Presbyopia Correction,

Differential Geometry,
and Free Boundary PDEs

Sergio Barbero and Marı́a del Mar González

We survey some mathematical tools involved in the de- correction. After giving a brief background in geometrical
sign of multifocal surfaces which are used in presbyopia optics, we consider the Willmore functional in differential
geometry, summarizing some classical results, and then
Sergio Barbero is an optics researcher at Instituto de Óptica from CSIC (Spain). explaining its significance in this particular optical appli-
His email address is [email protected]. cation.
Marı́a del Mar González is a professor of mathematics at the Universidad
Autónoma de Madrid (Spain). Her email address is mariamar.gonzalezn 1. A Basic Introduction to Geometrical Optics
@uam.es.
Modeling precisely how light propagates in a physical
Communicated by Notices Associate Editor Daniela De Silva.
medium has been, and still is, one of the great challenges
For permission to reprint this article, please contact:
of mathematical physics. Fortunately, in many cases of
[email protected].
practical interest, light propagation can just be modeled
DOI: https://doi.org/10.1090/noti2828

DECEMBER 2023 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 1763

by means of energy transported along rays; this is the crux Optical power. In any case, under the so-called paraxial ap-
of the so-called geometrical optics theory. Beyond geometri- proximation, i.e., when the sine of the angle is approxi-
cal optics, one has the wave optics model, a more sophis- mated by the angle itself in the refraction law, the imag-
ticated and accurate theory of light propagation which we ing system’s capacity of transforming wavefronts depends
will not consider here. For the interested reader, a classical on the radii of curvature of both the incoming (𝑟 𝑖 ) and the
text in optics is [BW65]. outgoing (𝑟𝑜 ) wavefronts by means of the vergence equation:
In general, rays are space curves in the Euclidean space,
𝑛1 𝑛
which obey a stationary action principle; namely, the = 𝑃 + 2,
𝑟0 𝑟𝑖
trajectories are stationary solutions of path length inte-
grals weighted by the refractive index of the propagating 𝑃 being the optical power of the imaging system, and 𝑛1
medium (this refractive index is related to the microscopic and 𝑛2 the refractive indexes at the incoming and outgoing
electromagnetic properties of the material of which the mediums. In consequence, if we have object points located
medium is made). Note that rays are straight lines when at different distances from the imaging system, its optical
the refractive index is constant. power must change accordingly.
Image formation. Within the geometrical optics framework, The optical power of the imaging system depends on
given a bundle of light rays generated by a punctual source, the geometrical properties of the optical element surfaces
there exists a normal surface to all the rays containing a and the refractive indexes of the mediums at each side (de-
fixed point, even if the bundle of rays has suffered multiple noted by 𝑛1 , 𝑛2 ). For instance, in the simple case of a single
refractions or reflections (Malus–Dupin theorem). This reflective/refractive spherical optical surface 𝑆, its optical
surface is called the wavefront. Therefore, light propagation (𝑛 −𝑛 )
power is 𝑃 = 2 1 , where 𝑟 𝑆 is the radius of curvature.
𝑟𝑆
can be modeled by propagating either rays or wavefronts.
Another simple example is given by traditional zoom cam-
Rays emanating from a point source, within a homoge-
era systems, which change the optical power through ax-
neous refractive index medium, form a spherical wavefront
ial distance shifts between two lenses. Our eyes, however,
which can be transformed after refraction or reflection by
change the optical power, when looking at near distances–
an optical element.
for instance, when reading a book—by means of bending
However, in practice, real objects are not point objects.
the crystalline lens with the help of the ciliary muscle, thus
We owe Ibn al-Haytham (c. 965–c. 1041), latinized as Al-
modifying 𝑟 𝑆 (see the scheme of a human eye in Figure 1.)
hacen, the idea that an extended object can be represented
by the union of multiple points and thus, from every point
of the surface of an object, light is emitted in all directions
in the form of rays. These rays can, for instance after cross-
ing a lens, converge onto another point, which is called fo-
cus. Following Alhacen, we say that a set of focal points is
the image of an object if rays emerging from each point of
the latter converge onto points of the former. Notice that
the focus is the only geometrical light concept we actually
see or measure; rays and wavefronts are merely convenient
mathematical tools.
Then, an imaging system is a set of lenses, mirrors, or
other types of optical elements that transform rays, and
hence wavefronts, onto converging rays/wavefronts that fo-
cus on a light detector—such as a camera sensor or the Figure 1. Scheme showing the optical elements of the human
retina of the human eye—forming a set of images of the ob- eye. Two lenses: the cornea and crystalline lens form images
ject’s points. Ideally, in order to have a “perfect image” of on the retina. The human eye sees nearby objects by means
an extended object, each spherical wavefront generated by of increasing the optical power of the crystalline lens.
a point of the extended object should be transformed, by
the imaging system, into an outgoing spherical wavefront. Presbyopia. When we get old we lose the capacity to bend
Such a system is called an absolute instrument. However, our crystalline lens because of the increase in the stiffness
as already noted by James Clerk Maxwell in 1858 [Max58], of the crystalline lens. This phenomenon is called presby-
there are some limitations to the existence of absolute in- opia. Then, as shown in Figure 1, our eyes (formed by two
struments and, indeed, forming an image of an extended lenses: the cornea and the crystalline lens) cannot achieve
object with depth into a planar light detector is a specially the required radius of curvature of the outgoing wavefront
difficult task. associated with near vision. The result is that, at the retina,

