Comput. Aided Geom. Des.

108 (2024) 102266

Contents lists available at ScienceDirect

Computer Aided Geometric Design

journal homepage: www.elsevier.com/locate/cagd

Conics in rational cubic Bézier form made simple

Javier Sánchez-Reyes
Department of Applied Mechanics, Universidad de Castilla-La Mancha, IMACI, ETS Ingeniería Industrial, 13071 Ciudad Real, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: We revisit the rational cubic Bézier representation of conics, simplifying and expanding previous
Base point works, elucidating their connection, and making them more accessible. The key ingredient is
Conic the concept of conic associated with a given (planar) cubic Bézier polygon, resulting from
Cubic
an intuitive geometric construction: Take a cubic semicircle, whose control polygon forms a
Rational Bézier curve
Rho-value
square, and apply the perspective that maps this square to the given polygon. Since cubic
Shape factor conics come from a quadratic version by inserting a base point, this conic admitting the
polygon turns out to be unique. Therefore, detecting whether a cubic is a conic boils down to
checking out whether it coincides with the conic associated with its control polygon. These two
curves coincide if they have the same shape factors (aka, shape invariants) or, equivalently,
the same oriented curvatures at the endpoints. Our results hold for any cubic polygon (with no
three points collinear), irrespective of its convexity. However, only polygons forming a strictly
convex quadrilateral define conics whose cubic form admits positive weights. Also, we provide
a geometric interpretation for the added expressive power (over quadratics) that such cubics
with positive weights offer. In addition to semiellipses, they encompass elliptical segments with
rho-values over the negative unit interval.

1. Introduction: conics in rational cubic Bézier form

Although conic segments have a quadratic rational representation, their cubic Bézier form deserves attention in CAGD for several
reasons:

• Compatibility: Some CAD systems employ cubic curves to represent all the geometry. Also, rational skinning (Piegl and Tiller,
1997) between conics and cubic curves requires a common cubic form.
• Estimation of parameters in cubics: Given only a cubic control polygon, the weights defining a conic constitute a sound default
(Farin, 1999). Also, Li et al. (2014) choose the two free parameters available in 𝐺2 Hermite interpolation with a rational cubic
to ensure conic precision.
• Detection of conics: Determining whether a given cubic is actually a conic (and finding out its quadratic version) facilitates
geometry processing (Barnhill, 1992).
• Higher expressive power (with positive weights): A cubic can describe larger circular arcs, spanning angles Δ ∈ (0, 4𝜋∕3) versus
Δ ∈ (0, 𝜋) for quadratics.

✩ Editor: Hartmut Prautzsch.

E-mail address: Javier.SanchezReyes@uclm.es.

https://doi.org/10.1016/j.cagd.2023.102266
Received 18 June 2023; Received in revised form 10 December 2023; Accepted 11 December 2023
Available online 12 December 2023
0167-8396/© 2023 The Author. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 1. Generating a cubic Bézier conic 𝐜(𝑢) associated with a polygon 𝐜, by applying a perspective transformation Φ to a cubic semicircle 𝐂(𝑢).

In a seminal article, Wang and Wang (1992) derived the necessary and sufficient conditions for a Bézier cubic, with positive
weights, to represent a conic. Fiorot et al. (1995) tackled the alternative BR-form, that is, curves controlled by a massic polygon via
Bernstein polynomials. Also, Section 8.1 of the book on NURBS by Farin (1999) addresses cubic conics, whereas Wang (1991) and
Chou (1995) analyze the particular case of circles. These works overlook that any cubic conic comes from a quadratic one simply by
inserting a base point, i.e., via generalized degree-elevation following the terminology introduced by Denker and Herron (1997). We
capitalize on this property, bringing together, expanding, and simplifying results on cubic conics using only a handful of compact
formulas.
The paper is thus structured as follows. Section 2 shows how to endow a given cubic polygon with the weights guaranteeing a
conic. This conic associated with the polygon results from embedding a cubic semicircle into a square and applying the perspective
map deforming the square into the polygon. Next, Section 3 derives the general cubic representation of an arbitrary nondegenerate
conic by inserting a base point in a properly parametrized quadratic curve, whereas Section 4 solves the reverse problem of finding
the conic generating a given cubic polygon. In Section 5, these results lead to the geometric characterization of conics in cubic Bézier
form. Section 6 explores the added expressive power of cubics, interpreting it as an extended range for the conic rho-values, and
discusses the freedom introduced by the base point. Section 7 particularizes our results to the notable case of cubic circles. Section 8
finishes with conclusions and final remarks.

2. Finding the weights of the conic associated with a given cubic polygon 𝐜

Suppose we are given a planar Bézier polygon 𝐜 = {𝐜𝑖 }3𝑖=0 . Can we assign suitable weights {𝑤𝑖 }3𝑖=0 such that the resulting rational
Bézier cubic 𝐜(𝑢), 𝑢 ∈ [0, 1] is actually a conic (𝐜)? Farin (1999) interpreted this problem as choosing sensible weights, solving it
for a convex 𝐜, albeit without providing explicit expressions. We introduce a simpler alternative that, in addition, does not require
convexity and readily yields the sought weights, coinciding with those obtained by Wang and Wang (1992) using involved algebraic
methods. Henceforth, we call (𝐜) the conic associated with a given polygon 𝐜.
The key idea (Fig. 1) is applying a suitable perspective map Φ to an initial semicircle 𝐂(𝑢) to generate a conic Φ(𝐂(𝑢)) with
polygon 𝐜. We choose a semicircle as a proto-segment since it admits a simple cubic form (Hoschek and Lasser, 1993; Piegl and
Tiller, 1997), with polygon 𝐂 forming a square and weights

