CAUSTICS BY REFRACTION OF CIRCLES AND LINES

FELIX RYDELL

A BSTRACT. This short note revisits a classical result that the complete caustic by refraction
of a circle is the evolute of a Cartesian oval. We provide additional details to the statement and
geometric proof of this fact, as presented in G. Salmon’s 1879 book ‘Higher Plane Curves’.
arXiv:2402.00475v1 [math.AG] 1 Feb 2024

We observe that as the circle tends to a line, this Cartesian oval collapses into an ellipse.
Finally, we discuss a computational method to find the complete caustics by refraction, inde-
pendent of Salmon’s proof.

I NTRODUCTION
The envelope of a family of curves in the plane is a curve that is tangent to each curve
of the family at some point, such that these points of tangency form the whole curve. As
Bruce and Giblin write, “These [curves] appear to cluster along another curve, which the
eye immediately picks out . . . The new curve is called the envelope” [BG92, Chapter 5],
see Figure 1 for illustrations. Important examples of envelopes are evolutes. The evolute
of a plane curve is the envelope of the family of normals of the curve, and in differential
geometry, they are the locus of centres of curvature [BG92, page 89]. Their study has a long
standing history in mathematics. From 200 BC by Apollonius [Too12], to Huygens [HB86]
and Salmon [Sal79] in the 19th century. Recently, Piene, Riener and Shapiro continued this
tradition [PRS21], by studying various numerical invariants of evolutes in the plane. Evolutes
of algebraic curves coincide with the classical Euclidean distance discriminant, the set of
points where the number of complex critical solution of the nearest point problems differs
from the Euclidean distance degree [DHO+ 16, Section 7].
In optics, a caustic in the plane is an envelope of light rays which have been reflected
or refracted by a curve. From the physics point of view, it is a curve of concentrated light.
Caustics gained interest in the 17th century by Tschirnhaus, La Hire and Leibniz [SS05]. In

F IGURE 1. On the Left: The black curve is the parabola y 2 = x. On the right:
The black curve is the ellipse x2 + 4y 2 = 4. The blue lines are the normals to
the black curves, and the red curves are their envelopes, called evolutes. On
the left: The red curve is 2(2x − 1)3 = 27y 2 . On the right: The red curve is
64x6 +48x4 y 2 +12x2 y 4 +y 6 −432x4 +756x2 y 2 −27y 4 +972x2 +243y 2 −729,
known as a superellipse or Lamé curve. The images have been taken from
Mathworld–A Wolfram Web Resource [Weib, Weia].
1
2 FELIX RYDELL

this paper, light rays are assumed to emanate from a fixed point A in the affine plane or from
a point at infinity (such as the sun,) called the radiant point. If A ∈ R2 , then the light rays
are lines through A, and if A is a point at infinity, the light rays are all parallel to some fixed
line LA determined by A. Consider now a plane curve C ⊆ R2 . Fix a smooth point X ∈ C
and let NX denote the normal space of C at X. In the first case, define the light ray L(X) as
the span of A and X, and in the other case, let L(X) := X + LA . Any two lines meeting at a
point X make two angles. We say that the angle of a line L with respect to a line L0 meeting
at X is the unique angle θ ∈ (−π/2, π/2] that L0 needs to be rotated around X to become L.
Denote by ∢in the angle of L(X) with respect to X + NX . Fix a refraction constant n ∈ R.
If n1 sin ∢in ∈ [−1, 1], then the refracted ray Rn (X) is the unique line whose angle ∢out with
respect to X + NX satisfies
sin ∢in
= n. (1)
sin ∢out
Reflected rays are a special case of refracted rays, namely R−1 (X). The caustic by refrac-
tion with respect to A, C and n, is the envelope of all refracted rays Rn (X). The complete
caustic by refraction is the envelope of all refracted rays Rn (X) and R−n (X). As we will
see, from the algebraic point of view, it is more natural to study complete caustics by refrac-
tion, as the they tend to be algebraic curves. Robert Ferréol’s “Encyclopedia of Remarkable
Mathematical Shapes” includes a friendly survey on caustics that is available on the web
https://mathcurve.com/courbes2d.gb/caustic/caustic.htm.
The aim of this paper is to revisit a classic result and its proof as it appears in Salmon’s
book [Sal79, pages 99-101] that complete caustics by refraction of circles are evolutes of
Cartesian ovals, see Figure 2. In Section 1, we recall classical real geometry. In Section 2,
we revisit the proof that appears in Salmon’s book, and add numerous details. In Section 3,
we show that the associated Cartesian ovals collaps into ellipses as the circles we refract
from degenerate into lines, yielding the complete caustics by refraction of lines. Finally, we
discuss computational approaches in Section 4 to determining these complete caustics, and
with the help of Macaulay2 [GS20], we find a general formula.