1764 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 70, NUMBER 11

instead of having an image point we perceive a blurred
disk. Of course, it is possible to correct that using reading
eyeglasses. However, these are monofocal, meaning that
they only provide a single optical power and entail shift-
ing between spectacles for far (in case of being myopic or
hypermetropic) and near vision tasks. An alternative to
that nuisance is to use multifocal instruments.

2. Multifocality
Multifocality is the property of a surface—which could be
a wavefront or an optical element—of providing light in-
tensity distributions, not concentrated around a single im-
age point (monofocality) but, instead, distributed along
different foci or an extended region. While a spherical sur-
face produces a single focus, multifocality requires a sur-
face with nonconstant curvature.
Geometrically, at any point of a smooth surface 𝑆 in Eu-
clidean space there are two principal curvatures 𝜅1 and 𝜅2 ,
which are the maximal and minimal normal curvatures,
respectively. Hence it is reasonable to take the mean curva-
ture 𝐻 = 𝜅1 + 𝜅2 as an average measurement of the inverse Figure 2. The effect of sunlight on clear sea water.
of 𝑟 𝑆 (radius of curvature) overall normal planes. Another
important quantity is the absolute difference between prin-
cipal curvatures 𝐶 = |𝜅1 − 𝜅2 |, known as cylinder or astig-
matism, which quantifies, at second order approximation,
the blurring of the image point due to that difference in
ray trajectories for each normal plane.
A caustic digression. Caustics, the geometrical location of
light energy concentration, is an ancient concept in geo-
metrical optics because of the caustic (hence the name)
properties of some burning lenses and mirrors known
since antiquity. Caustics are ubiquitous and provide us
with beautiful patterns (see Figure 2).
We underline here the work of Leonardo Da Vinci who,
fascinated with light as he was, drew the caustics generated
by a circular mirror (these are known as catacaustics): see
Figure 3 for the original drawings and Figure 4 for a real Figure 3. Da Vinci’s drawing of a reflected caustic by a
life example. spherical mirror.
For a planar curve, Tschirnhausen defined the caustic
curve as its evolute or the envelope of the normal family of concentration in the regions of interest. However, the two
rays to the original curve. In the case of having a wavefront caustic surfaces only coincide when the associated princi-
surface, the analog of the evolute of a plane curve is the so- pal curvatures are equal (𝜅1 = 𝜅2 ), i.e., at umbilical points
called focal set; then the caustic can be defined as the geo- where the cylinder vanishes.
metric locus of centers of principal curvatures of the wave-
Minkwitz theorem. It is known that, within a smooth Eu-
front [KO93]. Since there are two principal curvatures,
clidean surface, umbilicity is only possible either at iso-
each wavefront generates two caustic sheets. From a Rie-
lated points or lines. Günter von Minkwitz, a young
mannian geometry perspective, caustics are the set where
mathematician working on progressive addition lenses, de-
rays (geodesics in the optical Riemannian space) concen-
duced in 1963 what came to be known as Minkwitz’s the-
trate, i.e., the cut locus, the location where more than one
orem [Min63], which establishes that if a smooth surface
minimizing geodesic arrives.
contains an umbilical line, then the cylinder along the or-
Ideally, in a perfect multifocal wavefront, the two caus-
thogonal direction increases twice as quickly as the growth
tic sheets should coincide and degenerate into a set of
rate of the mean curvature along the umbilical line. The ef-
points, or a line, because this provides maximum intensity
fect of Minkwitz’s theorem is illustrated in Figure 5, which

