See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/366408582

Optimized Multi-Element Lens Design

Preprint · December 2022

DOI: 10.13140/RG.2.2.14970.36808

CITATIONS READS

0 247

1 author:

Isabelle Oliveira
Boston University
32 PUBLICATIONS 35 CITATIONS

SEE PROFILE

All content following this page was uploaded by Isabelle Oliveira on 19 December 2022.

The user has requested enhancement of the downloaded file.

 Optimized Multi-Element Lens Design

I developed this study in Optimized Multi-Element Lens Design to meet the specific
demand in Optical Engineering.

For this project it is necessary to use the optical design program, the ZEMAX Lens
Design software.

Lens design is a strong component of comprehensive optics education and a skill valued
by industries that employ optical and technical engineers.

Technical information about optical elements is described by their surfaces, which

include surface curvature, thickness and average refractive index.

In the optimization software we can insert prescriptions, assign variables and the order
in which these and other tasks can be performed.

Top lens data editor, 3D Layout, Diagram are ZEMAX's extensive analysis tools to
assess project quality and meet your specifications.

Multi-element lens design, The Cooke Triplet, three elements to define the opening
stop, F#, merit function construction, optimization and configuration variables.

Layout, RMS Wavefront Error vs. Field position, Modulation Transfer Function (MTF)
and Transverse Ray Fan Plot are examples of modeling tools for a 2 channel system
using coordinate breaks and multiple configurations.

An example of a specific project that uses wavelength-sensitive dichroic beam splitters

to separate two bands in an optical system, sending each wavelength band through a
different optical path, perpendicular to each other uses the lens data editor which is
controlled by multiple editor configuration.

In the multi-configuration editor, the values of the parameters of the first configuration
of the wavelength, curvatures of the lens surface, thicknesses and the optical element,
materials are displayed.

Thermal analysis of the system is done using multiple configurations in optical

modeling.

An example in the multi-configuration editor represents the system prescription for the
temperature range from 0 °C to 40 °C.

The first nominal configuration is modeled for a wavelength of 656nm at 20 °C and a

pressure of 1 atm.

The third and fourth configurations show the thermal effect on the optical system, at
656nm, when the temperature is 0 °C or 40 °C.

The second configuration is nominally modeled for a wavelength of 486nm at 20 °C

and a pressure of 1 atm.
The fifth and sixth configurations show the thermal effect on the optical system, at
486nm, when the temperature is at 0 °C or 40 °C.

Merit functions are optical modeling skills important to the project.

The merit function editor shows the specifications given for this system to obtain
diffraction-limited images for each wavelength path and have equal magnification for
both channels at the specified temperature.

This function of merit described was built to solve the problem of athermalization of the
optical system of the example mentioned.

Operand #3 controls the focal length of the system for the 656nm wavelength optical
path as defined in the multi-configuration editor.

Operand #10 blocks the position of the image plane for the marginal rays of optical
paths, 656nm and 486nm, at the nominal temperature of 20 °C, so that for the same field
angles in each wavelength path, the images of both settings have the same
magnification in each respective image plane, except that one is inverted.

Operands #20, 21, 30, 31, 32 and 33 control the athermic behavior of the optical system
for both channels.

The standard merit function follows operand #34.

The performance of the system is assessed using the Monte Carlo statistical tolerance
analysis in which each optical surface, thickness, material and optical element is
assigned a tolerance range.

The result of the tolerance analysis is that the appropriate tolerances are chosen so that
the elements of the system can be manufactured, assembled and aligned to meet the
general performance specification of the optical system during testing.

An example to perform the Monte Carlo tolerance statistical analysis, in the typical
tolerance data editor, some of the operands are:

• #1 the trim on the rear focus,

• #2 the test wavelength,

• #3-6 are for lightning tolerances.

Tolerances on surfaces and the positioning of the element are also included in the
tolerance data editor.

The worst performance point diagram of the system for 9 field angles for the system is
derived from the Monte Carlo Statistical Tolerance Analysis tool.

The optical systems of the software allow to model and control the propagation of
Gaussian beams of lasers and optical fiber based on optical systems.
Like a 16X Galilean beam expander, whose total irradiance from the propagation of
physical optics, the graph shows the distribution of irradiance and the beam size in the
surface test, the size and the position of the waist.

Integration of the optical design in the optomechanical model of the instrument evolves
during the Monte Carlo statistical tolerance analysis, where a material can be chosen for
the base plate and lens cases, applying an appropriate thermal expansion coefficient
(TCE) in the ZEMAX model in the spaces between optical components.

In Lens Data Editor, surfaces are the air gaps that will be affected by the thermal
environment between the optical elements.

The integration of ZEMAX engineering tools and SolidWork software significantly

increases the efficiency of the optics, instrument design process, reducing human error
and cost.