Acknoledgements. This paper is dedicated to Vahid Shahverdi, a great friend and mathe-
matician, who provided invaluable feedback for this project. The author would also like to
thank Kathlén Kohn for helpful discussions and for introducing me to this subject.
The author was supported by the Knut and Alice Wallenberg Foundation within their
WASP (Wallenberg AI, Autonomous Systems and Software Program) AI/Math initiative.

1. P RELIMINARIES
In this section, we recall some preliminary facts that we need. Firstly, however, we alge-
braically describe the refracted rays, given a real radiant point A ∈ R2 , a real curve C ⊆ R2
given by the equation G and a a refraction constant n ∈ R. Let u, v ∈ R2 be non-zero
vectors. If θ ∈ (−π, π] is the angle that u needs to be rotated to be parallel to v, then
ι(u) · v = sin θ∥u∥∥v∥, where ι : (u1 , u2 ) 7→ (−u2 , u1 ) is the rotation by π/2 radians, and · is
the standard inner product. Let ∢in ∈ (−π/2, π/2] be angles defined as in the introduction.
For a smooth X ∈ C, the sine of this angle is the sine between σ(A − X) and ∇G(X),
where σ ∈ {−1, 1} ensures that angle between them is within (−π/2, π/2], i.e. its the sign
of (A − X) · ∇G(X). By this logic, we have
σ(A − X) · ι(∇G(X)) σℓ · ι(∇G(X))
= sin ∢in , = sin ∢out , (1.1)
∥A − X∥∥∇G(X)∥ ∥ℓ∥∥∇G(X)∥
 CAUSTICS BY REFRACTION OF CIRCLES AND LINES 3

F IGURE 2. This illustration shows the refracted rays (in blue and orange)
given the black radiant point A, refracted on the black circle C. The red curve
is the complete caustic by refraction, and the green curves are the Caustic
ovals, whose evolute is the red curve. Here, A = (0, 0) and C has center (1, 0)
and radius r = 1/3. The blue lines are the refracted lines for the refraction
constant n = 1/2 and the orange lines are the refracted lines given n = −1/2.
See Example 2.3. The plot was created using [Inc].

where ℓ is a direction of the refracted ray; Rn (X) is parametrized by X + ℓt, t ∈ R. Using

the assumed identity sin ∢in / sin ∢out = n, we then get the following algebraic expression
 ∥ℓ∥ 2  (A − X) · ι(∇G(X)) 2
= n2 , (1.2)
∥A − X∥ ℓ · ι(∇G(X))
which does not care about the sign of n. In ℓ, this is a homogeneous equation (after clearing
denominators) of degree 2. Then for a fixed X, we expect two solutions in ℓ projectively,
corresponding to n and −n. Observe that the equality of this equation makes sense also over
the complex numbers, and so we take this to define the refracted rays at X.
If A is a point at infinity, then we simply replace A − X in (1.2) by a fixed direction v of
the fixed line LA as in the introduction.

1.1. Cartesian ovals. Given two distinct foci A, B ∈ R2 , and two real numbers s, t, a Carte-
sian oval is the set of points M satisfying
∥A − M ∥ + s∥B − M ∥ = t. (1.3)
This oval is contained in the degree-four plane curve
 2
∥A − M ∥2 − s2 ∥B − M ∥2 + t2 − 4t2 ∥A − M ∥2 = 0, (1.4)

which we get by multiplying together the four expressions

∥A − M ∥ ± s∥B − M ∥ ± t. (1.5)
Among the four ovals these equations define, precisely two have real solutions.
4 FELIX RYDELL

Putting s = 0 (or A = B) in (2.14), we get a circle, and putting s = 1, we get an ellipse.