DECEMBER 2023 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 1765

 Figure 6 illustrates this principle: of all rays (green ar-
row) emerging from a far object (butterfly) only those pass-
ing through the central part of the contact lens, converge
after refraction onto the retina. On the other hand, of the
rays coming from a near object (book) only those (red ar-
row) passing through the peripheral part of the contact
lens converge onto the retina, after passing through the
contact lens and the eye. This is possible because the cen-
tral and peripheral parts of the contact lens are spheres of
different radii of curvature: 𝑅1 and 𝑅2 respectively. Thus
the main question is how to design the transition zone be-
tween both spheres while keeping astigmatism under con-
trol.

Figure 4. Caustics in a cup of milk.

Figure 6. Scheme showing how a multifocal intraocular or

contact lens works (in this case a contact lens). Rays from a
far object (green arrow emerging from the butterfly) that pass
through the central part of the contact lens, where the radius
of curvature is 𝑅1 , converge onto the retina. On the other
hand, rays from a near object (red arrow emerging from the
book) only converge onto the retina if they pass through the
peripheral part of the contact lens because, there, the radius
of curvature (𝑅2 ) is different from that of the central part (𝑅1 ).

Figure 5. Mean power and cylinder of a real PAL surface.

In PALs, the latter solution (B), a surface is designed
containing a spatially varying mean curvature. Contrary to
shows the mean curvature and cylinder distribution across contact and intraocular lenses, in PALs, the eye can move
a circular domain of a real PAL surface [RY11]. independently of the lens, so through gaze movements it
Recently, a generalization of Minkwitz’s theorem has chooses, sequentially on time, each viewing area of the
been derived that provides the exact magnitude of the PAL providing the required optical power for: far distance
cylinder at any arbitrary point as a function of the geodesic vision (center of the lens) and near-view (low nasal por-
curvature and the ratio of change of a principal curvature tion). Between these two zones, the power varies progres-
along a principal line [BdMG20]. sively. Figure 7 shows this.
Presbyopia correction: A multifocal approach. There are two An ideal progressive surface is one with the prescribed
ways of applying multifocality for presbyopia correction: smooth progressive power (mean curvature) and with zero
A. Multifocal intraocular or contact lenses. astigmatism (cylinder) everywhere. But, as we have seen
B. Progressive addition spectacles (PALs, for short). above, some amount of astigmatism is unavoidable unless
the surface is a plane or a sphere, which does not provide
The former solution (A) is based on generating an out- progressive power. In any case, an archetypical model of
going wavefront with a varying mean curvature which, for a PAL is that of an elephant’s trunk (see Figure 8), where
different viewing distances, provides superimposed im- there is an umbilical line (red solid line) and the circles
ages on the retina. Then, the neural system is capable of obtained by cutting the trunk with parallel planes (dashed
selecting the correct focus at each moment; this is called
the simultaneous vision principle.

1766 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 70, NUMBER 11

 Figure 8. An elephant trunk is a primitive type of PAL surface
model with an umbilical line.

Willmore energy as
1
𝒲= ∫ 𝐻 2 𝑑𝑎, (3.1)
4
𝑆

where 𝐻 = 𝜅1 + 𝜅2 is the mean curvature of 𝑆 and 𝑑𝑎 the

Figure 7. Scheme showing how the eye changes its gazes (top surface area differential. The factor 1/4 in front is just a
figure) when using different parts of the PAL surface (bottom normalization constant.
figure). 𝐻1 and 𝐻2 denote mean curvatures for far and near
vision, respectively.
One can also consider the related functional

