multiplication for each ray step and can
be done in a parallelized fashion for tracing multiple rays at once. There are three
reasons why matrix multiplication can be leveraged. First, the GRIN definition is
readily stored in a two-dimensional coefficients matrix Θ. Second, the refractive
index is being linearly interpolated, which can be readily performed by matrix
multiplication. Third, the dynamic step size ∆t ensures that all rays take the same
number of steps for a GRIN with plane-parallel surfaces.
The described ray tracing procedure is validated by results obtained from CODE
V for different types of GRIN including those with freeform and/or axial index
change. With no axial variation, ∂n/∂z = 0, the results exactly match those of
CODE V, regardless of the transverse profile. With axial index change present,
∂n/∂z 6= 0, the ray trace errs from CODE V but on a sub-micron level for the ex-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 107

amples considered in this work. The sources of error for axial index variation are
due to the approximations inherent to Eq. (3.43) as well as the linear interpolation
along z.
The hybrid GRIN definition is also readily optimized. In this work, the opti-
mization cost function J is constructed from three parts,

J = JRA + J∆n + Jsmooth . (3.44)

The first cost term JRA is defined by the L2 norm (i.e., root sum square) of the
ray aberrations ~ε, either transverse or angular, summed over rays spanning the
field-of-view and pupil,
XX
JRA = wRA k~εk2 (3.45)
F OV pupil

where wRA is a user-defined weighting on the ray aberration contribution. The


second cost term J∆n constrains the GRIN total refractive index change,

 64
∆n
J∆n = w∆n , (3.46)
∆nmax

where w∆n is a user-defined weighting on the constrained index change ∆nmax .


The form of J∆n is the same as the penalty constraint (PTC) found in CODE V [72],
which is used for the polynomial design in Sec. 3.3.1. The third cost term Jsmooth
enforces smoothness in the refractive index between adjacent planes. This is done
by weighting the L2 norm of the GRIN coefficients between neighboring planes,
Θi and Θi+1 ,
N
X −1
Jsmooth = wsmooth kΘi+1 − Θi k2 (3.47)
i=1

where wsmooth is a user-defined weighting on the smoothness constraint.


Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 108

Parameter Value
Focal length, f [mm] 100.0
Entrance pupil diameter [mm] 25.0
Circular full field-of-view [deg] 10.0
Spectrum monochromatic

Center thickness, t [mm] 5.0


Total refractive index change, ∆n 0.156

Number of planes, N 51
Plane spacing, ∆z [mm] 0.1
Number of polynomial terms 4 (r2 , r4 , r6 , r8 )
Number of GRIN coefficients 204
Table 3.4: Specifications of the f /4 Wood lens starting point for the optimized design using
the hybrid polynomial-discrete model.

With ray tracing being performed by matrix multiplication, automatic differ-


entiation can be applied to calculate the derivatives of the cost function with re-
spect to the GRIN coefficients, ∇Θ J. Having access to the cost function gradient
enables derivative-informed optimization algorithms that offer improved conver-
gence when solving for Θ [117]. In this work, the python library PyTorch is used
to perform automatic differentiation, and the Broyden–Fletcher–Goldfarb–Shanno
(BFGS) algorithm is used for optimization [117].
Before applying the hybrid continuous-discrete model to GAL design, the de-
scribed ray tracing and optimization procedure is validated for the simpler case
of a Wood lens [18]. An f /4 Wood lens is used as the optimization starting point,
as shown with ray aberration plots in Fig. 3.28(a). From this starting point, axial
refractive index change as well as higher-order transverse terms are optimized for
improved performance. The design specifications are outlined in Table 3.4, includ-
ing the parameters used in defining the hybrid representation.
From the Wood lens starting point, a hyperparameter search is performed to
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 109

(a) Wood lens (b) Optimized design

Y-FAN X-FAN Y-FAN X-FAN


Field angle Field angle
(0.0°, 5.0°) (0.0°, 5.0°)

Field angle Field angle


(0.0°, 4.3°) (0.0°, 4.3°)

Field angle Field angle


(0.0°, 3.5°) (0.0°, 3.5°)

Field angle Field angle


(0.0°, 2.5°) (0.0°, 2.5°)

Field angle Field angle


(0.0°, 0.0°) (0.0°, 0.0°)

Figure 3.28: Refractive index profile and transverse ray aberration plots for (a) the
Wood lens defined in Table 3.4 and (b) the optimized design represented by the hybrid
polynomial-discrete model. Starting from the Wood lens design, the optimized design in-
corporates axial refractive index change and higher-order transverse variation to correct
spherical aberration and coma. The optimized design is ultimately limited by astigmatism.
The ray aberrations for both designs are shown on the same scale. The optimized design
has an increase of 0.039 in total refractive index change.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 110

identify appropriate values for the optimization weights wRA , w∆n , and wsmooth .
The constrained value for the total refractive index change is ∆nmax = 0.2, which
allows for a 28% increase compared with the starting design. The optimization
problem solves for 204 variables, and traces 1165 rays in evaluating JRA . The com-
putation time is 1.1 seconds per optimization cycle (iMac, Apple M1, 16 GB RAM),
and converges in 245 cycles. The optimized design is shown in Fig. 3.28(b). The
GRIN profile shows an axial refractive index change is added to the starting Wood
lens for further performance improvement. The optimized ∆n is shown to meet
with constrained value of ∆nmax = 0.2. The smoothness constraint also proves
effective as change between planes appears smooth. The optimized hybrid design
achieves a 2.4x reduction in transverse ray aberration kεk2 , as present in the cost
function term JRA . This performance improvement is also apparent in the trans-
verse ray aberration plots in Fig. 3.28(b) for the optimized design. The optimized
hybrid design incorporates axial refractive index change and higher-order trans-
verse variation to correct spherical aberration and coma. The design is now limited
by astigmatism. This improvement in Wood lens performance confirms the effec-
tiveness of the devised ray tracing and optimization scheme.
Using the same hybrid design process, a polynomial-discrete GAL is designed
for the same specifications as in Sec. 3.3.1. The optimization starting point is the
GAL listed in Table 3.1 with tilt minimizing the ∆n. The design is optimized over
five power configurations to minimize the angular ray aberrations with best fit
power removed for the on-axis field. The resultant design can be seen in Fig. 3.29.
As for the Wood lens, axial refractive index variation is imparted in the hybrid
design for further aberration correction. The axial change is also similar to that
of the polynomial design in Fig. 3.25. The optimized GAL design achieves a 12.8x
reduction in angular ray aberration kεk2 , as present in the cost function component
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 111

Figure 3.29: Optimized GAL elements represented by the hybrid polynomial-discrete


model. The optimized design achieves a 12.8x reduction in angular ray aberration com-
pared with the starting point base GAL.

JRA . This significant performance improvement demonstrates the effectiveness of


the devised ray tracing and optimization scheme in higher-order GAL design.
The hybrid GAL design offers more degrees of freedom for aberration correc-
tion than the polynomial design; however, incorporating them in larger optical
systems, such as with CODE V, proves difficult. For one, the dynamic ray trace step
size needed to trace between defined planes is not currently permitted in CODE V.
Second, the powerful ability to ray trace and optimize GRIN using matrix opera-
tions is not supported in CODE V since customized ray tracing algorithms are not
allowed. Lastly, there is currently a restriction on the number of variables that can
be optimized in CODE V, which limits the capability of a user-defined GRIN with
a hybrid definition. For these reasons, the hybrid design method is not considered
for the zoom riflescope design performed in Sec. 3.4.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 112

3.3.3 Discontinuous plate designs

A third option for optimized GALs is considered that combines the approaches
used in Secs. 3.3.1 and 3.3.2. The three-dimensional polynomial optimized in Sec.
3.3.1 is compatible in CODE V for larger system design. The hybrid polynomial-
discrete method in Sec. 3.3.2 offers additional degrees of freedom for improved
performance but is incompatible with CODE V.
In this section, an intermediate approach is used to design GALs in CODE V
using a series of thin GRIN plates for each element. Each thin plate possesses
refractive index that is only varying transversely. The described plate method is
unlike the hybrid design which defines the refractive index in infinitesimal planes
while linearly interpolating between plates. Also unlike the hybrid design, there
is no enforcement of smoothness between adjacent plates, meaning the resultant
elements may be discontinuous in z.
The plate GAL design is performed in CODE V with the same specifications
as in the previous two sections (see Table 3.1). Due to a restriction on the number
of variables that can be active in CODE V, each GAL element consists of 6 GRIN
plates, each of which has 500 µm thickness. The optimized design refractive in-
dex can be seen in Fig. 3.30. The axial refractive index change is discontinuous,
as expected since there is no active constraint on smoothness between adjacent
plates. Interestingly, the second element resembles those found in Secs. 3.3.1 and
3.3.2 while the first element is significantly different. The RMS WFE performance
versus field-of-view for this design is shown in Fig. 3.31 for three power configu-
rations. At full field, there is a 8x improvement in performance compared with the
polynomial design. This significant improvement yields nearly diffraction-limited
performance across the field.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 113

1.5603

Refractive index
1.5000

1.4397

Figure 3.30: Optimized GAL using a series of transversely varying plates with discontin-
uous axial refractive index change. The optimization is performed in CODE V. The design
performance shown in Fig. 3.31.

Figure 3.31: RMS WFE across a circular 10◦ full field-of-view for three different power
configurations for the discontinuous plate GAL design in Fig. 3.30. Diffraction-limited per-
formance is found across much of the field-of-view. RMS WFE is shown on the same scale
as in Fig. 3.10 for the base GAL design in Table 3.1.

While the excellent performance of the plate design is attractive, the discon-
tinuities along z and limitations in optimization make this design difficult to in-
corporate in the subsequent riflescope design in Sec. 3.4. Fabrication of a plate
design is also more challenging, even for sub-aperture techniques such as additive
manufacturing.
Both the hybrid and plate approaches offer additional degrees of freedom for
axial index change that prove very effective. While neither is used in the ultimate
riflescope design in Sec. 3.4, both options show promise. These new computational
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 114

tools for modeling and optimization, potentially beyond the scope of commercial
software such as CODE V, are important subjects for future work.

3.4 Zoom riflescope designs with GALs

A riflescope is an afocal visual optical system that is mounted on a rifle. The rifle-
scope pointing axis is aligned with that of the rifle barrel. A reticle overlaid in the
scope field-of-view allows for accurate aiming of the rifle at distant objects.
Conventional riflescope designs consist of an objective, a relay, and an eyepiece
(see Fig. 3.32). Intermediate images are found between the objective and relay
(first focal plane, or FFP) as well as between the relay and eyepiece (second focal
plane, or SFP). A reticle can be incorporated at either or both of these intermediate
conjugate planes. Furthermore, the relay module is required to present an erect
image to the eye (without it, a riflescope becomes a Keplerian telescope, which has
an inverted image). Consequently, its role can be served by relay lenses or by an
image inversion prism.
The pupil positions in a riflescope are important to ensure comfortable match-
ing with the human eye. The objective’s pupil is relayed to within the relay and

50 mm

Eyepiece
Objective Zoom Relay
FFP SFP Eyebox

reticle

TTL ER
Figure 3.32: Conventional zoom riflescope design consisting of objective, zoom relay, and
eyepiece. The X-ray image is of the Leupold Mark 6 1-6x20 riflescope, which is used to
devise the specifications in Table 3.5.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 115

then to beyond the eyepiece in image space where the exit pupil can be paired with
the eye. The eye relief ER is measured from the eyepiece exterior surface vertex to
the system exit pupil, also known as the eyebox (see Fig. 3.32). The eye relief is an
important design parameter for a riflescope and must be adequately long to avoid
collision with the eye upon rifle kickback.
Some riflescopes are zoom systems, which allow for a continuous range of an-
gular magnifications M . The image space field at the eye is held constant through
zoom, meaning the change in magnification imparts a change in object space field-
of-view. The zooming object space field presents to the eye different apparent sizes
of an object. The riflescope zoom is typically implemented by a finite conjugate
zoom relay module. Since the eyepiece field is held constant through zoom, the
zoom relay changes the FFP object size while the SFP image remains fixed in size.
If a reticle is positioned at the FFP, it will zoom in apparent size along with the
object space field-of-view, as is the case in Fig. 3.32.
Conventionally, the zoom in the relay is achieved by longitudinally translating
two or more lens groups independently. With the proper axial motions, a change in
transverse magnification is obtained while the object and image positions remain
fixed [111]. The length of a zoom relay must provide adequate clearance for these
longitudinally translating elements, making zoom riflescopes typically longer than
fixed magnification scopes.
Meanwhile, variable power components such as Alvarez lenses or liquids lenses
offer alternate zoom kernels where the elements are longitudinally fixed in posi-
tion [93, 94]. In these designs, the zoom’s power change and fixed conjugate posi-
tions are achieved with the proper element power variation, not by axial transla-
tion. For a finite conjugate zoom as found in a riflescope zoom relay, a variation in
transverse magnification m can be obtained with two elements of variable power
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 116

Φ1 and Φ2 satisfying [93]

l0
− 2 + l1 − t −l1 m + l20 + t
Φ1 = m , Φ2 = . (3.48)
l1 t l20 t

In Eq. (3.48), l1 is the object distance from the first variable power element, l20 is
the image distance from the second variable power element, and t is the element
spacing. In total, these distances make the object-to-image relay length, L = −l1 +
t+l20 . Interestingly, for some relay length L there is a continuum of unique solutions
for Φ1 and Φ2 depending on how l1 , t, and l20 are distributed. In fact, given L and a
magnification range m ∈ [mlow , mhigh ], there is a unique combination of l1 , t, and l20
that requires the least power variation from Φ1 and Φ2 .

3.4.1 Design goal and specifications

A zoom riflescope that incorporates a zoom relay consisting of axially station-


ary, variable power components is investigated. The design goal is to determine
whether a reduction in system total track length (T T L; measured from the first sur-
face vertex of the objective to the last surface vertex of the eyepiece, excluding ER)
is possible by eliminating the need for longitudinal element translation. Specif-
ically, this work applies GALs for the variable power components in the zoom
relay. As a result, the longitudinal translation of conventional zoom riflescopes is
traded for the transverse movement of the two GAL elements.
The design study goal and specifications are drawn from the Leupold Mark 6
1-6x20, an available high-performance zoom riflescope [120]. This scope is also the
design shown in the X-ray image in Fig. 3.32. The primary design goal is to achieve
a reduction in system length compared with the Leupold scope, T T L < 262.0 mm.
This T T L value is drawn from system specifications reported by Leupold [121].
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 117

A second driving specification is the zoom ratio between high and low magnifi-
cation. The performed design study seeks a 6x zoom design. Following Leupold’s
lead, the zoom ratio specification is 5.5x, rounded up to 6x. The zoom ratio is
derived from Leupold specifications [121] of a low magnification linear full field-
of-view of 35.20 m at 100 meters (9.98◦ half field-of-view) and a high magnification
linear full field-of-view of 6.40 m at 100 meters (1.83◦ half field-of-view), yielding
the 35.20/6.40 = 5.5 zoom ratio. At low magnification, the angular magnification
is taken to be unity, M = 1, to enable “eyes open” operation. A low magnification
object space half field-of-view of 9.98◦ ≈ 10◦ is assumed for simplicity. As a result,
the eyepiece circular half field-of-view across all zoom positions is also 10◦ .
Moreover, the eyebox diameter changes through zoom from 8 mm at low mag-
nification to 3 mm at high magnification. The system eye relief requirement is
ER ≥ 90 mm to ensure adequate clearance upon rifle kickback. ER is modeled
as constant through zoom; although, movement of the head and eye means there
is a longitudinal range of distances over which the eye sees through the system.
Lastly, only a monochromatic wavelength (λd = 587.56 nm) is considered with the
assumption that future chromatic aberration correction will not significantly affect
the achievable T T L. A complete list of design specifications is outlined in Table
3.5.

3.4.2 Design process

Riflescope designs are modular, meaning the objective, relay, and eyepiece can be
designed independently but with the intention of future integration. In this work,
the riflescope design process begins with the GAL zoom relay, followed by the in-
corporation of the eyepiece, followed by the incorporation of the objective. More-
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 118

Specification Low mag. High mag.


Angular magnification, M 1.0 5.5

Entrance pupil diameter [mm] 8.0 16.5


Eyebox diameter [mm] 8.0 3.0

Half field-of-view, object space [deg] 10.00 1.84


Half field-of-view, image space [deg] 10.00 10.00

monochromatic
Spectrum
(λd = 587.56 nm)
Total track length, T T L [mm] ≤ 262.0
Eye relief, ER [mm] ≥ 90.0
Table 3.5: Specifications for the 6x zoom riflescope design study. Low magnification,
M = 1.0, enables “eyes open” operation at the widest field-of-view zoom position. The
eye relief and total track length are constant across zoom. Specifications are drawn from
a conventional zoom riflescope (Leupold Mark 6 1-6x20).

over, the size and position of the fields and pupils must be co-designed between
all modules to ensure intercompatibility. Also, in this work, a reticle is positioned
at the FFP such that it zooms with the field-of-view. As a result, the FFP formed
between the objective and the relay must be well-corrected. On the other hand, the
SFP formed between relay and eyepiece does not need to be well-corrected, and to
improve performance, the relay and eyepiece are jointly optimized. A field stop
does need to be incorporated at the SFP, however, to present a fixed image space
field to the eye.
The GAL zoom relay is the most critical module in the riflescope’s design. The
goal of the design study is to achieve a reduction in system T T L by applying GALs
rather than longitudinally translating fixed power elements. As a result, the zoom
relay length L from object to image is of primary interest during the design. As
L decreases, the requisite power variation in Eq. (3.48) increases. The balance
between length and GAL power is the crux of the zoom relay design.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 119

In addition to reaching a length target, the relay at finite conjugate must also
introduce variable transverse magnification mrelay such that full system achieves
the desired 6x zoom ratio in angular magnification M . The angular and transverse
magnifications are related by

fo0
M =− mrelay (3.49)
fe0

where fo0 and fe0 are the focal lengths of the objective and eyepiece, respectively.
Clearly, a 6x zoom ratio in system angular magnification requires the same 6x
ratio in the relay transverse magnification, although the specific values of mrelay
changes with the ratio fo0 /fe0 . Initially, it is assumed that fo0 = fe0 , requiring mrelay ∈
[−1.0, −5.5] to meet the system specifications in Table 3.5.
Sufficient GAL power variation is needed to achieve a 6x zoom relay, yet there
are constraints on F-GRIN fabrication. Constraints on GAL refractive index change
are listed in Table 3.6 which are adhered to in the design study. For a clear aperture
of CA = 20 mm, the maximum power range from these constraints is ∆Φ = 36.9 D
according to Eq. (3.24) with tilt minimizing the ∆n and D/CA = 1.5.
Meanwhile, for the finite conjugate zoom kernel, the variable GAL powers Φ1
and Φ2 change monotonically with m according to Eq. (3.48). Φ1 changes non-
linearly with m while Φ2 changes linearly. This monotonic power variation also

Constraint Value
Refractive index range, n [1.4117, 1.6152]
Total refractive index change, ∆n ≤ 0.2035
Center thickness, t [mm] ≤ 6.0
Element diameter, D [mm] ≤ 40.0
Table 3.6: GRIN fabrication constraints in the zoom riflescope design study. Constraints
derive from current boundaries on Nanovox F-GRIN additive manufacturing.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 120

entails a monotonic change in element shift δ with zoom. Since monotonic, the
power range for these two components can be found from the values mrelay ∈
[mmin , mmax ],

l20 mmax − mmin l1


∆Φ1 = − · , ∆Φ2 = − · (mmax − mmin ) . (3.50)
l1 t mmax mmin l20 t

While the power range must be achieved by the GALs, the bias power is also a
consideration. There is no requirement that the zoom power variation in Eq. (3.48)
be centered on Φbias = 0 for a given magnification range. The introduction of bias
power in the GALs can compensate for this offset but at the cost of an increase in
∆n (and subsequent decrease in ∆Φ) as well as a decrease in performance (see Fig.
3.14). Instead, a conventional fixed power element is insert next to each GAL to
provide the necessary bias power. This allows the GALs to operate across a max-
imum power range of equal positive and negative powers. In addition to the two
fixed power elements, two field lenses are also included to facilitate pupil match-
ing with the objective and eyepiece, as found in conventional riflescope designs.
The optimal first-order configuration for the described zoom relay is solved by
optimization using CODE V. This relay solution includes as variables the GALs,
fixed power elements, and field lenses as well as all intermediate airspaces. In this
process, some GAL parameters are manually imposed based on the findings of
Secs. 3.2.4 and 3.3.1. The GALs are established with a total refractive index change
of ∆n ≈ 0.16 in order to leave some buffer for incorporating higher-order terms
for further performance improvement (see Fig. 3.24). The GAL elements’ center
thicknesses are then made t = 5.5 mm, kept as thin as possible for minimizing
induced aberrations (see Fig. 3.15). The GAL elements’ diameters are all set to the
maximum, D = 40 mm, in order to maximize D/CA for improved performance
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 121

(see Fig. 3.16).


The relay design is implemented in reverse in CODE V where the FFP serves
as the image plane. Meanwhile, the SFP object is allowed to curve since it does
not need to be a well corrected planar conjugate. The design is optimized for the
described parameters to reach the 6x zoom ratio with high performance and in a
compact axial package. The optimized fields must be defined in both x and y due
to the lack of rotational symmetry. For this reason, the magnification must also
be constrained independently in x and y to control anamorphosis. Moreover, the
GALs impart freeform aberrations, some of which are changing with zoom while
others remain constant. Both GALs are optimized with the select higher-order
polynomial terms identified in Sec. 3.3.1 and listed in green in Table 3.3. One fixed
power element with spherical surfaces is also made of F-GRIN for further freeform
aberration correction. For this fixed power element, the F-GRIN is not rotationally
symmetric but is constrained to maintain the system y-z symmetry. The optimized
relay design also constrains the entrance and exit pupil positions for subsequent
matching with the eyepiece and the objective. Lastly, the GAL clear apertures are
required to be centered on the optical axis along with the fixed power elements.
For boresight error correction, it may be advantageous to offset one or both GAL
pairs from the optical axis, although for simplicity this is not considered here.
The F-GRIN optimization for both GALs and the fixed power F-GRIN element
in the relay is performed in a similar manor to in Sec. 3.3.1. The GRIN ∆n is con-
strained using a function that samples the refractive index throughout a volume.
For the fixed power element with spherical surfaces, the index sampling takes into
account the surface sag. The GRIN coefficients that are varied are selected based on
effectiveness, and the coefficient derivatives are calculated with finite differences
for every optimization cycle.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 122

Upon obtaining the optimized zoom relay design in reverse, an eyepiece is de-
signed independently for pupil matching with the relay while maintaining ER ≥
90 mm. The eyepiece is also designed for a curved SFP intermediate image match-
ing that of the relay. Then, the eyepiece design is added in front of the relay, going
from eyebox to FFP, and the system is jointly optimized. The eyebox half field-of-
view is fixed at 10◦ while the image size at the FFP changes. The eyepiece, relay
system in reverse functions equivalently to that of a zoom hypercentric lens where
the entrance pupil is accessibly located in object space.
Next, an objective is designed that pupil matches with the reversed eyepiece-
relay system and generates the requisite image height at the FFP. The objective
must be independently well-corrected for reticle performance at the FFP. Although,
the on-axis reticle performance is weighted three times more than at full field
where the reticle serves little use. The objective is incorporated in the full sys-
tem, still in reverse, and the complete system is optimized where the field-of-view
and FFP are both optimized for performance. Lastly, the system is flipped to the
proper orientation, and the aperture stop is located at the eyebox.

3.4.3 Final designs

From the design process described in Sec. 3.4.2, a final 6x GAL zoom riflescope
design is obtained. All elements require custom optics, and there are five F-GRIN
elements (4 between the two GALs and 1 in a fixed power element; all located
in the relay module). The design total track length is T T L = 237 mm, achieving
a 25 mm reduction in length compared with the conventional riflescope specifica-
tions listed in Table 3.5. This final design successfully meets the goal of using GALs
for a significant (in this case, 10%) reduction in system length. The final design is
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 123

shown in Fig. 3.33 for three zoom positions spanning the zoom range. The angular
ray aberrations for these three zoom positions are plotted in Fig. 3.34 for fields
sampling both x and y.
An interesting feature of the GAL riflescope design in Fig. 3.33 is its relatively
large chief ray angle at the reticle, located between objective and relay. This chief
ray angle differs from conventional zoom scopes which are nearly telecentric at the
first focal plane. The strong negative field lens before the reticle is responsible for
this increase in chief ray angle as well as making the objective a telephoto design
form (telephoto ratio of 0.88). While the objective is telephoto, the GAL zoom
relay is the reason why this non-zero chief ray angle is necessary. Due to the fixed
positions of the GALs in the relay, an upward-sloping chief ray at the reticle is
required to limit the used clear aperture of the GAL pairs through zoom. A larger
GAL clear aperture restricts the extent of available element shift, yielding smaller
power variation ∆Φ, as apparent in Eqs. (3.23) and (3.24) for an increase in CA.
For this reason, a telephoto objective with an upward-sloping chief ray angle at
the reticle is more compatible with a high zoom ratio GAL relay. Also, for this
configuration, the pupil imaging between objective, relay, and eyepiece results in a
significantly changing entrance pupil position through zoom for a fixed exit pupil
position. Recall, eye relief is subject to change when the riflescope is in use based
on the user’s eye position and orientation.
The telephoto objective is also beneficial in reducing the system T T L. Mean-
while, the relay length measured between intermediate images is 135.5 mm, which
is only ≈ 5 mm shorter than the Leupold scope relay, as determined from the X-ray
image shown in Fig. 3.32. With a total 25 mm reduction in system T T L for the
design in Fig. 3.33, this relay length shows the T T L reduction is largely achieved
by the objective and eyepiece. These shorter objective and eyepiece lengths are
Eyepiece
Object space Objective Reticle Relay
Eyebox
M = 1.00

1.6164

M = 3.25 1.5145
Refractive index

1.4125

M = 5.50

50 mm

Figure 3.33: 6x zoom riflescope design using GALs and custom optics. The total track length T T L = 237 mm is a 25 mm
reduction compared with the conventional design listed in Table 3.5.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses
124
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 125

M = 1.00 M = 3.25 M = 5.50

Figure 3.34: Angular ray aberration plots through zoom for the custom 6x zoom riflescope
design. Note, both x and y fields are present, and the aberration scale is in units of
arcminutes.

largely due to the pupil matching requirements with the GAL relay, which differ
from those of conventional riflescope zoom relays.
Furthermore, the riflescope field-of-view is circular, meaning a rotation of the
system around its optical axis does not result in a change in observed field. How-
ever, the system has y-z symmetry rather than rotational symmetry, meaning the
image quality (both resolution and anamorphosis) may differ across the field de-
pending on the orientation. The system performance is symmetric across the GAL
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 126

element shift axis, which in this case is the y axis. Since riflescopes are most often
used to look horizontally (i.e., along the horizon rather than looking up at the sky
and down at the foreground), it is best for the GAL elements to shift vertically in
order to maintain symmetric field performance horizontally.
The MTF performance for the custom riflescope design is shown in Fig. 3.35.
The MTF is evaluated out to a spatial frequency of 0.5 cycles/arcmin, which corre-
sponds to the eye’s visual acuity cutoff in the foveal region [115]. While the design
is performed across the eyebox diameters listed in Table 3.5, the MTF performance
is evaluated for a 2.5 mm exit pupil, centered on the optical axis. This pupil size
matches that of the eye in normal lighting conditions [115]. The results show nearly
diffraction-limited performance on-axis while off-axis fields suffer a reduction in
contrast. The field performance necessary in riflescopes is different from other sys-
tems such as cameras. Riflescopes are primarily used about the on-axis field where
the center of the reticle is located. Rather than examining full field, the scope and
rifle are repositioned to center the target in the center of the field. Also, accom-
modation in the eye means field curvature in off-axis fields may be compensated
when in use.
Next, the element shift δ through zoom is shown in Fig. 3.36 for the two GALs
in the custom riflescope. By Eq. (3.20), GAL power is directly proportional to δ,
meaning Fig. 3.36 approximately shows the GAL power variation through zoom.
The connection to power is only approximate since higher-order terms are pre-
sented, making the analytical power relationships derived in Sec. 3.2.3 no longer
exact. The curves in Fig. 3.36 show that the first GAL changes nonlinearly with
magnification while the second GAL changes linearly. This dependence matches
the variable power zoom expressions listed in Eq. (3.48) where Φ1 is nonlinear with
m while Φ2 is linear.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 127

M = 1.00 M = 3.25 M = 5.50


1.00 1.00 1.00

0.75 0.75 0.75


Modulation

Modulation

Modulation
0.50 0.50 0.50

0.25 0.25 0.25

0.00 0.00 0.00


0.00 0.25 0.50 0.00 0.25 0.50 0.00 0.25 0.50
Spatial frequency (cycles/min) Spatial frequency (cycles/min) Spatial frequency (cycles/min)
Y Diffraction Limit F2: Y (0.0, -0.7) rel. fld. F4: Y (0.0, -1.0) rel. fld. F6: Y (0.7, 0.0) rel. fld.
X Diffraction Limit F2: X (0.0, -0.7) rel. fld. F4: X (0.0, -1.0) rel. fld. F6: X (0.7, 0.0) rel. fld.
F1: Y (0.0, 0.0) rel. fld. F3: Y (0.0, 0.7) rel. fld. F5: Y (0.0, 1.0) rel. fld. F7: Y (1.0, 0.0) rel. fld.
F1: X (0.0, 0.0) rel. fld. F3: X (0.0, 0.7) rel. fld. F5: X (0.0, 1.0) rel. fld. F7: X (1.0, 0.0) rel. fld.

Figure 3.35: Diffraction MTF performance for the custom 6x zoom riflescope design. MTF
is evaluated across a 2.5 mm eye pupil, calculated out to the eye’s spatial frequency cutoff,
and considered monochromatically at the design wavelength.

Figure 3.36: GAL element shift δ through zoom for the custom 6x zoom riflescope design.

Figure 3.37: Angular boresight error through zoom for the custom 6x zoom riflescope
design.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 128

GALs introduce variable boresight error (BSE), as described in Sec. 3.2.4. When
using a riflescope, care is taken to ensure the rifle and scope pointing axes corre-
spond for the same point in object space. Conventionally, the user calibrates this
link using elevation and windage adjustments of the reticle for a given instance.
Static BSE is compensated in this adjustment, yet variable BSE means elevation and
windage adjustment is only correct for a single zoom position. If the riflescope is
kept stationary, variable BSE results in the field-of-view and reticle changing po-
sition through zoom when viewed by the eye. The shift of the field and reticle is
coupled for the current design since the reticle is located at the FFP.
Since a riflescope is an afocal system, angular BSE is the metric of interest for
assessing its effects when viewed by the eye. In this work, the system angular BSE
is constrained to be ≤ 1 arcminute across the zoom range. The 1 arcminute target
corresponds to the eye’s visual acuity [115], beyond which variable BSE would be
imperceptible. Control over variable BSE is accomplished using axial refractive in-
dex variation in the GALs. For the final riflescope design, the angular BSE through
zoom is shown in Fig. 3.37 where it can be seen that the 1 arcminute target is met.
The 6x zoom riflescope design presented in Fig. 3.33 requires all custom opti-
cal elements, which are costly and may require significant lead times. For rapid
prototyping of a proof of concept GAL scope, an additional design is considered
that uses exclusively commercial off-the-shelf (COTS) components for the homo-
geneous elements. Also, the fixed power F-GRIN element in the custom design is
converted to homogeneous. As a result, only four F-GRIN elements are required
for constructing the two GALs, and all elements have plane-parallel surfaces to
simplify fabrication.
The COTS riflescope design procedure differs from the custom design. A COTS
photographic objective is selected for the riflescope objective, ensuring the FFP is
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 129

Lens element Manufacturer Part number


L1 Edmund Optics #86-574
L2 Thorlabs LE1234-A
L3 Thorlabs LBF254-050-A
L4 Edmund Optics #45-240
L5 Edmund Optics #45-031
L6 Edmund Optics #45-218
L7 Edmund Optics #63-593
Table 3.7: COTS elements used in the 6x zoom riflescope design. Lens element numbers
correspond to those noted in Fig. 3.38. Together, all COTS elements can be purchased for
$1,097.

well-corrected and of adequate size for the reticle. The eyepiece is designed from a
global search of all combinations of select COTS singlets and achromatic doublets.
The eyepiece design allows for two elements at most, of which there were 14,152
combinations when considering both orientations for each element. From these
combinations, the optimal configuration is selected. Then, the relay is designed
between the objective and eyepiece. Homogeneous elements are selected with sim-
ilar power and bendings to those in the custom riflescope relay. The GAL elements
are again optimized using higher-order polynomial terms. The final COTS ele-
ments including their manufacturer and part number are listed in Table 3.7. The
total cost of these elements is $1,097.
The final 6x zoom riflescope design using COTS elements can be seen in Fig.
3.38. The COTS design successfully achieves the 6x zoom ratio. GAL elements
of increased thickness t = 6 mm are used to satisfy a tighter F-GRIN fabrication
constraint ∆n ≤ 0.1827. As expected for a design made from COTS elements, per-
formance and specifications are lacking compared with the custom design. The
COTS design is longer with T T L = 300 mm, which exceeds the Leupold product
specifications in Table 3.5. (Note, the available product also uses all custom opti-
cal elements.) The COTS design also has lesser pupil control where the eye relief
Object space Objective Reticle Relay Eyepiece
Eyebox
M = 1.00

L1 L2 L3 L4 L5
L6 L7 1.6057

M = 3.25 1.5174
Refractive index

1.4291

M = 5.50

50 mm

Figure 3.38: 6x zoom riflescope design using GALs and the COTS elements listed in Table 3.7. The total track length T T L =
300 mm is longer than specified, and ER is changing through zoom.
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses
130
Chapter 3. Compact Zoom Riflescope using Gradient-Index Alvarez Lenses 131

changes through zoom, including at distances less than the 90 mm specification.


The design MTF performance is also decreased, achieving similar contrast to the
custom design at one-fifth the spatial frequency, 0.1 cycles/arcmin. Nevertheless,
the COTS design remains adequate as a proof of concept of a 6x zoom riflescope
using GALs.
Chapter 4

Gradient-Index Cover Glass for Field


Curvature Compensation

4.1 Field curvature (FC) aberrations

4.1.1 Definition

A core tenet of first-order optics is that imaging occurs between planar conjugates
from object to image. Meanwhile, optical aberrations introduce deviations from
first-order imaging, and the planar imaging condition is no exception. There is a
class of aberrations that will be referred to here as field curvature (FC) aberrations
where a planar object is imaged to a non-planar surface.
Historically, there have been different approaches to nomenclature for FC aber-
rations. Depending on the reference frame, the behavior of different FC aberra-
tions can be intertwined, such as for wavefront aberrations as compared to ray
aberrations. The convention established here for FC aberrations is chosen with
generality in mind since, in this chapter, FC aberrations are not limited to the pri-
mary monochromatic aberrations or even to those found in rotationally symmetric
systems.

132
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 133

Stigmatic
(a) Imaging x
System
y
z

Pupil
TS
y

Astigmatic T
(b) Imaging x
System S
y
z
Figure 4.1: Field curvature (FC) aberrations yield a non-planar image surface (bold black
lines) that can be either (a) stigmatic or (b) astigmatic. (a) Stigmatic FC results in all rays
in the pupil intersecting at an infinitesimal point for a given field, introducing no “blur” in
the image spot. (b) Astigmatic FC produces spatially extended image spots for a field, re-
gardless of the longitudinal image position, due to differing focal distances for different rays
in the pupil. Tangential (T) and sagittal (S) rays in the pupil focus at different longitudinal
positions.

One general way of classifying FC aberrations is as either stigmatic or astig-


matic. Stigmatic FC aberrations yield a non-planar image, yet for a given field-
of-view, all rays in the pupil intersect at an infinitesimal image point, as shown in
Fig. 4.1(a). Thus, stigmatic FC introduces no “blur” to the curved image. Although,
capturing a curved image on standard planar imaging detector reduces image res-
olution, as discussed in Sec. 4.1.3. An example of a stigmatic FC aberration is
Petzval field curvature, which is one of the primary monochromatic aberrations
that appears off-axis in rotationally symmetric systems.
On the other hand, astigmatic FC aberrations produce image spots of finite ex-
tent for a given field as well as with differing longitudinal positions between fields,
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 134

as shown in Fig. 4.1(b). An extended spot from astigmatic FC aberrations is caused


by different focal distances for different rays in the pupil. A complication with
astigmatic FC aberrations is that, since there is no infinitesimal image spot for a
given field, there is no single longitudinal position at which to assess the field cur-
vature. Instead, astigmatic focal distances are conventionally defined separately
for specific sets of rays in the pupil. For an object extending along the y-axis, tan-
gential rays are the rays in the pupil that are tangent to the object (i.e., the y-z pupil
cross section). Meanwhile, sagittal rays are the rays in the pupil oriented orthog-
onal to the tangential ray section (i.e., the x-z pupil section). Then, the tangential
image can be defined as the longitudinal position at which tangential rays focus
while the sagittal image is the longitudinal distance at which sagittal rays focus.
The tangential (T) and sagittal (S) focal surfaces are denoted in Fig. 4.1(b). A third
image surface of interest is the medial focus, which is located halfway between the
tangential and sagittal focal surfaces. An example of an astigmatic FC aberration
is primary astigmatism which appears off-axis in rotationally symmetric systems.
In designing an imaging system, both stigmatic and astigmatic FC aberrations
must be considered. In fact, their joint effect is ultimately what influences imag-
ing performance. The independent effects of stigmatic and astigmatic FC can also
be used to balance one another to minimize FC across both tangential and sagittal
images. For example, in the presence of undercorrected Petzval field curvature, in-
corporating overcorrected astigmatism to flatten the tangential field nets the least
overall FC [1].
In addition to the distinction between stigmatic and astigmatic, there are also
different types of FC aberrations. In rotationally symmetric systems, there are
higher-order FC aberrations that influence imaging performance and must be con-
sidered in design. For example, Petzval curvature and primary astigmatism both
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 135

increase quadratically with field-of-view h2 , yet there are additional stigmatic and
astigmatic FC aberrations in rotationally symmetric systems that increase with h4 ,
h6 , etc. [122].
Moreover, additional types of FC aberrations appear for imaging systems that
break rotational symmetry. Anamorphic imaging systems, which are non-rotationally
symmetric but maintain bi-planar symmetry, suffer from astigmatism that, unlike
primary astigmatism, appears on-axis [123]. The resultant anamorphic image sur-
faces are toroidal in form. Further breaking symmetry, tilting and/or decenter-
ing elements introduces freeform FC aberrations according to nodal aberration
theory [7–9]. Freeform optical elements may also introduce freeform FC aberra-
tions [10]. Freeform FC presents image surfaces that lack rotational symmetry (e.g.,
medial field curvature, field conjugate astigmatism, field asymmetric astigmatism)
and may posses field dependencies not found in rotationally symmetric systems
(e.g., h, h3 , etc.).

4.1.2 Impact on imaging optical design

FC aberrations play an important role in imaging optical design, regardless of sys-


tem geometry. In general, all classes of FC aberrations (e.g., stigmatic versus astig-
matic, rotationally symmetric versus freeform) degrade image performance. The
reason for this is that, conventionally, detectors have been restricted to planar form
factors (more on this in Sec. 4.1.4). Consequently, the need to correct FC aberra-
tions is one of the primary driving factors in a design. The diversity of design
forms across different imaging application spaces can largely be attributed to dif-
ferent strategies of correcting FC aberrations.
The impact of FC on imaging design can be demonstrated for the simplest case
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 136

of Petzval field curvature. Petzval is the lowest-order stigmatic FC aberration in


rotationally symmetric systems, and it increases quadratically with field-of-view,
h2 . The radius of curvature Rp of the Petzval image surface is governed by the
Petzval sum performed over all k surfaces in a system [1],

k
1 X φj
0
= − 0
, (4.1)
nk Rp j=1
n j nj

where nj and n0j are the refractive indices preceding and following surface j and
φj = n0j − nj /Rj is the surface power for radius of curvature Rj . The Petzval


sum, remarkably, has no dependence on ray tracing and is determined only by


these system parameters.
The design of rotationally symmetric systems is significantly influenced when
attempting to keep the Petzval sum at an acceptable level. For example, by Eq.
(4.1), both positive and negative elements are needed to reduce the Petzval sum.
In order to maintain system power, these elements of opposite sign power must
be separated in space. The requirement on reducing the Petzval sum is one of the
primary reasons why multiple lens elements are needed for most imaging appli-
cations. The need for elements of differing power is also responsible for many of
the prototypical design forms, such as the telephoto and reverse telephoto lenses.
Furthermore, a lens element known as a “field flattener” serves the specific
purposes of reducing the Petzval sum [124]. Field flatteners are thin, powered
elements that are incorporated near conjugate planes in a design. The power of
the field flattener is used to contribute a balancing term to the Petzval sum while
its position near a conjugate plane means it contributes minimally to the overall
system power and aberration balance. Field flatteners are commonly incorporated
in designs to reduce Petzval; although, there are often mechanical restrictions that
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 137

prevent field flatteners from being placed at conjugate planes.


In addition to Petzval, mitigating other types of FC aberrations also strongly
impacts the design of imaging systems. A primary means of controlling astigmatic
FC is with the position of the aperture stop. For example, any surface concentric
with a pupil introduces zero astigmatism since the chief ray is normally incident
on the surface. Generally, astigmatism derives from differing behavior between
tangential and sagittal rays in the pupil. One way of assessing the presence of
astigmatism is with Coddington’s equations, which offers a convenient means of
evaluating astigmatism using only a real chief ray trace [1].
Overall, the correction of FC aberrations, both stigmatic and astigmatic, is a
driving factor when designing any imaging system used with a planar detector.
The reason for this is the detrimental impact of FC on imaging performance.

4.1.3 Impact on imaging performance

The impact of FC on imaging performance can, again, be analyzed for the simplest
case of Petzval field curvature. For a system suffering from Petzval, a planar object
is imaged to a quadratic surface (see Fig. 4.2). Consequently, Petzval can also be
considered more intuitively as an image defocus that increases quadratically with
field-of-view, h2 . Thus, the negative effect of Petzval on image resolution can be
recognized as equivalent to that of a varying amount of defocus across the field.
An image captured by a planar detector in the presence of Petzval has only a single
image height in focus. (Although, the system depth of focus means there is always
some tolerable range of in-focus image heights.) For this quadratic image surface,
one option is to longitudinally position the planar detector at the on-axis focus,
yielding optimal performance on-axis while image quality decreases quadratically
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 138

y y

x x
(a) (b)
25.00 MM 25.00 MM

50% field focus


Position: 1 DHL 21-Mar-23 Position: 2 DHL 21-Mar-23

y y

x x
(c) (d)

Figure 4.2: Field curvature degrades image performance across the field-of-view when
restricted to a planar image detector. Transverse spot diagrams are shown across the
field for a telecentric imaging system suffering only from Petzval field curvature, a stigmatic
FC aberration. The dashed curve shows the stigmatic Petzval image surface. A planar
detector is longitudinally positioned at the ideal focus for the (a) on-axis, (b) mid-field, and
(c) full field image heights. (a)-(c) The image spots are extended for the out of focus fields,
reducing image resolution. (d) A curved image detector captures the stigmatic curved
image in focus across the full field-of-view.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 139

with field height, as shown in Fig. 4.2(a). Two other options are to position the
planar detector for in-focus imaging at mid-field or at full field, as shown in Fig.
4.2(b)-(c), respectively. In all three cases, the presence of FC reduces imaging per-
formance for some region of the field-of-view when restricted to a planar imaging
detector. For Petzval, which is a stigmatic FC aberration, a curved detector would
solve the problem of reduced image quality, capturing all fields in focus, as shown
in Fig. 4.2(d). For this reason, curved detectors offer new possibilities in FC cor-
rection that are unattainable with planar detectors.