Note that (1.4) becomes degree-two for s = 1. Indeed, we are left with
 2
∥A∥2 − ∥B∥2 + 2M · (B − A) + t2 − 4t2 ∥B − M ∥2 = 0. (1.6)
As we are interested in evolutes of Cartesian ovals, we study their normals below. For the
sake of this lemma, we consider angles in [0, π) instead of (−π/2, π/2], because that’s how
we apply it in Section 2.
Lemma 1.1. Consider a Cartesian oval
±∥A − M ∥ + s∥B − M ∥ = t. (1.7)
For a fixed point M on this oval, let NM be the normal of the oval at M . If the angle between
A − M and NM at M is denoted α ∈ [0, π] and the angle between B − M and NM at M is
denoted β ∈ [0, π], then
sin α
= |s|. (1.8)
sin β
Further, NM is the unique line through M satisfying this property.
Proof. As follows from the beginning of the section, |ι(u)·v| = sin θ∥u∥∥v∥, where θ ∈ [0, π]
is the angle between the lines spanned by u and v. Let G be the defining equation of (1.7).
Then
   
0 −1 0 −1
 
(A − M ) · ∇G(M ) (B − M ) · ∇G(M )
1 0 1 0
= sin α, = sin β.
∥A − M ∥∥∇G(M )∥ ∥B − M ∥∥∇G(M )∥
(1.9)
Using that
h i
M1 −A1 M1 −B1 M2 −A2 M2 −B2
∇GT (M ) = ± ∥A−M ∥
+ s ∥B−M ∥
, ± ∥A−M ∥
+ s ∥B−M ∥ , (1.10)
where both ± reference the same sign, and plugging it into (1.9), we get the desired equality
sin α/ sin β = |s|.
For the last part, define ∢AM B ∈ (0, π) to be the angle at M of the triangle defined by the
points A, M and B. We consider the expression
sin(∢AM B − γ)
, (1.11)
sin γ
for γ ∈ (0, π). If we can show that (1.11) monotonically decreases in this interval, then we
are done. Deriving with respect to γ, we arrive at
sin ∢AM B
− , (1.12)
sin2 γ
which is strictly negative. □

1.2. Classical real geometry. For the sake of completeness, we review classical results on
the real geometry used in Salmon’s book. Firstly, we need some notation. Given two distinct
points A and B, let LAB denote the line spanned by A and B. Given three distinct points
A, B, C, let ABC denote the triangle they form, and let ∢ABC ∈ [0, π] denote the inner
angle of the triangle at the point B. In this subsection, all points are real, and all circles have
positive radius.
 CAUSTICS BY REFRACTION OF CIRCLES AND LINES 5

F IGURE 3. Illustration for the inscrirbed angle theorem. Taken from https:
//en.wikipedia.org/wiki/Inscribed_angle.

Theorem 1.2 (Inscribed Angle Theorem). Let O be the center of a circle with distinct in-
scribed points A, B, C. We have that
(1) if A lies in the major arc defined by B and C, then ∢BOC = 2∢BAC ;
(2) if A lies in the minor arc defined by B and C, then ∢BOC = 2(π − ∢BAC ).
See Figure 3 for an illustration of this theorem.
Corollary 1.3. Let O be the center of a circle with distinct inscribed points A, B, C. The
line LBC divides R2 in two open half-planes, one containing A and one that does not. Let
D be any point in the latter half-plane that lies on the tangent of the circle at C. Then
∢BAC = ∢BCD .
Proof. Consider the two different cases from Theorem 1.2.
(1): We have that
∢OCB = ∢OBC = (π − ∢BOC )/2 = π/2 − ∢BAC (1.13)
Now, since LCD is tangent to the circle, ∢OCB +∢BCD = π/2, and we conclude that ∢BCD =
∢BAC .
(2): We have that
∢OCB = ∢OBC = (π − ∢BOC )/2 = −π/2 + ∢BAC (1.14)
In this case, however, we have ∢BCD = ∢OCB + π/2, since LCD is tangent to the circle.
From this, the equality follows directly. □
Corollary 1.4. Let O be the center of a circle with distinct inscribed points A, B, C. If E is
the intersection of LAB and the tangent of the circle at C, then ACE is similar to BCE, and
∥A − E∥ ∥C − E∥
= . (1.15)
∥A − C∥ ∥C − B∥
Proof. Both triangles ACE and BCE have at least one angle in common, namely ∢AEC =
∢BEC . It suffices to show that they have one more angle in common. We consider the two
cases according to Theorem 1.2.
(1): In this case we set D = E to apply Corollary 1.3. We show that ∢CAD = ∢BCD .
Since by construction, ∢CAD = ∢BAC , this follows directly from Corollary 1.3.
(2): We show that ∢EBC = ∢ACE . Let D be any point as in Corollary 1.3. Then
∢BCD + ∢ECA + ∢ACB = π and ∢BAC + ∢ECA + ∢ACB = π, (1.16)
6 FELIX RYDELL