˜ = ∫( 1 𝐻 2 − 𝐾) 𝑑𝑎,
𝒲
4
𝑆

blue lines) exhibit reducing radii of curvature when mov- for 𝐾 = 𝜅1 𝜅2 the Gauss curvature of the surface. Remark
ing from the upper to the lower part of the trunk. that an equivalent expression is given by
In general, to obtain designs for both solutions, (A) and 1
˜=
𝒲 ∫(𝜅1 − 𝜅2 )2 𝑑𝑎, (3.2)
(B), one should consider the combined action of the eye 4
𝑆
and the artificial lenses [Rub17]. However, since multifo-
cality is usually restricted to a single surface, both in in- which is the relevant formulation for our application to
traocular, contact, and PALs, the mathematical study can the design of multifocal surfaces as we are interested in
be, as a reasonable approximation, restricted to a single minimizing astigmatism (cylinder). Indeed, we can think
surface. Then, one may use the calculus of variations and ˜ as a measure of the deviation of 𝑆 from
of the functional 𝒲
PDEs to give optimal designs in both settings (A) and (B). ˜
being a sphere (note that, if 𝑆 is exactly a round sphere, 𝒲
Yet, there is a crucial difference between solutions vanishes.)
(A) and (B). In the former case (intraocular and contact Taking into account the Gauss–Bonnet formula
lenses), axial rotational symmetry is, in most cases, ap-
plied, which can be mathematically exploited. Then, the ∫ 𝐾 𝑑𝑎 = 2𝜋𝜒(𝑆),
𝑆
problem is finding a surface that interpolates between two
mean curvature values (for adjusting far and near vision, which relates the Euler characteristic 𝜒(𝑆) of the surface to
respectively) and minimizes astigmatism. On the other its total Gauss curvature, one sees that
hand, for PALs one instead introduces some weights in the 1
variational functional to include this spatial dependence. ˜=
𝒲 ∫ 𝐻 2 𝑑𝑎 − 2𝜋𝜒(𝑆).
4
𝑆
Both approaches will be explained in Section 4.
Since the last term in the formula above is a topological in-
3. The Willmore Functional variant, in minimization problems for closed surfaces both
The research of Thomas J. Willmore (1919–2005) was fun- 𝒲 and 𝒲 ˜ are equivalent.
damental in the understanding of the following functional, An important property of the Willmore functional is its
and this is the reason it bears his name [Wil93]. Let 𝑆 be conformal invariance (recall that conformal map in ℝ3 is a
a smooth, closed, embedded surface 𝑆 in ℝ3 . We define transformation of space that locally preserves angles.) To

DECEMBER 2023 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 1767

prove this statement, one observes that any such confor- with equality if the generating curve is a circle and the ratio
mal transformation can be decomposed into a composi- of the radii is 1/√2, which corresponds to the parametriza-
tion of translations, dilations, and inversions. The precise tion
calculation is due to White [Whi73] although the result
was known to Blaschke in 1929. 𝑆 0 ∶ ((√2 + cos 𝑢) cos 𝑣, (√2 + cos 𝑢) sin 𝑣, sin 𝑢),
The Willmore functional in elasticity. Describing elasticity in for 𝑢, 𝑣 ∈ ℝ. Note that 𝑆 0 can be conformally mapped to
terms of geometry is an old subject. In this regard, we un- the Clifford torus.
derline the works of Sophie Germain and Siméon D. Pois- Willmore conjectured in 1965 that this result could be
son in the 19th century, in which they modeled the bend- generalized to all surfaces in ℝ3 . However, this conjecture
ing energy of a thin plate as the integral with respect to the remained open until the seminal paper of Marques and
surface area of an even, symmetric function of the princi- Neves in 2014:
pal curvatures (note here that the Willmore energy is the
simplest possible example). Theorem 3.1 ([MN14a]). Every embedded compact surface
A classical generalization of the Willmore functional 𝑆 in ℝ3 with positive genus satisfies (3.3). Up to rigid motions,
comes from the theory of membranes. Indeed, most of the equality holds only for stereographic projections of the Clif-
the cell membranes of living organisms are made of a lipid ford torus.
bilayer, modeled as a surface minimizing the Canham–
Finally, a classical result of Li–Yau [LY82] states that
Helfrich bending energy, which is essentially
when an immersion Ψ ∶ 𝑆 ↪ ℝ3 has multiplicity 𝑘 (i.e.,
𝑘 points in 𝑆 get mapped to the same point by Ψ), then
∫ 𝑐(𝐻 − 𝑐 0 )2 𝑑𝑎,
𝑆 𝒲(Ψ(𝑆)) ≥ 𝑘 ⋅ 4𝜋.
where 𝑐 and 𝑐 0 are given constants. This approach explains,
for instance, the biconcave discoid shape of red blood cells. This settles the Willmore conjecture for all surfaces. For
(See [ZCOY99] for an exhaustive discussion on this topic.) further insight in this topic and additional references, we
refer to the beautiful survey [MN14b].
A variational point of view. Going back to the geometric
problem, the first question one may ask is how to find Willmore surfaces. The Euler–Lagrange equation for 𝒲 is
the infimum of the Willmore energy 𝒲 over the space of the fourth order quasi-linear elliptic PDE
closed surfaces 𝑆 in ℝ3 . Willmore himself showed that the 1
Δ𝑆 𝐻 + 2𝐻( 𝐻 2 − 𝐾) = 0, (3.4)
sphere is a minimizer; more precisely, 4