4.1.4 Compensation with curved image detectors

The impact of FC aberrations on the design and performance of imaging systems


using planar detectors is discussed in Secs. 4.1.2 and 4.1.3. For stigmatic FC, a
curved image detector, on the other hand, allows performance to be fully recov-
ered. The curved detector requires a surface form that matches the stigmatic im-
age surface, as shown in Fig. 4.2(d) for the simplest case of Petzval field curvature.
Meanwhile, astigmatic FC aberrations cannot be fully compensated by curved de-
tectors since they produce different image surfaces for different rays in the pupil
(e.g., tangential versus sagittal ray fans). Although, systems suffering from astig-
matic FC can be significantly improved using a curved detector that matches the
profile of the medial image surface, which lies midway between the tangential and
sagittal images. As described in Sec. 4.1.1, the impact of FC is found in the combi-
nation of stigmatic and astigmatic contributions based on the individual tangential
and sagittal image surfaces. Likewise, FC compensation with curved image detec-
tors must also account for the joint effect of stigmatic and astigmatic FC.
Due to their ability to mitigate FC aberrations, curved image detectors dramat-
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 140

ically change the design space for imaging systems [125, 126]. Stigmatic FC aber-
rations can be entirely neglected with their correction happening via the curved
detector rather than by optical components. For example, the separated positive
and negative elements are no longer required to reduce the Petzval sum in Eq.
(4.1), such as in the telephoto and reverse telephoto design forms. Restrictions
on astigmatic FC can also be relaxed due to their additional compensation with a
curved detector. Instead, new design forms can be accessed that reduce system
volume and weight [127–130] while also improving performance [126, 131, 132].
Curved detectors also offer improvements in image relative illumination [133]. It is
important to emphasize that these benefits from curved detectors are only achiev-
able in designs intended for a curved image. An imaging system designed with a
highly flat field for a planar detector will be negligibly improved by incorporating
a curved detector.
Curved detectors offer great promise in advanced imaging designs, yet practi-
cal challenges have limited their adoption. Early curved detectors employed film,
which due to its flexible nature, could be mechanically deformed into cylindrical
surfaces. Although, this deformation is not as easily achieved for rotationally sym-
metric image surfaces. One classical use of curved film detectors is in the Schmidt
telescope, which has a highly curved stigmatic image surface [134]. For this reason,
the Schmidt telescope is also the focus of a design study in Sec. 4.3.3.
Meanwhile, modern image detectors are semiconductor-based, such as the charge-
coupled device (CCD) and complementary metal-oxide semiconductor (CMOS)
detectors [135]. Silicon wafers are planar in format, and photolithographic pro-
cesses for fabricating detectors are not intended for printing on curved substrates
[63], a feat some have attempted with so-called “soft” lithography [136–139]. More-
over, the rigidity of semiconductor-based detectors means they cannot be easily
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 141

mechanically deformed to curved surfaces. This limitation proves especially diffi-


cult at the scale of mass-production for current semiconductor detectors. Nonethe-
less, different approaches have been presented for mechanically curving semicon-
ductor detectors. The primary technique used by several is to uniformly grind
the photosensitive material, either CCD [140–142] or CMOS [126, 143–146], until
thin enough to be adequately flexible. Then, the thinned detector can be curved
by applying a force, either mechanical [144–146] or pneumatic [126, 141, 142]. In
each of these approaches, detector curvature is limited by stresses and buckling
from two-dimensional curving of the thinned yet still rigid material. A different
option that avoids the need for bending silicon uses a planar detector coupled to
a curved fiber array in order to achieve a curved detector effect [128, 147, 148]. Al-
though, fiber-coupled curved detectors lack the high spatial resolution of standard
semiconductor-based detectors.
Among these various approaches, there remain significant obstacles before curved
image detectors can be fabricated at mass-production scales. For this reason, the
majority of current optical designs continue to assume a planar image detector.

4.2 Homogeneous detector cover glass

Unrelated to FC, semiconductor-based image detectors conventionally incorporate


a cover glass (CG) in the detector stack immediately prior to the surface of pho-
tosensitive elements [145, 149–151]. CG is a plane-parallel plate (PPP) made of
a homogeneous optical material. The purpose of CG is twofold. First, CG pro-
tects detector circuitry from mechanical or environmental damage [145]. Second,
CG presents a convenient location for incorporating a spectral filter. For example,
many image detectors designed for the visible spectrum include a high-pass filter
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 142

to reject any incident near-infrared light that may otherwise be detected [152].
While seemingly innocuous, the presence of detector CG must be included in
the design of an associated imaging system. CG is located in space where light is
converging, and as a result, aberrations are introduced, as they are for any PPP in
focusing space [111]. In fact, PPPs in focusing space generally introduce all pri-
mary monochromatic and polychromatic aberrations except for Petzval field cur-
vature. For CG, overcorrected spherical aberration typically presents the greatest
impact that must be accounted for in designing the accompanying imaging sys-
tem. Off-axis aberrations are absent for telecentric fields but increase with chief
ray angle, meaning highly non-telecentric systems may suffer significant effects
across the field. PPP aberration contributions also increase linearly with center
thickness t. Consequently, CG is often restricted to small values for t that are still
readily achievable in fabrication. Some ranges of CG thicknesses for different de-
tector manufacturers can be seen in Table 4.1. While some values for t are as small
as 0.5 mm, aberrations imparted by the CG must still be accounted for in any pre-
cision imaging design.
A second impact CG has image quality is that any defects in either surface of
the CG will be apparent in the captured image due to its proximity to the detec-

Manufacturer CG thickness, t [mm]


Canon 1.0 – 2.75
Fuji 2.0 – 2.2
Leica 0.5 – 0.8
Nikon 0.7 – 2.2
Olympus 3.8
Panasonic 4.1
Pentax 1.6
SONY 0.6 – 2.4
Table 4.1: Cover glass (CG) center thicknesses for different image detector manufacturers
[153]. Notice there is significant variation between manufacturers.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 143

tor. For this reason, CG must maintain very stringent scratch-dig tolerances. The
cleanliness of CG is also of important for avoiding unwanted image artifacts.
A third effect CG has on imaging systems is a longitudinal displacement of the
image. This feature is studied next in Sec. 4.2.1.

4.2.1 Longitudinal image displacement

An image formed by light converging or diverging through a PPP, such as CG, is


longitudinally displaced by some distance d (see Fig. 4.3). For a converging beam,
d can be solved exactly from Snell’s law and geometrical optics for a PPP of center
thickness t and homogeneous refractive index n [111],

" s #
1 − (N A)2
d=t 1− , (4.2)
n2 − (N A)2

n
Uak

t d
Figure 4.3: Longitudinal image displacement d for a beam converging through a PPP of
center thickness t and homogeneous refractive index n. For each the upper and lower
marginal rays, the entering and exiting ray segments are parallel but offset from one an-
other.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 144

Figure 4.4: Longitudinal image displacement d calculated according to Eq. (4.2) for in-
creasing refractive index n and for different numerical apertures N A. The image displace-
ment increases monotonically with refractive index. For this example, the PPP center
thickness is t = 2 mm.

0
where N A = |sin Uak | is the numerical aperture of the incident beam, assuming
the PPP is immersed in air. The behavior of Eq. (4.2) is plotted in Fig. 4.4 for
increasing values of refractive index n as well as for different numerical apertures.
While not immediately clear from Eq. (4.2), it can be seen from the plot that the
image displacement d increases monotonically with refractive index n. Based on
Eq. (4.2), d also increases linearly with PPP center thickness t.
The presence of a square root in Eq. (4.2) imposes a restriction on what image
displacements d are attainable. These boundaries are the result of the refractive
index being, theoretically, constrained within n ∈ [1, ∞). When n = 1, there is no
image displacement, d = 0, as expected when the PPP refractive index matches that
of the immersing medium. When n approaches infinity, the image displacement
approaches its maximum attainable value of d = t. Thus, the maximum range
of possible image displacements equals the thickness t of the PPP. This restriction
also affects the potential of GRIN CG, as described in Sec. 4.3.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 145

A simpler expression for the image displacement d can be obtained by assum-


ing the paraxial approximation where N A = 0. Applying the paraxial approxima-
tion to Eq. (4.2) yields
 
n−1
d=t . (4.3)
n

The paraxial image displacement is also plotted in Fig. 4.4 where the approxima-
tion’s accuracy can be seen compared with non-paraxial numerical apertures.
In deriving Eqs. (4.2) and (4.3), it is assumed that the incident beam has a
chief ray normally incident on the PPP, as shown in Fig. 4.3. For this scenario,
the focusing ray cone is oriented with its altitude normally incident on the PPP. In
a rotationally symmetric system, the on-axis field is required to have a normally
incident beam. Any off-axis field can also satisfy this criteria if telecentric in image
space (i.e., exit pupil located at infinity).
The image displacement due to a PPP must be accounted for in any imaging
system using a detector with CG. Otherwise, the image captured by assuming d =
0 will be out of focus. Fortunately, this adjustment is a simple refocusing of the
lens, meaning upon refocus there is no significant effect on image quality due to
the displacement. Significant emphasis has been placed on the image displacement
d, not due to its effect within homogeneous CG, but due to its implications for
inhomogeneous CG.

4.3 Inhomogeneous detector cover glass

4.3.1 FC compensation

In all embodiments, CG has been limited to homogeneous optical media. An inter-


esting and advantageous feature is discovered when considering an inhomogeneous
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 146

CG containing a spatially varying refractive index. For CG containing an arbitrary


two-dimensional GRIN profile n (x, y), according to Eq. (4.2) the longitudinal im-
age displacement now also becomes arbitrarily varying,

" s #
1 − (N A)2
d (x, y) ≈ t 1 − . (4.4)
n (x, y)2 − (N A)2

This scenario of a GRIN CG introducing a spatially varying image displacement is


depicted in Fig. 4.5.
While Eq. (4.2) is an exact expression for a homogeneous PPP, the image shift
due to a GRIN PPP in Eq. (4.4) is approximate due to several factors. First, rays
curve within inhomogeneous media according to Fermat’s principle [17], so the
straight rays transferring within the GRIN shown in Fig. 4.5 are approximations.
This assumption that rays undergo negligible curvature is most accurate for small
values of t, as is the typical form factor for CG.
Second, the refractive index and gradient experienced by the upper marginal
ray differs from that of the lower marginal ray. As a result, the trajectory of the

n(x,y)
Uak
Refractive index

t d(x,y)
Figure 4.5: Transversely varying image displacement d (x, y) for a beam converging
through a GRIN CG of center thickness t and spatially varying refractive index n (x, y).
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 147

two rays is no longer symmetrical. Meanwhile, in Eq. (4.4) the GRIN refractive
index n (x, y) is calculated at the chief ray position. This value for the refractive
index at the chief ray serves as a useful approximation by acting as the net index
experienced by both the upper and lower marginal rays, assuming the index is
slowly varying spatially. The chief ray net index approximation breaks down as
the numerical aperture increases. Thus, a paraxial form of Eq. (4.4) is also helpful,

 
n (x, y) − 1
d (x, y) ≈ t . (4.5)
n (x, y)

As for homogeneous PPPs, the longitudinal image displacement d (x, y) described


by Eqs. (4.4) and (4.5) applies specifically to the case of a normally incident chief
ray, either for the on-axis field in a rotationally symmetric system or for any tele-
centric field.
Furthermore, by rearranging Eq. (4.4), an expression for the refractive index
n (x, y) can be obtained that yields a certain image displacement d (x, y),

q
t2 + [d (x, y) − 2 t] d (x, y) (N A)2
n (x, y) ≈ . (4.6)
t − d (x, y)

In deriving Eq. (4.6), simplifications are performed based on the fact that n (x, y) >
0 and d (x, y) ≤ t, the latter of which can be gathered by inspecting Eq. (4.4).
Likewise, the same process of solving for n (x, y) can be performed within the
paraxial approximation. Rearranging Eq. (4.5) yields a simpler form for the requi-
site refractive index distribution n (x, y) that produces displacement d (x, y),

t
n (x, y) ≈ . (4.7)
t − d (x, y)
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 148

z(x,y)

Refractive index
d(x,y) d(x,y)

(a) (b)

Figure 4.6: FC introduction and compensation using GRIN CG. The image surface for the
imaging system preceding the CG is shown with a dotted line and the image surface upon
transmitting through the CG is shown with a solid line. (a) For an imaging system producing
a planar image, the addition of a GRIN CG yields a curved image surface, similar to in Fig.
4.5. (b) For an imaging system producing a curved image surface with sag z (x, y), the
incorporation of the proper GRIN CG compensates for FC aberrations and yields a planar
image surface.

For an imaging system that on its own produces a planar image, the addition
of a GRIN CG yields a curved image surface due to the spatially varying image
displacement, as shown in Fig. 4.6(a). This is a notable finding, especially when
considering the reverse procedure. Given an imaging system that in isolation pro-
duces a curved image, a GRIN CG can be incorporated to yield a planar image
by introducing the appropriate spatially varying image displacement, as shown in
Fig. 4.6(b). Thus, GRIN CG can be used to compensate FC aberrations in a way
that does not require adjustment of the preceding imaging system. In fact, the in-
corporation of GRIN CG into a planar detector stack, as shown in Fig. 4.7, means
an artificially curved image detector can be constructed, avoiding the difficulty of
mechanical curving the photosensitive surface.
For an imaging system with a curved image surface defined by sag z (x, y), the
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 149

Figure 4.7: CAD depiction of GRIN CG directly incorporated into a planar detector, yielding
an artificially curved image detector.

GRIN profile n (x, y) necessary to flatten the field can be solved using Eq. (4.6)
or (4.7). First, it is assumed that z (0, 0) = 0 for the image point along the optical
axis and that n (0, 0) = n0 for the refractive index along the axis. (Note that, at
this point, the refractive index is axially constant.) From inspecting Fig. 4.6(b), the
requisite image distance d (x, y) to flatten the field can now be expressed as

d (x, y) = d0 − z (x, y) , (4.8)

where d0 is the axial image displacement according to Eq. (4.4),

" s #
1 − (N A)2
d0 ≈ t 1 − , (4.9)
n20 − (N A)2

or paraxially according to Eq. (4.5),

 
n0 − 1
d0 ≈ t . (4.10)
n0

Then, to produce the required d (x, y) in Eq. (4.8), the refractive index n (x, y)
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 150

can be solved using Eq. (4.6),

q
t2 + [d0 − z (x, y) − 2 t] [d0 − z (x, y)] (N A)2
n (x, y) ≈ , (4.11)
t − d0 + z (x, y))

or assuming the paraxial approximation with Eq. (4.7),

t
n (x, y) ≈ . (4.12)
t − d0 + z (x, y)

Note the index expressions in Eqs. (4.11) and (4.12) are still based on the assump-
tion of telecentricity and the various approximations described previously.
It is important to emphasize that the form of z (x, y) in Eqs. (4.11) and (4.12) is
left entirely arbitrary. As a result, there is no limitation on what types of FC aber-
rations can be compensated using GRIN CG. For a rotationally symmetric system
z (x, y) is also rotationally symmetric while in a freeform system z (x, y) can be of
arbitrary form. Also, recall that the combination of stigmatic and astigmatic FC
aberrations dictates the optimal form of z (x, y) based on the medial image surface.
Finally, FC compensation using GRIN CG can be achieved at any conjugate plane
and is not restricted to the ultimate image captured by a detector. For example, an
intermediate image can be corrected using GRIN CG in a visual system such as a
riflescope.
Greater intuition can be gained from the results for n (x, y) by considering the
simpler paraxial case. Substituting the expression for d0 into Eq. (4.12) and simpli-
fying gives
t
n (x, y) ≈ t . (4.13)
n0
+ z (x, y)

Within the paraxial approximation, Eq. (4.13) provides the refractive index n (x, y)
needed to flatten an image surface sag z (x, y) in terms of only t and n0 . While
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 151

not immediately clear from Eq. (4.13), the change in index versus sag compensa-
tion can be seen by plotting the relationship, as done in Fig. 4.8. Two boundaries
can be recognized in Fig. 4.8(a) due to the restriction on theoretical values for the
refractive index, n ∈ [1, ∞). These index extrema occur at the periphery of a cor-
h i
responding image sag range, z ∈ − nt0 , t − nt0 . Notice the full extent of the image
sag range is equal to the GRIN CG thickness t, the same as found for the homoge-
neous CG in Sec. 4.2. On the other hand, Fig. 4.8(b) shows a sub-range centered

(a)

(b)

Figure 4.8: Refractive index n required to flatten an image surface of sag z calculated
according to Eq. (4.13). (a) The full theoretical range for n and z is shown within the
achievable bounds (red). (b) A sub-range is shown centered around z = 0 and n = n0
where the relationship is closer to linear. (a)-(b) The required index increases monoton-
ically with increasing sag, either positive or negative. For this example, the PPP center
thickness is t = 2 mm and the base refractive index is n0 = 1.5.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 152

about z = 0 and n = n0 where the relationship between n and z is monotonic


and closer to linear. This sub-range is the more realistic domain since values for n
rarely approach unity or infinity in GRIN media.
Another important conclusion can be drawn by solving for z in Eq. (4.13),

z (x, y) ≈ t n (x, y)−1 − n−1


 
0 . (4.14)

The extent of image sag that can be flattened increases linearly with the GRIN CG
thickness t. Unlike homogeneous CG which is ideally kept as thin as possible to
minimize imparted aberrations, larger GRIN CG thicknesses can compensate for
larger FC aberrations. While thicker GRIN CG may introduce greater aberrations,
the refractive index variation may also be used to correct some of these aberrations.
The notable exception is spherical aberration correction. Radially quadratic GRIN
terms are helpful in correcting spherical aberration [79], but the GRIN CG posi-
tion immediately next to a conjugate plane makes its spherical correcting abilities
minimal.
Finally, analytical expressions have been derived for the transverse refractive
index variation needed to flatten a curved image surface; however, these expres-
sions rely on various assumptions and approximations, such as telecentricity across
the field-of-view. As a result, the analytical form of the refractive index is not nec-
essarily the optimal form for overall GRIN CG performance. To compensate for
deviations from these assumptions in the design process, the GRIN CG can be ini-
tialized using the analytical expressions for n (x, y) and then optimized for further
performance improvement. The GRIN optimization serves to counteract any non-
telecentricity of the imaging system as well as other aberrations imparted due to
the GRIN CG thickness. Moreover, the incorporation of three-dimensional refrac-
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 153

tive index change n (x, y, z) offers additional degrees of freedom that prove valu-
able in correction, especially for highly non-telecentric imaging systems, such as
in Sec. 4.3.5.

4.3.2 Petzval field curvature

Before considering a full system example, it is important to consider the specific


case of Petzval field curvature. Petzval is the lowest-order stigmatic FC aberration
which plays a pivotal role in the design of rotationally symmetric imaging systems.
For example, it is described in Sec. 4.1.2 how elements of different sign power are
required to reduce the Petzval sum in Eq. (4.1) for obtaining an adequately flat
image.
Petzval FC produces a stigmatic image on a curved surface that increases quadrat-
ically with field-of-view. The sag of this image surface can be written in terms of
its curvature cP ,
x2 + y 2
z (x, y) = cP . (4.15)
2

Substituting z (x, y) into the paraxial expression in Eq. (4.13) for the corresponding
refractive index n (x, y) yields

t
n (x, y) = t x2 +y 2
. (4.16)
n0
+ 2
cP

In order to impart a quadratic image shift to correct Petzval FC, the requisite re-
fractive index profile in Eq. (4.16) is, notably, not quadratic. Instead, the refractive
index profile for correcting Petzval is a Lorentzian function of the radial coordi-
nate, r2 = x2 + y 2 . The nonlinear relationship between n and z depicted in Fig.
4.8(b) explains this quadratic versus Lorentzian difference between sag and index,
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 154

Figure 4.9: Comparison of exact and approximate GRIN CG refractive index for correcting
Petzval field curvature. In this example, GRIN CG thickness is t = 2 mm, and the Petzval
surface curvature is cP = 0.25 mm−1 .

respectively. When numerically evaluated, however, there is but a slight difference


between the exact Lorentzian GRIN CG profile and an approximate quadratic pro-
file of equivalent refractive index change, as shown in Fig. 4.9.
The Lorentzian GRIN profile in Eq. (4.16) is based on the paraxial refractive
index expression stemming from Eq. (4.12). The refractive index according to
the non-paraxial expression in Eq. (4.11) also deviates from quadratic when at-
tempting to flatten the quadratic Petzval surface. In fact, the deviation of n from
parabolic deviates more for the non-paraxial expression, and its deviation increases
with numerical aperture.
While the refractive index profile needed to flatten the Petzval surface is not
precisely radially quadratic, the difference from the exact expression is compara-
tively small. This fact is demonstrated in Fig. 4.9 even for a highly curved image
surface. Consequently, the GRIN CG refractive index can be fairly approximated
as quadratic for the specific case of correcting Petzval FC. Then, the profile can be
optimized for further improvement alongside correction for non-telecentricity and
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 155

other introduced aberrations. This procedure of initializing the GRIN CG with a


quadratic profile and optimizing is applied next in Sec. 4.3.3 to a full system design
study.

4.3.3 Schmidt telescope design study

The Schmidt telescope, also known as the Schmidt camera, is an ingenious cata-
dioptric design that uses a spherical primary mirror [134]. In this way, the Schmidt
telescope is unlike other canonical telescopes that require conic primary mirrors,
such as the Cassegrain telescope [46]. Most telescopes achieve on-axis stigmatic
imaging by using conics as Cartesian reflectors where the stigmatic imaging prop-
erty between conic foci is leveraged. Meanwhile, the Schmidt telescope corrects
the on-axis field with only a spherical mirror by using a nominally powerless as-
pheric corrector plate, known as a Schmidt plate, located in the aperture stop of the
system [1]. The Schmidt plate is a refractive optic, which is why the full Schmidt
telescope is a catadioptric system. Although, the dispersion from the Schmidt plate
introduces chromatic aberrations, namely spherochromatism, which is not found
in fully reflective designs.
In the Schmidt telescope, the aperture stop and Schmidt plate are positioned at
the center of curvature of the spherical mirror. The benefit of positioning the stop
at the mirror center is that the chief ray is normally incident on the mirror. As a re-
sult, no primary coma, astigmatism, or distortion is introduced [111]. Remarkably,
the only residual primary aberration in the Schmidt telescope is Petzval field cur-
vature, meaning it produces a stigmatic image but on a curved surface. (Since the
system is monocentric, the curved stigmatic image lies on a concentric sphere, yet
the parabolic Petzval surface captures the majority of the behavior while higher-
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 156

order stigmatic FC terms contribute more at the edge of the field.) Nevertheless, a
downside to the design form is its long track length equal to twice its focal length
due to the stop position requirement.
A design study is performed to demonstrate FC compensation using GRIN CG.
As described in Sec. 4.1.4, conventional designs are predicated on correcting FC
for use with a planar detector, so the application of a curved detector or GRIN
CG provides little improvement. Instead, to take full advantage of curved image
technology, design forms must be devised that deliberately incorporate residual
FC in the imaging system [126, 133]. To avoid the need for designing new classes
of imaging systems, the Schmidt telescope is a prime candidate for the GRIN CG
design study due to its inherently curved image surface and otherwise excellent
performance.
Within the design study, all Schmidt telescope designs abide by the same sys-
tem specifications listed in Table 4.2. Each design is optimized independently for
the optimal Schmidt plate form using three asphere terms. By re-optimizing the
Schmidt plate for each design, overcorrected spherical aberration imparted by the
GRIN CG can be compensated by undercorrected spherical aberration from the
Schmidt plate. Consequently, if the GRIN CG were removed, the Schmidt tele-

Specification Value
Half field-of-view (HFOV) 4◦
F-number f /2
monochromatic
Spectrum
(λd = 587.56 nm)
Focal length [mm] 100.0
Mirror radius of curvature [mm] −200.0
Schmidt plate center thickness [mm] 3.5
Schmidt plate glass N-BK7
Table 4.2: Specifications for the Schmidt telescope design study using GRIN CG to per-
form FC compensation.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 157

scope would possess undercorrected spherical. Realistically, any imaging design


incorporating GRIN CG would account for its effects in the full system just as ho-
mogeneous CG is account for in the design of conventional lens systems. For each
Schmidt telescope design, the spherical mirror position is also determined by op-
timization, including its distance from the stop with Schmidt plate as well as its
distance from the detector and GRIN CG.
First, a conventional Schmidt telescope design is performed for use with a pla-
nar detector. The Schmidt plate and mirror position are optimized as described.
The resultant Schmidt telescope design is shown in Fig. 4.10(a). The field curves
clearly show a curved yet stigmatic image surface. The planar detector position
obtained in optimization captures the curved image in focus for only the ≈ 70%
half field-of-view, similar to in Fig. 4.2(b). As a result, MTF performance suffers
greatly on-axis and at full field.
Second, the conventional Schmidt design in Fig. 4.10(a) is modified for use
with a curved detector. The image surface is made quadratic (conic constant k =
−1), and its curvature is obtained by optimization, along with a new best focus
position. The optimal image surface is identified as backwards-curving, which
is the opposite of conventional refractive imaging systems with positive power.
The reason for this is the presence of a single reflection. The Schmidt telescope
design with a curved detector is shown in Fig. 4.10(b). With the curved detector,
all fields achieve diffraction-limited MTF performance. The downside to this high-
performing design is that a curved image detector is required which, as explained
in Sec. 4.1.4, poses significant difficulties in fabrication.
Third, the conventional Schmidt design with a planar detector in Fig. 4.10(a) is
restored. Then, a GRIN CG is incorporated 100 µm from the planar detector and
with a center thickness of t = 3 mm. The selected thickness falls within the range of
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 158

HFOV [deg]
ST 4.0

(a) 2.0

100 mm -0.25 0.0 0.25


Focus [mm]

HFOV [deg]
ST4.0

(b) 2.0

100 mm -0.25 0.0 0.25


Focus [mm]

HFOV [deg]
ST4.0

(c) 2.0

100 mm -0.25 0.0 0.25


Focus [mm]

HFOV [deg]
ST4.0

(d) 2.0

100 mm -0.25 0.0 0.25


Focus [mm]

Figure 4.10: Schmidt camera design study using GRIN CG for FC compensation. System
specifications are listed in Table 4.2. Ray trace, astigmatic field curves, and MTF plot are
shown for each design. Different levels of FC compensation are used including (a) planar
detector at best focus, (b) parabolic curved detector, (c) GRIN CG with analytical refractive
index according to Eq. (4.13), and (d) GRIN CG with subsequently optimized refractive
index profile. (c)-(d) Refractive index profiles are shown separately in Fig. 4.11.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 159

commercially available homogeneous CG listed in Table 4.1. The GRIN CG refrac-


tive index profile is made radially quadratic to flatten a quadratic image surface.
As discussed in Sec. 4.3.2, a radially quadratic profile is a close approximation to
the Lorentzian profile dictated by the paraxial expression in Eq. (4.16). The index
change in the quadratic profile is set analytically based on a maximum image sag
of zmax = 243.6 µm in the planar design’s field curves in Fig. 4.10(a). This analytical
radially quadratic GRIN CG profile is

n (r) = n0 + N10 r2 (4.17)

where r2 = x2 + y 2 is the radial coordinate and N10 is a scalar coefficient. The


base refractive index is set arbitrarily to n0 = 1.6 to remain within conventional
refractive index limits for the visible spectrum. Then, the coefficient N10 can be
solved according to

nmax − n0 nz
N10 = 2
=−  0 max  = −3.128 × 10−3 mm−2 (4.18)
rmax r2 t
max n0 + zmax

where nmax is calculated from zmax according to Eq. (4.13) and rmax = 7.669 mm is
the maximum radial extent of the GRIN CG based on the telescope system spec-
ifications. Based on this semi-aperture, the total refractive index change for the
analytical GRIN CG is ∆n = 0.1861.
Upon incorporating the analytically derived refractive index profile, the mirror
distance from the planar detector and GRIN CG is obtained by optimization for
best-focus. The resultant design with the analytical GRIN CG for FC compensa-
tion is shown in Fig. 4.10(c). The CG refractive index variation can also be seen
in Fig. 4.11(i). From the field curves it is confirmed that the field has been signifi-
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 160

(i) (ii)

Figure 4.11: GRIN CG in Schmidt camera designs. (i) The GRIN profile used in the design
in Fig. 4.10(c). The refractive index is obtained analytically to flatten the Petzval surface,
as specified in Eqs. (4.17) and (4.18) to flatten the Petzval surface. (ii) The GRIN profile
used in the design in Fig. 4.10(d). The refractive index in this design is initialized with the
profile in (i) and then optimized for further correction with the form in Eq. (4.19).

cantly flatted by the GRIN CG. The corresponding MTF performance is also much
improved compared with the planar detector design in Fig. 4.10(a). Although, the
analytical GRIN CG design does not reach the field flatness or MTF performance
of the curved detector design in Fig. 4.10(b).
Fourth, the Schmidt telescope design with analytical GRIN CG in Fig. 4.10(c)
is revisited. As described in Sec. 4.3.1, the analytical GRIN CG relies on assump-
tions of telecentricity across the field-of-view as well as the paraxial approxima-
tion, which do not hold true in any physical system. To compensate for these
deviations from the analytical model, optimization of the GRIN CG profile can be
performed for further performance improvement. Starting from the design in Fig.
4.10(c), the GRIN CG profile is expanded to include four radial terms,

n (r) = n0 + N10 r2 + N20 r4 + N30 r6 + N40 r8 (4.19)


Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 161

where coefficients N10 , N20 , N30 , and N40 are obtained by optimization. Note that
axial refractive index terms could also be incorporated if larger chief ray angles
were present. The Schmidt telescope design in Fig. 4.10(a) has a maximum chief
ray angle of 4◦ , so the added complexity of axial refractive index change is deemed
unnecessary. The optimized GRIN CG also serves to compensate for other errors
including any spherical aberration imparted by the CG as well as differences be-
tween the derived Lorentzian and quadratic profiles (see Sec. 4.3.2). Within the
GRIN CG optimization, the image distance is also obtained by optimization for
best focus.
The final Schmidt telescope design with optimized GRIN CG for FC compen-
sation is shown in Fig. 4.10(d). The optimized CG refractive index profile is also
shown in Fig. 4.11(ii). The additional optimization step has further flattened the
field and achieves diffraction-limited performance across the field, matching that
of the curved detector design in Fig. 4.10(b). The total refractive index change
∆n = 0.2175 increases by 17% for the optimized GRIN CG compared with the
analytical solution.
The results of the Schmidt telescope design study clearly show how GRIN CG
can successfully compensate for FC aberrations. The optimized GRIN CG design
in Fig. 4.10(d) performs equally well to the curved detector design in in Fig. 4.10(b)
but without the need for mechanically curving the image detector. Instead, the
GRIN CG serves to artificially curve the image surface while still requiring only a
planar detector. Moreover, the GRIN CG is incorporated directly into the detector
stack, offering the same self-contained unit found in conventional detectors with
homogeneous CG. As a result, the advantages of curved detector technology can
be accessed without the fabrication difficulties described in Sec. 4.1.4 for mechani-
cal curving a CMOS or CCD detector.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 162

4.3.4 Connection to conventional field flatteners

The Schmidt telescope design study in Sec. 4.3.3 elucidates a connection between
GRIN CG and the conventional concept in lens design of a field flattener [124].
The GRIN CG refractive index profiles shown in Fig. 4.11 are reminiscent of Wood
lenses [18] due to their dominant radially quadratic term. The optical power φ pre-
sented by a Wood lens means, if positioned near a conjugate plane, it can serve as
a field flattener by contributing a balancing term to the Petzval sum in Eq. (4.1)
without significantly affecting other system behavior. Thus, the GRIN CG in the
Schmidt telescope designs can be recognized as essentially a GRIN-based field flat-
tener.
Although, due to its PPP form factor, GRIN CG can be positioned immediately
next to a planar image and incorporated directly in an image detector stack, re-
placing the ubiquitous homogeneous CG. Meanwhile, conventional homogeneous
field flatteners are incorporated in the design of the imaging system, not as part of
the detector system. This subtle difference is critical since it allows GRIN CG to
be used in systems where conventional field flatteners are disallowed due to me-
chanical constraints. For example, single-lens reflex (SLR) camera bodies possess a
flange distance of tens of millimeter that prevent conventional field flatteners from
being positioned adjacent to the image detector.
Finally, for the Schmidt telescope, GRIN CG takes a form reminiscent of a field
flattener by balancing the Petzval sum, yet it is reductive to claim GRIN CG is
simply a GRIN embodiment of a field flattener. The generality of the GRIN CG
framework devised in Sec. 4.3.1 allows for FC compensation of any order or sys-
tem symmetry. This framework is deliberately constructed to be applicable for an
arbitrary image surface sag z (x, y) with no restriction on symmetry or whether it is
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 163

a convex surface. In fact, GRIN CG for correcting non-monotonic or freeform im-


age surfaces offers capability that surpasses that of the conventional field flattener.
Two such examples of GRIN CG compensating higher-order and non-rotationally
symmetric FC aberrations is highlighted next in Sec. 4.3.5

4.3.5 Future work: higher-order application spaces

Many different applications spaces serve to benefit from FC compensation using


GRIN CG. As described for curved detectors in Sec. 4.1.4, new system config-
urations can now be achieved [126, 127, 133] when FC correction in the imaging
system is neglected for subsequent correction in the GRIN CG. The process of de-
vising these new system architectures is a much broader problem than the subject
covered in this chapter. For this reason, the pre-established Schmidt telescope is
the subject of the preceding design study in Sec. 4.3.3 since it inherently has a sig-
nificant amount of residual FC. Yet, the arbitrary image surface z (x, y) in the GRIN
CG formulation in Sec. 4.3.1 means higher-order FC can also be compensated in
more complex systems.
While no further design studies are performed, it is helpful to consider some
application spaces that leverage higher-order FC compensation offered by GRIN
CG. There are two main categories of interest: (1) rotationally symmetric, non-
monotonic image surfaces and (2) freeform image surfaces.

Rotationally symmetric, non-monotonic image surfaces

First, in rotationally symmetric imaging systems, different orders of FC aberra-


tions may present contributions of opposite sign in forming the ultimate image
surface. As a result, the image sag z may change non-monotonically with field-of-
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 164

view increasing from the optical axis while still remaining rotationally symmetric
about the axis (e.g., in Fig. 4.1). For these cases, a mechanically curved image
detector would also need to be deformed to a matching non-monotonic surface
for FC compensation. One example of non-monotonic FC requires a “gull-wing”
aspheric image detector, which requires even greater complexity in mechanically
curving the detector surface.
One of the most common occurrences of non-monotonic FC in rotationally sym-
metric systems is in the compact imagers used in mobile devices [154]. These sys-
tems employ several highly aspheric optical surfaces, and as a result, the image
surface is often non-monotonic and sometimes presents two or more inflection
points [155]. For a non-monotonic image sag z to be corrected, a non-monotonic
GRIN CG profile is required according to Eq. (4.13) and the n, z relationship plot-
ted in Fig. 4.8. Fortunately, new GRIN fabrication techniques such as additive
manufacturing [38] allow for creation of non-monotonic GRIN profiles.
Within the design space of mobile device imagers, it is standard for the sys-
tem track length to be highly constrained [154]. The reason for this is the imager
thickness directly impacts the device thickness. In some devices, the imager is long
enough that it protrudes from the device, resulting in a “bump” in the device con-
tour. Consequently, it is desirable to reduce system length without a subsequent
loss in image performance. With airspaces already being small, the most effective
way of reducing system length is by decreasing the number of elements. A con-
ventional mobile device imager is illustrated in Fig. 4.12(a) which uses six highly
aspheric elements, followed by a homogeneous CG. The first three elements in the
design are positive-negative-positive in power, which similar to in a Cooke triplet,
serves to flatten the field. The latter three elements then provide further correction
of FC as well as distortion and telecentricity control.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 165

z(x,y) n(x,y)

Refractive index
(a) (b) (c)

Figure 4.12: Concept illustration of applying GRIN CG to reduce track length in a mobile
device imager by performing FC compensation. (a) A conventional mobile device imager
applies six highly aspheric elements followed by a homogeneous CG [156]. (b) A reduction
from six to four elements decreases system track length while sacrificing field flatness with
curved image surface z (x, y). (c) The incorporation of GRIN CG compensates for FC while
maintaining the reduced system length. In this example, the GRIN CG three-dimensional
refractive index variation accommodates the highly non-telecentric system.

For the example mobile device imager in Fig. 4.12(a), a reduction in track length
can be illustrated by the addition of GRIN CG. Removing two elements, for in-
stance, reduces the system track length significantly, such as in the hypothetical
system shown in Fig. 4.12(b). Although, the reduced element count sacrifices field
flatness with a now non-monotonically curved image surface z (x, y). The curved
image surface decreases imaging performance when used with a planar detector.
Then, GRIN CG with a non-monotonically varying refractive index can be incor-
porated to flatten the image surface, as shown in Fig. 4.12(c). The GRIN profile
can subsequently be optimized for further performance improvement, as done for
the Schmidt telescope in Sec. 4.3.3. Mobile device imagers are typically highly
non-telecentric. As a result, the optimized GRIN CG may also benefit from axial
refractive index variation, as is depicted in Fig. 4.12(c).
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 166

Freeform image surfaces

A second application space worth considering is a freeform imaging system. Due


to the arbitrary form of z (x, y) in Eq. (4.13), the refractive index profile n (x, y)
providing FC correction is also left arbitrary. In freeform imaging systems, z (x, y)
is not constrained by any degree of symmetry, meaning non-rotationally symmet-
ric as well as non-monotonic image surfaces are allowed. As a result, correcting
freeform FC aberrations requires more complex GRIN CG. Moreover, since both
stigmatic and astigmatic FC must be accounted for in the medial image surface,
the determination of z (x, y) itself becomes more complicated.
As an illustration of mitigating freeform FC, a prototypical three-mirror freeform
telescope is shown in Fig. 4.13(a). The system configuration is such that the field
is sufficiently flat. To reduce the system volume, mirrors can be made stronger
and tilted further, as depicted in Fig. 4.13(b). Although, the resultant design now
possesses a curved image surface. The freeform image surface lacks rotational

z(x,y) n(x,y)

Refractive index

(a) (b) (c)

Figure 4.13: Concept illustration of GRIN CG used to compensate freeform FC aberra-


tions. (a) Conventional three-mirror freeform telescope has a sufficiently planar image
surface [11]. (b) System volume is reduced by increasing mirror power and surface tilt. As
a result, the image surface z (x, y) suffers from freeform FC. (c) The addition of GRIN CG
is used to compensate for freeform FC aberrations while keeping system volume reduced.
Chapter 4. Gradient-Index Cover Glass for Field Curvature Compensation 167

symmetry and is non-monotonic. The curved image reduces image quality when
captured by a planar detector. To solve this issue, GRIN CG is incorporated with
a freeform refractive index variation, as shown in Fig. 4.13(c). The GRIN CG miti-
gates freeform FC aberrations while keeping the system volume reduced.
Chapter 5

Design, Fabrication, and Metrology


of F-GRIN Prescribed Illumination
Optics

5.1 Prescribed illumination design

Illumination design seeks the solution to the inverse problem depicted in Fig. 5.1.
Given a light source and an illumination target, what intermediate optic or optical
system is required to perform the necessary transformation from source to target?
The source is defined by its étendue, i.e., its spatial and angular distribution of
emitted light, as well as its position in space relative to the illumination optic.
Similarly, the illumination target is specified by the étendue of incident light as
well as its position in space. Étendue conservation requires the target étendue to be
at most equal to that of the source with no diffraction, scattering, or amplification
occurs [157].
Solutions to this illumination design problem are generally obtained using the
principals of geometrical optics by applying refraction, reflection, and/or total in-
ternal reflection (TIR) [158]. Traditionally, designs have maintained a degree of
symmetry that is also reflected in the achievable target distributions. Also, illu-

168
Chapter 5. F-GRIN Prescribed Illumination Optics 169

Source Optic Target

y
x
z

Figure 5.1: The inverse problem central to illumination optics, including prescribed illumi-
nation design.

mination optics are classically made of homogeneous media, exerting their optical
influence by the form of their optical surfaces. There are, however, some prior ex-
amples of gradient-index materials used in non-imaging optics, primarily in the
spaces of beam mode conversion [35, 159] and solar concentration [160–162].
Enabled by new fabrication techniques, freeform optics have reshaped the land-
scape of illumination design [163] as they have done for imaging design [4]. Cer-
tain illumination targets previously unattainable with conventional optical compo-
nents are now achievable with the departure from symmetry offered by freeform
optics. These freeform illumination targets may have no degree of symmetry, pos-
sess higher spatial frequency features, and offer discontinuities such as holes or
sharp edges.
Freeform illumination optics have spurred the field of “prescribed” or “tai-
lored” illumination design where the goal is to generate an arbitrary illumination
target using a generalized design method [163]. Examples of prescribed illumina-
tion targets range from letters, symbols, and geometric shapes to photorealistic im-
ages [164–168]. Combined with the development of new solid-state light sources,
prescribed illumination optics have prompted new solutions in a variety of appli-
Chapter 5. F-GRIN Prescribed Illumination Optics 170

Surface sag
z(x, y)

y
x
z

Figure 5.2: Prescribed illumination generated using a freeform surface with sag z (x, y).
The black contour on the surface is a slope discontinuity needed to form the hole in the
illumination target.

cation spaces including automotive, street, architectural, and structured lighting.