using the equality ∢BAC = ∢BCD from Corollary 1.3. Next, since the angles of a triangle
add up to π,
∢BAC + ∢ABC + ∢ACB = π, (1.17)
and from (1.16) and (1.17), we directly conclude that ∢EBC = ∢ACE .
The last part can be easily checked in both cases, as a consequence of similarity. □
Theorem 1.5 (Ptolemy’s Theorem). Let A, B, C, D be vertices of a quadrilateral inscribed
in a circle, with A, C and B, D being opposite vertices. Then
∥A − C∥∥B − D∥ = ∥A − B∥∥C − D∥ + ∥B − C∥∥D − A∥. (1.18)
Theorem 1.6 (Law of Sines). Let O be the center of a circle with distinct inscribed points
A, B, C. Then
∥B − C∥ ∥A − C∥ ∥A − B∥
= = = 2r, (1.19)
∢CAB ∢ABC ∢BCA
where r is the radius of the circle.

2. C AUSTICS BY R EFRACTION OF C IRCLES

We are now ready to revisit the classic proof appearing in Salmon’s book [Sal79, pages 99-
101] that complete caustic by refraction of circles are evolutes of Cartesian ovals, with details
added. We first need a lemma that constitutes the basis of the proof of our main theorem.
Lemma 2.1. Let A be a point away from the circle C with center O and radius r > 0. For
each point R on this circle away from LAO , define C(R) to be the unique circle meeting A
and R and that is tangent to LRO at R. Then C(R) meets LAO at the point
 r2 
B := A + 1 − (O − A), (2.1)
∥A − O∥2
independently of R. Further, ∥A − O∥∥B − O∥ = r2 .
Observe that
 r2 
A=B+ 1− (O − B). (2.2)
∥B − O∥2
For the proof, we use the following identity for three points U, V, W :
∥U − W ∥2 = ∥V − U ∥2 + ∥W − U ∥2 − 2(V − U ) · (W − U ). (2.3)
Proof. Let E be the center of the circle C(R). By assumption, we have the following identi-
ties:
∥R − O∥2 = r2 (2.4)
2 2
∥R − E∥ = γ (2.5)
∥A − E∥2 = γ 2 (2.6)
(R − O) · (R − E) = 0. (2.7)
It follows by (2.3) to (2.5) and (2.7) that
∥E − O∥2 = γ 2 + r2 . (2.8)
Next, consider the identity
∥B − E∥2 = ∥O − B∥2 + ∥O − E∥2 − 2(O − B) · (O − E). (2.9)
 CAUSTICS BY REFRACTION OF CIRCLES AND LINES 7

F IGURE 4. On the left: An illustration for the proof that the caustic by re-
fraction of lines are the evolutes of ellipses. On the right: An illustration for
the proof that caustic by refraction of lines are the evolutes of Cartesian ovals.
The points A, B, R, O and M have the same meaning as in our proof. These
images are taken from [Sal79, page 100].

Since
r2
O−B = (O − A), (2.10)
∥O − A∥2
we have ∥O − B∥2 = r4 /∥O − A∥2 , and in particular ∥O − A∥∥O − B∥ = r2 . Putting
everything together, we have
r2
2(O − B) · (O − E) = 2(O − A) · (O − E) (2.11)
∥O − A∥2
r2
= (∥O − E∥2 + ∥O − A∥2 − ∥A − E∥2 ) (2.12)
∥O − A∥2
r4
= + r2 . (2.13)
∥O − A∥2
Then from (2.9) ∥B − E∥2 = γ 2 , which is precisely what we needed to show. □
Theorem 2.2. Given a radiant point A in the plane away from the circle C with center O
and radius r > 0, let B be defined as in Lemma 2.1. For a refraction constant n ̸= 0, the
refracted rays are normals to the Cartesian ovals defined by points M in the plane such that
∥A − O∥ ∥A − O∥∥A − B∥
±∥A − M ∥ + ∥B − M ∥ = . (2.14)
r r|n|
Since the proof below is very geometric in nature, we include the original illustration by
Salmon Figure 4 to assist the reader.
Proof. Take any point R on the circle C away from LAO and consider the unique circle
C(R) as in Lemma 2.1. For sin ∢in = n sin ∢out to hold, we must have that n1 sin ∢in ∈
[−1, 1], which holds at least for small enough ∢in . The refracted line meets the circle C(R)
in two distinct points, R itself and another point M . We get a quadrilateral in C(R) given by
the points A, B, R, M . To apply Ptolemy’s theorem, we must specify a cyclic order of the
vertices. We get
±∥R − B∥∥A − M ∥ + ∥R − A∥∥B − M ∥ = ∥A − B∥∥R − M ∥, (2.15)
8 FELIX RYDELL

with ± = + if A, B and R, M are the opposite vertices and ± = − otherwise. We rewrite