𝒲(𝑆) ≥ 4𝜋, where Δ𝑆 is the Laplacian on 𝑆. Solutions to this equation

are known as Willmore surfaces, even if not necessarily min-
among all such 𝑆, with equality if and only if 𝑆 is the round imizers.
sphere. The proof of this result is quite simple in geometric This is a highly nonlinear fourth-order PDE (see
terms: if we lay the surface on the floor, the point where 𝑆 [GGS10] for an introduction to fourth-order boundary
touches the ground (so the normal vector is vertical) must value problems.) An important remark here is that these
be elliptic, that is 𝐾 = 𝜅1 𝜅2 ≥ 0. But now we can repeat equations do not usually enjoy many of the nice properties
this process in any direction. So, for any direction of the of second-order elliptic problems:
normal, we can find a point on the surface 𝑆 such that 𝐾 ≥ • The first issue is that we do not have a priori bounds
0. This means that the image of the Gauss map 𝑁 ∶ 𝑆 → for the regularity of the solution.
𝕊2 covers the whole sphere 𝕊2 and, moreover, since the • The second point is the lack of a comparison princi-
Jacobian of the Gauss map is precisely the Gauss curvature, ple that says that any subsolution of the equation stays
we have that below any supersolution. For second-order equations
∫ 𝐾 𝑑𝑎 ≥ ∫ 1 = 4𝜋. with a variational formulation, the classical trick to
𝑆∩{𝐾≥0} 𝕊2 prove a maximum principle is to look at 𝑢+ , 𝑢− (pos-
1 itive and negative parts of 𝑢, respectively). However,
To conclude, just note that 𝐻 2 ≥ 𝐾, with equality iff 𝜅1 = these test functions do not have enough regularity in
4
𝜅2 . a fourth-order equation, even in a weak sense.
Willmore’s next question was to find a minimizer in a
class of fixed topology, for instance, in the class of tori. He 4. Practical Applications of Willmore-Type
proved, in particular, that any torus generated by a small Functionals
circle moving (perpendicularly) along a closed plane curve In the previous section, we considered closed surfaces, that
𝐶 fulfills is, without boundary. However, lenses have boundaries!
𝒲(𝑆) ≥ 2𝜋2 , (3.3) So we should let 𝑆 be a surface with boundary. The natural