Moreover, unlike alternative technologies such as engineered diffractives [169] or
scatterers [170], freeform prescribed illumination optics rely on the principles of
geometrical optics (i.e., refraction and reflection) for their optical influence. As a
result, their functionality is also independent of the coherence and polarization
state of incident light.
The vast majority of prescribed illumination literature features freeform optical
surfaces [163]. Several different design techniques have been presented where for
each method the surface sag z (x, y) is sought that generates the prescribed illumi-
nation, as depicted in Fig. 5.2. At the heart of each design method is either the law
of refraction or reflection (i.e., Snell’s law). In either case, the mathematical form
of ray tracing for homogeneous media is simple enough to support these various
design procedures.
One design technique obtains the requisite freeform surface by solving a non-
linear, second-order partial differential equation of Monge-Ampère (MA) type [171,
172]. In general, the degree of complexity in solving the MA equation requires
numerical techniques where various options have been demonstrated [164, 166–
Chapter 5. F-GRIN Prescribed Illumination Optics 171

168, 173, 174]. An alternate approach known as “ray mapping” determines a map-
ping from source to optic to target. Then, the optical surface is solved for that
satisfies the mapping according to geometrical optics. The key to ray mapping
approaches is to ensure the mapping is a favorable one, meaning it results in well-
behaved surfaces [175]. A common means of obtaining and solving the mapping
is by formatting the problem similar to the optimal transport problem [176–179].
A third design technique known as the supporting quadratic method (SQM) lever-
ages the stigmatic imaging of conic surfaces to discretely generate the prescribed
target [180–184].
Furthermore, there are several considerations one encounters when designing
prescribed illumination optics. These various factors prove relevant regardless of
design method or type of freeform optic (i.e., surface versus GRIN). One difference
is whether the illumination target is to be generated in the near-field or far-field.
Near-field targets are generated in terms of irradiance (spatial distribution) at a
near distance [180] while far-field targets are generated in terms radiant intensity
(angular distribution) at a large enough distance that the optic extent is negligible
[173, 178, 179].
Another consideration is the spatial extent of the source. Many design methods
consider an infinitesimal point source possessing no spatial extent. A point source
has zero étendue and is a nonphysical entity, but it offers a convenient approxima-
tion for light sources that are small relative to the optical element. A convenient
rule of thumb known as the “5 times” rule offers a metric that any source extent
less than 5 times the source distance can be approximated as a point source [185].
For optics designed with a point source, finite source extent can be seen as a per-
turbation that reduces illumination contrast, particularly at any sharp edges [186].
For small sources, the resultant contrast loss can be mitigated to a degree by ad-
Chapter 5. F-GRIN Prescribed Illumination Optics 172

justing the target discretization [187, 188]. Meanwhile, design methods capable
of accommodating large extended sources that surpass the “5 times” rule are far
fewer in number [186, 189, 190]. The downside to these methods is that multiple
freeform optics are required for tailoring the source which is now extended both
spatially and angularly. Many consider designing for extended sources to be the
greatest challenge facing prescribed illumination design.
A third factor in prescribed illumination design is the number of unique inten-
sity levels found in the target. For the case of white light, the number of gray levels
can also be thought of as the target’s “bit depth,” a term borrowed from the repre-
sentation of digital images. For example, a 1-bit target is a binary target with only
two intensity levels, white (100% relative intensity) and black (0% relative inten-
sity). Grayscale targets with greater bit depths can be achieved by incorporating a
concentrating effect in the optical element’s design.
Lastly, the spatial appearance of the target directly influences the appearance
of the corresponding prescribed illumination design. The spatial frequency con-
tent of the target is mirrored in the optical element itself. For instance, a freeform
surface generating the letter ‘S’ has the letter ‘S’ apparent in its surface sag. As
a result, targets such as photorealistic images require fabrication tolerances with
micron-scale features [168, 191].
Moreover, targets with discontinuities such as holes and sharp edges can only
be generated by illumination optics possessing discontinuities in their imparted
phase. The sharpness of an as-built discontinuity directly relates to the sharpness
of the produced target discontinuity. For freeform surfaces, the requisite disconti-
nuities are incorporated via discontinuities in the surface slope [165, 178, 179, 192].
These surface slope discontinuities take the form of cusps and pose a challenge
in fabrication. For the case of GRIN illumination optics, the required discontinu-
Chapter 5. F-GRIN Prescribed Illumination Optics 173

ities are within in the refractive index gradient. Fortunately, recent GRIN fabrication
techniques, namely additive manufacturing, have enable direct incorporation of
gradient discontinuities.

5.2 F-GRIN for prescribed illumination

The objective of prescribed illumination design can be accomplished using a dif-


ferent kind of freeform optic than the freeform surfaces applied previously, namely
freeform gradient-index (F-GRIN) optics [33,34,36,193]. The freeform optical influ-
ence of F-GRIN is imparted by its spatial variation in refractive index n (x, y, z), as
shown in Fig. 5.3, as opposed to the surface sag z (x, y) of freeform surfaces. As a
result, a unique characteristic of F-GRIN prescribed illumination optics is that they
require simply plane-parallel surfaces. That is not to say F-GRIN illumination op-
tics can only function with planar surfaces; rather, it is a feature that distinguishes
them from freeform surface designs.
Another difference compared with freeform surface illumination optics is how

Refractive index
n(x, y, z)

y
x
z

Figure 5.3: Prescribed illumination generated using a freeform gradient-index (F-GRIN)


optic with refractive index n (x, y, z). The black contour in the refractive index is a gradient
discontinuity needed to form the hole in the illumination target. Unlike for the freeform
surface in Fig. 5.2, the F-GRIN illumination optic requires only plane-parallel surfaces.
Chapter 5. F-GRIN Prescribed Illumination Optics 174

discontinuities are incorporated within the F-GRIN to generate illumination tar-


gets containing holes or sharp edges. Where freeform surfaces apply disconti-
∂z ∂z
nuities in the surface slope, ∂x
and ∂y
, F-GRIN discontinuities are implemented
by means of a discontinuity in the refractive index gradient, ∇n. This distinction
presents an advantage for F-GRIN designs due to recent innovations in the addi-
tive manufacturing of GRIN media [37, 38, 40]. By the nature of additive manu-
facturing where discrete points of refractive index are deposited, gradient discon-
tinuities can be readily incorporated by suddenly changing the gradient between
adjacent voxels.
A limitation of prescribed illumination design using F-GRIN is there is no equiv-
alent to the simple formulae of geometrical optics – Snell’s law and the law of re-
flection – used when designing with freeform surfaces. Consequently, freeform
surface design methods cannot be directly ported for use with F-GRIN. Although,
some techniques and concepts in freeform illumination design can be adapted for
use with F-GRIN. Specifically, the F-GRIN design process described in this chap-
ter is an amalgamation of the ray mapping and SQM freeform surface methods
described in Sec. 5.1.
As explained in Chapter 1, there is, in general, no analytical expression for the
ray path within F-GRIN due to its lacking degrees of symmetry [17]. Rather, ray
tracing is performed using numerical techniques that are iterative and serve as an
approximation that ideally provides adequate accuracy and precision [19–21]. In
order to enact the base concepts of prescribed illumination design that have been
devised for freeform surfaces, analytical tools of some extent are required for GRIN
design. For this work, linear GRIN, which has a linear change in refractive index
along an arbitrary axis, is leveraged as the analytical foundation for the design
process. Due its symmetry, the general form of the ray path within linear GRIN
Chapter 5. F-GRIN Prescribed Illumination Optics 175

can be derived analytically, as described in Sec. 5.3. Then, F-GRIN profiles can be
composed from a series of discrete linear GRIN sections, as covered in Sec. 5.4.2.
Finally, the linear GRIN arrays can subsequently be integrated for a piecewise-
continuous design, as discussed in Sec. 5.4.

5.2.1 Design scope

There are many different options when it comes to prescribed illumination design.
As described in Sec. 5.1, the different properties of the source, optic, and target all
serve as potential variables. The design work presented in this chapter limits its
scope to certain properties and configurations.
First, the system configuration is considered in three-dimensional Cartesian co-
ordinates and consists of the source, optic, and target all centered upon an axis, as
depicted in Fig. 5.3. This axis is referred to as the optical axis and is taken to
be coincident with the z-axis. The coordinate system origin is located at the front
surface vertex of the illumination optic.
The light source considered in the design process is an infinitesimal point source.
Lacking any spatial extent, the source has zero étendue and, thus, is nonphysical.
As a result, evaluation of a physical source containing some spatial extent yields
a loss in illumination target contrast, as expanded upon in Sec. 5.4.4. The source
position is defined relative to the coordinate origin at the front surface vertex of
the illumination optic. Also, the point source emits uniformly in angle into the ac-
ceptance cone of the optic. The acceptance cone of the optic is based on the source
distance and the optic’s clear aperture. Finally, the design process considers only a
point source, but the developed simulation tools do allow for ray trace evaluation
of a circular or rectangular extended source, as demonstrated in Sec. 5.4.4.
Chapter 5. F-GRIN Prescribed Illumination Optics 176

The F-GRIN illumination optic is limited to planar surfaces and in a plane-


parallel configuration that is centered on and normal to the optical axis. This
window-like form factor is desirable for certain packaging constraints and is un-
achievable with freeform surface designs. Plane-parallel surfaces are also the most
easily attainable option in fabrication. The clear aperture of the considered F-GRIN
designs are rectangular in shape and on the scale of millimeters to centimeters.
The F-GRIN designs are also intended to be fabricated by additive manufacturing,
meaning freeform index variation in three dimensions and gradient discontinu-
ities are both allowed. The effect of material properties and fabrication constraints
(e.g., ∆n) is discussed later in Sec. 5.5. While additive manufacturing allows for
three-dimensional index variation, in this work the F-GRIN index profile is lim-
ited to the two transverse dimensions while remaining axially constant, n (x, y).
This restriction offers a reduction in complexity while not limiting what targets are
achievable with a point source, as becomes clear in Sec. 5.4. Moreover, all designs
are evaluated monochromatically or, equivalently, are taken to be dispersion-less.
It is assumed that any chromatic effects that may appear when used with a poly-
chromatic source are negligible compared with other as-built effects from current
fabrication methods. Lastly, unlike for freeform surface designs, the F-GRIN de-
signs are, naturally, refractive in nature with no option for purely reflective de-
signs.
The illumination target is designed for observation in a plane that is centered
on and orthogonal to the optical axis. Illumination targets are only demonstrated
for far-field distances, although the design process can be easily adapted for near-
field targets. In either case, the target evaluation distance is measured from the
coordinate origin at the front surface vertex of the optic. Different target bit depths
are also considered, and designs for both binary as well as grayscale targets are
Chapter 5. F-GRIN Prescribed Illumination Optics 177

presented in Sec. 5.5. The design process presented in Sec. 5.4 readily adapts for
either option using the same procedure. Finally, no limitation is placed on target
discontinuities such as holes and sharp edges. This agrees with the allowance of
gradient discontinuities in F-GRIN by means of additive manufacturing.

5.3 Analytical linear GRIN ray tracing

Some degree of analyticity is advantageous when devising a design process for


F-GRIN prescribed illumination optics. Without it, numerical techniques such as
stepwise ray tracing [19–21] and optimization [117] must solely be relied upon.
Optimization techniques are likely to struggle with the F-GRIN illumination de-
sign problem due to its nonsmooth and highly nonlinear conditioning [117]. Also,
commercially available illumination software (e.g., LightTools [194]) does not eas-
ily support the modeling and optimization of such optics. For these reasons, hav-
ing an analytical backbone to inform the design process is highly desirable. That
is not to say an equation must be devised that directly yields the optimal illumina-
tion design as its solution. Rather, the design process presented next in Sec. 5.4 is
a hybrid of both analytical and numerical techniques.
The analytical foundation for the design process relies on linear GRIN media.
Linear GRIN has a refractive index variation that changes linearly along a one-
dimensional axis. The linear gradient direction can be defined in three-dimensional
Cartesian coordinates by the unit vector ĝ = gx x̂ + gy ŷ + gz ẑ. Mathematically, the
refractive index profile n (~r) can then be described by

n (~r) = n0 + α (~r · ĝ) (5.1)


Chapter 5. F-GRIN Prescribed Illumination Optics 178

where ~r is the position vector, n0 is the base refractive index located at the coordi-
nate origin, and α is the slope of the linear refractive index change per unit distance
along ĝ. This can be seen directly by considering the refractive index gradient ∇n
of Eq. (5.1),
∇n = α (gx x̂ + gy ŷ + gz ẑ) = αĝ. (5.2)

For linear GRIN, there is no requirement on the gradient direction ĝ. The gradi-
ent is free to have components in all three dimensions, as is the case in Fig. 5.4(a).
When parallel to the optical axis direction, ĝ = ẑ, the canonical axial gradient is
obtained, as in Fig. 5.4(b). However, in this work, ĝ is only considered orthog-
onal to ẑ, oriented transversely in the x-y plane, as in Fig. 5.4(c). Also, to avoid
redundancy, the refractive index slope α is restricted to positive values while the
gradient direction ĝ determines the direction of increasing slope.
Confined to the x-y plane, it becomes more convenient to define the gradient
direction in two dimensions by ĝ = sin θg x̂ + cos θg ŷ where θg is the gradient angle
with respect to the y-axis, as shown in Fig. 5.5(a). A new g-axis is defined along the
gradient direction, emanating from the origin located at the front surface vertex.

Refractive index

y
z
(a) (b) (c) x
Figure 5.4: Three different examples of linear GRIN with gradient directions ĝ (a) in three
dimensions, (b) parallel to z (i.e., axial gradient), and (c) orthogonal to z in the transverse
x-y plane. (c) Linear GRIN defined by its gradient in the x-y plane is what is used in the
prescribed illumination design process.
Chapter 5. F-GRIN Prescribed Illumination Optics 179

y
g θg g

Refractive index
x z
z

(a) (b)

Figure 5.5: Coordinates and geometry of linear GRIN used in the design process, shown
in cross-section. (a) The transverse x-y cross-section shows the gradient direction along
the g-axis, defined by the angle θg measured from the y-axis. (b) The longitudinal cross-
section shows the g-z plane containing the gradient and optical axis. The coordinate origin
is located at the front surface vertex.

The ray path in a linear GRIN can be derived analytically from Fermat’s princi-
ple for the described geometry. Fermat’s principle requires in a medium of refrac-
tive index n that the optical path integral

Z s1
P = n ds (5.3)
s0

is stationary with respect to variations in ray path s between start and end points,
s0 and s1 [16]. In homogeneous media, the familiar linear ray path derives from
this formal definition when n is spatially constant. On the other hand, for inho-
mogeneous media where n (~r) is spatially varying, the ray path satisfying Fermat’s
principle may be obtained from the calculus of variations using the Euler-Lagrange
equation. In Cartesian coordinates, the Lagrangian L of Eq. (5.3) can be written as

1/2
L = n (x, y, z) x0 2 + y 0 2 + z 0 2 (5.4)
Chapter 5. F-GRIN Prescribed Illumination Optics 180

where (· · · )0 indicates d/ds. For this Lagrangian, the Euler-Lagrange equation be-
comes, in vectorial form [17],
0
(n~r 0 ) = ∇n. (5.5)

There is no general solution to Eq. (5.5) for arbitrary refractive index distribu-
tions [17]. For this reason, numerical solutions to Eq. (5.5) are commonly used
in optical design software to accommodate any differentiable refractive index dis-
tribution [19–21]. Although, in some instances, symmetry in the refractive index
allows for simplification of Eq. (5.5) to a more manageable form.
Fortunately, the symmetry inherent in linear GRIN means an analytical solution
to the ray path is attainable. Recall, this is one of the primary reasons why linear
GRIN is leveraged in the design process. For the linear GRIN geometry in Fig.
5.5, the refractive index n (x, y) is unchanging along z. As a result, the optical path
integral in Eq. (5.3) can instead be written as

Z s1 1/2
P = n (x, y) 1 + ẋ2 + ẏ 2 dz (5.6)
s0

where a dot indicates d/dz. Thus, a different Lagrangian is recognized than the
one in Eq. (5.4),
1/2
L = n (x, y) 1 + ẋ2 + ẏ 2 , (5.7)

and Euler-Lagrange now gives two equations rather than three,

 
d ∂L ∂L
=
dz ∂ ẋ ∂x
  (5.8)
d ∂L ∂L
= .
dz ∂ ẏ ∂y

Proper manipulation of Eq. (5.8) by logarithmic differentiation gives a pair of cou-


Chapter 5. F-GRIN Prescribed Illumination Optics 181

pled nonlinear second-order differential equations, as presented by Moore [195]


and derived by Marchand in Appendix A of [17],

 
∂n ∂n
1 + ẋ2 + ẏ 2 = 0

nẍ + ẋ −
∂z ∂x
  (5.9)
∂n ∂n
1 + ẋ2 + ẏ 2 = 0.

nÿ + ẏ −
∂z ∂y

From Eq. (5.9), the ray path in a linear GRIN can be derived. Initially, the ray
path is considered in the g-z plane depicted in Fig. 5.5(b), containing the linear
gradient and the optical axis. Skew rays lying outside the g-z plane can ultimately
be traced using the same result but with some modification, as explained in Sec.
5.3.1. The plane symmetry about g-z and the lack of axial refractive index depen-
dence means Eq. (5.9) can be written as a single differential equation for the ray
path in the g-z plane,
∂n
1 + ġ 2 = 0.

ng̈ − (5.10)
∂g

Finally, the linear GRIN ray path g (z) can be solved from Eq. (5.10) as a function
of axial coordinate. The ray initial conditions must be specified, namely its initial
position and slope. Without loss of generality, the initial ray position is taken to be
the coordinate origin, g (z = 0) = 0, where the refractive index is n0 according to
Eq. (5.1). (Incidence heights g 6= 0 can be accommodated by simply adjusting the
value of n0 to match the refractive index at the point of incidence.) Meanwhile, the
initial ray slope is represented by β, making the two initial conditions

g (z = 0) = 0
(5.11)
ġ (z = 0) = β.

Then, the ray path g (z) can be derived from Eq. (5.10), as demonstrated in Ap-
Chapter 5. F-GRIN Prescribed Illumination Optics 182

pendix D, " ! #
p
n0 1 α 1 + β2
g (z) = p cosh z+c −1 (5.12)
α 1 + β2 n0

where
 p 
c = sinh−1 β = ln β + β 2 + 1 . (5.13)

Linear GRIN as applied in the illumination design process uses plane-parallel


surfaces with center thickness t, as depicted in Fig. 5.6. The linear GRIN is also
assumed to be immersed in air, nair = 1. Thus, the initial ray slope within the
linear GRIN β can subsequently be obtained by Snell’s law in terms of the angle
of incidence θ1 . With the ray incident on the coordinate origin where n0 is the
refractive index, β is found to be

sin θ1
β=p . (5.14)
n20 − sin2 θ1

From Eq. (5.12), the ray path within a linear GRIN is seen to be hyperbolic cosine
in form. Moreover, the section of the hyperbolic cosine function constituting the
Refractive index

θ1 θ2
θ1 θ2
g g(z)
z
t
Figure 5.6: Surface refraction at a linear GRIN. The ray is traced in the g-z plane containing
the gradient and the optical axis. The ray path g (z) is a section of the hyperbolic cosine
function determined by the ray initial conditions.
Chapter 5. F-GRIN Prescribed Illumination Optics 183

ray path is determined by matched boundary conditions for the angle of incidence
θ1 by means of the coefficient c.
Snell’s law must also be performed at the rear planar surface, located at axial
coordinate z = t, in order to know the ray exit angle θ20 . First, the ray slope can be
obtained by differentiating Eq. (5.12) with respect to z,

p !
dg (z) α 1 + β2
= sinh z+c . (5.15)
dz n0

Upon evaluating Eq. (5.15) at z = t, the angle of incidence θ2 is then known. The
angle of refraction in air θ20 following the linear GRIN is subsequently given by
Snell’s law, p !
n0 α 1 + β2
sin θ20 = p sinh t+c . (5.16)
1 + β2 n0

The expression in Eq. (5.16) for the exit angle θ20 is exact and also derived in Ap-
pendix D.
In addition to its analytical ray expressions, linear GRIN is excellently posi-
tioned for use in GRIN illumination design due to a very convenient feature. Ray
bundles incident on a linear GRIN are deflected in the direction of the gradient
but without taking on optical power or any significant aberration. In this way, a
linear GRIN acts very similar to that of a tilted planar mirror at glancing incidence,
redirecting light without significantly affecting the properties of the beam. Impor-
tantly, this holds true regardless of the state of the incident light. For example, the
beam deflecting property of linear GRIN is depicted in Fig. 5.7 for (a) collimated
light, (b) diverging light, and (c) converging light. Thus, linear GRIN can be di-
rectly applied in illumination design because the gradient direction θg and slope α
can be solved for to redirect incident light wherever prescribed.
Chapter 5. F-GRIN Prescribed Illumination Optics 184

(a)

Refractive index
(b)

(c)

Figure 5.7: Beam deflecting property of linear GRIN, demonstrated for (a) collimated light,
(b) diverging light, and (c) converging light. No significant power or aberration is introduced
by the linear GRIN.

5.3.1 Skew ray tracing

Before proceeding to the prescribed illumination design process, the added com-
plication of tracing skew rays in linear GRIN must be explained. As defined in
Sec. 5.3, skew rays propagate outside of the g-z plane containing the gradient and
optical axis. Meanwhile, the ray path derived in Sec. 5.3 abides by the geometry in
Fig. 5.8(a) where the ray travels within the g-z plane. As a result, skew rays cannot
be directly accounted for by these formulae since they are not restricted to the g-z
plane and, therefore, seemingly break the plane symmetry used in obtaining the
simplified form of Eq. (5.10). Fortunately, the results in the g-z plane can be gen-
eralized for use with any incident ray by following the proper procedure. The key
reason for this is that, upon refraction into the linear GRIN, a planar cross-section
can always be formed that contains the gradient and the initial ray direction, as
shown in Fig. 5.8(b). This planar cross-section continues to maintain a linear GRIN
Chapter 5. F-GRIN Prescribed Illumination Optics 185

(a) (b)

Figure 5.8: Ray tracing in a linear GRIN for (a) a ray incident in the g-z plane containing the
gradient and the optical axis and (b) a skew ray incident outside the g-z plane. The planar
cross-section for the skew ray obliquely dissects the volume but continues to experience a
linear gradient. (a)-(b) In these cases, the gradient is aligned along the y-axis, although this
not a requirement, and the same skew planar cross-section can be obtained regardless of
θg .

profile, allowing for the use of the derived ray formulae, even though the ray is not
confined to the global g-z plane. Practically, this adjustment is performed by the
proper changes in coordinate reference frame to realign the refracted ray with the
familiar 2D framework applied in Sec. 5.3.
Breaking from plane symmetry, skew rays must be described in three dimen-
sions with direction represented by a unit vector ~r = rx x̂ + ry ŷ + rz ẑ where rx , ry ,
and rz are the ray direction cosines. For any ray incident upon the front surface of
a linear GRIN, refraction is specified by the vectorial form of Snell’s law,

   
~ = n0 r~0 × N
n ~r × N ~ . (5.17)

~ at the point of ray-surface intersection is also a unit vector,


The surface normal N
~ = 1· ẑ.
and for linear GRIN with planar surfaces normal to the z-axis, it is simply N
Chapter 5. F-GRIN Prescribed Illumination Optics 186

For this simplified case, the refracted ray r~0 is found by

n0 r~0 = n~r + (n0 cos I 0 − n cos I) ẑ (5.18)

~ = rz and I 0 can be found from I using two-dimensional Snell’s


where cos I = ~r · N
law. Also, notice in Eq. (5.18) that it is often easiest to track the optical direction
cosine for the ray, which is the ray geometrical direction cosine multiplied by the
immersing refractive index, n~r = Lx̂ + M ŷ + N ẑ.
By vectorial Snell’s law, it is known that the incident ray ~r, refracted ray r~0 , and
~ must all be coplanar. There is no guarantee, however, that this
surface normal N
plane is tangent to the g-z plane constituted by the gradient and optical axis. Al-
though, the key realization that allows for convenient skew ray tracing is that it
is always possible to identify the planar cross-section of the linear GRIN that con-
tains r~0 as well as the gradient direction ĝ at the point of ray-surface intersection.
(The only exception to this is when the ray and gradient are parallel, although this
is not allowed in the given geometry where the gradient is necessarily perpendicu-
lar to the optical axis.) Thus, the same results from Sec. 5.3 can be used for tracing
skew rays since they are again confined to a two-dimensional plane, albeit a plane
that differs from the g-z plane. This alternate planar cross-section can be evaluated
instead via the proper changes in coordinate reference frame. One final note: while
within the linear GRIN, the skew ray is confined to the plane formed by its initial
ray direction r~0 and the gradient ĝ; however, when considering its transfer in air to
and from the GRIN, the skew ray’s total path does not lie in a plane and instead
must be considered three-dimensionally, regardless of change in coordinate refer-
ence. For this reason, the coordinate change is applied only to the ray’s path within
the linear GRIN and is reverted before transfer in air following the GRIN.
Chapter 5. F-GRIN Prescribed Illumination Optics 187

After performing refraction, the ray is within the linear GRIN with some initial
position, taken here to be the origin, and with some initial direction r~0 = rx0 x̂+ry0 ŷ +
rz0 ẑ. The first step in determining the skew ray path is to rotate the global reference
frame, including both the GRIN and the ray, such that the gradient direction ĝ is
oriented along a Cartesian axis such as the ŷ-direction. For the case of aligning
ĝ with ŷ, this operation is a roll about the z-axis of −θg , as shown in Fig. 5.9(a).
The corresponding GRIN rotation is as simple as reassigning θg = 0 for alignment
with the y-axis. The rotation of r~0 is performed by multiplication with the three-
dimensional rotation matrix,
    
 rx00
  cos (−θg ) sin (−θg ) 0    rx0
    
 r00  =  − sin (−θ ) cos (−θ ) 0   r0 
 y   g g  y  (5.19)
    
00 0
rz 0 0 1 rz

where r~00 is the ray direction after performing the roll.

g
g -θg teff

Refractive index
θs
y r y t
-θg θs x
x x z
r z z y

(a) (b) (c)


Figure 5.9: Coordinate transformations performed in tracing a linear GRIN skew ray inci-
dent outside the plane of the gradient. Unlike the gradient, the initial ray direction may have
components in all three dimensions and is not confined to the transverse x-y plane. (a)
The coordinate system is rolled around the z-axis by −θg . (b) Next, the coordinate system
undergoes a yaw about the ĝ = ŷ direction by θs such that the ray again falls in the g-z
plane. (c) The longitudinal cross-section depicts the effective propagation thickness tef f
found by the obliquity of the planar cross-section.
Chapter 5. F-GRIN Prescribed Illumination Optics 188

Next, the rotated coordinate system containing both GRIN and ray is rotated
again, this time in the form of a yaw about the ĝ = ŷ axis. Since the linear GRIN is
being rotated around its axis-symmetric gradient, no change occurs in the refrac-
tive index distribution. On the other hand, the ray direction is rotated about the
g-axis such that it lies exclusively in the g-z plane in the doubly-rotated reference
frame, as shown in Fig. 5.9(b). A yaw rotation of r~00 by θs can again be performed
by matrix multiplication with a rotation matrix,

    
 rx000   cos θs 0 − sin θs  rx00

    
 r000
 y
= 0
  1 0   r00 
 y  (5.20)
    
rz000 sin θs 0 cos θs rz00

where r~000 is the ray direction following both rotations.


Before solving for the needed value of θs , the coordinate reference roll and yaw
can conveniently be combined into a single transformation by multiplication of the
two rotation matrices,
    
 rx000   cos θg cos θs − sin θg cos θs − sin θs   rx0
    
 r000
 y
=
  sin θg cos θg 0   r0  .
 y  (5.21)
    
rz000 cos θg sin θs − sin θg sin θs cos θs 0
rz

Now, the value of the yaw rotation angle θs can be determined in terms of r~0
and θg based on the requirement that the ray after rotation must lie only in the g-z
plane, meaning rx000 = 0,

rx000 = cos θg cos θs rx0 − sin θg cos θs ry0 − sin θs rz0 = 0, (5.22)
Chapter 5. F-GRIN Prescribed Illumination Optics 189

where dividing by cos θs and solving for θs finds

cos θg rx0 − sin θg ry0


tan θs = . (5.23)
rz0

Currently, the reference frame has been rotated such that the ray again lies in
the g-z plane compatible with the derived ray formulae. One additional require-
ment due to the yaw rotation is the incorporation of an obliquity factor in the axial
propagation distance. For rays incident in the plane of the gradient, the center
thickness t serves as the axial propagation distance, as in Fig. 5.6. For skew rays,
however, an additional cos θs obliquity factor is required to account for the oblique
planar cross-section apparent in Fig. 5.8(b). This obliquity can be expressed by
an effective propagation thickness tef f , found according to the longitudinal cross-
section in Fig. 5.9(c),

t t
q 2
tef f = = 0 cos θg rx0 − sin θg ry0 + rz02 . (5.24)
cos θs rz

Note that cos θs can only take positive values since θs ∈ − π2 , π2 .


 

The skew ray trajectory can now be found by the same expression in Eq. (5.12),
but in the rotated reference frame and with the added axial obliquity factor,

" p ! #
n0 1 α 1 + β2
gskew (z) = p cosh z+c −1
α 1 + β2 n0 cos θs
p ! (5.25)
dgskew (z) α 1 + β2
= sinh z+c cos θs .
dz n0 cos θs

The initial ray slope β must also be defined within the rotated reference frame
where
ry000
β = 000 (5.26)
rz
Chapter 5. F-GRIN Prescribed Illumination Optics 190

for the considered case where the gradient is rolled to be oriented along y.
Within the rotated coordinate system, the traced ray position and direction can
be expressed in Cartesian coordinates by accounting for the yaw rotation θs ,

dx (z)
x (z) = z tan θs , = tan θs
dz (5.27)
dy (z) dgskew (z)
y (z) = gskew (z) , = .
dz dz

Likewise, the optical direction cosines (L, M, N ) for the traced skew ray are found
by

n [g (z)] dx
L= q ·
dx 2 dy 2 dz
 
1+ dz
+ dz
n [g (z)] dy
M=q · (5.28)
dx 2 dy 2 dz
 
1+ dz
+ dz
n [g (z)]
N=q .
dx 2 dy 2
 
1+ dz
+ dz

At this point, the skew ray trajectory has been determined, including the effects
of the yaw about ĝ. The one remaining task is to revert the reference frame to the
global coordinates of the incident ray by rolling all positions and direction cosines
about z by θg ,

    
 xglobal   cos θg sin θg 0   x 
    
 y  =  − sin θ cos θg 0   y 
 global   g  
    
zglobal 0 0 1 z
     (5.29)
 Lglobal   cos θg sin θg 0   L 
    
 =  − sin θ
.
 M cos θg 0   M 
 global   g 
    
Nglobal 0 0 1 N
Chapter 5. F-GRIN Prescribed Illumination Optics 191

Lastly, the skew ray position and direction is now known in global coordinates
at the rear planar surface of the linear GRIN. Vectorial Snell’s law in Eq. (5.17) can
be applied a second time to determine the ray refraction into air.

5.4 Design process for F-GRIN prescribed illumination optics

The design of F-GRIN prescribed illumination optics is a four-step process [33] that
is outlined in Fig. 5.10. The steps are a hybrid of analytical and numerical tech-
niques. The analytical backbone is the beam deflecting behavior of linear GRIN,
which acts similar to a planar mirror at grazing incidence (recall Fig. 5.7). The nu-
merical methods serve to identify a favorable mapping between the illumination
optic and target. Later, numerical integration is also used to obtain a piecewise-
continuous F-GRIN design from an array of discrete elements. The entirety of the
design process described in this section is performed in Python, including all ray
tracing.

1 Optic-Target
Mapping 2 Linear
Array
GRIN

Target-optic discretization Array formation

Linear assignment problem Linear GRIN optimization

3 F-GRIN
Integration 4 Ray Trace
Evaluation

Southwell reconstruction Monte Carlo ray tracing

Bicubic interpolation Heatmap evaluation

Figure 5.10: Flowchart of the four-step process used in F-GRIN prescribed illumination
design. The four steps are described in detail in Sec. 5.4.1 – 5.4.4.
Chapter 5. F-GRIN Prescribed Illumination Optics 192

From the scope laid out in Sec. 5.2.1, the design process only considers a ge-
ometry with the source, optic, and target all aligned along the z-axis. The source
is an infinitesimal point source, which with zero étendue, serves as an approxi-
mation for spatially small sources. The illumination target will be evaluated on a
planar surface, normal to the z-axis. As a result, the generated illumination will be
quantified in terms of irradiance (radiant flux per unit area).
Before specifically describing each step of the design process, a broad overview
is helpful to inform on the motivation for each subsequent step. The F-GRIN de-
sign is analogous in many ways to digital micromirror devices (DMDs), a type of
dynamic spatial light modulator commonly used in digital projection [196]. DMDs
perform digital spatial light modulation using a rectangular array of micromirrors
that can be electrically changed in angular orientation. Each micromirror consti-
tutes a single pixel and can be electrically addressed to toggle between “on” and
“off” states.
The F-GRIN design comes about in a similar way to DMD operation. The il-
lumination target is discretized, as is the F-GRIN optic, which is subdivided into
a rectangular array of elements. Each array element in the optic is prescribed to
generate a single “pixel” in the target. This is accomplished using the gradient
slope and orientation of linear GRIN to deflect incident light from the source to
its corresponding position in the target without introducing optical power or any
other aberration. Lastly, the array of linear GRIN elements is integrated in a way
that preserves necessary gradient discontinuities but still provides regions of con-
tinuous refractive index change. Design are then interrogated using Monte Carlo
ray tracing to assess the fidelity of the generated illumination field.
Chapter 5. F-GRIN Prescribed Illumination Optics 193

5.4.1 Optic-target mapping

The design process begins by discretizing the illumination target. Subsequently,


an array is formed for the optic that has the same number of elements as the target
has points.
The target is discretized both spatially as well as in intensity (i.e., “bit depth”).
The discretized target distribution can be conveniently formatted as a lossless,
monochrome raster image file with square pixels (e.g., TIFF) where both spatial
and intensity information is specified. Then, the total number of points N needed
to generate the target is the sum of the bit values for all spatial pixels. For example,
a binary target consisting of only black (0% intensity) and white (100% intensity)
regions has a number of illumination points N equal to the number of white pixels
in the target, as in Fig. 5.11(a). Meanwhile, a grayscale target consisting of three
levels – black (0% intensity), gray (50% intensity) and white (100% intensity) – has

N = 35 N = 47

1 1 2 2
1 1 1 1 2 2 2 2
1 1 1 1 2 2 2 2
1 1 2 2

1 1 1 1
1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1

(a) (b)

Figure 5.11: Number of points N in a discretized illumination target. (a) For a binary
target, N is simply the number of white pixels in the target. (b) For a grayscale target, N
is the sum of bit values for all pixels. In this grayscale example, there are three gray levels
with black (value 0), gray (value 1), and white (value 2) regions.
Chapter 5. F-GRIN Prescribed Illumination Optics 194

N equal to the sum of pixel intensity levels, as shown in Fig. 5.11(b). This can
intuitively be thought of as the white regions needing to be generated by twice
as many elements in the optic as the gray regions of half the intensity. The same
concept can, of course, be applied to any number of gray levels, although compu-
tational complexity increases rapidly with N , as will be explained. Nonetheless,
it is convenient that both binary and grayscale targets can be neatly discretized in
the same way.
Upon determining the value of N needed to generate the illumination target,
an array is formed at the optic consisting of N individual elements. The simplest
means of doing so is in a rectangular array of i rows and j columns where N =
ij. Rectangular arrays are the only optic discretization considered here, although
more advanced formulations such as Voronoi cells may be advantageous [165]. It
is important to note that depending on the target discretization, an unfavorable
value for N , such as a prime number, leads to rectangular array dimensions with
aspect ratios i/j deviating greatly from unity. An easy way of adjusting the value
for N is by a slight interpolated scaling of the target image file. For most designs,
the best course of action is to keep the array aspect ratio as close as possible to the
aspect ratio of the illumination target.
While the optic array may be rectangular where i 6= j, each array element is
held to be square. The beam deflecting behavior of linear GRIN depicted in Fig.
5.7 shows that a square element produces a nominally square irradiance patch on
the observation plane normal to the optical axis. Since the target is discretized in
an image file with square pixels, the square element shape is required for properly
generating the target distribution.
Next, spatial coordinates must be assigned to the center of each array element
and target pixel. The target distance zt from the optic’s front surface vertex is spec-
Chapter 5. F-GRIN Prescribed Illumination Optics 195

ified initially and serves as the evaluation distance for the ultimate illumination
profile. Then, the maximum spatial extent of the target Dt is specified for posi-
tion z = zt . Together, the distance and extent quantify a target “throw ratio,”
Rt = zt /Dt , of size per unit distance away from the optic. For example, a target
with a maximum dimension of Dt = 500 mm evaluated at a distance of zt = 1000
mm from the optic requires a throw ratio of Rt = Dt /zt = 0.5. The throw ratio
also entails that, neglecting diffraction, the evaluated target distribution linearly
scales in spatial extent with distance away from the optic. Finally, the coordinates
(xt , yt , zt ) for the center point of each pixel in the target can now be found in the
evaluation plane z = zt when scaled to extent Dt according to the throw ratio.
Each pixel within the target has its own throw ratio that depends on the spatial
resolution of the target. For the same example as before, at a distance zt = 1000
mm, 100 pixels across a target extent of Dt = 500 mm requires the size of each
square target pixel to be Dpx = 5 mm or, equivalently, have a throw ratio of Rpx =
Dpx /zt = 0.005.
Meanwhile, the source distance zs and the size of each element in the array De
jointly describe the numerical aperture subtended by a single element in the optic.
Since each array element is initially constituted by a linear GRIN, no power is
introduced, and incident light is redirected without being spread or concentrated.
As a result, the dimensions zs and De are also solely responsible for achieving
the necessary target pixel size Dpx at the evaluation plane, z = zt . The relationship
between array element size and target pixel size is found to be a simple geometrical
scaling, Dpx = mg De , according to the similar triangles depicted in Fig. 5.12. The
geometric magnification mg is the scale factor derived from zs , zt , as well as the
Chapter 5. F-GRIN Prescribed Illumination Optics 196

Target
Pixel
Individual
Array
Element Dpx /2
Source De /2
zs τ + zt

Figure 5.12: Different combinations of source distance zs and optic array element size De
for maintaining a fixed target pixel size Dpx and distance zt , i.e., throw ratio Rpx .

air-equivalent thickness of the illumination optic, τ = t/n0 ,

τ + zt
mg = 1 − . (5.30)
zs

The source distance zs is measured from the coordinate origin at the front surface
vertex of the optic, so for all real source positions, zs < 0 and mg > 0. Also, notice
that by Eq. (5.30), a collimated source, zs = ±∞, gives mg = 1, which means the
array element and target pixel sizes are equal, Dpx = De . As a result, the current
design method, which applies powerless elements, cannot increase the scale of the
target larger than the illumination optic when used with collimated light. (This
effect is empirically demonstrated later in Fig. 5.33.)
Given the target size, the geometric magnification can be used to choose a
source distance zs and element size De pair that is compatible with size constraints
Chapter 5. F-GRIN Prescribed Illumination Optics 197

on the system and element. For example, there may be a fabrication limit on the
illumination optic’s overall dimensions of height H = i · De and width W = j · De .
These constraints can often be met since there are different valid combinations of
zs and De that generate the same target pixel size Dpx (see Fig. 5.12). In doing so,
there is a clear trend where larger source distances require larges element sizes,
and vice versa.
After selecting a valid pair of zs and De that yield Dpx , the dimensions of the
illumination optic are set. This includes the center point coordinate of each element
in the array, (xo , yo , zo ), which along with the target pixel coordinates, (xt , yt , zt ), are
needed when determining the optic-target mapping.
Upon discretizing both the optic array and illumination target into N points of
known coordinates, a one-to-one mapping must be determined between points in
the two planes (see Fig. 5.13). The mapping describes which element in the optic
is responsible for generating which point in the target. Consequently, the optic-
target mapping is a critical step in the design process that significantly affects the
ultimate design.

Figure 5.13: Mapping between points in the discretized GRIN array and illumination target.
Chapter 5. F-GRIN Prescribed Illumination Optics 198

The goal is to obtain a mapping that produces a “favorable” final design. Al-
though, what is considered favorable depends largely on the application as well
as tolerances and constraints in fabrication. One option is to determine the map-
ping that minimizes the required GRIN ∆n. This mapping proves favorable when
restricted by ∆n in fabrication but may yield a large number of discontinuities in
the final integrated design. Recall, discontinuities in the refractive index gradient
are required to generate discontinuities in the illumination target, yet tolerances
in the fabrication of gradient discontinuities entail some degree of “sharpness” er-
ror. Fabricated optics possessing these rounded discontinuities yield artifacts in
the generated illumination, as found in Sec. 5.6. In fact, within the current design
framework of only transversely varying refractive index, n (x, y), artifacts appear
for perfectly sharp discontinuities as well. These artifacts are due to “crosstalk”
where rays incident near discontinuities erroneously cross a discontinuity while
propagating within the optic. Although, these effects can be compensated for by
an axial refractive index variation that is tapered to match the source numerical
aperture.
Instead, the mapping used in the design process is selected to encourage as
large of continuous regions in the refractive index as possible. In doing so, any nec-
essary gradient discontinuities are limited in number and form contiguous regions
that mirror the target discontinuities. To encourage continuous GRIN regions, the
optic-target mapping is sought that minimizes the sum of the Euclidean distances
between all N pairs. Mathematically, this mapping problem can be formatted as
minimizing the cost function C,

N q
X
C= (xo,m − xt,m )2 + (yo,m − yt,m )2 + (zo,m − zt,m )2 (5.31)
m=1
Chapter 5. F-GRIN Prescribed Illumination Optics 199

where recall (xo,m , yo,m , zo,m ) is the center point of each optic array element and
(xt,m , yt,m , zt,m ) is the center point of each illumination target pixel. Since the coor-
dinate origin is defined at the vertex of the front planar surface of the optic, zo,i = 0
for all array elements. Similarly, zt,m = zt for all target points when considered in
a plane normal to the axis.
Solving for the mapping that minimizes C is a form of the linear assignment
problem (LAP) [197, 198]. The objective of a LAP is to find the optimal set of pairs
that minimizes some predefined cost function. An example of a LAP is as follows:
if one employs N workers each of whom demands a different rate for N jobs, how
does one assign all workers to a job such that the total pay rate is minimized? It is
clear that the optic-target mapping problem embodies the same format as a LAP
where the mapping between optic and target points is sought that minimizes the
total Euclidean separation C. A two-dimensional example of this LAP is solved
in Fig. 5.14 for N = 4 where the black circles are the array elements and the col-
ored circles are the target points. Determining a ray mapping by solving a LAP
is a technique that is also used in freeform surface prescribed illumination design

(a) (b)

Figure 5.14: The linear assignment problem (LAP) demonstrated for N = 4 where the
objective is to minimize the Euclidean distance between optic (black) and target (colored)
points. (a) All pair combinations are shown. (b) The pairs that minimize the Euclidean
separation among all points is solved.
Chapter 5. F-GRIN Prescribed Illumination Optics 200

where the Monge-Kantorovich mass transportation problem in discrete form can


be posed as an LAP [178, 179, 199].
There are different numerical methods available for solving a LAP. Classically,
it has been solved with a class of optimization algorithms known as the Hungar-
ian algorithm [197, 200]. A variant of the Hungarian algorithm is used here with
O (N 3 ) computational complexity [201, 202]. Due to the cubic dependence on N ,
computation time increases rapidly with the number of illumination points, which
includes both the number of pixels as well as the bit depth. Designs performed
here are on the order of N ∼ 104 , which takes approximately 15 minutes on a
personal computer (MacBook Pro, Dual-Core i5 @ 2.9 GHz, 16 GB RAM). Solving
the LAP is the most time-consuming step in the design process. Consequently,
to handle higher resolution and larger bit depth targets in the future, it would be
advantageous to implement a parallelized version of the algorithm [203].