(2.15) to
∥R − A∥ ∥R − M ∥
±∥A − M ∥ + ∥B − M ∥ = ∥A − B∥ . (2.16)
∥R − B∥ ∥R − B∥
By Corollary 1.4, the triangles AOR and BOR are similar and
∥A − O∥ ∥R − O∥
= , (2.17)
∥R − A∥ ∥R − B∥
meaning that ∥R − A∥/∥R − B∥ is fixed and equal to ∥A − O∥/r. Next, by the law of sines,
∥R − A∥ ∥R − M ∥
= = twice the radius of C(R). (2.18)
sin ∢ABR sin ∢M BR
It follows by Corollary 1.3, that | sin ∢in | = sin ∢ABR and | sin ∢out | = sin ∢M BR . For
instance, ∢in ≡ ∢ABR modulo π. Then ∥R − A∥/∥R − M ∥ equals the absolute value of the
refraction constant n. By (2.17) and (2.18), we get
∥R − M ∥ ∥R − A∥ ∥R − M ∥ ∥A − O∥
= = . (2.19)
∥R − B∥ ∥R − B∥ ∥R − A∥ r|n|
By (2.16),(2.17) and (2.19), we get the equation of the statement.
By the law of sines and the above,
sin ∢AM R ∥R − A∥
= = ∥A − O∥/r =: s. (2.20)
sin ∢BM R ∥R − B∥
As a consequence of Lemma 1.1, LRM , i.e. the refracted line, is the normal of the Cartesian
ovals Equation (2.14), and we are done. □

Example 2.3. In Figure 2, we put A = (0, 0), O = (1, 0), r = 1/2 and n = 1/2. Then (2.14)
and (1.4) together provide the degree four equation in x, y
 2
2 2
− 72(x + y ) + 144x + 192 − 9216(x2 + y 2 ) = 0, (2.21)

whose real part is the two green ovals. Recall that the blue lines in Figure 2 are refracted
rays corresponding to n = 1/2 and the orange ones are refracted rays corresponding to
n = −1/2. Some of the blue lines are normal to the outer oval, and some are normals to the
inner oval. Likewise for the orange lines.
Observe that not all refracted rays exists, i.e. for certain points R on the circle, there is no
refracted ray. This is explored in detail next. ♢

Recall that in order for a refracted ray to exist, we need that sin ∢in ∈ [−n, n]. In what
follows, we investigate for which choices of R ∈ C, the caustic by refraction corresponds to
the positive oval in Equation (2.14) and which correspond to the negative one. We do this by
calculating the intersection of LM R with LAO . The refracted ray correspond to the positive
oval precisely when this intersection lies in the line segment defined by A, B. Let the radiant
point be A = (0, 0) and assume that the circle has center O = (1, 0) and radius r. Fix a point
R in the circle, i.e. R = O + r(cos θ; sin θ) for some angle θ. We compute sin θ1 and cos θ1 ,
where θ1 is the angle that (cos θ, sin θ) needs to be rotated round R to become LAR via the
 CAUSTICS BY REFRACTION OF CIRCLES AND LINES 9

following formulae:
    .  
−1 − r cos θ r cos θ −1 − r cos θ sin θ
sin θ1 = ·ι r∥ ∥= √ ,
−r sin θ r sin θ −r sin θ 1 + r2 + 2r cos θ
   .  
−1 − r cos θ r cos θ −1 − r cos θ −r − cos θ
cos θ1 = · r∥ ∥= √ .
−r sin θ r sin θ −r sin θ 1 + r2 + 2r cos θ
(2.22)
Note that sin ∢in = ± sin θ1 . Let θ2 be√the angle satisfying sin θ2 = sin θ1 /n such that
sin ∢out = ± sin θ2 , and with cos θ2 = σ 1 − sin θ2 , where σ := sgn(cos θ1 ), we have by
construction that (cos ∢out , sin ∢out ) = ±(cos θ2 , sin θ2 ). Note that the rotation matrix
 
cos θ2 − sin θ2
(2.23)
sin θ2 cos θ2
is up to scaling equal to
 p
σ n2 (1 + r2 + 2r cos θ) − sin2 θ p