1768 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 70, NUMBER 11

generalization is, thus, to replace the functional (3.1) by: we need to impose that this interpolation solution is
1
smooth up to second derivatives at each of the two bound-
𝒲𝜕 = ∫ 𝐻 2 𝑑𝑎 + ∫ 𝜅𝑔 𝑑𝑠, (4.1) aries (the two contact lines, let us denote them by ℓ1 and
4
𝑆 𝜕𝑆
ℓ2 , respectively). This imposes three boundary conditions
where 𝜅𝑔 is the geodesic curvature of the boundary curve at each ℓ1 and ℓ2 , which make equation (3.4) overdeter-
and 𝑑𝑠 the arclength. Some existence results, in the spirit mined and, as a consequence, one cannot prescribe ℓ1 and
of those of the closed case, are considered by [Sch10]. ℓ2 in advance. Remark that we should not expect higher
In the following, we will consider two direct applica- regularity than second derivatives since this is a fourth-
tions of the Willmore functional: designing a PAL surface order obstacle-type problem. For the interested reader, we
and a multifocal intraocular (or contact) lens. refer to a 2020 Notices paper by D. Danielli [Dan20] on
Modified Willmore functionals in PALs design. The Willmore this topic.
functional has a relevant industrial application in PAL de- In general, such a 𝑆 is called a free boundary Willmore
sign. On a PAL surface, one can first prescribe a desired surface. A free boundary problem is PDE (or ODE) de-
mean curvature distribution within the domain of the sur- fined on a domain whose boundary is a priori unknown
face to be designed, then measure the difference with the and it has to be determined from the solution of the prob-
true mean curvature, and finally minimize a functional lem a posteriori. The archetypal example of a free bound-
where both the mean curvature difference and the astig- ary model is that of understanding the interface between
matism are weighted [KR99, WGS03]. ice and water when studying the heat equation (Stefan prob-
More precisely, let us parameterize the surface as a graph lem).
of a function 𝑢 ∶ Ω → ℝ over a domain Ω in ℝ2 . We Unfortunately, the free boundary Willmore problem
denote the desired mean curvature distribution by 𝐻0 ∶ has not been solved yet in full generality. A partial answer
Ω → ℝ. One tries to minimize can be given if one assumes additional symmetries.
𝐼(𝑢) = ∫ {𝛼(𝑥, 𝑦)(𝜅1 − 𝜅2 )2 The one-dimensional problem. Assuming translation invari-
Ω (4.2) ance, one can reduce to finding a curve Γ that minimizes
2
the curvature functional for
+𝛽(𝑥, 𝑦)(𝐻 − 𝐻0 ) } √1 + |∇𝑢|2 𝑑𝑥𝑑𝑦,
𝒲(Γ) = ∫ 𝜅2 𝑑𝑠,
where 𝛼 and 𝛽 are prescribed weights. In fact, in the de- Γ
sign of ophthalmic lenses (recall Figure 7), certain regions where 𝜅 is the curvature of Γ and 𝑑𝑠 the arclength. Such
on the surface are required to have very small astigmatism Γ is called an elastic curve, and satisfies the Euler–Lagrange
(where we impose 𝛼 to be large), while in other regions its equation
more desirable that the lens has the correct power (where 1
𝜅𝑠𝑠 + 𝜅3 = 0 on Γ. (4.3)
we set 𝛽 large). Nevertheless, this technique involves pre- 2
scribing a varying mean curvature on the whole surface, If the curve is given by the graph of a function 𝑢 ∶
which may not be the optimal solution. [0, 1] → ℝ, then equation (4.3) reduces to the nonlinear
A relevant theoretical problem associated with both the ODE
functional (4.2) and Willmore energy, is whether it is pos- 1 𝑑 𝜅′ 1
( ) + 𝜅3 = 0 in (0, 1), (4.4)
sible to find its minimum value on open surfaces, as we √1 + 𝑢′2 𝑑𝑥 √1 + 𝑢′2 2
did for closed surfaces in Section 3. This would offer a
where
kind of Minkwitz theorem in the large. Knowing a lower 𝑢″ (𝑥)
bound for the amount of total cylinder achievable by a 𝜅(𝑥) = 3
.
PAL surface would offer useful insights to an optical de- (1 + 𝑢′ (𝑥)2 ) 2
signer about how close his/her PAL design is to an ideal There are various sets of boundary conditions that one may
one. impose. Since we are interested in controlling the curva-
A free boundary problem for Willmore surfaces. As mentioned ture at the endpoints, let us consider Navier conditions:
in Section 2, in intraocular and contact lens multifocal 𝑢(0) = 𝑢(1) = 0,
{ (4.5)
applications, we are interested in the following geomet- 𝜅(0) = 𝜅(1) = 𝛼.
ric problem: find a Willmore surface 𝑆 that interpolates
In the symmetric case 𝑢(𝑥) = 𝑢(1 − 𝑥) one knows the
between two spheres 𝑆 1 and 𝑆 2 of different radius 𝑅1 and
following bifurcation picture (see Figure 9)
𝑅2 , respectively. Of course, one would like to have a min-
imizer of Willmore energy, but for now, let us be satisfied Theorem 4.1 ([DG07]). There exists 𝛼𝑚𝑎𝑥 = 1.34 … such
with a solution of the Euler–Lagrange equation (3.4). that for 0 < |𝛼| < 𝛼𝑚𝑎𝑥 , the problem (4.4) with boundary
For practical reasons (that is, having sharp jumps on a conditions (4.5) has precisely two smooth solutions in the class
lens may introduce undesired visual effects for the wearer), of smooth symmetric functions. If |𝛼| = 𝛼𝑚𝑎𝑥 there exists only

DECEMBER 2023 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 1769

one such solution, for 𝛼 = 0 one only has the trivial solution by
and no such solutions exist for |𝛼| > 𝛼𝑚𝑎𝑥 .
𝑢(−1) = 𝑢(1) = 𝛼,
The proof of this result goes back to an observation by {𝐻(±1) = 𝛾 (4.7)
, for 𝛾 ∈ (0, 1).
Euler in 1952. He found that such a solution must satisfy 𝛼√1 + 𝑢′ (±1)2
the conservation law
1 Note that the simplest solution is the circular arc 𝑢(𝑥) =
𝜅(𝑥)(1 + 𝑢′ (𝑥)2 ) 4 = 𝑐𝑠𝑡. (4.6) √𝛼2 + 1 − 𝑥2 for 𝛾 = 1. In general we have:

Theorem 4.2 ([BDF10]). For each 𝛼 > 0 and 𝛾 ∈ [0, 1],

there exists a positive, smooth, and symmetric function 𝑢 such
that the surface generated satisfies (3.4) with conditions (4.7).