Upon completion of the system design approach and the associated analysis completed;
how athermalization and Monte Carlo statistical tolerance analysis; the project plan will
indicate:

• the prescription for the individual optical components with the tolerances,

• the tolerances of the optical system,

• the material and tolerances for assembling optical components,

• the alignment strategy based on the compensators chosen in Operand 1.

The ray tracing of the optical system with all optical components is exported from
ZEMAX as a STEP file for 3D SolidWorks.

The top-level assembly is shown with optical components imported from ZEMAX.

The lens holders are designed around the lenses with the appropriate tolerances derived
from Monte Carlo's statistical tolerance analysis.

The compensators are derived from this statistical tolerance analysis and are performed
on templates controlled by fine-tuning screws and spring plungers.

Lens design, optical system design and optomechanical system design use optimization
routinely in their daily practice.

Local optimization algorithms lead to the closest minimum.

For comprehensive research on lens architecture, the optimization software offers two
options for multi-extremum optimization: Global and Hammer.

They provide a number of solutions, depending on the designer's choice as a starting

point.
Both Global and Hammer optimization options are stochastic and cannot guarantee the
integrity of the result.

The multi-extremum optimization of optical architectures of moderate complexity are

located within extended merit function vouchers.

The minima of the merit function are separated by saddle points.

An algorithm traverses these valleys from a local minimum through a saddle point to
another minimum.

From that new minimum, a new valley is found, which passes through another saddle
point to another minimum and so on.

In a finite number of steps, a complete and mutually connected system of stationary

points, minimum points and saddle is revealed, guaranteeing the conclusion of the
research.

The Hammer optimizer is used to improve a lens that has already been optimized by the
local optimizer.

Hammer runs an optimization problem, when you are working on something else.

The hammer can be left to work overnight, on weekends or even weeks, if necessary.

Optimization objective is the unique characteristics of the merit function scenario.

As an example, a ZEBASEV Kfi002 20x microscope objective for optimization.

The optical prescription is the Schott type of the 2019 catalog.

The eight optimization parameters are marked with the “v” symbol - six spokes and two
air gaps.

In order to maintain the lens' manufacturing capacity, avoid thin edges and negative air
spaces and other problems, the optimization parameters have the following restrictions:

• abs (R2) > = 10.0mm,

• abs (R3) > = 7.7mm,

• abs (R4) > = 11.0mm,

• abs (R5) > = 8.0mm,

• abs (R7) > = 4.0mm,

• abs (R8) > = 3.0mm,

• 4.0mm < T6 < 8.2mm,

• T9 > = 0.3mm.

The opening stop, entry pupil is located on surface 1.

The diameter of the entrance pupil is 8mm and the objective operates with F# = 1.02 in
the image space.

Point diagrams in three image heights: -0.4mm, object height 8mm, -2mm, object height
4mm and 0mm.

To reduce the computational load associated with ray tracing derivatives, the test is
optimized on a single 0.xn--587m-nnd wavelength.

To estimate image quality, each beam consists of 79 rays, 5 rays in the pupil radius for
each field point drawn by ray.

The optimization criterion, C, is the sum of the square side aberrations plus the
restriction violation penalty function:

C = Σ {Σ [(79 = 1 - 00 ) ^ 2 + ( - 00 ) ^ 2]} + ;

Where:

• m is the beam number,

• coordinates of the x-ray in the image plane;

• yi - y coordinates;

• x00m - paraxial coordinate of the image x (for all beams x00m = 0.0mm);

• y00m - paraxial coordinates of the image y in the image plane (y000 = 0.0mm);

• y001 = - 0.2mm and y002 = - 0.4mm).

Criterion C depends on the side aberrations and the chain extension.

P is a penalty function that is zero within the area of restrictions and increases in the
event of a violation of restrictions.

For lightning:

• P = 0.0 within the restrictions;

• P = 0.25 * (RIConstraint - RI) ^ 2 outside the restrictions (3a).

For air gaps:

• P = 0.0 within the restrictions;

• P = (TIConstraint - TI) ^ 2 external restrictions (3b);

• RIConstraint - constraint radii;

• TIConstraint - restrict thickness.

For the ZEBASE Kfi002 lens, C = 6.08E - 4, indicating that the lens is well optimized.

Attempts to make further improvements with local ZEMAX, DSL or orthogonal descent
optimizations result in oscillations around this point without noticeable criteria
improvements.

A quadratic model of C in the vicinity of the starting point can help reveal the reason for
such local optimization behavior.

First and second derivatives of the criterion function are calculated using the finite
difference method.

The use of rays in optimization makes it impossible to overcome the signal barrier.

Curves and air gaps are used as optimization parameters.

The quadratic model of criterion C is:

C (xi) = C0 + gT (xi) * (∆xi) + 0.5 * (∆xiT) Q (∆xi);

Where:

• gT is the transposed vector of the first derivatives;

• ∆xi - vector of parameter increments;

• Q is the Hessian matrix of the second derivatives.