5.4.2 Linear GRIN array

Completion of the first step in the design process yields a mapping between coor-
dinates in the optic and target. This mapping is specific to the defined positions
and dimensions for the source, optic, and target. The GRIN optic currently consists
of an array of i rows and j columns, with N = ij total elements. Each element is
square and of the same size, De , granting full optic dimensions of height H = i · De
and width W = j · De .
Next, each element in the array can be represented by a linear GRIN oriented
with its gradient lying in the x-y plane. As outlined in Sec. 5.3, each linear GRIN
element is fully defined by three parameters: the base refractive index n0 , the re-
fractive index slope α, and the gradient direction θg measured from the y-axis in the
Chapter 5. F-GRIN Prescribed Illumination Optics 201

x-y plane. Then, the beam deflecting property of linear GRIN allows these three
parameters to be solved for such that incident light from the source is redirected
to its mapped target pixel. For each array element, only a single “ray of interest”
(ROI) is needed to determine the necessary linear GRIN parameters. The ROI em-
anates from the point source, intersects the array element center point (xo,m , yo,m , 0)
on the front planar surface of the optic and then, ideally, is redirected to the center
point (xt,m , yt,m , zt ) of its corresponding target pixel.
Between the three free parameters – n0 , α, and θg – the base refractive index n0
imparts the least influence over the redirection of incident rays. This effect can be
seen from the expression for the ray exit angle θ20 in Eq. (5.16) where the factor of
n0 is largely negated by n−1
0 within the hyperbolic sine function. Consequently, for

each linear GRIN element, the value for n0 is set to the midpoint of the refractive
index range that is achievable in fabrication, n0 = (nmin + nmax ) /2. In doing so,
the largest achievable value for ∆n = nmax − nmin can be employed in generating
the illumination target. Note that for fixed element dimensions, ∆n is directly
proportional to the refractive index slope α.
The two remaining parameters, α and θg , can be jointly used to redirect the
ROI to its target point. Increasing values of α lead to greater ray deflection along
the gradient direction. There is no design limit on α except by fabrication, both
in terms of ∆n and gradient steepness. The gradient angle θg presents less control
over the amount of deflection but strongly influences the direction of ray deflec-
tion. The gradient direction can take on any angle in the x-y plane spanning a
circle, θg ∈ [−π, π), meaning the beam can be redirected in any angular direction.
The combined use of α and θg means the ROI can be redirected to any point within
a large region in the target plane. For example, in Fig. 5.15 the region of attainable
ROI coordinates in the target plane can be seen up to α ≤ 0.1 mm−1 . With the
Chapter 5. F-GRIN Prescribed Illumination Optics 202

α [mm-1]
θg = 0°
45° 0.00
0.01
45° -45° θg = 0°
0.02
0.03
0.04
90° -90° -45° 0.05
0.06
90° 0.07
-90° 0.08
0.09
135° -135°
-135° 0.10
135°
180° 180°

(a) (b)

Figure 5.15: Regions of attainable ray coordinates at the target plane after traversing lin-
ear GRIN of different parameters. Radial contours (blue gradient) show ray coordinates for
constant values of the refractive index slope α. Azimuthal contours (black) show ray coor-
dinates for constant values of gradient angle θg . The center point shows the ray coordinate
for homogeneous media, α = 0. Rays are traced from an on-axis source with position
zs = −100 mm and evaluated in a target plane with position zt = 1000 mm. (a) The linear
GRIN element is centered on the optical axis with the ray normally incident on the front
planar surface. (b) The linear GRIN element is offset from the optical axis with center point
(30, 30, 0) mm, which yields a 23.0◦ ray angle of incidence.

source located on the z-axis, a linear GRIN also centered on the optical axis has
a normally incident ROI, and the corresponding region of ray coordinates in the
target is circularly symmetric, as shown in Fig. 5.15(a). As can be seen, the radius
of deflection increases quasi-linearly with α while θg controls the azimuthal coor-
dinate along each ring. The center point of the region shows where the ROI would
intersect the target if encountering a homogeneous element, α = 0.
Moreover, in the illumination optic, each array element differs in transverse
position with respect to the source. For a source positioned at (0, 0, zs ), an element
m with center point (xo,m , yo,m , 0) has a ROI initial direction ~rm of

xo,m x̂ + yo,m ŷ − zs ẑ
~rm = p 2 2
, (5.32)
xo,m + yo,m + zs2
Chapter 5. F-GRIN Prescribed Illumination Optics 203

which differs from element-to-element. As a result, the shape and size of the ray
coordinate region in the target differs between elements. For example, Fig. 5.15(b)
shows the target coordinate region for a linear GRIN shifted diagonally from the
z-axis. The same general trends for α and θg hold, although no longer being circu-
larly symmetric. In this case, the greatest deflection occurs for a gradient parallel
to the incident ray direction.
With this in mind, the goal is now to determine the unique pair of values for
αm and θg,m that generates the ROI for linear GRIN element m to its mapped target
coordinate (xt,m , yt,m , zt ). From Fig. 5.15, it is clear that if the target coordinate
falls within the shown regions, there is always a valid pair for αm and θg,m . The
only exception occurs when too large of ray departure is required that TIR occurs
at the rear surface of the linear GRIN. The two unknowns, αm and θg,m , can be
solved analytically from the two target ray coordinates, xt,m and yt,m , according
to the described skew ray trace procedure where all other system parameters are
defined.
In actuality, the analytical formulae for αm and θg,m are far too unwieldy. Rather,
numerically solving for the parameters by a simple optimization routine is prefer-
ably. Due to the well-behaved relationship between linear GRIN parameters and
ray coordinates shown in Fig. 5.15, the optimized result is obtained quickly, on the
order of milliseconds, and with high precision.
The optimization routine consists of first estimating a value for θg,m based on
the angle in the x-y plane of the target ray coordinate with respect to the homoge-
neous ray coordinate, α = 0 (center point of concentric regions in Fig. 5.15). Then,
the value for αm that minimizes the distance from the ROI to the target point is
found by the binary search algorithm. With that current value of αm , the value
for θg,m is updated by binary search until the ROI-target point separation is mini-
Chapter 5. F-GRIN Prescribed Illumination Optics 204

mized. The alternating process of updating αm followed by θg,m is repeated until


convergence within some error threshold (for the performed designs, error is set
to < 0.5 µm).
The linear GRIN parameters n0 , α, and θg can now be established for all N
elements in the array at the illumination optic. One such design consisting of a
linear GRIN array is shown in Fig. 5.16. Each element has been solved for such
that the ROI incident on its center point is redirected to its mapped target coordi-
nate. With all elements solved for, the current design successfully generates the
prescribed illumination. At this point, the design is formulated very similar to the
freeform surface SQM designs described in Sec. 5.2; discrete points on the optic
generate discrete points on the target using the stigmatic deflection property of
linear GRIN.

1.502

1.501

Refractive index
1.500

1.499

1.498

Figure 5.16: Design example of a linear GRIN array that generates prescribed illumination.
The transverse refractive index distribution n (x, y) is shown, including an exploded view of
individual array elements. Each element has a unique combination of n0 , α, and θg . In this
design, there are N = 104 elements in the array.
Chapter 5. F-GRIN Prescribed Illumination Optics 205

5.4.3 Integrated and interpolated piecewise-continuous F-GRIN

The current design as a linear GRIN array produces the illumination target accord-
ing to the principles and assumptions of geometrical optics as well as the assump-
tion of errorless fabrication. Although, there are several physical and practical
limitations that necessitate a further step in the design process that integrates the
linear GRIN array into larger regions of continuous refractive index. First, the pe-
riodicity and small feature size of the array may introduce noticeable diffraction
artifacts in the far-field. Second, even with errorless fabrication, crosstalk with rays
transferring between elements within the optic introduces error in the generated il-
lumination. As mentioned previously, the goal is to have the minimum number of
gradient discontinuities necessary to produce the specified target discontinuities.
Meanwhile, array designs present the largest number of gradient discontinuities
possible, which is not ideal. Finally, fabrication tolerances necessarily yield a re-
duction in sharpness of gradient discontinuities, which further introduce errors.
These effects may also be increased due to the potential small size of each array
element. For these reasons, a next step in the design process is needed to integrate
the array design into a piecewise-continuous F-GRIN.
The first step in integrating the array design is to realize that a linear GRIN el-
ement can be more succinctly represented by a single vector, namely its gradient
∇n. The gradient direction has been confined to the two-dimensional x-y plane
throughout the design process while its magnitude is known by the refractive in-
dex slope α. Although the gradient is the same at every point within a linear GRIN,
a single vector describing an element is arbitrarily chosen to be positioned at the
element center point (xo,m , yo,m , 0).
An array of linear GRIN elements, therefore, can be represented by a two-
Chapter 5. F-GRIN Prescribed Illumination Optics 206

Refractive index
(a) (b) (c)

Figure 5.17: Three steps in F-GRIN integration from (a) a linear GRIN array to (b) discrete
refractive index values from Southwell reconstruction to (c) continuous refractive index from
bicubic interpolation.

dimensional array of two-dimensional vectors. For example, a linear GRIN array


overlaid with its representative vectors can be seen in Fig. 5.17(a). While some
adjacent elements may have similar gradients and integrate nicely, there is no re-
striction in the design process on having have continuous or closely agreeing gra-
dients. In fact, these local differences in linear gradients are what are responsible
for generating discontinuities in the illumination.
The problem of integrating a linear GRIN array represented by a field of vec-
tors can be solved using techniques developed for several reminiscent but unre-
lated problems in optical metrology. Zonal reconstruction is a technique used for
integrating a field of vectors to reconstruct a surface normal to each vector. This
technique is most commonly applied to wavefront reconstruction where wavefront
slopes are locally measured. There are several modalities for performing these
wavefront slope measurements, such as shearing interferometry, screen tests, and
Shack-Hartmann wavefront sensing [204]. Another use for zonal reconstruction is
in image reconstruction for Knox-Thompson stellar speckle interferometry [205].
Due to a variety of factors such as measurement noise, there is no guarantee that
all constituent vectors integrate exactly to reconstruct the same surface. As a result,
Chapter 5. F-GRIN Prescribed Illumination Optics 207

different algorithms have been presented to handle this inexactness for an overall
best-fit reconstructed surface [204].
One of the earliest and most prominent classes of zonal reconstruction algo-
rithms is Southwell reconstruction [206]. Southwell presented an iterative algo-
rithm for solving a system of linear equations using the successive over-relaxation
method for convergence in the presence of measurement noise. From a rectangular
array of vectors, a system of linear equations is built upon continuity between the
four vectors above, below, to the left, and to the right of a point of interest. Edges of
the vector field must also be handled differently since all four surrounding vectors
are not defined.
A modified version of Southwell reconstruction is used for integrating linear
GRIN arrays to form piecewise-continuous F-GRIN designs. The modified algo-
rithm [207] allows for non-uniform vector fields, which enables the inclusion of
discontinuities in the reconstructed refractive index. Consequently, before per-
forming zonal reconstruction, discontinuities must be identified between adjacent
linear GRIN elements. Discontinuities are identified based on differences in α and
θg between adjacent elements that surpass some threshold. The threshold values
for α and θg are set empirically by the designer such that the appearance and num-
ber of discontinuities matches those of the illumination target as closely as pos-
sible. This process is aided by a routine that determines the number of insular
regions in the illumination target (e.g., consisting of a series of letters). From these
insular target regions, it is known that the mapped array elements should form
regions discontinuous from one another.
The result of zonal reconstruction is a discrete set of reconstructed values that
are located at the original vectors’ positions. For refractive index reconstruction,
this corresponds to a series of discrete refractive index values, defined at the center
Chapter 5. F-GRIN Prescribed Illumination Optics 208

of each linear GRIN element, as shown in Fig. 5.17(b). While the reconstructed
refractive index has been integrated from the array, the grid of discrete values does
not present high enough spatial resolution to characterize a design for fabrication.
Continuous definition of n and ∇n is also required for evaluating the generated
illumination using numerical ray tracing [19–21].
Instead, a final step is required to continuously define the refractive index be-
tween discrete reconstructed values. The chosen interpolation scheme must pre-
serve smoothness across the grid of values, which eliminates nearest-neighbor or
linear interpolation as possibilities. Rather, cubic interpolation [208] is used in the
design process to go from the discrete index values to a continuous definition, as
shown in Fig. 5.17(c). Since the index variation is restricted two-dimensional,
n (x, y), bicubic interpolation is required rather than tricubic interpolation [23],
which would be needed for three-dimensional designs, n (x, y, z).
The implemented bicubic interpolation scheme relies on a matrix formulation
for determining a given value of n (x, y). The method is also custom-designed to
interpolate around the predefined discontinuities. First, the surrounding discrete
refractive index values are identified for a given coordinate. For bicubic inter-
polation, the 4 × 4 grid of values enclosing the coordinate are required (see Fig.
5.18). The innermost 2 × 2 grid of enclosing points are used directly while the
additional two outer rows and columns are used for calculating derivatives to
maintain smoothness. The evaluation coordinate is also normalized, (xn , yn ), to
the inner 2 × 2 grid. Then, the refractive index value is found by a pair of matrix
Chapter 5. F-GRIN Prescribed Illumination Optics 209

yn n(0,1) n(1,1) yn n(0,1)

n(0,0) n(1,0) n(0,0) n(1,0)

xn xn

(a) (b)
Figure 5.18: Bicubic interpolation scheme for (a) a continuous 4 × 4 enclosing grid and
(b) a grid including predefined gradient discontinuities. The red ‘X’ shows the coordinate
at which to interpolate the refractive index. The yellows points are the innermost 2 × 2
grid of discrete refractive index values used directly in interpolating index values. The blue
points are the two outermost rows and columns of discrete refractive index values used in
calculating index derivatives. The gray points lying on the opposite side of a discontinuity
are not used in any capacity. The coordinates (xn , yn ) are normalized according to the
innermost 2 × 2 grid.

multiplications,

  
a00 a01 a02 a03   1 

a10
 
a11 a12 a13 
 yn 
  
n (xn , yn ) = 1 xn x2n 3  . (5.33)

xn 
 yn2 
a a21 a22 a23   
 20
  
a30 a31 a32 a33 yn3

Subsequently, the refractive index gradient ∇n = h dn , dn , dn i needed for ray tracing


dx dy dz
Chapter 5. F-GRIN Prescribed Illumination Optics 210

is also given by matrix multiplication,

  
a00 a01 a02 a03   1 
a10
 
a11 a12 a13 
 yn 

dn (xn , yn )  
= 2

0 1 2xn 3xn 
dx
 
 yn2 
a a21 a22 a23   
 20
  
a30 a31 a32 a33 yn3
  
(5.34)
a00 a01 a02 a03   0 
 a10
 
a11 a12 a13 
 1 

dn (xn , yn )  
= 1 xn x2n 3

xn 
dy
 
a a21 a22 a23 
 2yn 
 
 20
  
a30 a31 a32 a33 3yn2
dn (xn , yn )
= 0.
dz

Common to all calculations, the values for the aij matrix originate from the
discrete refractive index values,

   
a00 a01 a02 a03   n (0, 0) n (0, 1) ny (0, 0) n y (0, 1) 
   
a10 a11 a12 a13   n (1, 0) n (1, 1) ny (1, 0) ny (1, 1) 
 = MT M (5.35)
   
 
a a21 a22 a23  n (0, 0) n (0, 1) n (0, 0) n (0, 1)
 20   x x xy xy 
   
a30 a31 a32 a33 nx (1, 0) nx (1, 1) nxy (1, 0) nxy (1, 1)

where  
1 0 −3 2
 
0 0 3 −2
M = . (5.36)
 
0
 1 −2 1 

 
0 0 −1 1

Subscripts in Eq. (5.35) indicate a derivative, and nxy is the mixed partial deriva-
tive. Since index values are defined on a discrete grid, these derivative are cal-
Chapter 5. F-GRIN Prescribed Illumination Optics 211

culated from finite differences. In the presence of discontinuities, the derivative


calculation is modified to only consider values on the same side of the discontinu-
ity as the ray coordinate, as in Fig. 5.18(b). A similar modification is needed for
coordinates near the edges or corners of the optic.
Construction of the aij matrix is not a computationally efficient task. Consider-
ing how often the interpolated refractive index and its gradient is evaluated when
ray tracing, this imposes a severe restriction on computation time. Fortunately, the
collection of aij matrices can be precomputed for a given design. This way, ray
tracing can be performed by a lookup table rather than constructing aij for every
ray step.
Furthermore, limitations on GRIN fabrication impose constraints on the final
interpolated design that are not directly controllable within the design process.
Fortunately, both the midpoint refractive index nmid as well as the total refractive
range ∆n can be adjusted to meet fabrication constraints after obtaining an inte-
grated and interpolated design. The modified designs produce illumination distri-
butions with little to no loss in illumination fidelity. A change in refractive index
magnitude (i.e., adding a constant value to the refractive index at all points) has
very little effect on the illumination based on the range of materials available. As a
result, the value for nmid can be readily set for integrated and interpolated designs.
On the other hand, the total refractive index range ∆n is a driving constraint
when it comes to the design of F-GRIN prescribed illumination optics. A larger
value for ∆n allows for a larger target throw ratio, and since the presented illumi-
nation optics are powerless, a larger ∆n also allows for a larger source numerical
aperture. As a result, the ∆n can be scaled for a design, leading to a decrease in
throw ratio but with little impact on illumination fidelity. The ∆n can also be scaled
along with the element dimensions to maintain the same gradient across the optic
Chapter 5. F-GRIN Prescribed Illumination Optics 212

1.550

1.525

Refractive index
1.500

1.475

1.450

Figure 5.19: Design example of an integrated piecewise-continuous F-GRIN that gener-


ates prescribed illumination. The transverse refractive index distribution n (x, y) is shown,
including an exploded view of a gradient discontinuity (black lines). The linear GRIN array
form of this design is shown in Fig. 5.16.

(index change per unit distance). The source distance must also be scaled by the
same factor to maintain the same numerical aperture and ROI angles of incidence.
Since the illumination is generated in the far-field, this scaling of the entire system
leaves the target throw ratio nominally unchanged.
The flexibility in the design to scale both refractive index ranges and/or system
dimensions is very advantageous for meeting different illumination requirements
ranging from compact source distances to larger optics, all while abiding by fabri-
cation constraints. Finally, even larger ∆n designs can be accommodated by intro-
ducing additional discontinuities to wrap the refractive index, similar to Fresnel
lenses wrapping surface sag.
Following the linear GRIN array stage shown in Fig. 5.16, an example piecewise-
continuous F-GRIN design is shown in Fig. 5.19 upon completion of integration
and interpolation. All present gradient discontinuities are depicted by solid black
lines. Since the design originates from an array, the incorporated discontinuities
have a “stair step” appearance. The design is also adjusted to meet self-imposed
Chapter 5. F-GRIN Prescribed Illumination Optics 213

fabrication limits of n ∈ [1.45, 1.55] where nmid = 1.5 and ∆n = 0.1.

5.4.4 Monte Carlo ray trace evaluation

The final step in the design process is to simulate the illumination generated by
the designed F-GRIN illumination optic. The illumination distribution is evalu-
ated using Monte Carlo ray tracing, a common technique in assessing nonimaging
optics [158, 194]. The ray tracing is performed numerically using a standard itera-
tive Runge-Kutta method [19–21]. Any arbitrary refractive index distribution can
be considered as long as the refractive index n and its gradient ∇n are known con-
tinuously. For the piecewise bicubically interpolated F-GRIN designs, both n and
∇n are accessible. The generated irradiance is evaluated in a plane normal to the
optical axis at the evaluation distance zt based on the spatial distribution of “ray
hits.”
Different aspects of the F-GRIN illumination optics can be observed from Monte
Carlo ray tracing. The nominal illumination performance is assessed using the
zero-étendue approximate source that is assumed in the design process. Although,
the effect of physical sources with finite spatial extent can also be evaluated in ray
tracing. The change in illumination with evaluation distance can also be observed
from ray trace data. The optics are designed for an evaluation distance zt , although
the generated distribution possesses some “depth of field” over which the illumi-
nation is preserved. All of these features are demonstrated next in Sec. 5.5 for
different F-GRIN designs.
Chapter 5. F-GRIN Prescribed Illumination Optics 214

5.5 Example designs

Several different designs have been performed for F-GRIN prescribed illumination
optics. The design process is implemented in a fully automated program. With the
program in hand, the design itself consists of providing the illumination target in
a raster image file. The program then queries the designer for different preferences
in terms of system size and fabrication limits. The design process takes approxi-
mately 30 minutes on a personal computer (MacBook Pro, Dual-Core i5 @ 2.9 GHz,
16 GB RAM), not including ray trace evaluation. Although, computation time de-
pends greatly on the target resolution with the mapping step typically taking the
longest with O (N 3 ) dependence. After obtaining a design, ray trace evaluation
can more than an hour depending on the number of rays traced (e.g., tracing 106
rays takes ≈ 30 minutes).
From the different designs that have been performed, four examples are demon-
strated here. Two designs generating binary targets are shown in Fig. 5.20, and two
designs generating grayscale targets are shown in Fig. 5.21. For both binary de-
signs, targets with holes, sharp edges, and null background are generated using
discontinuities incorporated in the gradient represented by solid black lines in Fig.
5.20. Meanwhile, the grayscale target designs show a concentrating effect not seen
in the binary target designs that is needed to produce target points with varying
levels of irradiance.
For all four nominal designs different artifacts can be seen in the Monte Carlo
irradiance that degrades the generated illumination. Each type of artifact appears
for a different reason, some of which are correctable and/or nonphysical. There is
also a separate class of artifacts that are introduced due to fabrication tolerances,
which are analyzed in Sec. 5.6.
Chapter 5. F-GRIN Prescribed Illumination Optics 215

Target Design Ray Trace


1.55 1.0

Relative irradiance
Refractive index
(a) 1.50 0.5

100
50 mm 1 mm 100
50 mm
1.45 0.0

1.55 1.0

Relative irradiance
Refractive index
(b) 1.50 0.5

100 2
mmm 1 mm 100 mm
1.45 0.0

Figure 5.20: F-GRIN prescribed illumination optics generating binary illumination targets
containing (a) the Rochester “Flower City” logo and (b) the letters “U of R.” For each optic,
the illumination target, design refractive index n (x, y), and Monte Carlo irradiance are
shown. Both optics are designed by the process outlined in Sec. 5.4 with zs ∼ −25 mm,
t = 2 mm, zt = 1, 000 mm, Rt = 0.5, and N ∼ 104 . The designs are both shifted and scaled
to self-imposed refractive index fabrication limits of nmid = 1.5 and ∆n = 0.1.

Target Design Ray Trace


1.538 1.0

Relative irradiance
Refractive index

(a) 1.479 0.5

50 mm 2 mm 50 mm
1.420 0.0

1.538 1.0
Relative irradiance
Refractive index

(b) 1.475
0.1

2 mm
2m 2m
1.413

Figure 5.21: F-GRIN prescribed illumination optics generating grayscale illumination tar-
gets containing (a) the Mona Lisa and (b) a simple automobile headlamp distribution. For
each optic, the illumination target, design refractive index n (x, y), and Monte Carlo irra-
diance are shown. The automobile headlamp irradiance, both target and ray trace, are
shown on a log intensity scale to make dim regions more visible. Both designs were exe-
cuted by Rongze Xu using the provided design program.
Chapter 5. F-GRIN Prescribed Illumination Optics 216

Figure 5.22: Histogram of relative irradiance for the Monte Carlo ray trace in Fig. 5.20(b)
for the “U of R” design. Due to statistical noise inherent to Monte Carlo ray tracing, there is
a negligible yet non-zero number of points with relative irradiance greater than 60%. The
denoted 90th percentile for non-zero relative irradiance (i.e., points with at least one ray
hit) is used to scale the target for the cross-sectional analysis in Fig. 5.23.

First, by the random nature of Monte Carlo ray tracing, there is an inherent level
of statistical noise when tracing a finite number of rays, sampled over a discrete
number of pixels. This noise is a familiar concept when analyzing illumination
optics in commercial software [194]. Statistical noise is responsible for the speckle-
like appearance of solid regions in the relative irradiance in Figs. 5.20 and 5.21.
Also, statistical ray tracing means there may be individual points in the irradi-
ance that possess anomalously higher values than their neighbors. As a result, the
black-white contrast in the ray traced irradiance is reduced since it is normalized
to these anomalous maximum values. This effect can be seen in the histogram in
Fig. 5.22 that shows the distribution of relative irradiance values for the “U of R”
design. There is a negligible, although non-zero, number of points with relative ir-
Chapter 5. F-GRIN Prescribed Illumination Optics 217

radiance greater than 60%, which results in an artificial reduction in contrast when
normalized. All of these described effects stem from ray trace statistical noise, and
as such, they are absent in fabricated optics used with a continuously-emitting
source.
Another class of artifacts in these nominal designs originates from rays that
cross gradient discontinuities while propagating within the optic. These cases are
limited to rays incident about a discontinuity, and they are especially prevalent for
rays with larger angles of incidence. As shown in Fig. 5.23 for the “U of R” design,

Figure 5.23: Cross-section of the relative irradiance for the “U of R” design in Fig. 5.20(b).
The cross-sectional plots compare the irradiance with the target (black dashed line). The
target is scaled to have its maximum irradiance occur at the 90th percentile of ray traced
values in order to remove the contrast reduction from statistical noise (see Fig. 5.22).
Artifacts are noticeable from rays crossing gradient discontinuities, yielding bands in the
irradiance.
Chapter 5. F-GRIN Prescribed Illumination Optics 218

these artifacts erroneously introduce ray hits outside of the letters (e.g., within the
letter “U”) in contours matching the gradient discontinuities. The design process
is built upon an array of linear GRIN elements where it is assumed that rays are
restricted to each discrete unit. Although, a piecewise-continuous design does not
adhere to this restriction.
A similar artifact is found for rays incident at the edge of the optic. The result
is clipping of the irradiance at edges of the target. This effect is also visible in Fig.
5.23 where, for example, the left side of the “U” is thinner than the right due to
rays being clipped at the edge of the aperture. Both of these types of artifacts can
be mitigated by (1) reducing the element thickness, (2) incorporating an absorbing
material at gradient discontinuities, and/or (3) axially tapering the refractive index
to match the source numerical aperture.
Furthermore, the generated illumination can be analyzed at distances other
than the designed-for evaluation distance. One such example is shown in Fig.
5.24 for the “U of R” design presented in Fig. 5.20(b), which is originally designed
for a target distance of zt = 1 m. The binary target designs are free from optical
power (no concentration effect) and generate the illumination in the far field. As
a result, the irradiance fidelity is largely maintained for all distances beyond one
meter. The scale bars in Fig. 5.24 also indicate how the target linear scales in size
with distance according to the designed throw ratio Rt . Although, physically this
effect eventually diminishes due to diffraction at gradient discontinuities, which is
not accounted for in the ray trace. Also, it is interesting to notice how the “U of R”
pattern transforms for distances less than one meter. The letters morph from the
rear square surface of the optic and are recognizable at distances far shorter than
zt = 1 m. Accurate production of the target illumination is only attained beyond
500 mm, however. On the other hand, designs for grayscale targets also have some
Chapter 5. F-GRIN Prescribed Illumination Optics 219

z t = 10 mm z t = 25 mm z t = 50 mm z t = 100 mm

1 mm 2 mm 5 mm 10 mm

z t = 500 mm zt = 1 m z t = 10 m z t = 100 m

50 mm 100 mm 1m 10 m

Figure 5.24: Change in Monte Carlo irradiance with target evaluation distance zt for the
“U of R” design shown in Fig. 5.20(b). The “U of R” design is performed for evaluation
distance zt = 1 m, although the ray trace shows preservation of the target over larger
distances due to the far-field design process. Evaluation distances closer than the design
value show how the irradiance transforms from the square F-GRIN optic.

“depth of field” over which adequate illumination is maintained, although it is


shallower than for binary targets due to its focusing effects apparent in the refrac-
tive index in Fig. 5.21. Finally, for all designs, the absolute irradiance decreases with
evaluation distance according to the inverse-square law. This effect is not included
in any shown Monte Carlo irradiance, which are scaled to relative intensity.
The designs’ generated illumination can also be analyzed for extended sources.
While an infinitesimal point source is assumed in the design process, evaluating
the effects from extended sources serves to validate the assumption that small ex-
tended sources can also be accommodated with minimal loss in target fidelity. This
assumption is found to be valid where adequate illumination is generated for small
extended sources, although degradation continuously increases with source size.
For example, the irradiance distribution from a circular disk source of diameter
Ds is shown in Fig. 5.25 for the Rochester “Flower City” design presented in Fig.
Chapter 5. F-GRIN Prescribed Illumination Optics 220

Ds = 250 μm Ds = 500 μm Ds = 1 mm Ds = 1.5 mm

100 mm 100 mm 100 mm 100 mm

Ds = 2 mm Ds = 3 mm Ds = 4 mm Ds = 5 mm

100 mm 100 mm 100 mm 100 mm

Figure 5.25: Change in Monte Carlo irradiance with circular disk source diameter Ds for
the “U of R” design shown in Fig. 5.20(b). The design is performed for an infinitesimal
point source, although illumination fidelity adequate for many applications is maintained for
extended sources used at a distance zs = −25 mm.

5.20(a). Compared with the irradiance from a point source in Fig. 5.20(a), edge
sharpness progressively decreases with Ds . Although, smaller source diameters
(Ds ≤ 3 mm) yield edges that prove sharp enough for many illumination appli-
cations. Notice that the extended source’s effect is similar to a convolution of the
source with the target where edges become proportionally rounded based on the
source dimensions. Finally, source diameters are shown in Fig. 5.25 up to Ds = 5
mm, which is the maximum allowable size according to the “5 times” rule [185] for
a source distance of zs = −25 mm. The target is still recognizable at the boundary
of the “5 times” rule, although not convincingly so. For designs with disconti-
nuities and higher spatial frequency features, the largest allowable source varies
depending on the application requirements.
Chapter 5. F-GRIN Prescribed Illumination Optics 221

5.6 Fabricated and measured designs

F-GRIN prescribed illumination optics are now fabricated to confirm the validity
of the design process. Additive manufacturing is selected as the method of fabrica-
tion so that gradient discontinuities can be readily incorporated. Recall, one of the
key advantages of designing prescribed illumination optics using F-GRIN rather
than freeform surfaces is that additive manufacturing is available for fabricating
discontinuities.
Two different designs are fabricated by Nanovox LLC using drop-on-demand
inkjet printing techniques. Both optics consists of polymer with nanoparticle dop-
ing to impart refractive index variation. Each optic is designed to satisfy fabri-
cation constraints provided by Nanovox, as outlined in Table 5.1. Both designs
approach the boundaries for all constraints except for the element dimensions.
One fabricated design generates the Rochester “Flower City” logo, as shown in
Fig. 5.26. The optic has a square clear aperture with dimensions 18.2 mm×18.2 mm
and a center thickness t = 3.0 mm. The refractive index distribution has ∆n =
0.1242 and possesses several gradient discontinuities to produce the gaps and holes
in the Flower target.
The second design generates the Nanovox logo, as shown in Fig. 5.27. Like the

Parameter Value
Refractive index range, n (λd ) ∈ [1.4117, 1.5376]
Total refractive index change, ∆n ≤ 0.1259
Lateral x-y pitch [µm] 42.3
Axial z pitch [µm] 11.1
Center thickness [mm] ≤ 3.0
Element dimensions [mm] ≤ 40.0
Table 5.1: Fabrication constraints for F-GRIN prescribed illumination optics made by addi-
tive manufacturing.
Chapter 5. F-GRIN Prescribed Illumination Optics 222

1.5359

Refractive index
1.4738

1.4117

18.2 mm
18.2 mm

Figure 5.26: Fabricated F-GRIN prescribed illumination optic with Rochester “Flower City”
design. Top: illumination target and design refractive index. Bottom: illumination optic
fabricated by additive manufacturing.

1.5375

Refractive index

1.4752

1.4129
20.0 mm

15.0 mm

Figure 5.27: Fabricated F-GRIN prescribed illumination optic with Nanovox logo design.
Top: illumination target and design refractive index. Bottom: illumination optic fabricated
by additive manufacturing.
Chapter 5. F-GRIN Prescribed Illumination Optics 223

target illumination, the optic is rectangular with aperture dimensions 15.0 mm ×


20.0 mm and center thickness t = 3.0 mm. The refractive index range is ∆n =
0.1246 and, again, possesses gradient discontinuities to produce the holes in the
target.
After fabrication, the illumination generated by each design is measured us-
ing the experimental setup shown in Fig. 5.28. The light source used is a white
light LED (recall, all designs are performed monochromatically), which is posi-
tioned at the designed-for source distance zs . The source is immediately followed
by an iris with a minimum aperture of 600 µm, which is smaller than the diam-
eter of the LED. This minimum source extent serves as an approximation to the
point source used in the design process. The simulated effect of a 500 µm extended
source is shown in Fig. 5.25 where blurring of the target edges due to the source
extent is minimal. Following the source and iris, the light transmits through the

Screen

F-GRIN Optic
Source

Figure 5.28: Experimental setup used in measuring generated illumination. For the shown
“Flower City” configuration, the source distance is zs = −115 mm, and the illumination is
viewed on a screen with distance zt = 700 mm. In this configuration, the illumination profile
on the screen has dimensions of 150 mm × 150 mm.
Chapter 5. F-GRIN Prescribed Illumination Optics 224

illumination optic, which is masked to block light that passes outside the optic’s
clear aperture. Following the optic, a white illumination screen is positioned at
the designed-for target distance zt . The generated irradiance on the screen is then
captured by a camera positioned directly above the LED source. The entire exper-
imental setup is located within a dark box.
The optic is mounted by a weak compressive force in a rigid cardstock frame
wrapped in vinyl tape for friction. This unconventional mount is needed to hold
the optic by edge contact only without obscuring its clear aperture. The reason for
this is the printed optics have a clear aperture extending to the mechanical edge of
the element. In the future, this step can be avoided by printing a larger mechanical
aperture outside of the clear aperture that can be baffled and used for mounting.
Vinyl tape is also applied to the edge apertures of each part (see Fig. 5.26) to help
suppress total internal reflection for rays incident upon the edge aperture within
the optic.
The irradiance generated by the two fabricated optics is shown in Fig. 5.29. It
can be seen that both designs successfully generate the target illumination. There
are unwanted artifacts apparent in both irradiance measurements, yet the illu-
mination fidelity remains adequate for many applications. Illumination fidelity
should be expected to only improve given the very early stage of this technology,
both in terms of design and fabrication.
There are different illumination artifacts visible in Fig. 5.29 that reduce illu-
mination fidelity for each design. Several of these effects are in addition to the
ones described in Sec. 5.4.4 for the nominal designs’ ray traced irradiance. In the
fabricated optics, each artifact and its cause can be analyzed, going in order of de-
creasing impact on fidelity. While doing so, the irradiance cross-section shown in
Fig. 5.30 serves as a useful aid.
Chapter 5. F-GRIN Prescribed Illumination Optics 225

50 mm 50 mm

(a) (b)
Figure 5.29: Measured irradiance from fabricated F-GRIN prescribed illumination optics.
(a) The illumination generated by the Rochester “Flower City” design in Fig. 5.26. (b) The
illumination generated by the Nanovox logo design in Fig. 5.27. Qualitatively, both designs
generate the target illumination with adequate fidelity for many applications. A skewing
effect is found in both irradiances due to the line of sight of the measuring camera differing
from the source-optic-target axis.

Figure 5.30: Cross-section of the measured relative irradiance for the fabricated Nanovox
logo design in Fig. 5.27. The cross-sectional plots compare the irradiance with the target
(black dashed line). A skewing effect is found in the irradiance due to the line of sight of
the measuring camera differing from the source-optic-target axis.
Chapter 5. F-GRIN Prescribed Illumination Optics 226

The first type of artifact is mid-spatial frequency (MSF) error in the fabricated
refractive index. MSF error is responsible for the “grainy” appearance of the ir-
radiance in Fig. 5.29 and for producing the “noise” in the cross-sections in Fig.
5.30. Note that unlike the statistical ray tracing noise in the simulation in Fig. 5.23,
the irradiance irregularity due to MSF is a physical effect that is not capture in
the Monte Carlo ray trace evaluation of the nominal designs. MSF error is a very
common issue for different types of optics that are fabricated using sub-aperture
techniques [209]. The additive manufacturing process used in for both F-GRIN
optics is an example of sub-aperture fabrication. MSF in F-GRIN is imparted due
to both the miscibility of constituent materials as well as the curing time that each
layer of the optic undergoes. A longer curing time reduces MSF in F-GRIN addi-
tive manufacturing.
A second type of error is the irradiance introduced to the target’s null (i.e.,
black) regions. This artifact results in a loss of illumination contrast and is caused
by imperfectly sharp gradient discontinuities in the fabricated optics. The round-
ness of a fabricated discontinuity can be seen from the visible Mach-Zehnder inter-
ferogram in Fig. 5.31. An entirely sharp discontinuity would produce interference
fringes with a slope discontinuity, although the fringes in the exploded view of
Fig. 5.31 show a smooth curve instead. Discontinuity roundness is due to the
finite curing time in the additive manufacturing process. Opposite to the MSF
effects, discontinuity roundness is improved by shorter curing times. As a result,
there is a balance between controlling MSF and imparting sharp discontinuities
when determining the curing time. Moreover, null regions in the generated irra-
diance can be improved by tapering the axial refractive index to match the source
numerical aperture. Other forms of three-dimensional index variation may also
be helpful in improving illumination fidelity, taking advantage of a z-component
Chapter 5. F-GRIN Prescribed Illumination Optics 227

Figure 5.31: Mach-Zehnder interferogram of the fabricated Rochester “Flower City” design
in Fig. 5.26. The exploded view shows a rounded-off gradient discontinuity.

in the index gradient. An absorbing material may also be printed along disconti-
nuities to remove light that would otherwise cross a gradient discontinuity. Other
techniques such as edge-enhancement in the printer dithering algorithm may offer
improvement as well.
A third artifact occurs due to TIR off the edge apertures for light propagating
within the optic. The TIR results in stray light that falls in the prescribed null re-
gion outside of the target illumination. TIR can be improved with better baffling
as well as incorporating an absorbing material along the edge aperture (vinyl tape
has already been applied to the edge apertures for this very reason). Like the gra-
dient discontinuities, the optic’s edge aperture can also be tapered according to
the source numerical aperture to reduce the amount of light incident on the edge
aperture within the optic.
Fourth, a more minor defect is the chromatic effect of the fabricated illumina-
tion optics. In both Figs. 5.29 and 5.30, a blue color fringing can be noticed at the
periphery of the measured irradiance. Chromatic effects can be reduced by using a
Chapter 5. F-GRIN Prescribed Illumination Optics 228

less dispersive material. Currently, chromatic effects are ignored in the design pro-
cess, and there is no means of compensation other than adjusting the dispersion of
the constituent GRIN materials.
Fifth, scattering of the bulk GRIN material reduces contrast across the full il-
lumination profile but does not impart specific spatial features. In this scenario,
the impact of scattering is difficult to measure in isolation, but it is believed that
scattering is a minor effect compared with the other artifacts which are more pro-
nounced. It is, however, suspected that scattering is greater in the printed F-GRIN
optics than other bulk materials due to the presence of nanoparticles.
Sixth, the observed irradiance is captured with a camera. Due to the position
of the source, optic, and target, it is not possible to position the camera with a line
of sight matching the optical axis. The result is a skewing effect in the measured
irradiances in Fig. 5.29. This artifact is especially noticeable in the Nanovox logo
irradiance where the left and right edge lines are skewed rather than parallel. This
skewing is an artificial effect of the measurement setup in Fig. 5.28, and it does not
require correction in the design or fabrication of the optic.
In addition to the evaluated irradiance artifacts, the effect of finite source size
can also be measured. Recall from Sec. 5.4.4, the F-GRIN prescribed illumination
optics are designed for an infinitesimal point source, and the finite extent inherent
to any physical source results in a loss of illumination contrast (see Fig. 5.25). The
effect of increasing circular source diameter in the measured irradiance can be seen
in Fig. 5.32 for the fabricated Rochester “Flower City” design. Larger source sizes
reduce edge contrast, as expected. The increasing source diameter also increases
the absolute irradiance. The extended source does help lessen the impact of MSF
error, so in current fabricated designs, some source extent is beneficial in obtaining
more homogeneous illumination. Experimentally, the change in source diameter
Chapter 5. F-GRIN Prescribed Illumination Optics 229

increasing circular disk source diameter


Ds = 600 μm Ds = 3 mm

Figure 5.32: Change in measured relative irradiance with increasing circular disk source
diameter Ds for the fabricated Rochester “Flower City” design in Fig. 5.26. The increasing
source extent rounds the edges of the illumination target, and the increasing source area
increases the absolute irradiance. The changing source diameter is achieved by adjust-
ment of an iris in contact with the white light LED source.

Sun Optic
Measured
irradiance

Figure 5.33: Measured irradiance using a collimated source for the fabricated Rochester
“Flower City” design in Fig. 5.26. The sun serves as the light source and presents some
angular extent that reduces edge contrast similar to a spatially extended source.

is achieved by use of an iris in contact with the white light LED source. The end-
points of the iris diameter range are measured by aid of a microscope.
Finally, an interesting feature of the F-GRIN prescribed illumination optics is
that their design is based on zero optical power via linear GRIN elements. As a
result, a wide range of source distances can be accommodated with minimal effect
on the generated illumination. For an optic designed for a finite source distance,
this effect can be observed to the maximum possible extent using a collimated
source, zs = −∞. For example, the irradiance generated from a collimated source,
in this case the sun, is measured for the Rochester “Flower City” design, as shown
in Fig. 5.33. The fabricated optic is designed for a source distance of zs = −115 mm
yet continues to produce the prescribed illumination with a collimated source.
Chapter 6

Three-Dimensional Gradient-Index
Modal Reconstruction

6.1 Reconstruction overview

6.1.1 Objective

Accurate metrology of optical components is necessary for fabricating optical sys-


tems. The same requirement holds true when incorporating GRIN optics in imag-
ing or illumination systems. The spatial refractive index variation must be verified
to ensure the system functions properly within defined tolerances. Specialized
metrology is required for GRIN optics, and the dimensionality of refractive index
variation is of central importance.
Current GRIN metrology techniques primarily rely on measuring the relative
phase φr of a transmitted wavefront,

φr (x, y) = k · OP D (x, y) (6.1)

where k = 2π/λ is the wavenumber and OP D (x, y) is the spatially variant optical
path difference across the transmitted wavefront. Assuming the GRIN optic under

230
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 231

test is a plane-parallel plate of center thickness t, the OP D for a plane wave prop-
agating along z is found within the thin element approximation (TEA) according
to [22]
Z t
OP D (x, y) = nr (x, y, z) dz (6.2)
0

where nr (x, y, z) = n (x, y, z) − n0 is the refractive index variation relative to base


index value n0 . Within the TEA, Eq. (6.2) shows that the OP D is equal to the
relative refractive index nr integrated along the plane wave propagation direction.
For GRIN with two-dimensional (2D) refractive index variation transverse to
the propagation direction, nr (x, y), the OP D becomes simply

OP D (x, y) = nr (x, y) t. (6.3)

Consequently, within the TEA, the 2D relative refractive index variation nr can be
obtained directly from a measurement of relative phase φr for a wavefront propa-
gating along z,
OP D (x, y) φr (x, y)
nr (x, y) = = . (6.4)
t kt

On the other hand, three-dimensional (3D) refractive index variation such as


in F-GRIN media cannot be deterministically recovered from a single 2D relative
phase measurement. The reason for this is the integral in Eq. (6.2) may yield
the same OP D (x, y) for an infinite number of unique 3D refractive index profiles
n (x, y, z). To rectify this refractive index ambiguity, multiple independent relative
phase measurements with different wavefront propagation directions are required.
Alternatively, measurement of 3D refractive index change can be performed
destructively by slicing a sample into thin sections that are approximately axially
constant. Then, each slice can be measured independently to ultimately recover
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 232

OPD2

1
D
OP

OP
D3
Figure 6.1: Illustration of the 3D GRIN tomography problem. Multi-angular OP D data are
needed to gain volumetric information about the refractive index variation.

relative refractive index information throughout the original 3D volume. The clear
downside to this process is that it is a destructive measurement where the part
under test is sacrificed.
The goal of this work is to devise a reconstruction technique for non-destructively
determining a 3D refractive index variation from a set of 2D relative phase mea-
surements. Each phase measurement is performed for a transmitted planar wave-
front propagating along a unique direction. Multi-angular data are required to
determine 3D index variation since no information is gained about the index gra-
dient along the axis of wavefront propagation according to Eq. (6.2). Ultimately,
3D GRIN reconstruction is formatted as a phase tomography problem [25], as il-
lustrated in Fig. 6.1.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 233

In a variety of fields, prior work has been presented for accomplishing a sim-
ilar goal to the 3D GRIN metrology problem. Although, there is currently no ad-
equate method for reconstructing 3D GRIN media with strong gradients, in the
visible spectrum, and in the absence of scattering and/or diffraction. Sweeney
et al. present several Fourier-based methods for performing tomographic recon-
struction of GRIN fields based on the projection-slice theorem [210]. To fill Fourier
space, Sweeney’s methods rely on 360◦ angular coverage. These approaches are
also only applicable in the “refractionless” regime where the gradient is very weak
and rays can be approximated as linear. As a result, these methods are not appli-
cable to F-GRIN media with strong gradients.
Furthermore, there have been different approaches for measuring GRIN using
beam deflectometry. From the known entering and exiting angles of a laser beam,
information about the intermediate GRIN volume may be obtained. One technique
uses laser beam deflectometry to measure anisotropic (i.e., polarization depen-
dent) GRIN media, although it is assumed there is no axial index variation [211].
Another option attempts to reconstruct axial index change by assuming the form
of the beam path is cubic, yet the reconstruction is limited to a plane rather than
a volume [212]. Two central issues for laser beam deflectometry are how to mea-
sure the exiting beam trajectory and the time it takes to perform the measurement
where the beam must be scanned across the element.
Other modalities for volumetric refractive index reconstruction rely on diffrac-
tion and/or scattering. Optical diffraction tomography (ODT) reconstructs the
complex refractive index volume (i.e., including absorption) via measurement of
diffracted light for different angles of illumination [213, 214]. Optical coherence
tomography (OCT) uses the coherence of scattered light to determine refractive
index differentials within a volume [215]. ODT and OCT are often applied to bio-
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 234

logical tissue or cell samples where refractive index change may be discontinuous,
meaning diffraction and scattering effects are significant. Meanwhile, volumetric
GRIN refractive index is, ideally, smooth and has no imaginary component (ab-
sorption), making diffraction and scattering minimal. For these reasons, both ODT
and OCT are incompatible with the GRIN reconstruction problem.
Another option is computed tomography (CT), which is primarily used in med-
ical imaging. CT measures X-ray absorption for different angles of illumination to
reconstruct a volume’s structure [216]. While there may be an indirect connec-
tion between X-ray absorption and the refractive index in the visible, CT’s appli-
cation for volumetric GRIN measurement is not a valid solution [25]. Moreover,
phase-contrast X-ray imaging considers both the real and complex components to
the refractive index, taking into account X-ray refraction [217]. Although, as with
conventional absorption-based CT, the refractive index in the X-ray region of the
spectrum is not necessarily linked with the index in the visible spectrum.
Finally, perhaps the closest modality for achieving volumetric GRIN measure-
ment is transmission-based ultrasound computed tomography (USCT). Ultrasound
wave propagation is dependent on the “slowness” of the immersing medium.
Slowness is a scalar quantity which is mathematically equivalent to the reciprocal
of the refractive index but in the context of ultrasound rather than light. Conse-
quently, as for light rays, ultrasound rays obey Fermat’s principle with ray curva-
ture within inhomogeneous media [218]. Importantly, ultrasound ray curvature
can be incorporated in USCT using a “bent ray” model [218, 219]. While the USCT
problem is very similarly posed to 3D GRIN metrology, USCT cannot be directly
applied to GRIN reconstruction for two reasons. First, as with X-rays, the slowness
witnessed by ultrasound is not necessarily directly correlated to the refractive in-
dex. Second, transmission-based ultrasound measurements rely on time-of-flight
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 235

using a transmitter-received ring geometry that is incompatible with measurement


of GRIN components. Although, in the future, USCT bent ray algorithms [219] can
potentially be repurposed for zonal GRIN reconstruction (see Sec. 6.1.5) but in an
alternate measurement geometry.