−sgn(n) sin θ
Tθ := . (2.24)
sgn(n) sin θ σ n2 (1 + r2 + 2r cos θ) − sin2 θ
The refracted line is then parametrized as
   
cos θ 1 + r cos θ
λTθ + , (2.25)
sin θ r sin θ
and our goal is to determine its intersection xint with the x-axis. Doing the computations, we
find
sgn(n)r
xint := p + 1, (2.26)
sgn(n) cos θ + σ n2 (1 + r2 + 2r cos θ) − sin2 θ
and the refracted ray corresponds to the positive oval precisely when xint ∈ (0, 1).
Example 2.4. In Theorem 2.2, we start with a radiant point A, a circle C and a refraction
constant n, and derive Cartesian ovals. Conversely, let us begin with the Cartesian oval
defined by
∥A − M ∥ + 2∥B − M ∥ = 3, (2.27)
for A = (0, 0) and B = (1, 0). We wish to find a circle C of center O with radius r and
refraction constant n such that
∥A − O∥
= 2,
r (2.28)
∥A − O∥∥A − B∥
= 3.
rn
Then, the oval (2.27) corresponds to the caustic by refraction of the circle C given the radiant
point A. Note that ∥A − B∥ = 1, and so from the second equation, we deduce that n = 2/3.
We assume A, B and O to be collinear as in Lemma 2.1, and therefore we put O = (x, 0). By
∥A − O∥ = 2r, we have |x| = 2r and x2 = 4r2 . To find x, we solve (2.1), i.e.
r2 3
1 = (1 − )x = x, (2.29)
x2 4
which gives us x = 4/3. It follows that r = 2/3. ♢
Based on the proof of Theorem 2.2, we can write a Macaulay2 program that, given a
radiant point, a circle C and a refraction constant n, outputs the algebraic closure of the
associated Cartesian ovals (2.14), see Figure 5.
10 FELIX RYDELL

R = QQ[M1,M2,Cent1,Cent2,R1,R2,r,n,Rad]
-- M1, M2 are coordinates for the point on the oval
-- Cent1, Cent2 are coordinates for the center of the circle C(R)
-- R1, R2 are coordinates for the point on c
-- Rad is the radius of C(R)

A = matrix{{0},{0}} -- radiant point

Circ = matrix{{1},{0}} -- center of circle
B = A + (1-rˆ2/(transpose(A-Circ)*(A-Circ))_(0,0))*(Circ-A) -- B

I = ideal ((M1-Cent1)ˆ2+(M2-Cent2)ˆ2-Radˆ2,
(R1-Cent1)ˆ2+(R2-Cent2)ˆ2-Radˆ2,
(B_(0,0)-Cent1)ˆ2+(B_(1,0)-Cent2)ˆ2-Radˆ2,
(A_(0,0)-Cent1)ˆ2+(A_(1,0)-Cent2)ˆ2-Radˆ2,
(R1-Circ_(0,0))ˆ2+(R2-Circ_(1,0))ˆ2-rˆ2,
nˆ2*rˆ2*((M1-R1)ˆ2+(M2-R2)ˆ2)-
(transpose(A-Circ)*(A-Circ))_(0,0)*((B_(0,0)-R1)ˆ2+
(B_(1,0)-R2)ˆ2))

J = eliminate (I,{Cent1,Cent2,R1,R2,Rad})

decompose J

F IGURE 5. Computing the associated Cartesian ovals in Macaulay2.

3. C AUSTICS BY R EFRACTION OF L INES

Also found in [Sal79, pages 99-101], is a proof that the caustics by refraction of lines are
normals to ellipses. In this case, we have:
Theorem 3.1. Given a radiant point A in the plane and a line L not containing A, let B be
the reflection of A with respect to L. For a refraction constant n ̸= 0, the refracted lines are
normals to the oval defined by points M in the plane such that
∥A − B∥
±∥A − M ∥ + ∥B − M ∥ = , (3.1)
|n|
with ± = + if |n| < 1 and ± = − if |n| > 1.
Of the two ovals of (3.1), precisely one has real solutions. The proof for Theorem 3.1
works exactly the same as for Theorem 2.2, where the circle C(R) is uniquely defined by
containing A, B and being tangent to the normal line of L at R ∈ L. To see the last part
of the statement, note that since B is the reflection along the line x = 1, the line M, R are
diagonals of the quadrilateral of points A, B, M, R in C(R) if and only if |n| < 1.
Below, we make the observation that we get (2.14) as a limit of (3.1). Let P be the orthog-
onal projection of A onto L, or equivalently P := (A + B)/2. Then L divides R2 into two
half-planes, one that contains A and one that does not. Inside the latter, let Ok ∈ LAP be a
sequence of points tending to infinity. Let rk := ∥P − Ok ∥. The line L can be viewed as the
limit of circles Ck with centers Ok and radii rk . Consider the sequence Bk defined by (2.1).
We argue that Bk → B. This is easy to check for A = 0 and L = {x : x1 = 1}. Indeed, we
 CAUSTICS BY REFRACTION OF CIRCLES AND LINES 11