The Hessian matrix has diagonal symmetry, and its linear algebra states that the
rotations of the coordinate system make the matrix Q diagonal or:

C (ui) = C0 + Σ ∗ ∆ui8 = 1 + 0.5 ∗ ΣEi ∗ (∆ui) ^ 2

Where ∆ui parameters are incremented in the rotated coordinate system, gui are the first
derivatives in the rotated coordinate system and the values Ei from the Hessian matrix.

The last two eigenvalues are very small, indicating that the Hessian matrix is poorly
conditioned.

Among the 6 variables in the rotated coordinate system, the criterion function will be a
narrow parabola of rapid growth and over the last two variables, the criterion landscape
is some kind of valley that changes slowly.
Optimization strategy to operate in the context of confusing merit functions, in the
vicinity of each minimum is surrounded by surfaces of equimagnitude.

The surfaces of equimagnitude are kept to a minimum.

At some point S, with the value of criterion CS, the surfaces of equimagnitude will
meet.

If we go from normal to the surfaces of equimagnitude at point S in both directions, we

will enter the surfaces of equimagnitude surrounding the minimums.

The criterion value will be less than CS for any small step on both sides.

It is possible only if the gradient at point S is zero and the e1 eigenvalue parallel to
normal is negative.

At point S, both surfaces of equimagnitude have a common tangential hyperplane Ω.

All points on this hyperplane are located outside the surfaces of equimagnitude CA and
CB and have a criterion value greater than CS.

The eigenvalues in this hyperplane are positive.

At point S the first derivatives are zero, a negative eigenvalue while the others are
positive.

This point is the Morse type saddle point.

The saddle points are separating the attraction areas from the neighboring minimums.

The special saddle point roller in the stationary point network is displayed.

A gradient curve is a curve orthogonal to the surfaces of equimagnitude at any point.

The ASB gradient curve follows through the saddle point S.

All other gradient curves connecting the minima A and B, curve Ψ1 will leave the
surfaces of equimagnitude CA and CB and enter areas with criteria greater than CS.

The ASB curve has the lowest maximum criterion value in addition to the other gradient
curves connecting the minimum A and B.

Values of function of merit, criterion, are those paths of growth, descent, slower.

The gradient curve Ψ2 originating from the minimum A instead of ASB will not
connect the minimums.

Merit function vouchers (ASB) are the only reliable paths from minimum A to
minimum B.
Local optimization will take you to some point at the bottom of the valley closest to the
bottom of the valley until reaching the local minimum.

More trips over the valley floor lead to the saddle point that will lead to the next
minimum, where optimization will pave the valley to a new saddle point.

Upon reaching the restraint surface, the minimum closest to the surface will be
searched.

If this minimum is separated from the previous one with the saddle point, it will be an
entrance to a new valley on the way back to the optimization space, where the new
valley will lead to a new minimum.

Multi-extremum optimization should be done on the surface of the constraint and all
new vouchers investigated.

The journey over the valley floor will be carried out in repeated cycles, where each
cycle begins with the DLS correction for the valley floor.

In step DLS, criterion C0 will be calculated using raytracing, the vector of the finite
difference technique g of the derivatives and Hessian matrix Q of the Equation.

Then, using the MATLAB eigen function, where the eigenvectors Vi and eigenvalues
Ei will be calculated.

The DLS step will be applied to the first 6 orthogonalized parameters and will not be
applied to the last 2 reliable parameters, which will lead to the closest point at the
bottom of the valley, where the derivatives of the strong variables must be zero (vertex
of the fast parabolas), but the derived from weak variables with small eigenvalues may
have some small value (slow growth or decline).

The next operation will be a step on the proper vector providing the increment of lower
criterion that can be negative indicating descent to the minimum or it can be positive,
indicating a trip over the valley towards the saddle point.

The optimization cycle will be repeated, paving the way over the network of stationary
points.

The saddle point detection method (SPD) is where several arbitrary directions from the
local minimum are chosen.

In each direction a step will be given the minimum criterion in the orthogonal
hyperplane and the chosen direction will be found and a new stage will follow the
minimum in the hyperplane.

These minimums constitute the SPD curve and the maximum of the criterion on
successful SPD curves will be the saddle point.
 The results of the multi-extremum optimization of the microscope objective
demonstrated the efficiency of the proposed algorithms that are associated with
computer overload and cannot be recommended for lens designers for everyday use.

But the further progress in the computer's CPU clock speed, coupled with the
implementation of parallel multicore processing, will make it possible to use this global
optimization method in commercial lens design software.

Did you like this abstract?

This is an example of a project that can be made especially for your demand and may
contain: Methods, materials, results, risk, bibliography, graphics, metaphors, analogies,
article references and further reading.

You can have a complete article to publish in PDF for as little as USD 40.00.

No need to pay VAT (Brazil) or tax.

https://www.linkedin.com/pulse/optimized-multi-element-lens-design-isabelle-oliveira/
https://www.youtube.com/watch?v=9yt5KXBd-5Y

View publication stats