Obviously, we are interested in the nonsymmetric case.

Unfortunately, there are still no results on the associated
Figure 9. Solutions of the Navier boundary value problem free boundary problem, despite its potential applications
from Theorem 4.1 for 𝛼 = 0.2, 𝛼 = 1, and 𝛼 = 1.34 (left to right).
Taken from [DG07].
in the optics field.

ACKNOWLEDGMENTS. We would like to thank Jacob

There are several other pairs of conditions one can im- Rubinstein and Hans-Christoph Grunau for their valu-
pose, for this we refer to the same paper [DG07]. In partic- able comments. S. B. and M. G. are supported by the
ular, the nonsymmetric case may be obtained by scaling, Spanish Government grant PID2020113596GB-I00.
translating, and rotating the coordinate system.
Now, going back at our initial question of interpolating
between two circles of different radii by a Willmore curve, References
it is clear that we need to impose boundary conditions on [BDF10] Matthias Bergner, Anna Dall’Acqua, and Steffen
𝑢, 𝑢′ , and 𝑢″ at both endpoints of the interval, ℓ1 and ℓ2 . Fröhlich, Symmetric Willmore surfaces of revolution satis-
A possible strategy is, leaving the contact points ℓ1 and ℓ2 fying natural boundary conditions, Calc. Var. Partial Dif-
free, to use the solution from Theorem 4.1 in order to also ferential Equations 39 (2010), no. 3-4, 361–378, DOI
10.1007/s00526-010-0313-7. MR2729304
match the first derivatives of 𝑢 at the endpoints.
[BdMG20] Sergio Barbero and Marı́a del Mar González, Ad-
Willmore surfaces of revolution. In this particular case, the missible surfaces in progressive addition lenses, Opt. Lett. 45
surface 𝑆 is obtained by rotating a curve in the 𝑥𝑦-plane (2020), 5656–5659.
around the 𝑥 axis. We assume that this curve is given by [BW65] Max Born and Emil Wolf, Principles of optics: Electro-
the graph of a function magnetic theory of propagation, interference and diffraction of
light, Third revised edition, Pergamon Press, Oxford-New
𝑢 ∶ [−1, 1] → (0, ∞) York-Paris, 1965. With contributions by A. B. Bhatia, P. C.
Clemmow, D. Gabor, A. R. Stokes, A. M. Taylor, P. A. Way-
which, in some cases, can be taken to be symmetric 𝑢(𝑥) =
man and W. L. Wilcock. MR198807
𝑢(−𝑥). The surface of revolution 𝑆 is then parametrized by [Dan20] Donatella Danielli, An overview of the obstacle prob-
𝑆ᵆ ∶ (𝑥, 𝑢(𝑥) cos 𝜃, 𝑢(𝑥) sin 𝜃), 𝑥 ∈ [−1, 1], 𝜃 ∈ [0, 2𝜋]. lem, Notices Amer. Math. Soc. 67 (2020), no. 10, 1487–
1497, DOI 10.1090/noti. MR4201882
After some calculation, we arrive at [DG07] Klaus Deckelnick and Hans-Christoph Grunau,
Boundary value problems for the one-dimensional Willmore
𝑢″ (𝑥) 1 equation, Calc. Var. Partial Differential Equations 30
𝐻=− 3
+ ,
(1 + 𝑢′ (𝑥)2 ) 2 𝑢(𝑥)√1 + 𝑢′ (𝑥)2 (2007), no. 3, 293–314, DOI 10.1007/s00526-007-0089-6.
MR2332416
𝑢″ (𝑥)
𝐾=− , [GGS10] Filippo Gazzola, Hans-Christoph Grunau, and
𝑢(𝑥)(1 + 𝑢′ (𝑥)2 )2 Guido Sweers, Polyharmonic boundary value problems: Pos-
which reveals the highly nonlinear character of (3.4). Un- itivity preserving and nonlinear higher order elliptic equa-
fortunately, conservation laws such as (4.6) are no longer tions in bounded domains, Lecture Notes in Mathematics,
vol. 1991, Springer-Verlag, Berlin, 2010, DOI 10.1007/978-
available in this case.
3-642-12245-3. MR2667016
There are many works that explore rotationally sym- [KO93] Yu. A. Kravtsov and Yu. I. Orlov, Caustics, catastro-
metric Willmore surfaces with boundaries, imposing dif- phes and wave fields, 2nd ed., Springer Series on Wave Phe-
ferent restrictions at the boundary. We recall some exis- nomena, vol. 15, Springer-Verlag, Berlin, 1999. Translated
tence results when natural boundary conditions (from the from the Russian by M. G. Edelev, DOI 10.1007/978-3-642-
variational point of view) are considered. These are given 59887-6. MR1723808