6.1.2 Application and scope

The primary application of 3D GRIN reconstruction is for refractive index metrol-


ogy. This work focuses specifically on the algorithm used in the reconstruction.
GRIN measurement data is given as input to the reconstruction algorithm. Then,
based on a model of the GRIN volume, the reconstruction attempts to solve for the
refractive index distribution that reproduces the input measured data. The vari-
ous factors needed to successfully achieve this reconstruction are the focus of this
work.
Separate from the algorithm, performing the experiment to obtain data used
in the reconstruction presents an additional set of challenges. These challenges
include noise and systematic error inherent to any experiment. The most signif-
icant sources of error are those that deviate from the system modeled in the re-
construction process. For example, the reconstruction assumes the GRIN volume
has plane-parallel surfaces, as explained below. In measurement, any wedge or
surface figure deviation from planar introduces error that alters the reconstructed
GRIN volume. Other sources of experimental error may affect the measurement
registration as well as the illumination incident angle (i.e., accurate adjustment of
GRIN tip/tilt).
For this work focused on the reconstruction algorithm, overcoming the de-
scribed experimental challenges is left for future work, and no empirically derived
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 236

data are considered here. Instead, data used in testing the reconstruction algorithm
are simulated from an established model and are formatted in a way that mirrors
experimental results. It is, however, important to evaluate how an algorithm per-
forms in the presence of noise, as demonstrated in Sec. 6.2.3.
In addition to metrology, a second application of the reconstruction process is
in GRIN inverse design for imaging or illumination. Instead of solving for the
GRIN that matches measured data, the same reconstruction process can be used to
solve for the GRIN volume that produces some desired OP D contribution versus
field angle. As discussed further in Sec. 6.2.2, the reconstruction can also be refor-
matted in terms of ray angle and/or coordinate rather than OP D. Since these ray
quantities are more commonly applied to optical design, this alternative may be
better suited for GRIN inverse design with the reconstruction method.
The scope of the 3D GRIN reconstruction problem must be outlined before
considering solutions. First, only continuous and smooth 3D GRIN profiles will
be considered. The introduction of discontinuities in the index or gradient, such
as in the illumination optics of Chapter 5, greatly complicates the reconstruction
problem by requiring a zonal GRIN representation (see Sec. 6.1.5). Second, it is as-
sumed that a base refractive index value n0 can be readily obtained in order to con-
vert from relative refractive index nr to absolute refractive index n. For instance, n0
can be measured at a point on an exterior surface (e.g., using a Metricon). As such,
the reconstruction process focuses on determining nr from which n = nr + n0 is as-
sumed. Third, only cylindrical GRIN volumes are evaluated. Cylindrical elements
are commonplace in GRIN optical design for systems with circular pupils. Impor-
tantly, in the reconstruction process, light is restricted to transmit only through the
cylinder’s plane-parallel faces. On the other hand, transmitted wavefront mea-
surement through a cylindrical edge aperture is typically infeasible since it is not
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 237

an optically finished surface and the surface’s cylindrical power imparts signifi-
cant OP D that obfuscates the GRIN’s effect.
Next, the data used in the reconstruction process consist of 2D optical path
difference (OP D) measurements. (Recall, these “measurements” can be either em-
pirically obtained or, in this case, modeled.) The OP D is recorded as a function
of plane wave incidence angle. Since plane wave transmittance is restricted to the
plane-parallel surfaces of the cylindrical GRIN volume, full 360◦ angular coverage
is not achievable. The angular restriction of OP D data is a key limitation compared
with CT and USCT, reducing the axial accuracy of the reconstruction. Different
modalities for empirically measuring OP D are discussed next in Sec. 6.1.3.

6.1.3 OP D measurement modalities

Experimentally, there are several means of obtaining transmitted wavefront OP D


data. While no measured OP D data are used in this work, it is important to high-
light how the data underlying the reconstruction process can be measured. It is
also helpful if the reconstruction relies on as direct of a data source as possible to
limit any assumptions or sources of error.
First, relative phase information, or equivalently OP D at a given wavelength,
is accessible from phase-shifting interferometry. A Mach-Zehnder interferome-
ter provides the necessary interferograms for a transmitted wavefront measure-
ment [45]. By phase-shifting, a wrapped phase is obtained, after which a phase
unwrapping procedure is executed to yield φr .
Wavefront OP D data are also available from wavefront sensors. A Shack-
Hartmann wavefront sensor (SHWFS) has been used for measuring the transmit-
ted wavefront following an F-GRIN optic [116]. Wavefront sensors avoid the need
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 238

for phase-shifting or phase unwrapping. By tracking movement of a SHWFS spot


field, the direction cosine of the centroid ray can also be directly obtained.
Third, fringe deflectometry offers a robust means of obtaining OP D from im-
ages of fringes taken through a GRIN element. The captured fringe shape in the
image provides wavefront slope data along the axis perpendicular to the fringes.
Deflectometry is a tool which has also been used for 2D GRIN metrology [116].
Some advantages of deflectometry include a faster measurement operation and no
requirement on source coherence.
Any of the three described OP D measurement modalities may be used within
the reconstruction process described in Sec. 6.2. Each option formats the OP D
data differently and provides different advantages and disadvantages from which
the reconstruction algorithm can be tailored.

6.1.4 Comparing ray and wave propagation

An important decision central to the GRIN reconstruction process is which domain


of optical phenomena to consider. The two options of primary interest are ray op-
tics and wave optics. The two domains are related by the fact that rays are, by
definition, perpendicular to an associated wavefront. Meanwhile, wave optics ac-
counts for wave phenomena such as interference and diffraction, making it a more
physically accurate but more computationally complex option. The computational
simplicity of ray optics is a primary reason why GRIN optical design typically re-
lies on ray tracing.
The prior work described in Sec. 6.1.1 mainly considers wave phenomena. As
a result, one might conclude that wave optics is the optimal domain for GRIN
reconstruction. While this is a valid option, there are additional factors one must
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 239

consider when performing wave propagation in GRIN. The primary challenge is in


GRIN with strong gradients where there is non-negligible ray curvature. In wave
optics parlance, strong gradients result in a spatial variation in wavevector ~k across
the wavefront.
One common approach to wave propagation within inhomogeneous media is
the beam propagation method (BPM) [220]. For propagating a scalar field, BPM
applies a two step process. First, angular spectrum propagation is performed as if
traveling through homogeneous media. Then, the propagated field is modified by
an inhomogeneous phase modulation. The imparted phase is determined within
the TEA according to Eqs. (6.1) – (6.3) by simply the propagation distance and the
transverse inhomogeneous refractive index. For propagation through 3D index
variation, the BPM method can be performed iteratively through sections of the
volume with approximately axially constant index change. For example, some
ODT methods apply a multi-slice BPM model where a volume is segmented, and
wave propagation is performed via BPM slice-by-slice [214].
BPM is very computationally efficient since it only requires one Fourier trans-
form and one inverse Fourier transform per step. Although, there are several
assumptions inherent to BPM that make it inaccurate for strong index gradients.
BPM assumes the TEA as well as the paraxial approximation and the slowly vary-
ing envelope approximation [220]. Inhomogeneity is not accounted for in the prop-
agation step but rather is only incorporated following propagation by multiplica-
tion with a phase modulation in the spatial domain. Consequently, BPM fails to
capture changes in ~k across the wavefront. Without capturing ray curvature, BPM
is not accurate when used with strong index gradients and thus invalid for the
reconstruction process. For example, BPM for Maxwell’s fisheye is shown in Fig.
6.2(a) where inaccurate focusing is demonstrated due to the lack of ~k variation.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 240

Scalar field intensity


(a) (b)

Figure 6.2: Comparison of wave propagation through Maxwell’s fisheye for (a) BPM and
(b) WPM. WPM succeeds in capturing ray curvature where BPM fails. Adapted with per-
mission from [220] © The Optical Society.

Instead, to capture ray curvature, a second option is the wave propagation


method (WPM) [220]. WPM includes refractive index inhomogeneity within the
propagation step rather than following the propagation as in BPM. For WPM, the
Fourier transform is taken of an input scalar field, and a phase modulation is ap-
plied that is specific to each decomposed plane wave frequency. The resultant
integral to revert to the spatial domain possesses both spatial and frequency de-
pendence. As a result, the integral is not simply an inverse Fourier transform, but
rather, each frequency must be integrated individually. This makes WPM a very
computationally expensive process where the number of integrals performed is
proportional to the wavefront sampling. Moreover, like for BPM, WPM assumes
axially constant index change, so a volume of 3D variation must be propagated in
slices, further adding to the computational load. Nevertheless, WPM accurately
accounts for ray curvature and variation in ~k across the wavefront due to a strong
index gradient. The accuracy of WPM can be seen in Fig. 6.2(b) for Maxwell’s
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 241

fisheye where the focus is correctly obtained, unlike for BPM.


WPM accurately models strong GRIN media, but its computational expense
makes a reconstruction process relying on WPM very time-consuming. Computa-
tional efficiency can be improved somewhat by a matrix formulation, as has been
demonstrated in some work on ODT [221, 222]. Nonetheless, for GRIN media that
are ideally free from scattering and diffraction (except at the clear aperture), mod-
eling of wave optical phenomena is unnecessary in the reconstruction process. As
a result, the GRIN reconstruction in Sec. 6.2 operates in the ray optics domain,
obeying Fermat’s principle as is familiar in GRIN optical design. (While not dis-
cussed previously, there is also a third domain of interest that is relevant to future
work. A hybrid ray-wave propagation method has been presented [223] that relies
on probe rays for constructing an ABCD matrix, followed by appropriate wave
propagation.)
The GRIN ray tracing scheme used for the modal reconstruction is the fourth-
order Runge-Kutta method presented by Sharma [19]. The Runge-Kutta method is
iterative where multiple ray steps are taken per ray trace. The number of steps is
based on a user-defined step size. From a ray trace, the optical path length (OP L)
is computed using the trapezoidal method along with the four-term expansion, as
also provided by Sharma [20].
An important feature of the ray tracing method is that it is differentiable. In
other words, the differential change in ray path and OP L with a change in refrac-
tive index distribution can be calculated analytically. This differentiable property
of the ray trace is needed for the iterative optimization procedure described in Sec.
6.2.2.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 242

6.1.5 Comparing modal and zonal reconstruction

Modal reconstruction and zonal reconstruction are two general approaches that
can be used for 3D GRIN reconstruction. The terms “modal” and “zonal” are bor-
rowed from 2D wavefront reconstruction terminology, such as from SHWFS 2D
slope information [224]. The two options have advantages and disadvantages.
Modal reconstruction obtains a 3D refractive index distribution via a globally
defined 3D polynomial basis. The 3D polynomial is governed by a series of coeffi-
cients that are obtained in the reconstruction process to define a specific volumetric
refractive index. The comparatively small number of modal variables reduces the
numerical problem size, improving convergence and allowing for more robust re-
construction. Defined via a polynomial, a modal reconstruction also provides a
refractive index that is continuous, smooth, and in functional form. The functional
form of the index allows for immediate calculation of n and ∇n, as needed for ray
tracing [19].
The primary drawback of modal reconstruction is that refractive index distribu-
tions are limited to those allowable by the finite number of terms in a polynomial
basis. As a result, even for a complete polynomial basis, an arbitrary 3D refrac-
tive index cannot be reconstructed unless an infinite number of terms are consid-
ered. Typically, polynomial terms are added to a reconstruction with increasing
polynomial order. Consequently, the modal reconstruction is often a best-fit re-
sult where mid- and high-spatial frequency index change may be neglected. For
a first attempt at a 3D GRIN reconstruction, the loss of higher spatial frequency
information is acceptable.
Meanwhile, zonal reconstruction obtains a 3D refractive index distribution as a
set of locally defined refractive index values positioned at discrete points in space.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 243

By a zonal reconstruction, the primary advantage is that there is no restriction on


what refractive index distributions can be obtained. This allows for capturing any
order of spatial frequency index change.
The various disadvantages of zonal reconstruction are all due to numerical
complexity. By defining discrete index values, a much larger number of variables
are needed in the reconstruction compared with the modal reconstruction. For ex-
ample, sampling a one millimeter cubic volume with 10 micron pitch requires one
million discrete index values! The vast numerical problem size makes reconstruc-
tion by iterative optimization very time-consuming and unpredictable. (Note, this
same obstacle of problem size is encountered in Sec. 3.3 for local GRIN defini-
tions in optical design.) A further disadvantage is that an interpolation scheme
is needed to continuously define the refractive index between sampled values. 3D
interpolation for n and ∇n from discrete values is not a trivial task and is necessary
for ray tracing. In addition, if a smooth 3D index distribution is desired, additional
smoothness constraints are required in the reconstruction.
In this work, modal reconstruction is selected due to the far reduced numerical
complexity. While restricted in what 3D GRIN profiles can be reconstructed, the
modal approach is a simpler first step in solving the reconstruction problem. Ulti-
mately, a combination of the two approaches – modal reconstruction followed by
zonal reconstruction – is a robust option that may be leveraged in future work.
The 3D polynomial basis used in the modal reconstruction in Sec. 6.2 is the F-
GRIN representation introduced by Yang et al. [14]. The F-GRIN basis uses Zernike
polynomials Ui (x, y) to describe the transverse refractive index change and Legen-
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 244

dre polynomials Vj (z) to capture the longitudinal variation,

NZ X
X NL
n (x, y, z) = n0 + ∆n Θi,j Ui (x, y) Vj (z) (6.5)
i=1 j=1

where NZ is the number of fringe Zernike terms used, NL is the number of Legen-
dre terms used, and Θi,j are coefficients specifying the contribution of each spatial
term. The 3D polynomial basis in Eq. (6.5) is complete, orthogonal and normalized
over the unit cylinder.

6.1.6 Modal reconstruction and mid-spatial frequency errors

The central caveat of modal reconstruction is a limitation in what GRIN profiles


can be obtained. For example, the reconstruction polynomial basis in Eq. (6.5)
consists of the product of a finite number of Zernike and Legendre polynomial
terms. While Eq. (6.5) constitutes a complete 3D polynomial, all GRIN profiles
can only be generated when using an infinite number of terms. For the physical
case of a finite number of terms, refractive index variation that is of higher spatial
frequency than the modal polynomial terms cannot be captured.
Mid-spatial frequency (MSF) error is a notable example of refractive index vari-
ation that cannot be encompassed using a small number of terms in Eq. (6.5).
MSF error in optical component is typically introduced by sub-aperture fabrica-
tion techniques. In particular, GRIN media fabricated by additive manufacturing
are known to possess significant MSF error (e.g., the fabricated illumination optics
in Sec. 5.6). For additive manufacturing, the GRIN MSF error is on the order of the
deposition voxel size (10’s of microns), which is much higher in spatial frequency
than lower-order polynomial terms. Another factor is that additive manufacturing
often relies on a raster deposition pattern for the transverse GRIN variation, which
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 245

is not easily captured by the Zernike polynomials.


For these reasons, it may be difficult to reconstruct GRIN MSF error using the
modal reconstruction technique presented here. For future work, it would be wise
to apply a series of reconstruction processes of increasing complexity. For example,
a model reconstruction followed by a zonal reconstruction is an efficient process
for identifying both lower and higher spatial frequency variation, gaining the ben-
efits of both options outlined in Sec. 6.1.5.

6.2 Modal reconstruction process

The 3D GRIN modal reconstruction process consists of two steps. First, a starting
point guess is generated to approximate the solution as closely as possible based
on available OP D data (see Sec. 6.2.1). Then, the starting point is iteratively opti-
mized to further reduce the difference between the measured data and the current
best guess for the GRIN volume (see Sec. 6.2.2). The algorithms used in the recon-
struction are outlined in detail in Appendix E.
The reconstruction requires no a priori knowledge, meaning any arbitrary 3D
GRIN can be pursued. For the reconstruction, OP D data must be provided across
a range of plane wave angles of incidence. Ideally, the angular OP D sampling is
as large as possible. Besides the OP D data, the only other required information
is the dimensions of the cylindrical volume. Although, a priori data such as the
nominal GRIN prescription, profile symmetry due to the fabrication method, and
the approximate total index change ∆n may also be included in the reconstruction
to improve convergence.
The result of the modal reconstruction process is a two-dimensional coefficient
array Θ of size NZ × NL with Nc total coefficients. The coefficients correspond to
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 246

the Zernike and Legendre polynomial terms outlined in Sec. 6.1.5. Given values
for Θ, the reconstructed 3D GRIN profile is defined continuously by the form of
the polynomial basis in Eq. (6.5).

6.2.1 Starting point generation

The reconstruction process ultimately relies on iterative local optimization, as dis-


cussed next in Sec. 6.2.2. As for all local optimization routines, the starting point
is of critical importance for obtaining reliable convergence to the global minimum
[117].
The best starting point that should be used, if available, is the prescription of
the GRIN. By starting from the prescription, the nominal profile of the GRIN un-
der test is obtained, and the only difference remaining is due to any fabrication
defects. The reconstruction process can subsequently be used to identify these dif-
ferences between prescription and fabrication. In this work where no empirically
measured data is considered, starting from the ground truth prescription in the
reconstruction process leads to a trivial result where the starting point and final
reconstructed result are equivalent for all intents and purposes. The reason for
this is the OP D data simulated from the ground truth matches exactly with that
of the prescription since no manufacturing defects are considered in simulation.
Consequently, the following reconstructions consider the GRIN prescription to be
unknown, testing the capability to reconstruct with no a priori knowledge.
Besides using the GRIN prescription, techniques can be devised for generating
a starting point with no a priori knowledge. These more general options allow
for any unknown GRIN volume to be reconstructed. In this work, three different
methods are considered for starting point generation with no a prior knowledge.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 247

The three options are presented in increasing order of quality but also in increasing
order of complexity.
The first option is to start from a homogeneous starting point. This starting
point is the simplest but is not informed by any measured OP D data. As a re-
sult, the homogeneous starting point is the least fruitful. A homogeneous starting
point does have the property of being the most universally comparable between
different GRIN volumes and different reconstruction algorithms.
The second option is to generate a starting point from the on-axis OP D mea-
surement, assuming the refractive index is axially constant. This starting point is
equivalent to the 2D GRIN metrology described in Sec. 6.1.1. For GRIN volumes
that are axially constant or have a weak axial gradient, this starting point provides
a result that closely matches the optic. Generating an axially constant starting point
from the on-axis measurement is a significant improvement over the homogeneous
starting point but lacks any information about OP D field dependence indicative
of axial index change.
The third option generates a starting point from all measured field-dependent
OP D data. This starting point is generated from a straight-ray approximation
(SRA) that neglects any ray curvature. While non-physical, SRA benefits from an
analytical OP L obtained by a line integral through the cylindrical volume. For a
ray incident at coordinate (x0 , y0 , z0 ) on the front surface of the cylindrical GRIN
volume, vectorial Snell’s law is performed assuming some homogeneous base re-
fractive index n0 . Then, the refracted ray direction is used to construct a linear ray
path between (x0 , y0 , z0 ) and coordinate (x1 , y1 , z1 ) on the rear surface of the GRIN.
This linear ray path ~r can be parameterized over t ∈ [0, 1] by

~r (t) = x (t) x̂ + y (t) ŷ + z (t) ẑ (6.6)


Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 248

where

x (t) = x0 + (x1 − x0 ) t

y (t) = y0 + (y1 − y0 ) t (6.7)

z (t) = z0 + (z1 − z0 ) t.

Then, the OP L line integral for this ray ~r simplifies to

Z 1 Z 1
0
OP L = n [~r (t)] k~r (t)kdt = L n [x (t) , y (t) , z (t)] dt. (6.8)
0 0

q
where L = (x1 − x0 )2 + (y1 − y0 )2 + (z1 − z0 )2 is the linear ray segment length,
which is independent of t. Using the polynomial expression for n in Eq. (6.5), this
OP L integral can be written as

NZ X
X NL
OP L = n0 L + Θi,j Di,j (6.9)
i=1 j=1

where
Z 1
Di,j = ∆n L Ui [x (t) , y (t)] Vj [z (t)] dt. (6.10)
0

The first term in Eq. (6.9) is the bulk OP L for linear ray segment length L through
base index n0 . Meanwhile, the summation in Eq. (6.9) is the optical path difference
OP D, which is the quantity of interest in the reconstruction,

NZ X
X NL
OP D = OP L − n0 L = Θi,j Di,j . (6.11)
i=1 j=1

Notably, OP D is a linear expression where Θi,j are the unknown coefficients to be


solved for and Di,j are scalar constants for a given linear ray path. The definite
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 249

integrals in Eq. (6.10) are performed using numerical integration.


Meanwhile, the measured optical path difference used in the reconstruction is
(a)
denoted OP Dm where the superscript a indicates a specific ray in the measure-
ment. The initial conditions of each measured ray are known including its initial
position and direction in the pupil (how measurement angle and pupil sampling
is performed is detailed in Sec. 6.2.2 and shown in Fig. 6.4). Within the SRA, these
initial conditions determine the linear ray path between (x0 , y0 , z0 ) and (x1 , y1 , z1 ).
For Nr total number of rays in the measurement, a linear system of Nr equations
can now be formed by equating the straight-ray OP D expression in Eq. (6.11) with
the measured data OP Dm ,

    
(1) (1) (1) (1)
 D1,0 ···
D2,0 DNZ ,NL   Θ1,0   OP Dm 
 (2) (2) (2)
    (2)

 1,0 D2,0 · · ·
D DNZ ,NL 
  Θ2,0   OP Dm 
   
 . .. ..   ..   = .. . (6.12)
...
   
 .. . .  .   . 
 
    
(N ) (N ) (Nr ) (Nr )
D1,0r D2,0r · · · DNZ ,NL ΘNZ ,NL OP Dm

The form of this linear system is XΘ = y where X is a matrix of size Nr × Nc , Θ


is a vector containing the Nc GRIN coefficients to be solved for, and y is a vector
containing measured optical path difference for Nr number of rays across mea-
surement angle and pupil.
The ordinary least squares (OLS) solution to the linear system in Eq. (6.12) can
be solved using the normal equation with regularization [117],

−1
Θ = X T X + ΛI XT y (6.13)

where Λ is a unitless regularization parameter used to avoid overfitting the data


and I is the identity matrix. (The value of regularization is discussed further in
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 250

Ground Truth n SRA Starting Point n Residual n


1 1.5190 1 1.5190 1 0.0054

1.5115 1.5115 0.0031


y [mm]

y [mm]

y [mm]
0 1.5040 0 1.5040 0 0.0007

1.4965 1.4965 -0.0016

-1 1.4890 -1 1.4890 -1 -0.0040


-1 0 1 -1 0 1 -1 0 1
x [mm] x [mm] x [mm]
n n n
1 1.5190 1 1.5190 1 0.0130

1.5115 1.5115 0.0079


y [mm]

y [mm]

y [mm]
0 1.5040 0 1.5040 0 0.0027

1.4965 1.4965 -0.0024

-1 1.4890 -1 1.4890 -1 -0.0075


0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0
z [mm] z [mm] z [mm]

Figure 6.3: Starting point for 3D GRIN reconstruction generated using the straight-ray
approximation (SRA). The ground truth, SRA generated starting point, and residual are
shown for cross-sections in x-y (z = 0.5 mm) and y-z. The starting point is obtained from
simulated OP D ray trace data and a regularization parameter Λ = 50. The measurement
angle and pupil sampling used in the SRA is shown in Fig. 6.4. The SRA result does
not exactly produce the axial refractive index change but rather is meant to provide an
advantageous starting point for subsequent iterative optimization.

Sec. 6.2.2.) Since the term in parentheses in Eq. (6.13) may be a singular matrix,
the Moore-Penrose pseudoinverse is needed.
The described process generates a starting point guess Θ by the SRA. The ap-
proach takes into account all available OP D measurements rather than only the
on-axis measurement. As a result, this process allows for an initial guess at the
axial refractive index change. An example starting point generated with the SRA
method is shown in Fig. 6.3, compared with the ground truth GRIN profile (NZ =
16, NL = 4). The axial refractive index change does not exactly match that of the
ground truth; however, the purpose of the starting point guess to provide an ad-
vantageous starting point for iterative optimization, as described next in Sec. 6.2.2.
The difficulty with the SRA integral approach is that its accuracy decreases
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 251

rapidly with ray curvature. Depending on the gradient direction, ray curvature
may either increase or decrease the OP D compared with the straight-ray result.
Consequently, SRA starting point generation works best for thin and/or weak
GRIN volumes. Although, ultimately it is the reconstructed result following sub-
sequent iterative optimization that determines the quality of a starting point guess.

6.2.2 Iterative optimization

Following generation of a starting point guess of coefficients Θ, iterative optimiza-


tion is performed to minimize the difference between the measured OP Dm and
the simulated OP Ds (Θ). The simulated OP Ds (Θ) is computed for the current
guess Θ using the Runge-Kutta ray tracing routine mentioned in Sec. 6.1.4. The
optimization routine seeks to iteratively reduce the difference between measured
and simulated results according to a cost function. Two different nonlinear itera-
tive optimization algorithms considered for this work are the Broyden–Fletcher–
Goldfarb–Shanno (BFGS) and conjugate gradient methods [117].
If optimization succeeds, the final reconstructed result for Θ produces the same
field dependence of optical path difference as the measured data. Notice, the 3D
GRIN reconstruction is obtained indirectly via OP D data rather than directly from
refractive index values, which are inaccessible. As a result, the reconstruction
process relies on an underlying assumption that 3D refractive index volumes are
uniquely paired with field-dependent OP D data. This assumption is only strictly
true for full 360◦ angular measurement coverage.
While not demonstrated here, the same iterative optimization approach can be
applied to reconstruct 3D refractive index using ray coordinates and/or direction
cosines rather than OP D data. Optimization would instead be used to minimize
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 252

the difference between simulated and measured (or desired) ray quantities. For
example, for inverse design, a 3D GRIN can be reconstructed that redirects rays to
focus at an image plane across the field-of-view.
As with any optimization, the key to success is the cost function J (Θ), which
returns a scalar value that is sought to be minimized. It is important that the cost
function accurately reflects the goal of the optimization. In this case, the cost func-
tion consists of two components,

J (Θ) = JOP D (Θ) + Jreg (Θ) . (6.14)

The first term in the cost function JOP D (Θ) is the cost due to differences in OP D
between the measured data OP Dm and the simulated ray trace result OP Ds (Θ) for
the current coefficient guess Θ. JOP D (Θ) is calculated by the mean squared error
(MSE) between these two data sets, summed across measurement field angles f
and rays in the pupil p,

Nf Np
1 XX (f, p) 2
OP Ds(f, p)(Θ) − OP Dm

JOP D (Θ) = (6.15)
2Nf Np f =1 p=1

where Nf is the total number of measurement angles and Np is the total number
rays in the pupil. The field and pupil sampling establishing Nf and Np is critical
and must ensure adequate coverage of the 3D GRIN volume.
OP D sampling across different measurement field angles f is the central premise
of the 3D GRIN reconstruction process. As described in Sec. 6.1.1, multiple unique
angular measurements are needed to obtain information about 3D refractive in-
dex change. To reconstruct a 3D GRIN profile with no assumed symmetry, OP D
measurements are required with 2D angular sampling, which is quantified here in
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 253

Nf = 129 Np = 207

(a) (b)
Figure 6.4: OP D ray sampling used for the cost function summation in Eq. (6.15). (a) Field
angle sampling is performed on a polar grid where each point indicates a plane wave
angle of incidence across which OP D is measured. (b) Pupil sampling is determined by
Gaussian quadrature [225, 226] where, for a given field angle, each point indicates the
entrance pupil coordinate of a traced ray for which OP D is measured.

terms of field angles θx and θy . In θx – θy space, sampling is performed on a polar


grid based on azimuthal and radial sampling out to a maximum field angle θmax .
Since little information is gained for the index gradient along the propagation di-
rection, reconstruction is aided by using as large of θmax as feasible to gain as much
OP D angular diversity as possible. For example, with the extreme limiting case
of θmax = 0◦ (i.e., only the normally incident measurement), no axial index infor-
mation can be gained. An example of field sampling with 5◦ radial steps and 22.5◦
azimuthal steps out to a maximum field angle of θmax = 40◦ is shown in Fig. 6.4(a),
yielding Nf = 129 total angular measurements.
Within each angular measurement, the OP D is sampled by rays across a 2D cir-
cular entrance pupil. These rays spanning the pupil are also summed over in the
OP D cost function term in Eq. (6.15). The pupil sampling is defined using Gaus-
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 254

sian quadrature [225, 226], specified in terms of a number of “rings” and a number
of “spokes” in the pupil. For example, pupil sampling is shown in Fig. 6.4(b)
with 10 rings and 10 spokes, yielding Np = 207 rays total for a given measurement
angle.
A fundamental difficulty for 3D GRIN reconstruction is the “missing cone”
problem, a challenge also found in other angularly restricted tomography prob-
lems [227]. The cylindrical geometry of the GRIN reconstruction volume requires
that wavefronts only transmit through the plane-parallel surfaces. As a result,
full 360◦ angular coverage of the GRIN volume is not attainable. Moreover, edge
portions of the GRIN volume are sampled less due to vignetting of rays in the
entrance pupil, as depicted in Fig. 6.5. In the reconstruction, any ray exiting the
edge aperture of the cylindrical GRIN volume is vignetted and not included in
the cost function. Consequently, the number of transmitted rays is a function of
field angle, Np (f ). In fact, the only measurement angle that samples the full GRIN
volume with no vignetting is the normally incident plane wave, as in Fig. 6.5(a).
Meanwhile, the non-normally incident measurements in Fig. 6.5(b)-(c) experience
vignetting where edges of the GRIN volume are not “seen” by the measurement.
This effect in 3D Fourier space corresponds to a cone of missing frequencies, hence
the name of the missing cone problem [227].
In addition to the OP D cost, the second term in the full cost function in Eq.
(6.14) is a regularization term,

NZ XNL
Λ X
Jreg (Θ) = Θ2 (6.16)
2NZ NL i=1 j=1 i,j

where Λ is a user-defined regularization parameter. Regularization provides a


dampening effect that prevents the optimized result Θ from overfitting to other
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 255

(a) x

(b) x

(c) x

Figure 6.5: The missing cone problem in 3D GRIN reconstruction, illustrated for three
plane wave measurement angles. The GRIN cylindrical volume under test is embedded
within a larger element with a homogeneous outer annulus. Circular apertures mask the
clear aperture on front and back surfaces. (a) The normally incident measurement cap-
tures OP D information for the entire cylindrical GRIN volume, and the entrance pupil is
filled. (b)-(c) Non-normally incident measurements capture OP D information for only a
subset of the volume, and the entrance pupil is underfilled due to vignetting.

components in the cost function. Regularization is essential when performing any


optimization on data that may possess error and/or noise that yields outlier data
points deviating from what is possible in simulation. The regularization cost in
Eq. (6.16) achieves this goal via the sum squared of Θ, meaning there is a cost to
the optimizer arbitrarily incorporating minimally impactful terms that overfit the
data.
For the given optimization problem, regularization prevents Θ from overfit-
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 256

ting to any measured OP Dm data that cannot agree with the simulated OP Ds re-
sult. For example, different sources of error and noise in experimentally obtained
OP Dm requires regularization to ensure optimization does not overfit the refrac-
tive index to erroneous (i.e., nonphysical) outlier data points. A balance must be
struck by ignoring erroneous outliers while still considering valid data points in
OP Dm , a problem known as the bias-variance trade-off. This balance is achieved
using a proper value for the regularization parameter Λ in Eq. (6.16). The optimal
value for Λ is problem-specific and must be carefully tuned. Determination of Λ
is done by a hyperparameter search using a test data set, a time-consuming but
critically important process.
In addition to the cost function value J, it is also helpful for optimization con-
vergence to provide the gradient of the cost function with respect to the optimiza-
tion variables, ∇Θ J [117]. This all-important cost function gradient can be solved
analytically by
∇Θ J = ∇Θ JOP D + ∇Θ Jreg (6.17)

where

Nf Np
1 XX
OP Ds(f, p)(Θ) − OP Dm
(f, p)
∇Θ OP Ds(f, p)(Θ)

∇Θ JOP D =
Nf Np f =1 p=1
(6.18)
NZ XNL
Λ X
∇Θ Jreg = Θi,j .
NZ NL i=1 j=1

For determining ∇Θ JOP D , it is found that the gradient with respect to Θ of the
(f, p)
simulated optical path difference is required, ∇Θ OP Ds (Θ). This means for a
given field and pupil ray (f, p), the optical path difference found by Runge-Kutta
ray tracing must be differentiated with respect to all GRIN coefficients found in
Θ. This relationship can be derived since the OP D and ray trace depends on n
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 257

and ∇n, both of which are dependent on Θ by means of Eq. (6.5). Although, the
differentiation itself is no trivial task with the Runge-Kutta method where each
ray takes several steps within the GRIN volume. Surface refraction according to
Snell’s law must also be accounted for in the OP D gradient with respect to Θ.
(f, p)
There are two options for calculating ∇Θ OP Ds (Θ). The first is implementing
the Runge-Kutta ray trace with automatic differentiation software. The alternative
is to manually calculate this analytical derivative and to check the accuracy of the
result using finite differences. In this work, the latter option is used for greater
control over the derivative calculation.
Calculating J and ∇Θ J can be a very time-consuming process, especially con-
sidering evaluation of these functions occurs multiple time per optimization cycle.
The cost function sums over potentially 100’s of measurement angles with 100’s
of rays per measurement, 10’s of steps per ray, and 4 coordinate samples of n and
∇n per ray step. In total, several hundred thousand calculations, at minimum, are
required per evaluation of J, after which even further computation is needed for
∇Θ J.
The computation time needed for evaluation of J and ∇Θ J is greatly reduced
using parallel computing. Parallel computing distributions the computational load
across multiple central processing units (CPUs) to calculate different components
of the cost function simultaneously. Calculations are performed independently for
each angular measurement, which can be executed in parallel on separate CPUs.
Then, the results from all CPUs are gathered to constitute J and ∇Θ J before pro-
ceeding with the optimization. The reduction in computation time with paral-
lelization is approximately proportional to the number of available CPUs. For the
iterative optimization, parallelization is performed across 24 to 56 CPUs from the
BlueHive computing cluster located at the University of Rochester’s Center for
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 258

Integrated Research Computing. An additional advantage of using the BlueHive


cluster is that multiple reconstructions can be performed simultaneously, which
proves very helpful in tuning the regularization parameter Λ and experimenting
with different optimization algorithms.
An iterative optimization is performed to demonstrate 3D GRIN modal recon-
struction. The ground truth 3D refractive index variation from Sec. 6.2.1 is an-
alyzed once again. The optimization starting point is generated by the SRA ap-
proach with the result shown in Fig. 6.3. This starting point is generated using
four axial index terms, NL = 4. Now, for iterative optimization, three additional
axial terms are included, NL = 7, for higher-order axial fitting. (Note the ground
truth has no axial terms higher than NL = 4.) The same number of transverse
Zernike terms is used as in the SRA result, NZ = 16. Thus, the total number of co-
efficients being solved for is Nc = 112. From the SRA starting point, the described
iterative optimization is performed. The field angle and pupil sampling constitut-
ing in the OP D cost function are those depicted in Fig. 6.4. The data OP Dm used
in the reconstruction are simulated using ray tracing for these field and pupil sam-
plings. The computation time for the reconstruction is 4 hours and 38 minutes on
the BlueHive cluster using 48 CPUs.
The final 3D GRIN reconstruction is shown in Fig. 6.6 compared with the
ground truth for x-y and y-z cross-sections. The residuals show refractive index
reconstruction error of 0.0011 for transverse variation and 0.0053 for axial varia-
tion. The axial index variation is almost fully captured in the reconstructed re-
sult, a feat unattainable with conventional techniques relying on only the normally
incident measurement. Deviations from ground truth are likely due to informa-
tion loss from the missing cone problem and the angular measurement restriction
θmax = 40◦ . Without 360◦ angular coverage, different axial terms may have approx-
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 259

Ground Truth n Reconstruction n Residual n


1 1.5167 1 1.5167 1 0.0005

1.5098 1.5098 0.0002


y [mm]

y [mm]

y [mm]
0 1.5028 0 1.5028 0 -0.0001

1.4959 1.4959 -0.0004

-1 1.4890 -1 1.4890 -1 -0.0006


-1 0 1 -1 0 1 -1 0 1
x [mm] x [mm] x [mm]
n n n
1 1.5167 1 1.5167 1 0.0029

1.5098 1.5098 0.0016


y [mm]

y [mm]

y [mm]
0 1.5028 0 1.5028 0 0.0003

1.4959 1.4959 -0.0010

-1 1.4890 -1 1.4890 -1 -0.0024


0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0
z [mm] z [mm] z [mm]

Figure 6.6: 3D GRIN reconstruction obtained using iterative optimization. The ground
truth, final reconstruction, and residual are shown for cross-sections in x-y (z = 0.5 mm)
and y-z. Reconstruction is performed using simulated OP D ray trace data and a regu-
larization parameter Λ = 6.11 × 10−2 . The optimization starting point is generated by the
SRA result shown in Fig. 6.3. The measurement angle and pupil sampling used in the
reconstruction is shown in Fig. 6.4.

Figure 6.7: Visualization of 3D GRIN coefficients Θ for the ground truth and reconstructed
result. Dominant terms in the ground truth are successfully identified by the reconstruction.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 260

imately similar OP D contributions.


The ground truth and reconstructed coefficients Θ are also visually depicted in
Fig. 6.7 for the Nc = 112 different Zernike and Legendre terms. By comparison, it
is seen that the reconstruction closely reproduces the dominant coefficients in the
ground truth. Recall, the ground truth coefficients are zero for all Legendre (axial)
terms greater than fourth-order. Meanwhile, the reconstructed result allows for
up to seventh-order Legendre terms, yet the optimized result correctly excludes
significant contributions from axial terms greater than fourth-order.
The iterative optimization used in the reconstruction seeks to minimize the cost
function in Eq. (6.14) which consists of OP D as well as regularization components.
As shown in Fig. 6.8(a), the optimization successfully reduces the cost function
monotonically across all optimization cycles. Although, the error of reconstructed

Figure 6.8: Change with optimization cycle of (a) cost function and (b) RMS error of recon-
structed refractive index. The reconstruction result shown in Fig. 6.6 is plotted for the final
optimization cycle 73. The RMS error is calculated by 3D sampling on a 10 µm grid. For
all cycles, the iterative optimization successfully reduces the cost function, which from Eq.
(6.14) is based on both OP D and regularization. Although, the RMS error of reconstructed
refractive index does not decrease for all cycles while ultimately reaching a minimum upon
completion of optimization.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 261

refractive index is the ultimate metric of success. As shown in Fig. 6.8(b), after
optimization, the RMS error of reconstructed refractive index is, like the cost func-
tion, at a minimum with a value of 5.5 × 10−4 . It is interesting to note that, unlike
the cost function, the RMS error of reconstructed refractive index does not decrease
monotonically for all optimization cycles. Instead, there are some intermediate cy-
cles where the RMS error increases to then later decrease, reaching the minimum in
RMS error for the final optimization cycle. The regularization cost is believed to be
the reason for this where the transition point Jreg > JOP D affects the behavior of
the optimizer.