have Ok = (xk , 0) with xk → ∞ and rk = xk − 1, and the formula (2.1) gives us

2 − x1k
 
Bk = → B. (3.2)
0
Since ∥A − Ok ∥/rk → 1, we see that
∥A − Ok ∥ ∥A − Bk ∥∥A − Ok ∥
±∥A − M ∥ + ∥Bk − M ∥ = . (3.3)
rk rk n
tends to (3.1).

4. S YMBOLIC C OMPUTATION OF E NVELOPES

Let F (x, y, t) be a non-zero polynomial in three variables that for fixed t defines a curve in
R2 . As heuristically explained in [CLOS94, Section 3.4], envelopes can be computed purely
algebraically by finding the solutions in x, y to the system
F (x, y, t) = 0,
∂F (4.1)
(x, y, t) = 0.
∂t
Viewing F as a polynomial in t with coefficients in x, y, the envelope is given by the deter-
minant of the Sylvester matrix of F and ∂F/∂t.
Example 4.1. Consider the parabola of Figure 1. It is parametrized by (t2 , t). The normal
line at (t2 , t) is parametrized by (t2 , t) + λ(1, −2t). Therefore we define
 
x − t2 1
F (x, y, t) := det = −2tx − y + 2t3 + t. (4.2)
y − t −2t
The envelope of Vt := {(x, y) : F (x, y, t) = 0} is the solution set in x, y to
−2tx − y + 2t3 + t = 0, (4.3)
−2x + 6t3 + 1 = 0. (4.4)
From (4.4), we get x = 3t3 + 1/2 and plugging this into (4.4), we get y = −4t3 . Eliminating
t by hand yields the equation 2(x − 1)3 = 27y 2 . ♢
In the setting of Section 2, let C is a circle with center O and radius r, parametrized by
R(t). Then Nt = R(t) − O is the normal of the circle direction at any point t. Using (1.2),
we consider the following polynomial in C[x, y, t, r, n],
 2
Fr,n (x, y, t) := numerator of ∥(x; y) − R(t)∥2 (A − R(t)) · ι(Nt ) −
 2 (4.5)
2 2
n ∥A − R(t)∥ ((x; y) − R(t)) · ι(Nt ) .
Observe that in the study of caustics by refraction, we may without loss of generality
fix A = (0, 0) and O = (1, 0): Refracted rays are naturally equivariant with respect to
translation, rotation and scaling. Viewing Fn,r as a polynomial in t with coefficients polyno-
mial in x, y, r and n, this allows to calculate the determinant of the Sylvester matrix using
Macaulay2 [GS20] via the code in Figure 6, which took 2 hours, 19 minutes and 26 sec-
onds to run on a 13th Gen Intel(R) Core(TM) i9-13900H CPU running at 2.60 GHz. The
computed polynomial consists of the factors
n4 , (r − 1)2 , (r + 1)2 , y2, ((x − 1)2 + y 2 − r2 )2 ,
(4.6)
(x − 1)2 (r2 n2 + n2 − 1) − (y − r)2 ,
12 FELIX RYDELL

and a polynomial of 1765 terms that is degree 12 in the four variables x, y, r and n:
x12 r12 n12 + 6x10 y 2 r12 n12 + 15x8 y 4 r12 n12 + · · · − 6n2 − 12x + 1. (4.7)
As mentioned in the introduction, evolutes of plane curves are ED discriminants. For our
running example Figure 2, we compute the evolute using Figure 7. It is a degree-twelve
equation of 49 terms:
16384000x12 + 351768768x10 y 2 + · · · − 167215104x + 11943936. (4.8)
This polynomial is equal to (4.7), when specifying r = 1/3 and n = 1/2; (4.7) is the compo-
nent that corresponds to the complete caustic by refraction. Keeping r and n as variables in
this code resulted in a program that did not terminate in reasonable time.