1770 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY VOLUME 70, NUMBER 11

[KR99] Dan Katzman and Jacob Rubinstein, Method for Credits
the design of multifocal optical elements, 1999, US Patent The opening image is courtesy of Fermate via Getty.
6302540B1. Figures 1, 4, 5, 6, 7, and 8 are courtesy of Sergio Barbero.
[LY82] Peter Li and Shing Tung Yau, A new conformal invari- Figure 2 is courtesy of Brocken Inaglory, CC-BY-SA 3.0.
ant and its applications to the Willmore conjecture and the
Figure 3 is courtesy of The British Library, folio 87r Codex
first eigenvalue of compact surfaces, Invent. Math. 69 (1982),
Arundel.
no. 2, 269–291, DOI 10.1007/BF01399507. MR674407
[Max58] J. C. Maxwell, On the general laws of optical instru- Figure 9 is courtesy of Klaus Deckelnick and Hans-
ments, Q. J. Pure and Appl. Maths (1858), no. 2, 233–247. Christoph Grunau, Boundary value problems for the one-
[Min63] G. Minkwitz, Über den ber den flchenastigmatismus bei dimensional Willmore equation, Calc. Var. Partial Differ-
gewissen symmetrischen asphren, Optica Acta: International ential Equations 30 (2007), no. 3, 293–314, https://
link.springer.com/article/10.1007/s00526
Journal of Optics 10 (1963), no. 3, 223–227.
-007-0089-6. Reprinted with permission.
[MN14a] Fernando C. Marques and André Neves, Min-
max theory and the Willmore conjecture, Ann. of Math. Photo of Sergio Barbero is courtesy of Natalia Barcenilla.
(2) 179 (2014), no. 2, 683–782, DOI 10.4007/an- Photo of Marı́a del Mar González is courtesy of Jose Pedro
nals.2014.179.2.6. MR3152944 Moreno.
[MN14b] Fernando C. Marques and André Neves, The
Willmore conjecture, Jahresber. Dtsch. Math.-Ver. 116
(2014), no. 4, 201–222, DOI 10.1365/s13291-014-0104-8.
MR3280571
[Rub17] Jacob Rubinstein, The mathematical theory of multi-
focal lenses, Chinese Ann. Math. Ser. B 38 (2017), no. 2,
647–660, DOI 10.1007/s11401-017-1088-3. MR3615509
[RY11] T. W. Raasch, L. Su, and A. Yi, Whole-surface character-
ization of progressive addition lenses, Optometry and Vision
Science 88 (2011), no. 2, E217–E226.
[Sch10] Reiner Schätzle, The Willmore boundary problem, Calc.
Var. Partial Differential Equations 37 (2010), no. 3-4, 275–
302, DOI 10.1007/s00526-009-0244-3. MR2592972
[WGS03] Jing Wang, Robert Gulliver, and Fadil Santosa,
Analysis of a variational approach to progressive lens design,
SIAM J. Appl. Math. 64 (2003), no. 1, 277–296, DOI
10.1137/S0036139902408941. MR2029135
[Whi73] James H. White, A global invariant of conformal map-
pings in space, Proc. Amer. Math. Soc. 38 (1973), 162–164,
DOI 10.2307/2038790. MR324603
[Wil93] T. J. Willmore, Riemannian geometry, Oxford Sci-
ence Publications, The Clarendon Press, Oxford University
Press, New York, 1993. MR1261641
[ZCOY99] Zhong-Can Ou-Yang, Ji-Xing Liu, and Yu-Zhang
Xie, Geometric methods in the elastic theory of membranes in
liquid crystal phases, World Scientific, 1999.

Sergio Barbero Marı́a del Mar

González

DECEMBER 2023 NOTICES OF THE AMERICAN MATHEMATICAL SOCIETY 1771