6.2.3 Effect of measurement noise

The reconstruction process is demonstrated using simulated OP D data; however,


any use of the reconstruction algorithm with empirically obtained data would
encounter noise and error in the measurements. To assess the effects of noise
on reconstruction accuracy, the reconstruction is repeated where random Gaus-
sian noise is introduced to all OP D data. The magnitude of the added noise is
3σ = ±1/4 λ at λ = 632.8 nm, meaning 99.7% of OP D data points experience noise
within plus-or-minus a quarter-wave (see Fig. 6.9).
The noisy reconstruction result is shown in Fig. 6.10 where it can be seen that
the process is still largely successful, albeit the residual is slightly larger compared
with the zero noise case in Fig. 6.6. From the noisy reconstruction, the obtained 3D
refractive index profile has an RMS error of 6.6 × 10−4 , a 20% increase compared
with the zero-noise RMS error of 5.5 × 10−4 . Nevertheless, 10−4 refractive index
accuracy is acceptable for many applications.
The coefficients Θ for the ground truth and result reconstructed from noisy data
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 262

Figure 6.9: Distribution of random noise introduced to OP D data used in the noisy recon-
struction. The noise bandwidth is 3σ = 1/4 λ at λ = 632.8 nm, meaning 99.7% of OP D
data points experience noise within plus-or-minus a quarter-wave

Ground Truth n Reconstruction n Residual n


1 1.5167 1 1.5167 1 0.0005

1.5098 1.5098 0.0001


y [mm]

y [mm]

y [mm]

0 1.5028 0 1.5028 0 -0.0003

1.4959 1.4959 -0.0007

-1 1.4890 -1 1.4890 -1 -0.0010


-1 0 1 -1 0 1 -1 0 1
x [mm] x [mm] x [mm]
n n n
1 1.5167 1 1.5167 1 0.0034

1.5098 1.5098 0.0022


y [mm]

y [mm]

y [mm]

0 1.5028 0 1.5028 0 0.0009

1.4959 1.4959 -0.0003

-1 1.4890 -1 1.4890 -1 -0.0015


0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0
z [mm] z [mm] z [mm]

Figure 6.10: 3D GRIN reconstruction obtained using iterative optimization for OP D data
containing Gaussian noise (3σ = ±1/4 λ). The ground truth, final reconstruction, and
residual are shown for cross-sections in x-y (z = 0.5 mm) and y-z. Reconstruction is
performed using simulated, noisy OP D ray trace data and a regularization parameter Λ =
5.50 × 10−1 . The optimization starting point is generated by the SRA result shown in Fig.
6.3. The measurement angle and pupil sampling used in the reconstruction is shown in
Fig. 6.4.
Chapter 6. Three-Dimensional Gradient-Index Modal Reconstruction 263

Figure 6.11: Visualization of 3D GRIN coefficients Θ for the ground truth and the result
reconstructed with noise. Dominant terms in the ground truth are successfully identified by
the reconstruction.

are also visually depicted in Fig. 6.11 for the Nc = 112 different Zernike and Leg-
endre terms. By comparison, it is seen that the reconstruction, again, reproduces
the dominant coefficients in the ground truth. As for the noise-free reconstruction
shown in Fig. 6.7, higher-order axial terms are not captured as well as transverse
terms.
Overall, the 3D GRIN modal reconstruction is shown to successfully obtain ax-
ial refractive index information, including in the presence of measurement noise.
The reconstruction process enables non-destructive 3D GRIN measurement for the
first time. For future work, the next step is to overcome the experimental chal-
lenges outlined in Sec. 6.1.1 and perform the modal reconstruction on empirically
obtained OP D data.
Chapter 7

Conclusion and Future Research

7.1 Concluding remarks

Empowered by continuously advancing fabrication options, F-GRIN media offers


invaluable new DOFs to optical design. This work has examined a selection of F-
GRIN application spaces for both imaging and illumination. New system geome-
tries and novel optical influence are obtained through F-GRIN while also achiev-
ing desired optical performance. F-GRIN enables creative alternatives for further
improvements in optical design while reducing SWAP-C.
Three dissertation objectives are outlined in Sec. 1.4 that are now reexamined.
The first is to apply F-GRIN for obtaining desirable system geometries that are
unattainable with conventional optics. The second is to realize novel optical func-
tionality using F-GRIN for improved optical design. The third is to tackle the
3D GRIN reconstruction problem facing non-destructive 3D GRIN metrology. All
three objectives have been accomplished across the presented chapters.
Chapter 2 explores the design of monolithic AFLs using F-GRIN media. High
system specifications (f /1.5) are achieved in a highly compact system geometry
via the available F-GRIN DOFs. For polychromatic imaging, the GRIN AFL dis-

264
Chapter 7. Conclusion and Future Research 265

persion is engineered using fictitious materials for chromatic aberration correction,


enabling high broadband performance across a 10◦ full field-of-view. The multi-
material design meets or surpasses the MTF specifications of the Sigma 135 mm
f /1.8 DG HSM, a high-performing conventional telephoto lens. The work in this
chapter addresses the first dissertation objective where a desirable system geom-
etry is obtained without sacrificing performance. The second objective for unique
optical functionality is also demonstrated where the engineered GRIN dispersion
achieves color correction in a scenario where a single homogeneous material can-
not.
Chapter 3 examines GRIN Alvarez lenses (GALs) and their application to a
compact zoom riflescope design. A first-order framework is developed for the
GAL power variation. Then, GAL optical performance is evaluated for both wave-
front error and boresight error. GAL performance is assessed as a function of
different free design variables including power range, bias power, element thick-
ness, diameter to clear aperture ratio, and GRIN tilt. Next, optimized higher-order
GAL designs are considered. By incorporating axial refractive index variation,
the diffraction-limited field-of-view is increased and RMS wavefront error is more
than halved at the edge of a 10◦ full field-of-view. From these findings, a novel
zoom riflescope design is executed using a pair of longitudinally fixed GAL pairs.
Without longitudinally translating elements, the system length is reduced by al-
most 10%. The work in this chapter further demonstrates the first dissertation
objective where F-GRIN allows for more compact systems. Toward the second
objective, the new optical capability of variable power GAL elements is also pre-
sented.
Chapter 4 applies a plane-parallel GRIN element at a conjugate plane to present
an artificially curved focal surface to a preceding optical system. This artificially
Chapter 7. Conclusion and Future Research 266

curved image provides a means for field curvature compensation using GRIN
cover glass. The relationship between the refractive index variation and imparted
image surface is analytically derived by considering in a new light the image dis-
placement through a homogeneous plate. Field curvature compensation is demon-
strated for the Schmidt telescope where image sag of 244 µm is reduced to < 6 µm
using GRIN cover glass. Field curvature compensation using GRIN enables more
compact imaging systems, meeting the first dissertation objective. The planar form
factor for the GRIN element is also unattainable with conventional homogeneous
optics relying on surface sag, supporting the second objective.
Chapter 5 introduces a new technique for generating prescribed illumination
using an F-GRIN optic. The design procedure is described where an array of linear
GRIN elements is integrated to yield a piecewise-continuous F-GRIN profile. Sev-
eral designs are presented along with two fabricated elements that serve as proofs
of concept. The second dissertation objective is supported by this work where a
plane-parallel optic is used to generate the prescribed illumination while in prior
work non-planar freeform elements are required that rely on surface sag.
Chapter 6 describes a 3D GRIN reconstruction technique. The modal recon-
struction algorithm first generates an intelligent starting point using multi-angular
relative phase data. Then, iterative optimization is applied using GRIN differen-
tiable ray tracing to minimize the difference between measured and modeled data.
An example reconstruction is performed where the RMS error of the 3D recon-
structed index is 5.5 × 10−4 with no noise in the measurement data. A second re-
construction finds the RMS error of the reconstructed refractive index is 6.6 × 10−4
with ±λ/4 Gaussian noise introduced to the data. The reconstruction process is
multi-purpose and can be useful in the inverse design of imaging or illumination
optics. The modal reconstruction also addresses the third dissertation objective by
Chapter 7. Conclusion and Future Research 267

offering a technique for performing non-destructive 3D GRIN metrology.

7.2 Suggestions for future research

The field of F-GRIN design and metrology is in its infancy. As a result, there is
an abundance of future research topics that are worth pursuing. There are several
unexplored opportunities that present significant potential in the spaces of F-GRIN
modeling, optimization, design forms, and metrology.
Current modeling of F-GRIN media relies primarily on global polynomial rep-
resentations. Polynomial definitions offer a simple approach with comparatively
few variables. On the other hand, local index representations offer significantly
more flexibility by allowing for higher spatial frequency refractive index variation.
However, the increased DOFs of local representations come at the cost of increased
computational complexity. In this work, hybrid global-local representations are
used with success for the optimized GALs (see Sec. 3.3.2) and the piecewise-
continuous illumination optics (see Sec. 5.4.3). These hybrid approaches offer the
best of both options where a manageable number of variables are introduced to
the design process yet plentiful DOFs allow for significant performance improve-
ment over purely global representation designs. In the immediate future, hybrid
representations offer the greatest promise, yet new modeling techniques must be
developed, including those outside of commercial software. With more advanced
computational methods, purely local GRIN definitions offer even greater DOFs in
the future. A good first option for local GRIN representation is with 3D B-splines,
which by limiting the number of control points are computationally simpler than
alternatives like 3D non-uniform rational B-splines (NURBS). Deep learning is also
an attractive option where a matrix based local representation can be evaluated
Chapter 7. Conclusion and Future Research 268

with computational efficiency using matrix operations.


Another challenge in F-GRIN modeling that remains largely unexplored is per-
forming a physically accurate sensitivity analysis. With sub-aperture fabrication
techniques such as additive manufacturing, the scale of refractive index error is
of mid- and high-spatial frequency. Sensitivity analysis is necessary for reliable
as-built performance and must be performed in accordance with the fabrication
method in use. This sensitivity analysis is needed for establishing much-needed
fabrication tolerances, particularly on the MSF effects from additive manufactur-
ing witnessed in Sec. 5.6.
Beyond modeling, improvements are also needed in F-GRIN optimization. Cur-
rent techniques often struggle with high-dimensional solution spaces where poten-
tially hundreds of parameters must be solved. Even if eventually converging to the
global minimum, computation time poses a limitation on effective search and op-
timization. New optimization algorithms, including those for global optimization,
are greatly needed for effective F-GRIN design. This requirement is especially true
for future work on hybrid and local GRIN representations that present an even
greater number of optimization variables.
One reason for currently inefficient F-GRIN optimization is that, generally, deriva-
tive calculation is performed by finite differences. A solution to this problem is
to use differentiable ray tracing, although this is not currently an option in com-
mercial software. Recall, GRIN differentiable ray tracing is applied in Chapter 6
for performing the 3D GRIN reconstruction via automatic differentiation software.
The same process can also be applied to GRIN optimization for inverse design.
Moreover, differentiable ray tracing is an advantageous tool for use in machine
learning-based design. Machine learning for GRIN design is an interesting alter-
native to iterative optimization where the computation time can be front-loaded.
Chapter 7. Conclusion and Future Research 269

Third, there are countless design forms that have not yet been explored that
may benefit from incorporating F-GRIN. Of particular interest are catadioptric sys-
tems (i.e., systems that also incorporate reflective surfaces), as examined with in
Chapter 2 for AFLs. Upon reflection, double-passing through the same F-GRIN
volume offers additional DOFs without the need for fabricating a second F-GRIN
element. For instance, an F-GRIN Mangin mirror design presents interesting pos-
sibilities.
Design forms incorporating GRIN cover glass for field curvature compensation
also require further examination. First, a simple system with known field curva-
ture such as the Schmidt camera or Petzval portrait lens can be joined with GRIN
cover glass to demonstrate its field flattening capability. A comparison with other
methods for field curvature compensation such as curved detectors and homoge-
neous lenses is also warranted. After, more complex designs such as mobile device
imagers and freeform systems can be analyzed to determine what track length re-
duction is possible (see Sec. 4.3.5). Another technique of interest is to use GRIN
cover glass for decreasing chief ray angles of incidence on the sensor for improved
image relative illumination (i.e., relocating the system exit pupil). Since located at
a conjugate plane, a limitation of GRIN cover glass is that fully pupil-dependent
aberrations cannot be controlled.
Lastly, F-GRIN metrology is severely lagging behind design and fabrication
techniques. It is critical that metrology be improved in order for as-built F-GRIN
systems to progress. An immediate task at hand is to experimentally perform a
3D reconstruction using the algorithm presented in Chapter 6. There are several
additional challenges associated with constructing and executing the experiment
including sources of error and noise, measurement fiducials, and accompanying
surface metrology. The reconstruction process may need to be modified for im-
Chapter 7. Conclusion and Future Research 270

proved regularization to accommodate these empirical effects. A means of per-


forming error analysis must also be devised, and a reconstruction validation study
must be performed in comparison with destructive metrology.
Future 3D reconstruction techniques may achieve higher refractive index fi-
delity using a hybrid modal-zonal approach. By following the modal reconstruc-
tion with a subsequent zonal process, higher spatial frequency features may be
detected. This procedure has previously been used with success for ultrasound
computed tomography (USCT) by the “bent ray” model. Future 3D GRIN recon-
struction may also considered the spatial variation in dispersion. This can be done
by repeating 3D metrology at multiple wavelengths for the same GRIN optic.
Bibliography

[1] R. Kingslake, Lens Design Fundamentals (Academic Press, 1978).

[2] G. W. Forbes, “Shape specification for axially symmetric optical surfaces,”


Optics Express 15, 5218–5226 (2007).

[3] T. Stone and N. George, “Hybrid diffractive-refractive lenses and achro-


mats,” Applied Optics 27, 2960–2971 (1988).

[4] J. P. Rolland, M. A. Davies, T. J. Suleski, C. Evans, A. Bauer, J. C. Lambropou-


los, and K. Falaggis, “Freeform optics for imaging,” Optica 8, 161–176 (2021).

[5] N. Yu and F. Capasso, “Flat optics with designer metasurfaces,” Nature Ma-
terials 13, 139–150 (2014).

[6] D. T. Moore, “Gradient-index optics: a review,” Applied Optics 19, 1035–1038


(1980).

[7] K. Thompson, “Description of the third-order optical aberrations of near-


circular pupil optical systems without symmetry,” Journal of the Optical Soci-
ety of America A 22, 1389–1401 (2005).

[8] J. P. Rolland, A. M. Bauer, K. H. Fuerschbach, T. Schmid, and K. P. Thomp-


son, “Roland V. Shack’s discovery of nodal aberration theory, the expansion
into the aberrations of freeform optics, and impact in optical design,” Proc.
SPIE 11479, Roland V. Shack Memorial Session: A Celebration of One of the Great
Teachers of Optical Aberration Theory, 114790G (2020).

[9] K. P. Thompson and J. P. Rolland, “A page from ”the drawer”: how Roland
Shack opened the door to the aberration theory of freeform optics,” Proc.
SPIE 9186, Fifty Years of Optical Sciences at The University of Arizona, 91860A
(2014).

[10] K. Fuerschbach, J. P. Rolland, and K. P. Thompson, “Theory of aberration


fields for general optical systems with freeform surfaces,” Optics Express 22,
26585–26606 (2014).

271
Bibliography 272

[11] A. Bauer, E. M. Schiesser, and J. P. Rolland, “Starting geometry creation and


design method for freeform optics,” Nature Communications 9, 1756 (2018).

[12] L. W. Alvarez, “Two-element variable-power spherical lens,” U.S. patent


3,305,294 (1967).

[13] A. J. Yee, W. Song, N. Takaki, T. Yang, Y. Zhao, Y. H. Ni, S. Y. Bodell, J. L.


Bentley, D. T. Moore, and J. P. Rolland, “Design of a freeform gradient-index
prism for mixed reality head mounted display,” Proc. SPIE 10676, Digital
Optics for Immersive Displays, 106760S (2018).

[14] T. Yang, N. Takaki, J. Bentley, G. Schmidt, and D. T. Moore, “Efficient repre-


sentation of freeform gradient-index profiles for non-rotationally symmetric
optical design,” Optics Express 28, 14788–14806 (2020).

[15] D. H. Lippman, N. Kochan, T. Yang, G. Schmidt, J. Bentley, and D. Moore,


“Freeform gradient-index media: a new frontier in freeform optics,” Optics
Express 29, 36997–37012 (2021).

[16] D. T. Moore, “Aberration Correction using Index Gradients,” MS thesis, Uni-


versity of Rochester, The Institute of Optics (1970).

[17] E. W. Marchand, Gradient Index Optics (Academic Press, 1978).

[18] R. W. Wood, Physical Optics (Macmillan, 1905).

[19] A. Sharma, D. V. Kumar, and A. K. Ghatak, “Tracing rays through graded-


index media: a new method,” Applied Optics 21, 984–987 (1982).

[20] A. Sharma, “Computing optical path length in gradient-index media: a fast


and accurate method,” Applied Optics 24, 4367–4370 (1985).

[21] A. Sharma and A. K. Ghatak, “Ray tracing in gradient-index lenses: compu-


tation of ray–surface intersection,” Applied Optics 25, 3409–3412 (1986).

[22] F. Wyrowski and J. Turunen, “Wave-optical Engineering,” in International


Trends in Applied Optics, (SPIE Press, 2002).

[23] F. Lekien and J. Marsden, “Tricubic interpolation in three dimensions,” In-


ternational Journal for Numerical Methods in Engineering 63, 455–471 (2005).

[24] L. A. Piegl and W. Tiller, The NURBS Book (Springer, 1997).

[25] P. W. McCarthy, “Gradient-Index Materials, Design, and Metrology for


Broadband Imaging Systems,” PhD thesis, University of Rochester, The In-
stitute of Optics, Rochester, NY (2015).
Bibliography 273

[26] D. P. Ryan-Howard and D. T. Moore, “Model for the chromatic properties of


gradient-index glass,” Applied Optics 24, 4356–4366 (1985).

[27] D. H. Lippman, R. Chou, A. X. Desai, N. S. Kochan, T. Yang, G. R. Schmidt,


J. L. Bentley, and D. T. Moore, “Polychromatic annular folded lenses using
freeform gradient-index optics,” Applied Optics 61, A1–A9 (2022).

[28] A. X. Desai, G. R. Schmidt, and D. T. Moore, “Multi-material freeform


gradient-index spectrometers,” Optics Express 30, 42912–42922 (2022).

[29] D. J. Fischer, “Gradient-Index Ophthalmic Lens Design and Polymer Ma-


terial Studies,” PhD thesis, University of Rochester, The Institute of Optics
(2002).

[30] S. D. Campbell, J. Nagar, and D. H. Werner, “Multi-element, multi-frequency


lens transformations enabled by optical wavefront matching,” Optics Express
25, 17258–17270 (2017).

[31] D. H. Lippman, G. R. Schmidt, J. L. Bentley, D. T. Moore, H. Akhavan, J. P.


Harmon, and G. M. Williams, “Gradient-index Alvarez lenses,” Applied Op-
tics 62, 3485–3495 (2023).

[32] Y. Zhu, C. Xu, Q. Mao, C. Guo, and W. Song, “Imaging stretching and dis-
placement using gradient-index elements during the lens design process,”
Optics Express 30, 47879–47895 (2022).

[33] D. H. Lippman and G. R. Schmidt, “Prescribed irradiance distributions with


freeform gradient-index optics,” Optics Express 28, 29132–29147 (2020).

[34] D. H. Lippman, R. Xu, and G. Schmidt, “Freeform gradient-index optics for


prescribed illumination,” Proc. SPIE 12220, Nonimaging Optics: Efficient De-
sign for Illumination and Solar Concentration XVIII, 1222005 (2022).

[35] W. M. Kunkel and J. R. Leger, “Gradient-index design for mode conversion


of diffracting beams,” Optics Express 24, 13480–13488 (2016).

[36] S. Maruca Donnelly, E. Jones, A. Granmoe, M. Wlodawski, T. Fenske, M. Ka-


mal, D. Dantsker, D. Marianucci, G. Fischer, C. Dupuy, H. Akhavan, S. P.
Grimm, and F. Long, “Generation and propagation of Airy beams and one
inch diameter focusing optics using 3D printed polymer optics,” Proc. SPIE
12219, Polymer Optics and Molded Glass Optics: Design, Fabrication, and Mate-
rials 2022, 122190A (2022).

[37] S. D. Campbell, D. E. Brocker, D. H. Werner, C. Dupuy, S.-K. Park, and


P. Harmon, “Three-dimensional gradient-index optics via injket-aided ad-
ditive manufacturing techniques,” in 2015 IEEE International Symposium on
Bibliography 274

Antennas and Propagation & USNC/URSI National Radio Science Meeting, 605–
606 (2015).

[38] G. M. Williams, H. Akhavan, C. Dupuy, and P. Harmon, “Additive Manufac-


turing of Freeform Optics for Defense Applications,” in 2021 IEEE Research
and Applications of Photonics in Defense Conference (RAPID), 1–2 (2021).

[39] D. M. Schut, C. G. Dupuy, and J. P. Harmon, “Inks for 3D printing gradient


refractive index (GRIN) optical components,” U.S. patent 9,447,299 (2016).

[40] R. Dylla-Spears, T. D. Yee, K. Sasan, D. T. Nguyen, N. A. Dudukovic, J. M.


Ortega, M. A. Johnson, O. D. Herrera, F. J. Ryerson, and L. L. Wong, “3D
printed gradient index glass optics,” Science Advances 6, eabc7429 (2020).

[41] S. Ji, K. Yin, M. Mackey, A. Brister, M. Ponting, and E. Baer, “Polymeric


nanolayered gradient refractive index lenses: technology review and intro-
duction of spherical gradient refractive index ball lenses,” Optical Engineer-
ing 52, 112105 (2013).

[42] A. C. Urness, K. Anderson, C. Ye, W. L. Wilson, and R. R. McLeod, “Arbitrary


GRIN component fabrication in optically driven diffusive photopolymers,”
Optics Express 23, 264–273 (2015).

[43] C. R. Ocier, C. A. Richards, D. A. Bacon-Brown, Q. Ding, R. Kumar, T. J.


Garcia, J. van de Groep, J.-H. Song, A. J. Cyphersmith, A. Rhode, A. N. Perry,
A. J. Littlefield, J. Zhu, D. Xie, H. Gao, J. F. Messinger, M. L. Brongersma, K. C.
Toussaint, L. L. Goddard, and P. V. Braun, “Direct laser writing of volumetric
gradient index lenses and waveguides,” Light: Science & Applications 9, 196
(2020).

[44] K. A. Richardson, M. Kang, L. Sisken, A. Yadav, S. Novak, A. Lepicard,


I. Martin, H. Francois-Saint-Cyr, C. M. Schwarz, T. S. Mayer, C. Rivero-
Baleine, A. J. Yee, and I. Mingareev, “Advances in infrared gradient refractive
index (GRIN) materials: a review,” Optical Engineering 59, 112602 (2020).

[45] A. J. Yee and D. T. Moore, “Free-space infrared Mach–Zehnder interferom-


eter for relative index of refraction measurement of gradient index optics,”
Optical Engineering 56, 111707 (2017).

[46] D. Korsch, Reflective Optics (Academic Press, 1991).

[47] E. J. Tremblay, R. A. Stack, R. L. Morrison, and J. E. Ford, “Ultrathin cameras


using annular folded optics,” Applied Optics 46, 463–471 (2007).

[48] E. J. Tremblay, R. A. Stack, R. L. Morrison, J. H. Karp, and J. E. Ford, “Ultra-


thin four-reflection imager,” Applied Optics 48, 343–354 (2009).
Bibliography 275

[49] P. Erdös, “Mirror Anastigmat with Two Concentric Spherical Surfaces,” Jour-
nal of the Optical Society of America 49, 877–886 (1959).

[50] C.-w. Lee, K.-w. Lee, S.-t. Jung, and D.-s. Kim, “Solid immersion mirror type
objective lens and optical pickup device adopting the same,” U.S. patent
6,801,492 B2 (2004).

[51] J. E. Webb, “Catadioptric imaging system for high numerical aperture imag-
ing with deep ultraviolet light,” U.S. patent 7,564,633 B2 (2009).

[52] M. Shenker, “High speed catadioptric objective in which three corrector ele-
ments define two power balanced air lenses,” U.S. patent 3,252,373 (1966).

[53] B. A. Cameron and G. R. Sturiale, “Solid catadioptric lens,” U.S. patent


5,793,538 (1998).

[54] T. Tsunashima, “Catadioptric lens,” U.S. patent 6,169,637 B1 (2001).

[55] Z. Maresse, “Ultra compact mono-bloc catadioptric imaging lens,” U.S.


patent 7,391,580 B2 (2008).

[56] R. K. Kerschner, “Catadioptric lens system for a scanning device,” U.S.


patent 6,639,203 B1 (2003).

[57] E. J. Tremblay, I. Stamenov, R. D. Beer, A. Arianpour, and J. E. Ford, “Switch-


able telescopic contact lens,” Optics Express 21, 15980–15986 (2013).

[58] A. Arianpour, G. M. Schuster, E. J. Tremblay, I. Stamenov, A. Groisman,


J. Legerton, W. Meyers, G. A. Amigo, and J. E. Ford, “Wearable telescopic
contact lens,” Applied Optics 54, 7195–7204 (2015).

[59] L. Li, D. Wang, C. Liu, and Q.-H. Wang, “Ultrathin zoom telescopic objec-
tive,” Optics Express 24, 18674–18684 (2016).

[60] M. Yamakawa, “Photoelectric sensor having a folded light path,” U.S. patent
4,978,843 (1990).

[61] Y. Seko, “Positional measurement system and lens for positional measure-
ment,” U.S. patent 7,554,676 B2 (2009).

[62] J. W. Goodman, Introduction to Fourier Optics (McGraw-Hill, 1996).

[63] H. Gross, F. Blechinger, and B. Achtner, Handbook of Optical Systems, Volume


4: Survey of Optical Instruments (Wiley-VCH, 2008).

[64] E. L. O’Neill, “Transfer Function for an Annular Aperture,” Journal of the


Optical Society of America 46, 285–288 (1956).
Bibliography 276

[65] E. L. O’Neill, “Errata: Transfer function for an annular aperture,” Journal of


the Optical Society of America 46, 1096 (1956).

[66] M. Born and E. Wolf, Principles of Optics: Electromagnetic Theory of Propagation,


Interference and Diffraction of Light (Cambridge University Press, 1959).

[67] M. Galan, M. Strojnik, and Y. Wang, “Design method for compact, achro-
matic, high-performance, solid catadioptric system (SoCatS), from visible to
IR,” Optics Express 27, 142–149 (2019).

[68] B. Zhang, M. Piao, and Q. Cui, “Achromatic annular folded lens with
reflective-diffractive optics,” Optics Express 27, 32337–32348 (2019).

[69] M. Piao, B. Zhang, and K. Dong, “Design of achromatic annular folded lens
with multilayer diffractive optics for visible and near-IR waveband,” Optics
Express 28, 29076–29085 (2020).

[70] D. H. Lippman, R. Chou, A. X. Desai, N. S. Kochan, T. Yang, G. R. Schmidt,


J. L. Bentley, and D. T. Moore, “Design of annular folded lenses using
freeform gradient-index optics,” Proc. SPIE 12078, International Optical De-
sign Conference 2021, 120781S (2021).

[71] Lensrentals, “Sigma 135mm f/1.8 Art MTF Charts (and a Look Behind the
Curtain),” https://www.lensrentals.com/blog/2017/04/sigma-
135mm-f1-8-art-mtf-charts-and-a-look-behind-the-curtain
(2017).

[72] CODE V documentation, Synopsys Inc., version 2022.03 (Pasadena, CA).

[73] J. M. Rodgers, “Control of packaging constraints in the optimization of unob-


scured reflective systems,” Proc. SPIE 0751, Reflective Optics, 143–149 (1987.

[74] W. Heller, “Remarks on Refractive Index Mixture Rules,” The Journal of Phys-
ical Chemistry 69, 1123–1129 (1965).

[75] P. Tripathy, P. K. Mandal, A. De, and K. R. Sahu, “Practical Feasibility of


Arago-Biot and Lorentz-Lorenz Theory Through Variation of Refractive In-
dex of Typical Binary Liquid Mixtures Employing Low-Cost Experimental
Setup,” Biointerface Research in Applied Chemistry 12, 3762–3779 (2021).

[76] P. McCarthy and D. T. Moore, “Optical Design with Gradient-Index Ele-


ments Constrained to Real Material Properties,” in OSA Imaging and Applied
Optics Technical Papers, OTu4D.2 (2012).

[77] A. X. Desai, G. R. Schmidt, and D. T. Moore, “Achromatization of multi-


material gradient-index singlets,” Optics Express 30, 40306–40314 (2022).
Bibliography 277

[78] W. T. Welford, Aberrations of Optical Systems (CRC Press, 1986).

[79] D. T. Moore, “Design of Singlets with Continuously Varying Indices of Re-


fraction,” Journal of the Optical Society of America 61, 886–894 (1971).

[80] T. Yang, D. H. Lippman, R. Y. Chou, N. S. Kochan, A. X. Desai, G. R. Schmidt,


J. L. Bentley, and D. T. Moore, “Material optimization in the design of broad-
band gradient-index optics,” Proc. SPIE 12078, International Optical Design
Conference 2021, 120780Z (2021).

[81] D. T. Moore and R. T. Salvage, “Radial gradient-index lenses with zero Petz-
val aberration,” Applied Optics 19, 1081–1086 (1980).

[82] A. J. Visconti, J. A. Corsetti, K. Fang, P. McCarthy, G. R. Schmidt, and D. T.


Moore, “Eyepiece designs with radial and spherical polymer gradient-index
optical elements,” Optical Engineering 52, 112102 (2013).

[83] D. T. Moore, “Design of single element gradient-index collimator,” Journal of


the Optical Society of America 67, 1137–1143 (1977).

[84] J. N. Mait, G. Beadie, R. A. Flynn, and P. Milojkovic, “Dispersion design


in gradient index elements using ternary blends,” Optics Express 24, 29295–
29301 (2016).

[85] M. K. Crawford, “Tolerance Analysis of Axial Gradient-index Lenses,” PhD


thesis, University of Rochester, The Institute of Optics, Rochester, NY (1999).

[86] Optimax Systems Inc., “Manufacturing Tolerance Chart,” https://www.


optimaxsi.com/optical-manufacturing-tolerance-chart/
(2023).

[87] L. W. Alvarez and W. E. Humphrey, “Variable-power lens and system,” U.S.


patent 3,507,565 (1970).

[88] L. W. Alvarez, “Development of variable-focus lenses and a new refractor,”


Journal of the American Optometric Association 49, 24–29 (1978).

[89] A. W. Lohmann, “Improvements relating to lenses and to variable optical


lens systems formed of such lenses,” G.B. patent 998,191 (1965).

[90] A. W. Lohmann, “A New Class of Varifocal Lenses,” Applied Optics 9, 1669–


1671 (1970).

[91] S. Barbero and J. Rubinstein, “Adjustable-focus lenses based on the Alvarez


principle,” Journal of Optics 13, 125705 (2011).
Bibliography 278

[92] M. Peloux and L. Berthelot, “Optimization of the optical performance of


variable-power and astigmatism Alvarez lenses,” Applied Optics 53, 6670–
6681 (2014).

[93] A. Mikš and J. Novák, “Analysis of two-element zoom systems based on


variable power lenses,” Optics Express 18, 6797–6810 (2010).

[94] A. Mikš and J. Novák, “Three-component double conjugate zoom lens sys-
tem from tunable focus lenses,” Applied Optics 52, 862–865 (2013).

[95] J. Babington, “Alvarez lens systems: theory and applications,” Proc. SPIE
9626, Optical Systems Design 2015: Optical Design and Engineering VI, 962615
(2015).

[96] Yongchao Zou, Wei Zhang, Feng Tian, Fook Siong Chau, and Guangya Zhou,
“Development of Miniature Tunable Multi-Element Alvarez Lenses,” IEEE
Journal of Selected Topics in Quantum Electronics 21, 100–107 (2015).

[97] C. Hou, Q. Xin, and Y. Zang, “Optical zoom system realized by lateral shift
of Alvarez freeform lenses,” Optical Engineering 57, 045103 (2018).

[98] C. Hou, Y. Ren, Y. Tan, Q. Xin, and Y. Zang, “Compact optical zoom camera
module based on Alvarez elements,” Optical Engineering 59, 025104 (2020).

[99] J. Schwiegerling and C. Paleta-Toxqui, “Minimal movement zoom lens,” Ap-


plied Optics 48, 1932–1935 (2009).

[100] S. S. Rege, T. S. Tkaczyk, and M. R. Descour, “Application of the Alvarez-


Humphrey concept to the design of a miniaturized scanning microscope,”
Optics Express 12, 2574–2588 (2004).

[101] E. Roth, H. Scheibe, T. Koehler, and C. Schindler, “Building Challenging Op-


tical Systems with Alvarez Lenses,” in OSA Optical Design and Fabrication
2019, FW4B.6 (2019).

[102] Y. Zou, W. Zhang, F. S. Chau, and G. Zhou, “Miniature adjustable-focus


endoscope with a solid electrically tunable lens,” Optics Express 23, 20582–
20592 (2015).

[103] Y. Zou, F. S. Chau, and G. Zhou, “Ultra-compact optical zoom endoscope


using solid tunable lenses,” Optics Express 25, 20675–20688 (2017).

[104] A. Wilson and H. Hua, “Design and demonstration of a vari-focal optical


see-through head-mounted display using freeform Alvarez lenses,” Optics
Express 27, 15627–15637 (2019).
Bibliography 279

[105] P. J. Smilie, “Design and characterization of an infrared Alvarez lens,” Optical


Engineering 51, 013006 (2012).

[106] P. J. Smilie and T. J. Suleski, “Variable-diameter refractive beam shaping with


freeform optical surfaces,” Optics Letters 36, 4170–4172 (2011).

[107] S. Shadalou, W. J. Cassarly, and T. J. Suleski, “Tunable illumination for LED-


based systems using refractive freeform arrays,” Optics Express 29, 35755–
35764 (2021).

[108] I. A. Palusinski, J. M. Sasián, and J. E. Greivenkamp, “Lateral-shift variable


aberration generators,” Applied Optics 38, 86–90 (1999).

[109] S. Barbero, “The Alvarez and Lohmann refractive lenses revisited,” Optics
Express 17, 9376–9390 (2009).

[110] A. Grewe, M. Hillenbrand, and S. Sinzinger, “Aberration analysis of opti-


mized Alvarez–Lohmann lenses,” Applied Optics 53, 7498–7506 (2014).

[111] W. J. Smith, Modern Optical Engineering (McGraw Hill, 2008).

[112] E. Acosta and J. Sasian, “Micro-Alvarez lenses for a tunable-dynamic-range


Shack–Hartmann wavefront sensor,” Japanese Journal of Applied Physics 53,
08MG04 (2014).

[113] J. A. Shultz, “Design, Tolerancing, and Experimental Characterization of Dy-


namic Freeform Optical Systems,” PhD thesis, University of North Carolina
at Charlotte, Charlotte, NC (2017).

[114] S. Shadalou and T. J. Suleski, “General design method for dynamic freeform
optics with variable functionality,” Optics Express 30, 19974–19989 (2022).

[115] B. H. Walker, Optical Design for Visual Systems (SPIE Press, 2000).

[116] N. S. Kochan, “Theory and Design of Freeform Gradient Index Optics and
Applications in Progressive Addition Lenses,” PhD thesis, University of
Rochester, The Institute of Optics, Rochester, NY (2022).

[117] J. Nocedal and S. J. Wright, Numerical optimization (Springer, 2006).

[118] M. P. Chrisp, B. Primeau, and M. A. Echter, “Imaging freeform optical sys-


tems designed with NURBS surfaces,” Optical Engineering 55, 071208 (2016).

[119] O. Cakmakci, B. Moore, H. Foroosh, and J. P. Rolland, “Optimal local shape


description for rotationally non-symmetric optical surface design and anal-
ysis,” Optics Express 16, 1583–1589 (2008).
Bibliography 280

[120] Leupold & Stevens, Inc., “Mark 6 1-6x20,” part #114337, (Beaverton, OR).

[121] Leupold & Stevens, Inc., “Mark 6 1-6×20mm (34mm) M6C1 white paper,”
https://www.leupold.com/hunting-shooting/whitepaper/
mark-6-1-6x20mm-34mm-m6c1 (2016).

[122] H. Gross, H. Zügge, M. Peschka, and F. Blechinger, Handbook of Optical Sys-


tems, Volume 3: Aberration Theory and Correction of Optical Systems (Wiley-
VCH, 2007).

[123] S. Yuan, “Aberrations of Anamorphic Optical Systems,” PhD thesis, Univer-


sity of Arizona, Tucson, AZ (2008).

[124] E. I. Betensky, “Postmodern lens design,” Optical Engineering 32, 1750–1756


(1993).

[125] S.-B. Rim, P. B. Catrysse, R. Dinyari, K. Huang, and P. Peumans, “The op-
tical advantages of curved focal plane arrays,” Optics Express 16, 4965–4971
(2008).

[126] B. Guenter, N. Joshi, R. Stoakley, A. Keefe, K. Geary, R. Freeman, J. Hundley,


P. Patterson, D. Hammon, G. Herrera, E. Sherman, A. Nowak, R. Schubert,
P. Brewer, L. Yang, R. Mott, and G. McKnight, “Highly curved image sensors:
a practical approach for improved optical performance,” Optics Express 25,
13010–13023 (2017).

[127] D. Shafer, “Lens designs with extreme image quality features,” Advanced Op-
tical Technologies 2, 53–62 (2013).

[128] I. Stamenov, A. Arianpour, S. J. Olivas, I. P. Agurok, A. R. Johnson, R. A.


Stack, R. L. Morrison, and J. E. Ford, “Panoramic monocentric imaging using
fiber-coupled focal planes,” Optics Express 22, 31708–31721 (2014).

[129] D. Reshidko and J. Sasian, “Optical analysis of miniature lenses with curved
imaging surfaces,” Applied Optics 54, E216–E223 (2015).

[130] X. Chen, D. S. Gere, and M. C. Waldon, “Small form factor high-resolution


camera,” U.S. patent 9,244,253 B2 (2016).

[131] E. Muslimov, E. Hugot, W. Jahn, S. Vives, M. Ferrari, B. Chambion, D. Henry,


and C. Gaschet, “Combining freeform optics and curved detectors for wide
field imaging: a polynomial approach over squared aperture,” Optics Express
25, 14598–14610 (2017).
Bibliography 281

[132] S. Lombardo, T. Behaghel, B. Chambion, S. Caplet, W. Jahn, E. Hugot,


E. Muslimov, M. Roulet, M. Ferrari, C. Gaschet, and D. Henry, “Curved de-
tectors for astronomical applications: characterization results on different
samples,” Applied Optics 58, 2174–2182 (2019).

[133] C. Gaschet, W. Jahn, B. Chambion, E. Hugot, T. Behaghel, S. Lombardo,


S. Lemared, M. Ferrari, S. Caplet, S. Gétin, A. Vandeneynde, and D. Henry,
“Methodology to design optical systems with curved sensors,” Applied Op-
tics 58, 973–978 (2019).

[134] B. Brixner, “A 3048-mm, f/2.5, Flat-Field Schmidt Telescope Design,” Applied


Optics 6, 1069–1072 (1967).

[135] N. Waltham, “CCD and CMOS sensors,” in Observing Photons in Space,


(Springer, 2013), pp. 423–442.

[136] P. Ruchhoeft, M. Colburn, B. Choi, H. Nounu, S. Johnson, T. Bailey, S. Damle,


M. Stewart, J. Ekerdt, S. V. Sreenivasan, J. C. Wolfe, and C. G. Willson, “Pat-
terning curved surfaces: Template generation by ion beam proximity lithog-
raphy and relief transfer by step and flash imprint lithography,” Journal of
Vacuum Science & Technology B: Microelectronics and Nanometer Structures 17,
2965–2969 (1999).

[137] H.-C. Jin, J. R. Abelson, M. K. Erhardt, and R. G. Nuzzo, “Soft lithographic


fabrication of an image sensor array on a curved substrate,” Journal of Vac-
uum Science & Technology B: Microelectronics and Nanometer Structures 22,
2548–2551 (2004).

[138] Z. Li, Y. Gu, L. Wang, H. Ge, W. Wu, Q. Xia, C. Yuan, Y. Chen, B. Cui, and
R. S. Williams, “Hybrid Nanoimprint–Soft Lithography with Sub-15 nm Res-
olution,” Nano Letters 9, 2306–2310 (2009).

[139] S. Snow and S. C. Jacobsen, “Microfabrication processes on cylindrical sub-


strates – Part II: Lithography and connections,” Microelectronic Engineering
84, 11–20 (2007).

[140] O. Iwert and B. Delabre, “The challenge of highly curved monolithic imaging
detectors,” Proc. SPIE 7742, High Energy, Optical, and Infrared Detectors for
Astronomy IV, 774227 (2010).

[141] O. Iwert, D. Ouellette, M. Lesser, and B. Delabre, “First results from a novel
curving process for large area scientific imagers,” in High Energy, Optical,
and Infrared Detectors for Astronomy V, vol. 8453 of Proc. SPIE (Amsterdam,
Netherlands, 2012), p. 84531W.
Bibliography 282

[142] J. A. Gregory, A. M. Smith, E. C. Pearce, R. L. Lambour, R. Y. Shah, H. R.


Clark, K. Warner, R. M. Osgood, D. F. Woods, A. E. DeCew, S. E. Forman,
L. Mendenhall, C. M. DeFranzo, V. S. Dolat, and A. H. Loomis, “Develop-
ment and application of spherically curved charge-coupled device imagers,”
Applied Optics 54, 3072–3082 (2015).
[143] D. Dumas, M. Fendler, N. Baier, J. Primot, and E. le Coarer, “Curved focal
plane detector array for wide field cameras,” Applied Optics 51, 5419–5424
(2012).
[144] E. Hugot, W. Jahn, B. Chambion, L. Nikitushkina, C. Gaschet, D. Henry,
S. Getin, G. Moulin, M. Ferrari, and Y. Gaeremynck, “Flexible focal plane
arrays for UVOIR wide field instrumentation,” Proc. SPIE 9915, High Energy,
Optical, and Infrared Detectors for Astronomy VII, 99151H (2016).
[145] B. Chambion, C. Gaschet, E. Hugot, S. Gétin, T. Behaghel, W. Jahn, S. Caplet,
A. Vandeneynde, D. Henry, S. Lombardo, and M. Ferrari, “Curved sensors
for compact high-resolution wide-field designs: prototype demonstration
and optical characterization,” Proc. SPIE 10539, Photonic Instrumentation En-
gineering V, 1053913 (2018).
[146] W. Jahn, M. Bailly, and G. Hein, “Wafer-level curved sensor manufacturing
process for enhanced optical system designs,” in OSA Optical Design and Fab-
rication Congress Abstract, ITu2A.4 (2021).
[147] T. W. Stone, “Optical field flatteners,” U.S. patent 7,428,106 B1 (2008).
[148] K. Chrzanowski, “Review of night vision technology,” Opto-Electronics Re-
view 21, 153–181 (2013).
[149] A. Theuwissen, “CMOS image sensors: State-Of-the-art and future perspec-
tives,” in IEEE ESSDERC 2007 - 37th European Solid State Device Research Con-
ference, 21–27 (2007).
[150] T. Matthias, G. Kreindl, V. Dragoi, M. Wimplinger, and P. Lindner, “CMOS
image sensor wafer-level packaging,” in IEEE 2011 12th International Confer-
ence on Electronic Packaging Technology and High Density Packaging, 1–6 (2011).
[151] Y. Y. Kee, K. Lin Sek, L. Zhu, Y. Hua, and X. Li, “A Novel CMOS Image Sen-
sor Package Cover Glass White Stain Material Identification Metrology by
TOF-SIMS,” in 2021 IEEE International Symposium on the Physical and Failure
Analysis of Integrated Circuits (IPFA), 1–3 (2021).
[152] J. Crowther, “Monochrome Camera Conversion: Effect on Sensitivity for
Multispectral Imaging (Ultraviolet, Visible, and Infrared),” Journal of Imag-
ing 8, 54 (2022).
Bibliography 283

[153] Lensrentals, “Sensor Stack Thickness: When Does It Matter?”


https://www.lensrentals.com/blog/2014/06/sensor-stack-
thickness-when-does-it-matter/ (2014).

[154] P. P. Clark, “Mobile platform optical design,” Proc. SPIE 9293, International
Optical Design Conference 2014, 92931M (2014).

[155] B. Ma, K. Sharma, K. P. Thompson, and J. P. Rolland, “Mobile device camera


design with Q-type polynomials to achieve higher production yield,” Optics
Express 21, 17454–17463 (2013).

[156] R. Mercado, “Camera lens system,” U.S. patent 9,874,721 B2 (2018).

[157] R. J. Koshel, Illumination Engineering: Design with Nonimaging Optics (Wiley-


IEEE Press, 2013).

[158] J. Muschaweck and H. Rehn, Designing illumination optics (SPIE Press, 2022).

[159] J. M. Stagaman and D. T. Moore, “Laser diode to fiber coupling using


anamorphic gradient-index lenses,” Applied Optics 23, 1730–1734 (1984).