R = QQ[x,y,t,r,n]

iota = (V) -> matrix{{-V_(1,0)},{V_(0,0)}}

A = matrix{{0},{0}}
Circ = matrix{{1},{0}}

R = matrix{{r*(2*t)/(1+tˆ2)},{r*(tˆ2-1)/(1+tˆ2)}}+Circ
Y = matrix{{x},{y}}

diffAR = transpose(A-R)*(A-R)
diffYR = transpose(Y-R)*(Y-R)
normal = R-O
dotAxnormal = transpose(iota(A-R))*normal
dotyxnormal = transpose(iota(Y-R))*normal

F = (diffYR*dotARnormalˆ2-nˆ2*diffAR*dotYRnormalˆ2)_(0)_(0)
Fpoly = numerator(F)

S = QQ[x,y,r,n][t]
Fpoly = sub(Fpoly,S)

factF = (factor(Fpoly))#1#0
SMAT = transpose(sylvesterMatrix(factF,diff(t,factF),t))
Final = det SMAT

F IGURE 6. Computational approach to finding the complete caustic by re-

fraction in Macaulay2.

R EFERENCES
[BG92] James William Bruce and Peter J Giblin. Curves and Singularities: a geometrical introduction to
singularity theory. Cambridge university press, 1992.
[CLOS94] David Cox, John Little, Donal O’Shea, and Moss Sweedler. Ideals, varieties, and algorithms.
American Mathematical Monthly, 101(6):582–586, 1994.
[DHO+ 16] Jan Draisma, Emil Horobeţ, Giorgio Ottaviani, Bernd Sturmfels, and Rekha R Thomas. The eu-
clidean distance degree of an algebraic variety. Foundations of computational mathematics, 16:99–
149, 2016.
 CAUSTICS BY REFRACTION OF CIRCLES AND LINES 13

R = QQ[x0,y0,x,y];

r = 1/2
n = 1/2
A = matrix{{0},{0}}
Circ = matrix{{1},{0}}

B = matrix{{1-rˆ2},{0}}
Y = matrix{{x0},{y0}}

s2 = (transpose(A-Circ)*(A-Circ))_(0,0)/rˆ2
t2 = ((transpose(A-Circ)*(A-Circ))*
(transpose(A-B)*(A-B)))_(0,0)/(rˆ2*nˆ2)

diffAY = transpose(A-Y)*(A-Y)
diffBY = transpose(B-Y)*(B-Y)

F = (diffAY-s2*diffBY-t2)ˆ2-4*s2*t2*diffBY
EX = ideal(F) + minors(2,matrix {{x0-x,y0-y},
{diff(x0,F),diff(y0,F)}})
detEX = det submatrix(jacobian EX,{0,1},{0,1})

I = EX + ideal(detEX)
toString first entries gens eliminate(I,{x0,y0})

F IGURE 7. Computation of the evolute of the Cartesian ovals in Equa-

tion (2.14) for Macaulay2.

[GS20] Daniel R. Grayson and Michael E. Stillman. Macaulay2, a software system for research in algebraic
geometry. Available at https://macaulay2.com/, 2020.
[HB86] C. Huygens and R.J. Blackwell. Christiaan Huygens’ the Pendulum Clock, Or, Geometrical
Demonstrations Concerning the Motion of Pendula as Applied to Clocks. History of Science and
Technology Series. Iowa State University Press, 1986.
[Inc] Wolfram Research, Inc. Mathematica, Version 14.0. Champaign, IL, 2024.
[PRS21] Ragni Piene, Cordian Riener, and Boris Shapiro. Return of the plane evolute. arXiv preprint
arXiv:2110.11691, 2021.
[Sal79] G. Salmon. A Treatise on the Higher Plane Curves. Dublin, 1879. Available on the web at http:
//archive.org/details/117724690.
[SS05] Giovanni Mingari Scarpello and Aldo Scimone. The work of tschirnhaus, la hire and leibniz on
catacaustics and the birth of the envelopes of lines in the 17th century. Archive for history of exact
sciences, 59:223–250, 2005.
[Too12] Gerald J Toomer. Apollonius: Conics Books V to VII: The Arabic Translation of the Lost Greek
Original in the Version of the Banū Mūsā, volume 9. Springer Science & Business Media, 2012.
[Weia] Eric W Weisstein. Ellipse Evolute. From MathWorld–A Wolfram Web Resource. Available on the
web at https://mathworld.wolfram.com/EllipseEvolute.html.
[Weib] Eric W Weisstein. Parabola Evolute. From MathWorld–A Wolfram Web Resource. Available on
the web at https://mathworld.wolfram.com/ParabolaEvolute.html.