[160] P. Kotsidas, V. Modi, and J. M. Gordon, “New vistas in solar concentration


with gradient-index optics,” Proc. SPIE 8124, Nonimaging Optics: Efficient De-
sign for Illumination and Solar Concentration VIII, 81240A (2011).

[161] N. Vaidya and O. Solgaard, “Immersion graded index optics: theory, design,
and prototypes,” Microsystems & Nanoengineering 8, 69 (2022).

[162] D. T. Moore, G. R. Schmidt, and M. Brown, “Light collecting and emitting


apparatus, method, and applications,” U.S. patent 9,246,038 B2 (2016).

[163] R. Wu, Z. Feng, Z. Zheng, R. Liang, P. Benı́tez, J. C. Miñano, and F. Duerr,


“Design of Freeform Illumination Optics,” Laser & Photonics Reviews 12,
1700310 (2018).

[164] R. Wu, L. Xu, P. Liu, Y. Zhang, Z. Zheng, H. Li, and X. Liu, “Freeform
illumination design: a nonlinear boundary problem for the elliptic
Monge–Ampére equation,” Optics Letters 38, 229–231 (2013).

[165] Y. Schwartzburg, R. Testuz, A. Tagliasacchi, and M. Pauly, “High-contrast


computational caustic design,” ACM Transactions on Graphics 33, 1–11 (2014).

[166] K. Brix, Y. Hafizogullari, and A. Platen, “Designing illumination lenses and


mirrors by the numerical solution of Monge–Ampère equations,” Journal of
the Optical Society of America A 32, 2227–2236 (2015).
Bibliography 284

[167] J. ten Thije Boonkkamp, C. Prins, W. IJzerman, and T. Tukker, “The Monge-
Ampère Equation for Freeform Optics,” in OSA Imaging and Applied Optics
2015, FTh3B.4 (2015).

[168] C. Bösel and H. Gross, “Single freeform surface design for prescribed input
wavefront and target irradiance,” Journal of the Optical Society of America A
34, 1490–1499 (2017).

[169] M. D. Himel, R. E. Hutchins, J. C. Colvin, M. K. Poutous, A. D. Kathman,


and A. S. Fedor, “Design and fabrication of customized illumination patterns
for low-k1 lithography: a diffractive approach,” Proc. SPIE 4346, Optical Mi-
crolithography XIV, 1436–1442 (2001).

[170] G. M. Morris and T. R. M. Sales, “Structured screens for controlled spreading


of light,” U.S. patent 7,033,736 (2006).

[171] J. S. Schruben, “Formulation of a Reflector-Design Problem for a Lighting


Fixture,” Journal of the Optical Society of America 62, 1498–1501 (1972).

[172] H. Ries and J. Muschaweck, “Tailored freeform optical surfaces,” JOSA A 19,
590–595 (2002).

[173] L. B. Romijn, J. H. M. ten Thije Boonkkamp, and W. L. IJzerman, “Freeform


lens design for a point source and far-field target,” Journal of the Optical Soci-
ety of America A 36, 1926–1939 (2019).

[174] W. Pohl, C. Anselm, C. Knoflach, A. L. Timinger, J. A. Muschaweck, and


H. Ries, “Complex 3D-tailored facets for optimal lighting of facades and
public places,” Proc. SPIE 5186, Design of Efficient Illumination Systems, 133–
142 (2003).

[175] C. Bösel and H. Gross, “Ray mapping approach for the efficient design of
continuous freeform surfaces,” Optics Express 24, 14271–14282 (2016).

[176] Z. Feng, B. D. Froese, and R. Liang, “Freeform illumination optics con-


struction following an optimal transport map,” Applied Optics 55, 4301–4306
(2016).

[177] C. Gannon and R. Liang, “Ray mapping with surface information for
freeform illumination design,” Optics Express 25, 9426–9434 (2017).

[178] L. L. Doskolovich, D. A. Bykov, A. A. Mingazov, and E. A. Bezus, “Opti-


mal mass transportation and linear assignment problems in the design of
freeform refractive optical elements generating far-field irradiance distribu-
tions,” Optics Express 27, 13083–13097 (2019).
Bibliography 285

[179] D. A. Bykov, L. L. Doskolovich, A. A. Mingazov, and E. A. Bezus, “Optimal


mass transportation problem in the design of freeform optical elements gen-
erating far-field irradiance distributions for plane incident beam,” Applied
Optics 58, 9131–9140 (2019).

[180] V. Oliker, “Mathematical Aspects of Design of Beam Shaping Surfaces in Ge-


ometrical Optics,” in Trends in Nonlinear Analysis, (Springer, 2003), pp. 193–
224.

[181] F. R. Fournier, W. J. Cassarly, and J. P. Rolland, “Fast freeform reflector gen-


eration using source-target maps,” Optics Express 18, 5295–5304 (2010).

[182] L. L. Doskolovich, K. V. Borisova, M. A. Moiseev, and N. L. Kazanskiy, “De-


sign of mirrors for generating prescribed continuous illuminance distribu-
tions on the basis of the supporting quadric method,” Applied Optics 55, 687–
695 (2016).

[183] V. Oliker, “Controlling light with freeform multifocal lens designed with
supporting quadric method (SQM),” Optics Express 25, A58–A72 (2017).

[184] D. A. Bykov, L. L. Doskolovich, E. V. Byzov, E. A. Bezus, and N. L. Kazanskiy,


“Supporting quadric method for designing refractive optical elements gen-
erating prescribed irradiance distributions and wavefronts,” Optics Express
29, 26304–26318 (2021).

[185] I. Moreno, C.-C. Sun, and R. Ivanov, “Far-field condition for light-emitting
diode arrays,” Applied Optics 48, 1190–1197 (2009).

[186] C. Bösel and H. Gross, “Compact freeform illumination system design for
pattern generation with extended light sources,” Applied Optics 58, 2713–
2724 (2019).

[187] F. R. Fournier, W. J. Cassarly, and J. P. Rolland, “Designing freeform reflectors


for extended sources,” Proc. SPIE 7423, Nonimaging Optics: Efficient Design for
Illumination and Solar Concentration VI, 742302 (2009).

[188] R. Wester, G. Müller, A. Völl, M. Berens, J. Stollenwerk, and P. Loosen, “De-


signing optical free-form surfaces for extended sources,” Optics Express 22,
A552–A560 (2014).

[189] P. Benı́tez, “Simultaneous multiple surface optical design method in three


dimensions,” Optical Engineering 43, 1489–1502 (2004).

[190] S. Sorgato, J. Chaves, H. Thienpont, and F. Duerr, “Design of illumination


optics with extended sources based on wavefront tailoring,” Optica 6, 966–
971 (2019).
Bibliography 286

[191] B. G. Assefa, T. Saastamoinen, M. Pekkarinen, V. Nissinen, J. Biskop, M. Kuit-


tinen, J. Turunen, and J. Saarinen, “Realizing freeform lenses using an optics
3D-printer for industrial based tailored irradiance distribution,” OSA Con-
tinuum 2, 690–702 (2019).

[192] R. Wu, P. Benı́tez, Y. Zhang, and J. C. Miñano, “Influence of the characteris-


tics of a light source and target on the Monge–Ampére equation method in
freeform optics design,” Optics Letters 39, 634–637 (2014).

[193] W. M. Kunkel and J. R. Leger, “Numerical design of three-dimensional gra-


dient refractive index structures for beam shaping,” Optics Express 28, 32061–
32076 (2020).

[194] LightTools documentation, Synopsys Inc., version 2022.03 (Pasadena, CA).

[195] D. T. Moore, “Ray tracing in gradient-index media,” Journal of the Optical


Society of America 65, 451–455 (1975).

[196] L. J. Hornbeck, “Spatial light modulator and method,” U.S. patent 5,061,049
(1991).

[197] H. W. Kuhn, “The Hungarian method for the assignment problem,” Naval
Research Logistics Quarterly 2, 83–97 (1955).

[198] J. Munkres, “Algorithms for the Assignment and Transportation Problems,”


Journal of the Society for Industrial and Applied Mathematics 5, 32–38 (1957).

[199] D. A. Bykov, L. L. Doskolovich, A. A. Mingazov, E. A. Bezus, and N. L.


Kazanskiy, “Linear assignment problem in the design of freeform refractive
optical elements generating prescribed irradiance distributions,” Optics Ex-
press 26, 27812–27825 (2018).

[200] H. W. Kuhn, “Variants of the hungarian method for assignment problems,”


Naval Research Logistics Quarterly 3, 253–258 (1956).

[201] D. F. Crouse, “On implementing 2D rectangular assignment algorithms,”


IEEE Transactions on Aerospace and Electronic Systems 52, 1679–1696 (2016).

[202] Scipy python library, optimization: linear sum assignment, version 1.10.0
(2023).

[203] E. Balas, D. Miller, J. Pekny, and P. Toth, “A parallel shortest augmenting


path algorithm for the assignment problem,” Journal of the ACM 38, 985–1004
(1991).

[204] D. Malacara, Optical Shop Testing (Wiley-Interscience, 2007).


Bibliography 287

[205] K. T. Knox, “Image retrieval from astronomical speckle patterns*,” Journal of


the Optical Society of America 66, 1236–1239 (1976).
[206] W. Southwell, “Wave-front estimation from wave-front slope measure-
ments,” Journal of the Optical Society of America 70, 998–1006 (1980).
[207] X. Li and W. Jiang, “Comparing zonal reconstruction algorithms and modal
reconstruction algorithms in adaptive optics system,” Proc. SPIE 4825, High-
Resolution Wavefront Control: Methods, Devices, and Applications IV, 121–130
(2002).
[208] R. Keys, “Cubic convolution interpolation for digital image processing,”
IEEE Transactions on Acoustics, Speech, and Signal Processing 29, 1153–1160
(1981).
[209] G. W. Forbes, “Never-ending struggles with mid-spatial frequencies,” Proc.
SPIE 9525, Optical Measurement Systems for Industrial Inspection IX, 95251B
(2015).
[210] D. W. Sweeney and C. M. Vest, “Reconstruction of Three-Dimensional Re-
fractive Index Fields from Multidirectional Interferometric Data,” Applied
Optics 12, 2649–2664 (1973).
[211] H. Ohno and K. Toya, “Localized gradient-index field reconstruction using
background-oriented schlieren,” Applied Optics 58, 7795–7804 (2019).
[212] D. Lin and J. R. Leger, “Retrieval of refractive index fields in two-
dimensional gradient-index elements from external deflectometry data,”
Journal of the Optical Society of America A 33, 396–403 (2016).
[213] S. Chowdhury, W. J. Eldridge, A. Wax, and J. Izatt, “Refractive index tomog-
raphy with structured illumination,” Optica 4, 537–545 (2017).
[214] S. Chowdhury, M. Chen, R. Eckert, D. Ren, F. Wu, N. Repina, and L. Waller,
“High-resolution 3D refractive index microscopy of multiple-scattering sam-
ples from intensity images,” Optica 6, 1211–1219 (2019).
[215] A. F. Fercher, W. Drexler, C. K. Hitzenberger, and T. Lasser, “Optical coher-
ence tomography - principles and applications,” Reports on Progress in Physics
66, 239–303 (2003).
[216] D. Fleischmann and F. E. Boas, “Computed tomography—old ideas and new
technology,” European Radiology 21, 510–517 (2011).
[217] S. C. Mayo, A. W. Stevenson, and S. W. Wilkins, “In-Line Phase-Contrast
X-ray Imaging and Tomography for Materials Science,” Materials 5, 937–965
(2012).
Bibliography 288

[218] A. Hormati, I. Jovanović, O. Roy, and M. Vetterli, “Robust ultrasound travel-


time tomography using the bent ray model,” Proc. SPIE 7629, Medical Imaging
2010: Ultrasonic Imaging, Tomography, and Therapy, 76290I (2010).

[219] R. Ali, S. Hsieh, and J. Dahl, “Open-source Gauss-Newton-based meth-


ods for refraction-corrected ultrasound computed tomography,” Proc. SPIE
10955, Medical Imaging 2019: Ultrasonic Imaging and Tomography, 1095508
(2019).

[220] K.-H. Brenner and W. Singer, “Light propagation through microlenses: a


new simulation method,” Applied Optics 32, 4984–4988 (1993).

[221] X. Ma, W. Xiao, and F. Pan, “Optical tomographic reconstruction based


on multi-slice wave propagation method,” Optics Express 25, 22595–22607
(2017).

[222] D. Suski, J. Winnik, and T. Kozacki, “Fast multiple-scattering holographic to-


mography based on the wave propagation method,” Applied Optics 59, 1397–
1403 (2020).

[223] G. N. Lawrence and S.-H. Hwang, “Beam propagation in gradient refractive-


index media,” Applied Optics 31, 5201–5210 (1992).

[224] D. M. Topa, “Wavefront reconstruction for the Shack-Hartmann wavefront


sensor,” Proc. SPIE 4769, Optical Design and Analysis Software II, 101–115
(2002).

[225] G. W. Forbes, “Optical system assessment for design: numerical ray tracing
in the Gaussian pupil,” Journal of the Optical Society of America A 5, 1943–1956
(1988).

[226] N. Takaki, G. Forbes, and J. P. Rolland, “Schemes for cubature over the unit
disk found via numerical optimization,” Journal of Computational and Applied
Mathematics 407, 114076 (2022).

[227] J. Lim, K. Lee, K. H. Jin, S. Shin, S. Lee, Y. Park, and J. C. Ye, “Comparative
study of iterative reconstruction algorithms for missing cone problems in
optical diffraction tomography,” Optics Express 23, 16933–16948 (2015).
Appendix A

CODE V User-Defined
Gradient-Index Source Code

A.1 “University of Rochester” linear, two-material composition

The following C++ source code compiled as a dynamic-link library (.dll) is used
for the CODE V user-defined GRIN in the design of GRIN AFLs in Chapter 2. The
code is written by Tianyi Yang.

/* Build directives */
/* DLLMAIN
*/
/*
##################################################################
##################################################################
OPTICAL RESEARCH ASSOCIATES
PROPRIETARY SOFTWARE NOTIFICATION - (C) OPTICAL RESEARCH
ASSOCIATES.
UNPUBLISHED-RIGHTS RESERVED UNDER THE COPYRIGHT LAWS OF THE
UNITED STATES.
THIS SOFTWARE IS THE PROPRIETARY AND CONFIDENTIAL PROPERTY OF
OPTICAL RESEARCH ASSOCIATES AND CANNOT BE MODIFIED, DUPLICATED,
OR COPIED IN ANY FORM WITHOUT PRIOR WRITTEN CONSENT OF OPTICAL
RESEARCH ASSOCIATES.
ORA, CODEV, AND LIGHTTOOLS ARE REGISTERED TRADEMARKS OF
OPTICAL RESEARCH ASSOCIATES, PASADENA CALIFORNIA
##################################################################
##################################################################

289
Appendix A. CODE V User-Defined Gradient-Index Source Code 290

*/
/*
* Purpose: Evaluates n and n*GRAD(n) of a user-defined index of
* refraction gradient. The function is of the form
* n(x,y,z), where (x,y,z) are the cartesian coordinates
* of a point in the gradient. The origin of this
* coordinate system is the vertex of the surface to which
* the gradient is attached. GRAD(n) is the three
* derivatives of the function n, viz.,
* (dn/dx, dn/dy, dn/dz).
*
* Parameters:
* The following is a brief description of the parameters in the
* call list. If the parameter is designated as "input", its
* value is passed to the subroutine by the calling program; if
* it is designated as "output", its value is supposed to be
* calculated or set by this subroutine and passed back to the
* calling program.
*
* KERROR - Error code (output): This is set to zero before this
* subroutine is entered. If there are any error
* conditions which prevent normal completion of
* the calculation (eg., negative SQRT, etc.),
* KERROR should be set to a non-zero value.
*
* BRIND - The base index (input): This is entered in the
* private catalog as the index of refraction.
*
* COEF - The gradient coefficients (input): these are entered
* in the private catalog as the coefficients. For
* example, COEF(1) is entered with the command UDG C1,
* COEF(2) is entered with the command UDG C2, etc. some
of
* the coef might be redefined for constrain purpose.
*
* S - The position vector (input): S is an array
* containing the three components, (X,Y,Z), of the
* position of a ray as it is traced; Z is the optical
* axis.
*
* RINDX - The calculated index of refraction (output): This is
* the calculated value of the index of refraction at
* (X,Y,Z) using the equation programmed in this
* subroutine.
*
* XNGRAN - The calculated n*GRAD(n) (output): XNGRAN is an
* array containing the three components of the
* calculated values of n*GRAD(n) at (X,Y,Z) using
* the equation programmed in this subroutine. Since
* n*GRAD(n) is equivalent to 0.5*GRAD(n**2), either
* form can be used, depending on which is more
* convenient.
################################################################
* Modified from Yang Zhao’s monomial profile into
orthogonal basis
Appendix A. CODE V User-Defined Gradient-Index Source Code 291

* revised by Tianyi Yang 02/15/2018


*/
#include <math.h>
#if _MSC_VER >= 1000
# define DLLEXPORT __declspec(dllexport)
#else
# define DLLEXPORT
#endif

void cv_udg_maturgrin2(long *kerror,


double *brind,
double *coef,
double *s,
double *rindx,
double *xngran)
{
/* Local variables */
double r2, z, nd1, nd2, var, refractiveIndex, temp, nomrad,
nomz;

/*
* Note: The following statements defining the type of the
* variables in the call list must be left in the subroutine
*
* NOTE: The index of refraction is always calculated as a
* positive quantity regardless of the direction of the
* light. The Z coordinate is positive in the direction of
* the material regardless of the direction of the light.
*

/* Parameter adjustments */
--xngran;
--s;
--coef;

/* function body */
//////////////////////////////////////////////////
// Whoever want to edit r2, please notice this is PUPIL
COORDINATES!
// s[3] is also normalized
//////////////////////////////////////////////////
nomrad = coef[53]; // normalization radius
nomz = coef[54]; // normalization thickness
r2 = (s[1]*s[1] + s[2]*s[2])/(nomrad*nomrad);
z = s[3]/nomz;

nd1 = coef[51]; // base index of material 1


nd2 = coef[52]; // base index of material 2
Appendix A. CODE V User-Defined Gradient-Index Source Code 292

// Calculate index of two other wavelengths using partial and


dispersion
// This is how Synopsys optimize UR GRIN notation. From
LTUDGRIN.vb in lightools
var = z * (coef[1] + z * (coef[2] + z * (coef[3] + z *
coef[4])));
var += r2 * (coef[10] + r2 * (coef[20] + r2 * (coef[30] + r2 *
coef[40]))); // spatial variational part of the profile
refractiveIndex = nd1 + (nd2 - nd1) * var;
/* Index of refraction */
*rindx = refractiveIndex;
/* n*GRAD(n) */
temp = (nd2 - nd1) * (coef[10] + r2 * (2 * coef[20] + r2 * (3
* coef[30] + r2 * 4 * coef[40])));
xngran[1] = 2 * s[1] * temp/(nomrad*nomrad) * refractiveIndex;
// notice the additional normalization
xngran[2] = 2 * s[2] * temp/(nomrad*nomrad) * refractiveIndex;
temp = (nd2 - nd1) * (coef[1] + z * (2 * coef[2] + z * (3 *
coef[3] + z * 4 * coef[4])));
xngran[3] = temp/nomz * refractiveIndex;
}
/*
* Return GRIN UMR type for this routine
*/
DLLEXPORT void cvumrtype(char *umrType)
{
strcpy(umrType, "GRN");
}
/*
* Return a value that indicates that cv_udg_cross is threadsafe
*/
DLLEXPORT int cvisthreadsafe()
{
return 1;
}
Appendix A. CODE V User-Defined Gradient-Index Source Code 293

A.2 XYZ polynomial

The following C++ source code compiled as a dynamic-link library (.dll) is used
for the CODE V user-defined GRIN in the design of the GRIN Alvarez lenses and
other F-GRIN elements in Chapter 3.

/*
##################################################################
##################################################################
OPTICAL RESEARCH ASSOCIATES
PROPRIETARY SOFTWARE NOTIFICATION - (C) OPTICAL RESEARCH
ASSOCIATES.
UNPUBLISHED-RIGHTS RESERVED UNDER THE COPYRIGHT LAWS OF THE
UNITED STATES.
THIS SOFTWARE IS THE PROPRIETARY AND CONFIDENTIAL PROPERTY OF
OPTICAL RESEARCH ASSOCIATES AND CANNOT BE MODIFIED, DUPLICATED,
OR COPIED IN ANY FORM WITHOUT PRIOR WRITTEN CONSENT OF OPTICAL
RESEARCH ASSOCIATES.
ORA, CODEV, AND LIGHTTOOLS ARE REGISTERED TRADEMARKS OF
OPTICAL RESEARCH ASSOCIATES, PASADENA CALIFORNIA
##################################################################
##################################################################
*/
/*
* Purpose: Evaluates n and n*GRAD(n) of a user-defined index of
* refraction gradient. The function is of the form
* n(x,y,z), where (x,y,z) are the cartesian coordinates
* of a point in the gradient. The origin of this
* coordinate system is the vertex of the surface to which
* the gradient is attached. GRAD(n) is the three
* derivatives of the function n, viz.,
* (dn/dx, dn/dy, dn/dz).
*
* Parameters:
* The following is a brief description of the parameters in the
* call list. If the parameter is designated as "input", its
* value is passed to the subroutine by the calling program; if
* it is designated as "output", its value is supposed to be
* calculated or set by this subroutine and passed back to the
* calling program.
*
* KERROR - Error code (output): This is set to zero before this
* subroutine is entered. If there are any error
* conditions which prevent normal completion of
* the calculation (eg., negative SQRT, etc.),
* KERROR should be set to a non-zero value.
Appendix A. CODE V User-Defined Gradient-Index Source Code 294

*
* BRIND - The base index (input): This is entered in the
* private catalog as the index of refraction.
*
* COEF - The gradient coefficients (input): these are entered
* in the private catalog as the coefficients. For
* example, COEF(1) is entered with the command UDG C1,
* COEF(2) is entered with the command UDG C2, etc. some
of
* the coef might be redefined for constrain purpose.
*
* S - The position vector (input): S is an array
* containing the three components, (X,Y,Z), of the
* position of a ray as it is traced; Z is the optical
* axis.
*
* RINDX - The calculated index of refraction (output): This is
* the calculated value of the index of refraction at
* (X,Y,Z) using the equation programmed in this
* subroutine.
*
* XNGRAN - The calculated n*GRAD(n) (output): XNGRAN is an
* array containing the three components of the
* calculated values of n*GRAD(n) at (X,Y,Z) using
* the equation programmed in this subroutine. Since
* n*GRAD(n) is equivalent to 0.5*GRAD(n**2), either
* form can be used, depending on which is more
* convenient.
################################################################
* Modified from Yang Zhao’s monomial profile into
orthogonal basis
* revised by Tianyi Yang 02/15/2018
################################################################
* Adapted for TrueFreeformGrin by Tianyi Yang 08/06/2018
*
*/

void cv_udg_xyzpoly(long *kerror,


double *brind,
double *coef,
double *s,
double *rindx,
double *xngran)
{
/* Local variables */
double xp, yp, zp;
double xp_dx, yp_dy, zp_dz;
double x0, x1, x2, x3, x4, x5, x6, x7;
double x1_dx, x2_dx, x3_dx, x4_dx, x5_dx, x6_dx, x7_dx;
double y0, y1, y2, y3, y4, y5, y6, y7;
double y1_dy, y2_dy, y3_dy, y4_dy, y5_dy, y6_dy, y7_dy;
double z0, z1, z2, z3, z4, z5, z6, z7;
double z1_dz, z2_dz, z3_dz, z4_dz, z5_dz, z6_dz, z7_dz;
double dn_dx, dn_dy, dn_dz;
Appendix A. CODE V User-Defined Gradient-Index Source Code 295

/*
* Note: The following statements defining the type of the
* variables in the call list must be left in the subroutine
*
* NOTE: The index of refraction is always calculated as a
* positive quantity regardless of the direction of the
* light. The Z coordinate is positive in the direction of
* the material regardless of the direction of the light.
*
* NOTE: From this point on, you will typically substitute
* your C code for the particular gradient being
* programmed
*/
/* Parameter adjustments */
--xngran;
--s;
--coef;
/*
* Coefficient definitions
*
* coef[1] = base refractive index, n0
* coef[2] = spatial scaling coefficient, dn
* coef[3] = normalization half-width in x (lens units)
* coef[4] = normalization half-width in y (lens units)
* coef[5] = normalization thickness (lens units)
* coef[6] = origin shift in x from optical axis (lens units)
* coef[7] = origin shift in y from optical axis (lens units)
* coef[8] = origin shift in z from midpoint of normalization
thickness (lens units)
* coef[9] through coef[128] are spatial term coefficients
*
*/
// Define change in coordinates
xp = (s[1] - coef[6]) / coef[3];
yp = (s[2] - coef[7]) / coef[4];
zp = 2 * (s[3] - coef[8]) / coef[5] - 1;
// Define reused polynomial terms
x0 = 1.;
x1 = x0 * xp;
x2 = x1 * xp;
x3 = x2 * xp;
x4 = x3 * xp;
x5 = x4 * xp;
x6 = x5 * xp;
x7 = x6 * xp;
y0 = 1.;
y1 = y0 * yp;
y2 = y1 * yp;
y3 = y2 * yp;
Appendix A. CODE V User-Defined Gradient-Index Source Code 296

y4 = y3 * yp;
y5 = y4 * yp;
y6 = y5 * yp;
y7 = y6 * yp;
z0 = 1.;
z1 = z0 * zp;
z2 = z1 * zp;
z3 = z2 * zp;
z4 = z3 * zp;
z5 = z4 * zp;
z6 = z5 * zp;
z7 = z6 * zp;
// Define reused polynomial derivative terms
xp_dx = 1. / coef[3];
yp_dy = 1. / coef[4];
zp_dz = 2. / coef[5];
x1_dx = 1. * x0 * xp_dx;
x2_dx = 2. * x1 * xp_dx;
x3_dx = 3. * x2 * xp_dx;
x4_dx = 4. * x3 * xp_dx;
x5_dx = 5. * x4 * xp_dx;
x6_dx = 6. * x5 * xp_dx;
x7_dx = 7. * x6 * xp_dx;
y1_dy = 1. * y0 * yp_dy;
y2_dy = 2. * y1 * yp_dy;
y3_dy = 3. * y2 * yp_dy;
y4_dy = 4. * y3 * yp_dy;
y5_dy = 5. * y4 * yp_dy;
y6_dy = 6. * y5 * yp_dy;
y7_dy = 7. * y6 * yp_dy;
z1_dz = 1. * z0 * zp_dz;
z2_dz = 2. * z1 * zp_dz;
z3_dz = 3. * z2 * zp_dz;
z4_dz = 4. * z3 * zp_dz;
z5_dz = 5. * z4 * zp_dz;
z6_dz = 6. * z5 * zp_dz;
z7_dz = 7. * z6 * zp_dz;
// Calculate refractive index from xyz polynomial terms with
coefficients

*rindx = coef[1] + coef[2] * (coef[9] * x0 * y0 * z0 +


coef[10] * x1 * y0 * z0 +
coef[11] * x0 * y1 * z0 +
coef[12] * x0 * y0 * z1 +
coef[13] * x2 * y0 * z0 +
coef[14] * x1 * y1 * z0 +
coef[15] * x1 * y0 * z1 +
coef[16] * x0 * y2 * z0 +
coef[17] * x0 * y1 * z1 +
Appendix A. CODE V User-Defined Gradient-Index Source Code 297

coef[18] * x0 * y0 * z2 +
coef[19] * x3 * y0 * z0 +
coef[20] * x2 * y1 * z0 +
coef[21] * x2 * y0 * z1 +
coef[22] * x1 * y2 * z0 +
coef[23] * x1 * y1 * z1 +
coef[24] * x1 * y0 * z2 +
coef[25] * x0 * y3 * z0 +
coef[26] * x0 * y2 * z1 +
coef[27] * x0 * y1 * z2 +
coef[28] * x0 * y0 * z3 +
coef[29] * x4 * y0 * z0 +
coef[30] * x3 * y1 * z0 +
coef[31] * x3 * y0 * z1 +
coef[32] * x2 * y2 * z0 +
coef[33] * x2 * y1 * z1 +
coef[34] * x2 * y0 * z2 +
coef[35] * x1 * y3 * z0 +
coef[36] * x1 * y2 * z1 +
coef[37] * x1 * y1 * z2 +
coef[38] * x1 * y0 * z3 +
coef[39] * x0 * y4 * z0 +
coef[40] * x0 * y3 * z1 +
coef[41] * x0 * y2 * z2 +
coef[42] * x0 * y1 * z3 +
coef[43] * x0 * y0 * z4 +
coef[44] * x5 * y0 * z0 +
coef[45] * x4 * y1 * z0 +
coef[46] * x4 * y0 * z1 +
coef[47] * x3 * y2 * z0 +
coef[48] * x3 * y1 * z1 +
coef[49] * x3 * y0 * z2 +
coef[50] * x2 * y3 * z0 +
coef[51] * x2 * y2 * z1 +
coef[52] * x2 * y1 * z2 +
coef[53] * x2 * y0 * z3 +
coef[54] * x1 * y4 * z0 +
coef[55] * x1 * y3 * z1 +
coef[56] * x1 * y2 * z2 +
coef[57] * x1 * y1 * z3 +
coef[58] * x1 * y0 * z4 +
coef[59] * x0 * y5 * z0 +
coef[60] * x0 * y4 * z1 +
coef[61] * x0 * y3 * z2 +
coef[62] * x0 * y2 * z3 +
coef[63] * x0 * y1 * z4 +
coef[64] * x0 * y0 * z5 +
coef[65] * x6 * y0 * z0 +
coef[66] * x5 * y1 * z0 +
coef[67] * x5 * y0 * z1 +
coef[68] * x4 * y2 * z0 +
coef[69] * x4 * y1 * z1 +
coef[70] * x4 * y0 * z2 +
coef[71] * x3 * y3 * z0 +
coef[72] * x3 * y2 * z1 +
coef[73] * x3 * y1 * z2 +
Appendix A. CODE V User-Defined Gradient-Index Source Code 298

coef[74] * x3 * y0 * z3 +
coef[75] * x2 * y4 * z0 +
coef[76] * x2 * y3 * z1 +
coef[77] * x2 * y2 * z2 +
coef[78] * x2 * y1 * z3 +
coef[79] * x2 * y0 * z4 +
coef[80] * x1 * y5 * z0 +
coef[81] * x1 * y4 * z1 +
coef[82] * x1 * y3 * z2 +
coef[83] * x1 * y2 * z3 +
coef[84] * x1 * y1 * z4 +
coef[85] * x1 * y0 * z5 +
coef[86] * x0 * y6 * z0 +
coef[87] * x0 * y5 * z1 +
coef[88] * x0 * y4 * z2 +
coef[89] * x0 * y3 * z3 +
coef[90] * x0 * y2 * z4 +
coef[91] * x0 * y1 * z5 +
coef[92] * x0 * y0 * z6 +
coef[93] * x7 * y0 * z0 +
coef[94] * x6 * y1 * z0 +
coef[95] * x6 * y0 * z1 +
coef[96] * x5 * y2 * z0 +
coef[97] * x5 * y1 * z1 +
coef[98] * x5 * y0 * z2 +
coef[99] * x4 * y3 * z0 +
coef[100] * x4 * y2 * z1 +
coef[101] * x4 * y1 * z2 +
coef[102] * x4 * y0 * z3 +
coef[103] * x3 * y4 * z0 +
coef[104] * x3 * y3 * z1 +
coef[105] * x3 * y2 * z2 +
coef[106] * x3 * y1 * z3 +
coef[107] * x3 * y0 * z4 +
coef[108] * x2 * y5 * z0 +
coef[109] * x2 * y4 * z1 +
coef[110] * x2 * y3 * z2 +
coef[111] * x2 * y2 * z3 +
coef[112] * x2 * y1 * z4 +
coef[113] * x2 * y0 * z5 +
coef[114] * x1 * y6 * z0 +
coef[115] * x1 * y5 * z1 +
coef[116] * x1 * y4 * z2 +
coef[117] * x1 * y3 * z3 +
coef[118] * x1 * y2 * z4 +
coef[119] * x1 * y1 * z5 +
coef[120] * x1 * y0 * z6 +
coef[121] * x0 * y7 * z0 +
coef[122] * x0 * y6 * z1 +
coef[123] * x0 * y5 * z2 +
coef[124] * x0 * y4 * z3 +
coef[125] * x0 * y3 * z4 +
coef[126] * x0 * y2 * z5 +
coef[127] * x0 * y1 * z6 +
coef[128] * x0 * y0 * z7);
Appendix A. CODE V User-Defined Gradient-Index Source Code 299

// Calculate refractive index gradient from xyz polynomial


terms with coefficients
dn_dx = coef[2] * (coef[10] * x1_dx * y0 * z0 +
coef[13] * x2_dx * y0 * z0 +
coef[14] * x1_dx * y1 * z0 +
coef[15] * x1_dx * y0 * z1 +
coef[19] * x3_dx * y0 * z0 +
coef[20] * x2_dx * y1 * z0 +
coef[21] * x2_dx * y0 * z1 +
coef[22] * x1_dx * y2 * z0 +
coef[23] * x1_dx * y1 * z1 +
coef[24] * x1_dx * y0 * z2 +
coef[29] * x4_dx * y0 * z0 +
coef[30] * x3_dx * y1 * z0 +
coef[31] * x3_dx * y0 * z1 +
coef[32] * x2_dx * y2 * z0 +
coef[33] * x2_dx * y1 * z1 +
coef[34] * x2_dx * y0 * z2 +
coef[35] * x1_dx * y3 * z0 +
coef[36] * x1_dx * y2 * z1 +
coef[37] * x1_dx * y1 * z2 +
coef[38] * x1_dx * y0 * z3 +
coef[44] * x5_dx * y0 * z0 +
coef[45] * x4_dx * y1 * z0 +
coef[46] * x4_dx * y0 * z1 +
coef[47] * x3_dx * y2 * z0 +
coef[48] * x3_dx * y1 * z1 +
coef[49] * x3_dx * y0 * z2 +
coef[50] * x2_dx * y3 * z0 +
coef[51] * x2_dx * y2 * z1 +
coef[52] * x2_dx * y1 * z2 +
coef[53] * x2_dx * y0 * z3 +
coef[54] * x1_dx * y4 * z0 +
coef[55] * x1_dx * y3 * z1 +
coef[56] * x1_dx * y2 * z2 +
coef[57] * x1_dx * y1 * z3 +
coef[58] * x1_dx * y0 * z4 +
coef[65] * x6_dx * y0 * z0 +
coef[66] * x5_dx * y1 * z0 +
coef[67] * x5_dx * y0 * z1 +
coef[68] * x4_dx * y2 * z0 +
coef[69] * x4_dx * y1 * z1 +
coef[70] * x4_dx * y0 * z2 +
coef[71] * x3_dx * y3 * z0 +
coef[72] * x3_dx * y2 * z1 +
coef[73] * x3_dx * y1 * z2 +
coef[74] * x3_dx * y0 * z3 +
coef[75] * x2_dx * y4 * z0 +
coef[76] * x2_dx * y3 * z1 +
coef[77] * x2_dx * y2 * z2 +
coef[78] * x2_dx * y1 * z3 +
coef[79] * x2_dx * y0 * z4 +
coef[80] * x1_dx * y5 * z0 +
coef[81] * x1_dx * y4 * z1 +
coef[82] * x1_dx * y3 * z2 +
Appendix A. CODE V User-Defined Gradient-Index Source Code 300

coef[83] * x1_dx * y2 * z3 +
coef[84] * x1_dx * y1 * z4 +
coef[85] * x1_dx * y0 * z5 +
coef[93] * x7_dx * y0 * z0 +
coef[94] * x6_dx * y1 * z0 +
coef[95] * x6_dx * y0 * z1 +
coef[96] * x5_dx * y2 * z0 +
coef[97] * x5_dx * y1 * z1 +
coef[98] * x5_dx * y0 * z2 +
coef[99] * x4_dx * y3 * z0 +
coef[100] * x4_dx * y2 * z1 +
coef[101] * x4_dx * y1 * z2 +
coef[102] * x4_dx * y0 * z3 +
coef[103] * x3_dx * y4 * z0 +
coef[104] * x3_dx * y3 * z1 +
coef[105] * x3_dx * y2 * z2 +
coef[106] * x3_dx * y1 * z3 +
coef[107] * x3_dx * y0 * z4 +
coef[108] * x2_dx * y5 * z0 +
coef[109] * x2_dx * y4 * z1 +
coef[110] * x2_dx * y3 * z2 +
coef[111] * x2_dx * y2 * z3 +
coef[112] * x2_dx * y1 * z4 +
coef[113] * x2_dx * y0 * z5 +
coef[114] * x1_dx * y6 * z0 +
coef[115] * x1_dx * y5 * z1 +
coef[116] * x1_dx * y4 * z2 +
coef[117] * x1_dx * y3 * z3 +
coef[118] * x1_dx * y2 * z4 +
coef[119] * x1_dx * y1 * z5 +
coef[120] * x1_dx * y0 * z6);
dn_dy = coef[2] * (coef[11] * x0 * y1_dy * z0 +
coef[14] * x1 * y1_dy * z0 +
coef[16] * x0 * y2_dy * z0 +
coef[17] * x0 * y1_dy * z1 +
coef[20] * x2 * y1_dy * z0 +
coef[22] * x1 * y2_dy * z0 +
coef[23] * x1 * y1_dy * z1 +
coef[25] * x0 * y3_dy * z0 +
coef[26] * x0 * y2_dy * z1 +
coef[27] * x0 * y1_dy * z2 +
coef[30] * x3 * y1_dy * z0 +
coef[32] * x2 * y2_dy * z0 +
coef[33] * x2 * y1_dy * z1 +
coef[35] * x1 * y3_dy * z0 +
coef[36] * x1 * y2_dy * z1 +
coef[37] * x1 * y1_dy * z2 +
coef[39] * x0 * y4_dy * z0 +
coef[40] * x0 * y3_dy * z1 +
coef[41] * x0 * y2_dy * z2 +
coef[42] * x0 * y1_dy * z3 +
coef[45] * x4 * y1_dy * z0 +
coef[47] * x3 * y2_dy * z0 +
coef[48] * x3 * y1_dy * z1 +
coef[50] * x2 * y3_dy * z0 +
Appendix A. CODE V User-Defined Gradient-Index Source Code 301

coef[51] * x2 * y2_dy * z1 +
coef[52] * x2 * y1_dy * z2 +
coef[54] * x1 * y4_dy * z0 +
coef[55] * x1 * y3_dy * z1 +
coef[56] * x1 * y2_dy * z2 +
coef[57] * x1 * y1_dy * z3 +
coef[59] * x0 * y5_dy * z0 +
coef[60] * x0 * y4_dy * z1 +
coef[61] * x0 * y3_dy * z2 +
coef[62] * x0 * y2_dy * z3 +
coef[63] * x0 * y1_dy * z4 +
coef[66] * x5 * y1_dy * z0 +
coef[68] * x4 * y2_dy * z0 +
coef[69] * x4 * y1_dy * z1 +
coef[71] * x3 * y3_dy * z0 +
coef[72] * x3 * y2_dy * z1 +
coef[73] * x3 * y1_dy * z2 +
coef[75] * x2 * y4_dy * z0 +
coef[76] * x2 * y3_dy * z1 +
coef[77] * x2 * y2_dy * z2 +
coef[78] * x2 * y1_dy * z3 +
coef[80] * x1 * y5_dy * z0 +
coef[81] * x1 * y4_dy * z1 +
coef[82] * x1 * y3_dy * z2 +
coef[83] * x1 * y2_dy * z3 +
coef[84] * x1 * y1_dy * z4 +
coef[86] * x0 * y6_dy * z0 +
coef[87] * x0 * y5_dy * z1 +
coef[88] * x0 * y4_dy * z2 +
coef[89] * x0 * y3_dy * z3 +
coef[90] * x0 * y2_dy * z4 +
coef[91] * x0 * y1_dy * z5 +
coef[94] * x6 * y1_dy * z0 +
coef[96] * x5 * y2_dy * z0 +
coef[97] * x5 * y1_dy * z1 +
coef[99] * x4 * y3_dy * z0 +
coef[100] * x4 * y2_dy * z1 +
coef[101] * x4 * y1_dy * z2 +
coef[103] * x3 * y4_dy * z0 +
coef[104] * x3 * y3_dy * z1 +
coef[105] * x3 * y2_dy * z2 +
coef[106] * x3 * y1_dy * z3 +
coef[108] * x2 * y5_dy * z0 +
coef[109] * x2 * y4_dy * z1 +
coef[110] * x2 * y3_dy * z2 +
coef[111] * x2 * y2_dy * z3 +
coef[112] * x2 * y1_dy * z4 +
coef[114] * x1 * y6_dy * z0 +
coef[115] * x1 * y5_dy * z1 +
coef[116] * x1 * y4_dy * z2 +
coef[117] * x1 * y3_dy * z3 +
coef[118] * x1 * y2_dy * z4 +
coef[119] * x1 * y1_dy * z5 +
coef[121] * x0 * y7_dy * z0 +
coef[122] * x0 * y6_dy * z1 +
coef[123] * x0 * y5_dy * z2 +
Appendix A. CODE V User-Defined Gradient-Index Source Code 302

coef[124] * x0 * y4_dy * z3 +
coef[125] * x0 * y3_dy * z4 +
coef[126] * x0 * y2_dy * z5 +
coef[127] * x0 * y1_dy * z6);
dn_dz = coef[2] * (coef[12] * x0 * y0 * z1_dz +
coef[15] * x1 * y0 * z1_dz +
coef[17] * x0 * y1 * z1_dz +
coef[18] * x0 * y0 * z2_dz +
coef[21] * x2 * y0 * z1_dz +
coef[23] * x1 * y1 * z1_dz +
coef[24] * x1 * y0 * z2_dz +
coef[26] * x0 * y2 * z1_dz +
coef[27] * x0 * y1 * z2_dz +
coef[28] * x0 * y0 * z3_dz +
coef[31] * x3 * y0 * z1_dz +
coef[33] * x2 * y1 * z1_dz +
coef[34] * x2 * y0 * z2_dz +
coef[36] * x1 * y2 * z1_dz +
coef[37] * x1 * y1 * z2_dz +
coef[38] * x1 * y0 * z3_dz +
coef[40] * x0 * y3 * z1_dz +
coef[41] * x0 * y2 * z2_dz +
coef[42] * x0 * y1 * z3_dz +
coef[43] * x0 * y0 * z4_dz +
coef[46] * x4 * y0 * z1_dz +
coef[48] * x3 * y1 * z1_dz +
coef[49] * x3 * y0 * z2_dz +
coef[51] * x2 * y2 * z1_dz +
coef[52] * x2 * y1 * z2_dz +
coef[53] * x2 * y0 * z3_dz +
coef[55] * x1 * y3 * z1_dz +
coef[56] * x1 * y2 * z2_dz +
coef[57] * x1 * y1 * z3_dz +
coef[58] * x1 * y0 * z4_dz +
coef[60] * x0 * y4 * z1_dz +
coef[61] * x0 * y3 * z2_dz +
coef[62] * x0 * y2 * z3_dz +
coef[63] * x0 * y1 * z4_dz +
coef[64] * x0 * y0 * z5_dz +
coef[67] * x5 * y0 * z1_dz +
coef[69] * x4 * y1 * z1_dz +
coef[70] * x4 * y0 * z2_dz +
coef[72] * x3 * y2 * z1_dz +
coef[73] * x3 * y1 * z2_dz +
coef[74] * x3 * y0 * z3_dz +
coef[76] * x2 * y3 * z1_dz +
coef[77] * x2 * y2 * z2_dz +
coef[78] * x2 * y1 * z3_dz +
coef[79] * x2 * y0 * z4_dz +
coef[81] * x1 * y4 * z1_dz +
coef[82] * x1 * y3 * z2_dz +
coef[83] * x1 * y2 * z3_dz +
coef[84] * x1 * y1 * z4_dz +
coef[85] * x1 * y0 * z5_dz +
coef[87] * x0 * y5 * z1_dz +
Appendix A. CODE V User-Defined Gradient-Index Source Code 303

coef[88] * x0 * y4 * z2_dz +
coef[89] * x0 * y3 * z3_dz +
coef[90] * x0 * y2 * z4_dz +
coef[91] * x0 * y1 * z5_dz +
coef[92] * x0 * y0 * z6_dz +
coef[95] * x6 * y0 * z1_dz +
coef[97] * x5 * y1 * z1_dz +
coef[98] * x5 * y0 * z2_dz +
coef[100] * x4 * y2 * z1_dz +
coef[101] * x4 * y1 * z2_dz +
coef[102] * x4 * y0 * z3_dz +
coef[104] * x3 * y3 * z1_dz +
coef[105] * x3 * y2 * z2_dz +
coef[106] * x3 * y1 * z3_dz +
coef[107] * x3 * y0 * z4_dz +
coef[109] * x2 * y4 * z1_dz +
coef[110] * x2 * y3 * z2_dz +
coef[111] * x2 * y2 * z3_dz +
coef[112] * x2 * y1 * z4_dz +
coef[113] * x2 * y0 * z5_dz +
coef[115] * x1 * y5 * z1_dz +
coef[116] * x1 * y4 * z2_dz +
coef[117] * x1 * y3 * z3_dz +
coef[118] * x1 * y2 * z4_dz +
coef[119] * x1 * y1 * z5_dz +
coef[120] * x1 * y0 * z6_dz +
coef[122] * x0 * y6 * z1_dz +
coef[123] * x0 * y5 * z2_dz +
coef[124] * x0 * y4 * z3_dz +
coef[125] * x0 * y3 * z4_dz +
coef[126] * x0 * y2 * z5_dz +
coef[127] * x0 * y1 * z6_dz +
coef[128] * x0 * y0 * z7_dz);
// Calculate D = n * Grad(n)
xngran[1] = *rindx * dn_dx;
xngran[2] = *rindx * dn_dy;
xngran[3] = *rindx * dn_dz;
}
/* Mark the DLL as thread safe */
__declspec(dllexport) int cvisthreadsafe()
{
return 1;
}
/* Identify the type of the UMR DLL: GRN */
__declspec(dllexport) void cvumrtype(char* umrType)
{
strcpy(umrType, "GRN");
}
Appendix B

Annular Folded Lenses Grid Search


Code

The following CODE V MacroPLUS code is used for the nine-dimensional grid
search of the homogeneous annular folded lens (AFL) solution space.

! Initialize variables
LCL NUM ˆhfov(3) ˆfno(3) ˆepd(3) ˆobs(4) ˆtele_rat(4)
ˆsto_surf(4) ˆrefr_surf(2) ˆoutput_array(10)
GBL NUM ˆn_medium ˆoverage_13 ˆoverage_24
ˆefl == 100
ˆhfov(1) == 2.5
ˆhfov(2) == 5
ˆhfov(3) == 7.5
ˆfno(1) == 4
ˆfno(2) == 2
ˆfno(3) == 1.5
ˆepd(1) == ˆefl / ˆfno(1)
ˆepd(2) == ˆefl / ˆfno(2)
ˆepd(3) == ˆefl / ˆfno(3)
ˆobs(1) == 0.8
ˆobs(2) == 0.7
ˆobs(3) == 0.6
ˆobs(4) == 0.5
ˆtele_rat(1) == 0.8
ˆtele_rat(2) == 0.7
ˆtele_rat(3) == 0.6

304
Appendix B. Annular Folded Lenses Grid Search Code 305

ˆtele_rat(4) == 0.5
ˆsto_surf(1) == 1
ˆsto_surf(2) == 2
ˆsto_surf(3) == 3
ˆsto_surf(4) == 4
ˆrefr_surf(1) == 1 ! planar refractive surfaces (enter and exit)
ˆrefr_surf(2) == 2 ! spherical refractive surfaces (enter and
exit)
ˆnum_sp == 5
ˆnum_hfov == 3
ˆnum_epd == 3
ˆnum_obs == 4
ˆnum_tele_rat == 4
ˆnum_refr_surf == 2
ˆnum_sto_surf == 4
ˆcount == 0
ˆfid == ’sp_’
ˆoverage_13 == 2
ˆoverage_24 == 2
! Set working directory
CD "C:\CVUSER\Origami\Two Bounce"
! Ensure @JMRCC and @OBSRAY functions are present
IF ISFCT("JMRCC") = 0
OUT N
IN CV_MACRO:define_jmrcc
OUT Y
END IF
IF ISFCT("OBSRAY") = 0
OUT N
in define_obsray ! UDF to find "inner" marginal ray just
beyond obscuration
OUT Y
END IF
! Find empty buffer
ˆbuf == 1
WHILE NOT (BUF.EMP Bˆbuf)
ˆbuf == ˆbuf + 1
END WHILE
! Write header
ˆcount == ˆcount + 1
BUF PUT Bˆbuf Iˆcount J1 "Sol Num"
Appendix B. Annular Folded Lenses Grid Search Code 306

BUF PUT Bˆbuf Iˆcount J2 "EFL"


BUF PUT Bˆbuf Iˆcount J3 "SP Num"
BUF PUT Bˆbuf Iˆcount J4 "HFOV"
BUF PUT Bˆbuf Iˆcount J5 "FNO"
BUF PUT Bˆbuf Iˆcount J6 "OBS"
BUF PUT Bˆbuf Iˆcount J7 "TTL constraint"
BUF PUT Bˆbuf Iˆcount J8 "TTL"
BUF PUT Bˆbuf Iˆcount J9 "STO"
BUF PUT Bˆbuf Iˆcount J10 "WL"
BUF PUT Bˆbuf Iˆcount J11 "JMRCC" ! surface clearance constraint
BUF PUT Bˆbuf Iˆcount J12 "RST" ! refractive surface type
BUF PUT Bˆbuf Iˆcount J13 "POW" ! surface power
BUF PUT Bˆbuf Iˆcount J14 "ERF"
FOR ˆf 1 (NUM F)
ˆcol == 14 + ˆf
BUF PUT Bˆbuf Iˆcount Jˆcol CONCAT("RMS SPO F", NUM_TO_STR(ˆf))
END FOR
! Loop over starting point designs
FOR ˆh 1 ˆnum_sp
! Loop over number of HFOVs
FOR ˆi 1 ˆnum_hfov
! Loop over EPDs
FOR ˆj 1 ˆnum_epd
! Loop over obscurations
FOR ˆk 1 ˆnum_obs
! Loop over telephoto ratios
FOR ˆl 1 ˆnum_tele_rat
! Loop over stop surfaces
FOR ˆm 1 ˆnum_sto_surf
! Loop over mono/polychromatic
FOR ˆn 1 2
! Loop over JMRCC surface clearance constraint
on/off
FOR ˆo 1 2
! Loop over refractive surface types
FOR ˆp 1 ˆnum_refr_surf
! Increment count
Appendix B. Annular Folded Lenses Grid Search Code 307

ˆcount == ˆcount + 1
! Assign variables
ˆsol_num == ˆcount - 1
ˆss == ˆsto_surf(ˆm)
! Load one of five pre-defined starting
point design designs
OUT N
IN CONCAT(ˆfid, NUM_TO_STR(ˆh))
OUT Y
! Set title
TIT CONCAT(’Solution #’,
NUM_TO_STR(ˆsol_num))
! Set HFOV
YAN 0 0.4*ˆhfov(ˆi) 0.7*ˆhfov(ˆi)
0.85*ˆhfov(ˆi) 1*ˆhfov(ˆi)
! Set EPD
EPD ˆepd(ˆj)
! Set wavelengths, if necessary (all
starting points are monochromatic)
IF ˆn = 2
WL 656.2725 587.5618 486.1327
WTW 1 1 1
END IF
! Set stop surface
STO Sˆss
! Delete any pre-existing obscuration
aperture
FOR ˆs 1 ((NUM S) - 1)
DEL APE Sˆs OBS
END FOR
! Set vignetting
IN CV_MACRO:setvig
! Set obscuration aperture based on
percentage of stop surface’s maximum
aperture (MAP)
Appendix B. Annular Folded Lenses Grid Search Code 308

CIR SS OBS ˆobs(ˆk)*(MAP SS)


! Vary appropriate curvatures and thicknesses
CCY S’Refl_1_Rear’ 0
CCY S’Refl_2_Front’ 0
THC S’Enter’ 0
THC S’Exit’ 0
! Vary refractive surfaces, if necessary
(all starting points are planar)
IF ˆp = 2
CCY S’Enter’ 0
CCY S’Exit’ 0
END IF
! Determine TTL constraint based on
telephoto ratio
ˆttl == ˆtele_rat(ˆl) * ˆefl
! Optimize
AUT
! Interactive mode
INT N
! Set rectangular ray grid density (chosen
using autogrid.seq)
DEL 0.050094
! Set aperture weighting to somewhat
emphasize edge of pupil
WTA 0.25
! Set obscuration in optimization ray grid
(see autogrid.seq)
OBS ˆobs(ˆk)
! Draw lens
DRA Y
! EFL
EFL = ˆefl
! TTL
OAL S1..i < ˆttl
Appendix B. Annular Folded Lenses Grid Search Code 309

! Hold positive center thickness


THI S’Enter’ > 0
! Surface clearance constraint, if necessary
IF ˆo = 1
! Get refractive index of medium
ˆn_medium == (IND S1)
! CONSTRAIN S1/S3 CLEARANCE
! Constraining surface clearance using the
distance between the on-axis
! Y coordinate of R2 on the entrance
refractive surface with the Y position
! of R2 from the third surface (second
reflection) for either the on-axis and
! fullfield bundles, depending on the stop
position. For stop on S1, S2, S3
! need to look at the full field R2. For
stop on S4, need to look at the
! on-axis R2
! Calculate Y surface clearance with
@JMRCC function
! Must take into account immersing medium
refractive index
IF (STO) < 4
! For full field
@clearance_13 == @JMRCC(2, (NUM F), 3,
2, 1, 1, 0, 0) / ˆn_medium
else
! For on-axis
@clearance_13 == @JMRCC(2, 1, 3, 2, 1,
1, 0, 0) / ˆn_medium
END IF
! Calculate delta of enter beam below
upper marginal ray (R2 F1)
@delta_1 == (Y R2 F1 S1) - @OBSRAY(1, (NUM
F), 1)
! Calculate minimum allowable clearance
with overage
Appendix B. Annular Folded Lenses Grid Search Code 310

@min_clearance_13 == @delta_1 + ˆoverage_13


! Calculate clearance difference (needed
to update each optimization cycle)
@diff_clearance_13 == @clearance_13 -
@min_clearance_13
! Constrain clearance
@diff_clearance_13 > 0
! CONSTRAIN S2/S4 CLEARANCE
! Constraining surface clearance using the
distance between the on-axis
! Y coordinate of R2 on the second surface
(first reflection) with the Y position
! of R2 from the third surface (second
reflection) for either the on-axis and
! fullfield bundles, depending on the stop
position. For stop on S1, S2, S3
! need to look at the full field R2. For
stop on S4, need to look at the
! on-axis R2
! Calculate Y surface clearance with
@JMRCC function
! Must take into account immersing medium
refractive index
IF (STO) < 4
! For full field
@clearance_24 == @JMRCC(2, (NUM F), 3,
2, 1, 2, 0, 0) / ˆn_medium
else
! For on-axis
@clearance_24 == @JMRCC(2, 1, 3, 2, 1,
2, 0, 0) / ˆn_medium
END IF
! Calculate delta of enter beam below
upper marginal ray (R2 F1)
! Using the OBSRAY in the bottom half
(negative Y) side of the pupil
! Otherwise a new negative field would
need to be defined
@delta_2 == (Y R2 F1 S2) - (-1 *
Appendix B. Annular Folded Lenses Grid Search Code 311

@OBSRAY(2, (NUM F), -1))


! Calculate minimum allowable clearance
with overage
@min_clearance_24 == @delta_2 + ˆoverage_24
! Calculate clearance difference (needed
to update each optimization cycle)
@diff_clearance_24 == @clearance_24 -
@min_clearance_24
! Constrain clearance
@diff_clearance_24 > 0
END IF
GO
! Optimization complete
! Store error function before performing any
changes
ˆerf == (AUT.ERF)
! Set vignetting
IN CV_MACRO:setvig
! Reset obscuration as percentage of stop
surfaces maximum aperture (MAP)
CIR SS OBS ˆobs(ˆk)*(MAP SS)
! Write to results buffer
BUF PUT Bˆbuf Iˆcount J1 ˆsol_num
BUF PUT Bˆbuf Iˆcount J2 ˆefl
BUF PUT Bˆbuf Iˆcount J3 ˆh
BUF PUT Bˆbuf Iˆcount J4 ˆhfov(ˆi)
BUF PUT Bˆbuf Iˆcount J5 ˆfno(ˆj)
BUF PUT Bˆbuf Iˆcount J6 ˆobs(ˆk)
BUF PUT Bˆbuf Iˆcount J7 ˆttl
BUF PUT Bˆbuf Iˆcount J8 (OAL S1..I)
BUF PUT Bˆbuf Iˆcount J9 ˆsto_surf(ˆm)
IF ˆn = 1 ! monochromatic
BUF PUT Bˆbuf Iˆcount J10 "MONO"
else ! polychromatic
BUF PUT Bˆbuf Iˆcount J10 "POLY"
END IF
IF ˆo = 1 ! surface clearance constraint on
Appendix B. Annular Folded Lenses Grid Search Code 312

BUF PUT Bˆbuf Iˆcount J11 "ON"


else ! surface clearance constraint off
BUF PUT Bˆbuf Iˆcount J11 "OFF"
END IF
IF ˆp = 1 ! planar refractive surfaces
BUF PUT Bˆbuf Iˆcount J12 "PLANAR"
else ! spherical refractive surfaces
BUF PUT Bˆbuf Iˆcount J12 "SPHERICAL"
END IF
! Evalulate surface power sign
ˆsign == ""
IF (CUY S’Enter’) > 0
ˆsign == CONCAT(ˆsign, "P")
else IF (CUY S’Enter’) < 0
ˆsign == CONCAT(ˆsign, "N")
END IF
IF (CUY S’Refl_1_Rear’) < 0
ˆsign == CONCAT(ˆsign, "P")
else IF (CUY S’Refl_1_Rear’) > 0
ˆsign == CONCAT(ˆsign, "N")
END IF
IF (CUY S’Refl_2_Front’) > 0
ˆsign == CONCAT(ˆsign, "P")
else IF (CUY S’Refl_2_Front’) < 0
ˆsign == CONCAT(ˆsign, "N")
END IF
IF (CUY S’Exit’) < 0
ˆsign == CONCAT(ˆsign, "P")
else IF (CUY S’Exit’) > 0
ˆsign == CONCAT(ˆsign, "N")
END IF
BUF PUT Bˆbuf Iˆcount J13 ˆsign
! Write final optimization error function
BUF PUT Bˆbuf Iˆcount J14 ˆerf
! Evaluate spot size for all fields
FOR ˆf 1 (NUM F)
ˆdummy == SPOTDATA(1, ˆf, 1, 0, "DEF", 0,
0, ˆoutput_array)
ˆcol == 14 + ˆf
BUF PUT Bˆbuf Iˆcount Jˆcol
ˆoutput_array(1)
END FOR
! Write to sequence file
Appendix B. Annular Folded Lenses Grid Search Code 313

ˆfid_out == concat("SEQs/sol_",
NUM_TO_STR(ˆsol_num))
WRL ˆfid_out
END FOR
END FOR
END FOR
END FOR
END FOR
END FOR
END FOR
END FOR
END FOR
! Print buffer to command window
BUF SEP ","
BUF LIS Bˆbuf
! Write buffer to text file
BUF EXP Bˆbuf SEP "origami_output.csv"
! Delete buffer
BUF DEL Bˆbuf
Appendix C

Analytical ∆n of Gradient-Index
Alvarez Lens Elements

The refractive index profiles for the two elements in the base (i.e., unoptimized)
GAL are defined in Eq. (3.9). Considering only a single element, the index function
for a GAL laterally translating along y is

y3
 
2
n (x, y) = n0 + A + x y + By (C.1)
3

where n0 is the base refractive index, A is the scalar coefficient governing the
Alvarez power variation, and B is the scalar tilt coefficient. In this work, the
GAL elements are defined over a circular clear aperture of diameter D, meaning
2
x2 + y 2 ≤ D2 in Eq. (C.1).
When designing a GAL, an element’s total refractive index change, ∆n = nmax −
nmin , is an important quantity due to constrains on GRIN fabrication and metrol-
ogy. As shown previously in Fig. 3.5, the coordinates containing the refractive
index minima and maximum change with tilt B. As a result, the function ∆n (B)
is piecewise and has four different subdomains, as depicted in Fig. C.1. The piece-
wise expression for ∆n (B) can be determined by first identifying the coordinates

314
Appendix C. Analytical ∆n of Gradient-Index Alvarez Lens Elements 315

I II III IV
B<-|γ| -|γ|≤ B<-|B0| -|B0|≤ B<|γ| |γ|≤ B

Figure C.1: GAL element total refractive index change ∆n for varying refractive index tilt B.
The Alvarez coefficient A and element diameter D are held constant. The function ∆n (B)
is piecewise with four different subdomains. Example refractive index profiles are shown
for each subdomain with index minima and maxima marked with white and black points,
respectively. This example is evaluated for Alvarez coefficient A = 8.485 × 10−4 mm−3 and
element diameter D = 10 mm.

of refractive index extrema and then calculating the difference between maximum
and minimum.
The refractive index minima and maxima are marked in the index profiles in
Fig. C.1 by white and black points, respectively, for the four different subdomains.
It can be seen that there are two categories of coordinates: (1) those lying along
the y-axis and (2) those lying along the clear aperture edge. Both categories can be
determined by differentiating Eq. (C.1) and equating to zero.
First, the refractive index extrema lying along the y-axis can be found by

∂n1 (x, y)
= Ay 2 + B = 0. (C.2)
∂y x=0
Appendix C. Analytical ∆n of Gradient-Index Alvarez Lens Elements 316

Solving for the values of y yielding zero slope yields the coordinates of local ex-
trema, r
B
x = 0, y=± − . (C.3)
A

The square root in Eq. (C.3) provides the requirement that B must be of opposite
sign of A in order for local extrema to exist along y. There is also the requirement
that these extrema along y must fall within the clear aperture radius,

r
B D
− ≤ . (C.4)
A 2

Rearranging and noting that B/A ≤ 0 gives a refined domain over which local
extrema exist along y within the clear aperture,



−γ ≤ B ≤ 0,
 A≥0
(C.5)

0 < B ≤ −γ,
 A<0

where
D2 A
γ= . (C.6)
4

Satisfying the condition in Eq. (C.5) means there are local extrema along y within
the clear aperture, such as for the GAL element shown for subdomain II in Fig.
C.1. Although, currently there is no guarantee that these extrema determine nmin
and nmax since other extrema may exist along the clear aperture. For example, as
shown for B/B0 = 0.5 in Fig. 3.5, there are local extrema along y yet the ∆n is still
governed by the second category of extrema.
The second class of refractive index extrema along the clear aperture edge can
be identified by converting the refractive index in Eq. (C.1) from Cartesian to polar
Appendix C. Analytical ∆n of Gradient-Index Alvarez Lens Elements 317

coordinates,

cos3 φ
 
3 2
n1 (ρ, φ) = n0 + Aρ + sin φ cos φ + Bρ cos φ
3
    (C.7)
2 2 − cos 2φ
= n0 + ρ cos φ Aρ +B
3

where ρ is the radial coordinate and φ is the azimuthal coordinate measured rela-
tive to the positive y direction. Then, differentiating Eq. (C.7) with respect to φ and
equating to zero provides the extrema along the circular clear aperture of radius
ρ = D/2,
∂n1 (ρ, φ) D
sin φ D2 A cos 2φ − 4B = 0.

= (C.8)
∂φ ρ= D 8
2

Due to the symmetry of GAL elements about y, only azimuthal values of φ ∈ [0, π]
need be considered. Moreover, since extrema along y are being handled as a sep-
arate case, a revised domain of polar angles along which extrema may occur is
φ ∈ (0, π). Applying this domain to Eq. (C.8) supplies the corresponding roots,

4B B
cos 2φ = 2
= . (C.9)
D A γ

The specific azimuthal angle φ can be solved from Eq. (C.9) using the trigonometric
relationship cos 2φ = 2 cos2 φ − 1 = 1 − 2 sin2 φ,

r ! r !
1 2B 1 2B
φ = cos−1 ± + 2 = sin−1 ± − 2 . (C.10)
2 D A 2 D A

Then, the Cartesian coordinates of the extrema along the clear aperture are

r
D D 1 2B
x = sin φ = ± − 2 ,
2 2 r2 D A
(C.11)
D D 1 2B
y = cos φ = ± + 2 .
2 2 2 D A
Appendix C. Analytical ∆n of Gradient-Index Alvarez Lens Elements 318

Since cos 2φ in Eq. (C.9) is bounded by ±1, valid refractive index extrema along the
clear aperture must satisfy
− |γ| ≤ B ≤ |γ| (C.12)

where γ is defined in Eq. (C.6).


There is overlap between the cases of extrema along y defined in Eq. (C.5)
and extrema along the clear aperture edge defined in Eq. (C.12). The transition
between these two cases occurs at the boundary of subdomains II and III in Fig.
C.1. This boundary occurs when the refractive index at the local extrema along y
equals the index extrema along the clear aperture. This tilt value where these two
subdomains meet is denoted B = B0 , and it also corresponds to the value of B
offering the minimum possible ∆n for certain values of A and D. This important
tilt value, B = B0 , can be solved by equating the refractive index in Eq. (C.1) for
the coordinates in Eqs. (C.3) and (C.11) describing the two categories of extrema,

r ! r r !
B D 1 2B D 1 2B
n 0, − =n − , +
A 2 2 D2 A 2 2 D2 A
r √ r
2 B 2 2
 4B
n0 + B − = n0 + D D A + 4B 1+ 2
3 A 24 D A
r √  2 r 2
2 B 2 D A + 4B D A + 4B
− = D
3 A 24 B D2 A
s

 2   
D A + 4B A D2 A + 4B (C.13)
8 2=D −
B B D2 A
3/2
D2 A + 4B

7/2
2 = −
B
2
D A
27/3 = − −4
B
D2 A
4 1 + 21/3 = −

.
B

Solving for B and multiplying by its conjugate to eliminate all the radical in the
Appendix C. Analytical ∆n of Gradient-Index Alvarez Lens Elements 319

denominator gives

D2 A 1 + 22/3 − 21/3 1 + 22/3 − 21/3 2


 
B=− · =− D A. (C.14)
4 (1 + 21/3 ) 1 + 22/3 − 21/3 12

Thus, the tilt coefficient B0 yielding the minimum GRIN ∆n is

B0 = −κD2 A = −4κγ (C.15)

where κ is a constant defined as

1 + 22/3 − 21/3
κ= ≈ 0.1106. (C.16)
12

This specific value of B serves as the transition between subdomains II and III for
the function ∆n (B), as plotted in Fig. C.1.
With B0 obtained, the piecewise function ∆n (B) can be fully determined by
taking the difference between refractive index maxima and minima using the ob-
tained extrema coordinate positions. These coordinates differ depending on the
subdomain. As can be seen in the GRIN profiles in Fig. C.1., the extrema for sub-
domains I and IV occur at coordinates 0, ± D2 . For subdomain II, the extrema


coordinates are internal to clear aperture along y, as defined in Eq. (C.3). For sub-
domain III, the extrema coordinates fall along the clear aperture edge, as defined
in Eq. (C.11).
Combining the results for the four subdomains, the piecewise total refractive
Appendix C. Analytical ∆n of Gradient-Index Alvarez Lens Elements 320

index change ∆n as a function of tilt coefficient B can be constructed,


3
− D12|A| − BD, B < −|γ|







 q
 4 − B3 , −|γ| ≤ B < −|B0 |


3 |A|
∆n (B · sgnA) =  3 r   (C.17)
D |A| BD 4B
, −|B0 | ≤ B < |γ|



 12
+ 3 2 1+ D2 |A|




 D3 |A| + BD,

|γ| ≤ B

12

where sgn (· · · ) is the sign function (i.e., A · sgnA = |A|), γ is defined in Eq. (C.6)
and B0 is defined in Eq. (C.15). The sign function is needed since A may take
either positive or negative values, reversing the order of the subdomains. The only
coefficient restricted in sign is the element diameter which may only take positive
values, D > 0.
The two specific cases of interest, B = {0, B0 }, are examined in Chapter 3.
Evaluating Eq. (C.17) for these scenarios yields simplified expressions. For zero
tilt, the ∆n is found within subdomain III,


2 3
∆n (B = 0) = D |A|. (C.18)
12

For tilt B0 minimizing the refractive index change, ∆n is defined at the boundary
of subdomains II and III,

4κ3/2 3
∆n (B = B0 ) = D |A|. (C.19)
3

In both cases, ∆n increases linearly with the magnitude of A and cubically with
element diameter D.
Appendix D

Derivation of the Linear


Gradient-Index Ray Path

For linear GRIN with the gradient direction ĝ = sin θg x̂ + cos θg ŷ lying in the trans-
verse plane, no axial refractive index variation and symmetry about the g-z plane
yields the simplified form of the Euler-Lagrange equation presented in Eq. (5.10),

∂n
1 + ġ 2 = 0

ng̈ − (D.1)
∂g

where a dot indicates d/dz.


Within this simplified framework, the linear GRIN profile is considered in only
two dimensions,

n (g) = n0 + αg (D.2)

where n0 is the base refractive index along the optical axis and α is the slope of
the linear refractive index change. Recall, α is restricted to positive values to avoid
redundancy with the gradient direction ĝ, as determined by θg .
The derivation for the linear GRIN ray path presented here assumes the ray
to be confined to this g-z plane containing the gradient and the optical axis. As

321
Appendix D. Derivation of the Linear Gradient-Index Ray Path 322

described in Sec. 5.3.1, this two-dimensional case can be generalized to skew rays
falling outside of the g-z plane by recognizing that a planar cross-section can al-
ways be formed that contains ĝ and the initial ray direction.
Toward deriving the ray path, initial conditions must first be specified for the
ray’s position and slope. Without loss of generality, the ray is taken to start at the
origin and with some initial slope β in the g-z plane,

g (z = 0) = 0
(D.3)
ġ (z = 0) = β.

To start, the linear GRIN refractive index n (g) is inserted into Eq. (D.1),

(n0 + αg) g̈ − α 1 + ġ 2 = 0.

(D.4)

Notice this yields a nonlinear, inhomogeneous, second-order ordinary differential


equation. Then, solving for g̈ gives

1 + ġ 2
g̈ = (D.5)
κ+g

where κ ≡ n0 /α.
By letting v ≡ ġ, now g̈ can be expressed in terms of v as

dġ dv dv dg dv
g̈ = = = =v . (D.6)
dz dz dg dz dg

Then, Eq. (D.5) can be recast in terms of v using this form of g̈,

dv 1 + v2
v = . (D.7)
dg κ+g
Appendix D. Derivation of the Linear Gradient-Index Ray Path 323

Now, performing separation of variables and integrating yields

Z Z
v dv dg
= . (D.8)
1 + v2 κ+g

Next, performing change of variables u ≡ v 2 grants the integral on the left-hand


side,
Z Z
1 du dg
=
2 1+u κ+g
1
ln (1 + u) = ln (κ + g) + c1
2 (D.9)
2[ln(κ+g)+c1 ]
1+u=e

u = (κ + g)2 e2c1 − 1.

Reverting from u to v and from v to g gives

ġ 2 = (κ + g)2 γ −2 − 1 (D.10)

where a new coefficient is defined, γ ≡ e−c1 . Subsequently pulling out γ −2 and


expanding provides a quadratic expression that can be factored,

ġ 2 = γ −2 g 2 + 2κg + κ2 − γ 2


= γ −2 g 2 + 2κg + (κ − γ) (κ + γ) (D.11)
 

= γ −2 (g + κ − γ) (g + κ + γ) ,

or with the square root taken,

ġ = ±γ −1 [(g + κ − γ) (g + κ + γ)]1/2 . (D.12)


Appendix D. Derivation of the Linear Gradient-Index Ray Path 324

Once again, performing separation of variables and integrating gives

Z Z
−1/2 −1
dg [(g + κ − γ) (g + κ + γ)] = ±γ dz (D.13)

Focusing on the left-hand side integral, adding γ − γ = 0 to the one of the factors
in parentheses provides a more favorable expression,

Z Z
−1/2
dg [(g + κ − γ) (g + κ + γ)] = dg [(g + κ − γ) (g + κ − γ + 2γ)]−1/2
Z
−1/2
dg (g + κ − γ)2 + 2γ (g + κ − γ)

=
  −1/2
g+κ−γ
Z
= dg 2γ (g + κ − γ) 1 + .

(D.14)

Performing the substitution

r
g+κ−γ 1
u=± , du = [2γ (g + κ − γ)]−1/2 (D.15)
2γ 2

miraculously yields for this integral

Z Z
−1/2 du
dg [(g + κ − γ) (g + κ + γ)] =2 √
1 + u2
= 2 sinh−1 (u) (D.16)
 r 
−1 y+κ−γ
= 2 sinh ± .

Returning to Eq. (D.13), the right-hand side integral is far simpler,

Z    
−1 z z + γc
±γ dz = ± +c =± (D.17)
γ γ
Appendix D. Derivation of the Linear Gradient-Index Ray Path 325

where c is a new constant of integration containing a factor of γ −1 .


Combining the result of each of these two integrals in Eq. (D.13) provides a
preliminary expression for the ray path g (z),

 r   
−1 g+κ−γ z + γc
2 sinh ± =±
2γ γ
r   
g+κ−γ z + γc
± = sinh ± (D.18)
2γ 2γ
  
2 z + γc
g (z) = 2γ sinh ± − κ + γ.

The g expression in Eq. (D.18) is solely in terms of z and the two coefficients γ
and c. Also, the plus-or-minus can be dropped since the square of hyperbolic sine
is an even function. This expression can also be refined further by manipulating in
z+γc
terms of exponentials, letting η = 2γ
,

 2
 
1 +η −η
g (z) = 2γ e −e −κ+γ
2
+ e−2η
 +2η 
e
=γ −1 −κ+γ
2
= γ [cosh (2η) − 1] − κ + γ (D.19)

= γ cosh (2η) − κ
 
z
= γ cosh + c − κ.
γ

Lastly, expressions for the two coefficients γ and c must be solved for that sat-
isfy the initial conditions in Eq. (D.3) defined at the front surface, z = 0. First, γ is
determined by applying these initial conditions to Eq. (D.10),

β 2 = κ2 γ −2 − 1
κ (D.20)
γ=p
1 + β2
Appendix D. Derivation of the Linear Gradient-Index Ray Path 326

where the plus-or-minus sign for γ is known to be positive based on its definition,
γ ≡ e−c1 , and the definition κ ≡ n0 /α.
Next, c is obtained from Eq. (D.19) using the initial condition on ray slope β,

ġ (z = 0) = sinh (c) = β ⇒ c = sinh−1 β. (D.21)

Finally, substituting the expression for γ and the definition of κ gives the final
form of the linear GRIN ray path g (z) in Eq. (5.12),

" p ! #
n0 1 α 1 + β2
g (z) = p cosh z+c −1 (D.22)
α 1 + β2 n0

where
 p 
c = sinh−1 β = ln β + β 2 + 1 . (D.23)

Differentiating Eq. (D.22) with respect to z gives the ray slope throughout the
linear GRIN, which will prove useful in determining refraction at the rear surface,

p !
α 1 + β2
ġ (z) = sinh z+c . (D.24)
n0

As verification, the ray initial position can be verified by Eq. (D.22),

" #
n0 1
cosh sinh−1 β − 1 = 0

g (z = 0) = p X (D.25)
α 1+β 2

while the ray initial slope can be verified by Eq. (D.24),

ġ (z = 0) = sinh sinh−1 β = β.

X (D.26)

The initial ray slope β can also be recast in terms of the angle of refraction θ10
Appendix D. Derivation of the Linear Gradient-Index Ray Path 327

following the front planar surface, β = tan θ10 . Using Snell’s law, θ10 can be obtained
from the angle of incident θ1 at the front surface. Without loss of generality, the ray
is taken to be incident at the front surface vertex where the refractive index is n0 .
This makes Snell’s law for a linear GRIN immersed in air, nair = 1,

sin θ1
sin θ10 = (D.27)
n0

where by the trigonometric relationship

  
sin θ1 sin θ1
β= tan θ10 = tan sin −1
=p . (D.28)
n0 n0 − sin2 θ1
2

p
Similarly, examining the common term 1 + β 2 in Eq. (D.22) finds

p n0
1 + β2 = p , (D.29)
n20 − sin2 θ1

which provides an alternate form for the ray path and slope in terms of θ1 ,

"p ! #
n0 n20 − sin2 θ1 α
g (z) = cosh p z+c −1
α n0 n20 − sin2 θ1
! (D.30)
α
ġ (z) = sinh p z+c .
n20 − sin2 θ1

For a normally incident ray, θ1 = 0, these expressions simplify to a hyperbolic


cosine with its global minimum shifted to the origin,

   
n0 α
g (z) = cosh z −1
α n0
  (D.31)
α
ġ (z) = sinh z .
n0
Appendix D. Derivation of the Linear Gradient-Index Ray Path 328

This simplest form for the ray in a linear GRIN offers the clearest picture of its
dependence on base index n0 and refractive index slope α. The leading factor of
n0 /α in the g (z) expression is less influential than the factor of α/n0 within the
hyperbolic cosine. As a result, the ray travels farthest from the optical axis (i.e.,
curves the most) when α is large and/or n0 is small.
It is also of interest to perform Snell’s law at the rear surface after propagating
through the linear GRIN. Doing so provides the ray exit angle θ20 that is used in the
illumination design process covered in Sec. 5.4. For plane-parallel surfaces with
center thickness t, the angle of incidence on the rear surface θ2 is given from the
ray slope ġ, !
p
α 1 + β2
tan θ2 = ġ (z = t) = sinh t+c . (D.32)
n0

Then, Snell’s law is performed for θ20 in air,

sin θ20 = n [g (z = t)] · sin θ2 . (D.33)

The refractive index is required at the point of ray-surface intersection on the rear
surface of the linear GRIN,

p !
n0 α 1 + β2
n [g (z = t)] = n0 + α · g (z = t) = p cosh t+c . (D.34)
1 + β2 n0


Next, sin θ2 in Snell’s law can be found from tan θ2 since sin (tan−1 x) = x/ 1 + x2 ,

 √ 
α 1+β 2
sinh n0
t +c p !
α 1 + β2
sin θ2 = s  √  = tanh t+c . (D.35)
α 1+β 2
n0
1 + sinh2 n0
t+c

Returning to Eq. (D.33), inserting the two obtained expressions for the right-hand
Appendix D. Derivation of the Linear Gradient-Index Ray Path 329

side yields the final result presented in Eq. (5.16),

p !
n0 α 1 + β2
sin θ20 = p sinh t+c , (D.36)
1 + β2 n0

or in terms of front surface angle of incidence θ1 ,

!
α
q
sin θ20 = n20 − sin2 θ1 sinh p t+c . (D.37)
n20 − sin2 θ1

For some applications, it is also helpful to know the optical path length (OPL)
for a ray traversing GRIN media. For linear GRIN of center thickness t, the OPL
can be found analytically by integrating along the ray path in Eq. (D.22),

Z Z t
OP L = n ds = n [g (z)] dz. (D.38)
0

Evaluating the integrand finds

n [g (z)] = n0 + αg (z)
( " p ! #)
n0 1 α 1 + β2
= n0 + α cosh z+c −1
(D.39)
p
α 1 + β2 n0
p !
n0 α 1 + β2
=p cosh z+c .
1 + β2 n0
Appendix D. Derivation of the Linear Gradient-Index Ray Path 330

Then, the OPL integral becomes

p !
t
1 + β2
Z
n0 α
OP L = p cosh z + c dz
1 + β2 0 n0
!
t
p
n20 α 1 + β2
= sinh z + c (D.40)
α (1 + β 2 ) n0 0
" p ! #
n20 α 1 + β2
= sinh t + c − sinh c ,
α (1 + β 2 ) n0

or in terms of θ1 instead of β,
" ! #
n20 − sin2 θ1 α sin θ1
OP L = sinh p t+c −p . (D.41)
α n20 − sin2 θ1 n20 − sin2 θ1

These various analytical expressions for the ray trajectory in a linear GRIN are
all that are needed for the F-GRIN illumination design process presented in Sec.
5.4.
Appendix E

Algorithms used for Gradient-Index


Modal Reconstruction

In this appendix, two algorithms are detailed that are used for the GRIN modal
reconstruction in Chapter 6. All code used in this work, including all ray tracing, is
written in Python (version 2.7.16) and executed on the BlueHive computing cluster
located at the University of Rochester’s Center for Integrated Research Computing.
For the modal reconstruction process, the user provides as input:

• Measured optical path difference data, OP Dm (f, p), following transmission


through the GRIN under test. OP Dm (f, p) is defined for rays with initial ray
direction cosines f = (α0 , β0 , γ0 ) and initial pupil coordinates p = (x0 , y0 , z0 )
at the front planar surface of the GRIN. In total, OP Dm is defined across Nf
measurement field angles and Np rays sampling the pupil for each measure-
ment angle. In total, there are Nr number of rays sampling the GRIN with
different spatial and angular initial conditions. Nr < Nf × Np due to vi-
gnetting for non-normally incident fields. For example, the reconstruction in
Chapter 6 uses Nf = 129 and Np = 207 (see Fig. 6.4) but Nr = 22,392 rays.

• Center thickness t of the cylindrical reconstruction volume.

331
Appendix E. Algorithms used for Gradient-Index Modal Reconstruction 332

• Diameter D of the cylindrical reconstruction volume.

• GRIN absolute refractive index n0 at the front surface vertex of the GRIN (de-
fined as the coordinate origin). This value can be estimated and has minimal
impact on the reconstruction of the relative refractive index variation, which
is the main objective of the reconstruction.

• Ray trace step size ∆t used in the fourth-order Runge-Kutta method pre-
sented by Sharma [19–21]. For example, the reconstruction in Chapter 6 uses
∆t = 0.02 mm for t = 1 mm.

• Regularization parameter Λsp used for the starting point generated by the
straight-ray approximation (SRA). For example, the reconstruction in Chap-
ter 6 uses Λsp = 50, as found by a hyperparameter search.

• Regularization parameter Λio used for the iterative optimization. For exam-
ple, the reconstruction in Chapter 6 uses Λio = 6.11 × 10−2 , as found by a
hyperparameter search.

• Number of fringe Zernike polynomial terms NZ governing reconstructed


transverse refractive index variation, as defined by Eq. (6.5). For example,
the reconstruction in Chapter 6 uses NZ = 16.

• Number of Legendre polynomial terms NL governing reconstructed axial re-


fractive index variation, as defined by Eq. (6.5). For example, the reconstruc-
tion in Chapter 6 uses NL = 7.

From the reconstruction, the user obtains as output:

• GRIN coefficients Θ in a matrix of size Nc = NZ × NL . The coefficients de-


scribe the reconstructed GRIN by the polynomial basis in Eq. (6.5).
Appendix E. Algorithms used for Gradient-Index Modal Reconstruction 333

The first step in the reconstruction process is to generate a starting point guess
Θsp by ordinary least squares (OLS) based on the straight-ray approximation (SRA),
as discussed in Sec. 6.2.1. This process relies on numerical integration performed
by the QUADPACK Fortran library, implemented with the scipy.integrate.
quad function. From these line integrals, a system of linear equations can be
formed and solved. Starting point generation is outlined in Algorithm 1.

Algorithm 1 OLS starting point generated assuming SRA


input:
y . column vector of size Nr containing OP Dm
initialize:
X . matrix of size Nr × Nc
for a ← 1 to Nr do . loop over measurement rays
perform vectorial Snell’s law for ray a at front planar surface, z = 0,
assuming homogeneous refractive index n0
from refracted ray direction, determine ray coordinates on rear surface, z = t,
assuming linear ray transfer
if ray exits cylindrical reconstruction volume then
skip current ray
remove current row a from matrix X
else
calculate linear ray segment length L within GRIN
for b ← 1 to Nc do . loop over modal polynomial terms
perform OP L line integral through polynomial term b of Eq. (6.5),
using scipy.integrate.quad function
X [a, b] ← OP L − n0 L . convert to OP D
end for
end if
end for
initialize:
I . identity matrix of size Nc × Nc
I [0, 0] ← 0 . remove bias term from regularization
obtain Θsp by OLS via the normal equation with regularization,
−1 T
Θsp ← X T X + Λsp I X y . result is vector of size Nc
reshape vector Θsp to matrix of size NZ × NL
return Θsp
Appendix E. Algorithms used for Gradient-Index Modal Reconstruction 334

The second step in the reconstruction is to perform iterative optimization of


coefficients Θ to minimize a cost function, as discussed in Sec. 6.2.2. The starting
value for Θ is initialized from Θsp obtained from Algorithm 1.
As explored further in Sec. 6.2.2, the cost function J (Θ) consists of the mean
squared error (MSE) between the measured OP D and simulated OP D for the cur-
rent guess of Θ. With the update of Θ for each optimization cycle, it is sought that
the simulated OP D converges on the measured OP D data. The cost function also
includes a regularization term, controlled by the regularization parameter Λio .
The cost function can be programmed using automatic differentiation software.
In doing so, the optimization algorithm can readily be provided with the gradient
of the cost function with respect to the optimization variables, ∇Θ J (Θ), to aid in
convergence.
The cost function can also be written using parallelization software. With paral-
lelization, the simulated ray trace and OP D calculation can be performed indepen-
dently for different sets of rays on different central processing units (CPUs). For
the reconstruction performed in Chapter 6 with Nr = 22,392 rays traced per eval-
uation of the cost function, parallelization drastically reduces computation time.
The optimization algorithm in use is the Broyden–Fletcher–Goldfarb–Shanno
(BFGS) algorithm, as implemented with the scipy.optimize.minimize func-
tion. A second optimization algorithm that has also been considered is the conju-
gate gradient method.
Modal reconstruction by iterative optimization is outlined in Algorithm 2.
Appendix E. Algorithms used for Gradient-Index Modal Reconstruction 335

Algorithm 2 Modal reconstruction by iterative optimization


function S IMULATED OPD(Θ, f , p)
implementation of Sharma ray tracing with OP D calculation [19–21],
evaluated for ray with initial conditions (f, p) using current guess of Θ
return OP Ds
end function
function C OST(Θ)
JOP D ← 0 . optical path difference MSE cost
for f ← 1 to Nf do
for p ← 1 to Np do
JOP D ← JOP D + [S IMULATED OPD (Θ, f, p) − OP Dm (f, p)]2
end for
end for
JOP D ← 12 · JOP
Nr
D
. take the mean
Jreg ← 0 . regularization cost
for i ← 1 to NZ do
for j ← 1 to NL do
Jreg ← Jreg + Θ [i, j]2
end for
end for
Jreg ← Λ2io · NJZreg
NL
. take the mean
J ← JOP D + Jreg
return J
end function
input:
Θsp . matrix of size NZ × NL
Θ ← BFGS M INIMIZE [C OST (Θsp )] . iterative optimization of Θsp
return Θ

You might also like