{𝑊𝑖 }3𝑖=0 = {1, 13 , 13 , 1}. (1)

Next, we construct the unique perspective map Φ (Wolberg, 1990) such that Φ(𝐂𝑖 ) = 𝐜𝑖 . Sederberg and Zheng (2015) observed that
Φ can be written as a degree-(1,1) rational Bézier map, with control points coinciding with 𝐜𝑖 and weights equal to the signed area
𝐴𝑖 of the triangle {𝐜𝑘 }3𝑘=0,𝑘≠𝑖 that results from dropping 𝐜𝑖 in the list 𝐜. By the projective invariance of the rational Bézier form (Farin
and Hansford, 2000), the conic 𝐜(𝑢) = Φ(𝐂(𝑢)) has a cubic Bézier polygon 𝐜 and weights 𝑤𝑖 = 𝑊𝑖 𝐴𝑖 :
{ }3 { }
𝑤𝑖 𝑖=0
= 𝐴0 , 13 𝐴1 , 13 𝐴2 , 𝐴3 . (2)
Note two remarkable properties of 𝐜(𝑢), preserved from 𝐂(𝑢) by a perspective Φ:

1
• The diagonals 𝐜0 𝐜2 and 𝐜1 𝐜3 meet at the point 𝐜( ) on the conic.
2
• The tangent at this point and the lines extending the segments 𝐜0 𝐜3 , 𝐜1 𝐜2 are concurrent (at infinity, for parallel lines).

We have expressed 𝐂(𝑢) in standard form, with unit end weights (1). However, we could submit 𝐂(𝑢) to a Moebius reparametriza-
tion, inherited by 𝐜(𝑢), and that alters the weights while keeping the conic shape and control polygon. To incorporate this possibility,
from the weights 𝑤𝑖 (2) we derive the shape invariants, aka shape factors (Piegl and Tiller, 1997), that guarantee a conic:

𝑤21 𝐴21 𝑤22 𝐴22

𝜎1 = = , 𝜎2 = = . (3)
𝑤0 𝑤2 3𝐴0 𝐴2 𝑤3 𝑤1 3𝐴1 𝐴3

2
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 2. Conic associated with a cubic polygon 𝐜: Different types of quadrilaterals (𝐜) formed by 𝐜.

1 1
A Moebius reparametrization of 𝐂(𝑢) resulting in weights {1, − , , −1} generates the complementary semicircle, so the shape factors
3 3
(3) actually define two complementary conic segments, making up a whole conic (𝐜). In Fig. 2, we have plotted (thick line) the
segment 𝐜(𝑢), 𝑢 ∈ [0, 1] obtained by mapping 𝐂(𝑢), as well as that (thin line) generated by the complementary semicircle.
To avoid any 𝐴𝑖 = 0 and thus vanishing denominators (3), we assume a nondegenerate polygon 𝐜 (no three points 𝐜𝑖 collinear).
We place no further restriction on 𝐜, which could be concave or even crossed (self-intersecting), as that in Fig. 2 and already appear-
ing in (Barrowclough, 2014). Unsurprisingly, such geometries lead to negative weights. Although software packages for symbolic
computation, such as Mathematica, admit them, most CAD systems do not, so we must identify the geometry resulting in a (𝐜)
expressible with positive weights. This characterization turns out to be straightforward in terms of the quadrilateral (𝐜) formed
by 𝐜, since a strictly convex (𝐜) is tantamount to areas 𝐴𝑖 (and hence weights) sharing a common sign. Positive shape factors also
characterize a convex (𝐜) because, in the nonconvex (concave or crossed) case, the areas have different (and non-alternating) signs,
which implies at least one negative shape factor:

Lemma 1. For the conic (𝐜) associated with a cubic polygon 𝐜, the following properties are equivalent:

(i) The quadrilateral (𝐜) defined by 𝐜 is strictly convex.

(ii) (𝐜) admits a parametrization 𝐜(𝑢) with positive weights (in Bézier form with polygon 𝐜).
(iii) (𝐜) has positive cubic shape factors (3) 𝜎1 , 𝜎2 > 0.

3. Generalized degree-elevation of a quadratic conic (inserting a base point)

The preceding Section has shown that any planar cubic polygon 𝐜 defines an associated conic. We find now what polygons 𝐜 a
given Bézier conic  can generate. First, we recall some basics from algebraic geometry, adopting the notation (Farin, 1999) whereby
homogeneous curves and control points are underlined to distinguish them from their affine counterparts.
Given a nondegenerate conic  (parabola, ellipse or hyperbola), which means a properly parametrized quadratic curve 𝐛(𝑢), there
exist two procedures (Sederberg, 1984) for obtaining higher-degree representations of 𝐛(𝑢):

(i) Nonlinear rational parameter substitution 𝑢(𝑣) of degree 𝑚 ≥ 2, resulting in an improperly parameterized representation (Seder-
berg, 1986) of even degree 𝑛 = 2𝑚.
(ii) Multiplying the homogeneous curve 𝐛(𝑢) by a degree-𝑚 polynomial 𝑎(𝑢), which raises the degree to 𝑛 = 2 + 𝑚.

Clearly, procedure (i) yields 𝑛 ≥ 4, so the only alternative to generate a cubic is (ii), with 𝑚 = 1:

𝐜(𝑢) = 𝑎(𝑢)𝐛(𝑢), 𝑎(𝑢) = 𝑎0 (1 − 𝑢) + 𝑎1 𝑢, (4)

where we have written the linear function 𝑎(𝑢), 𝑢 ∈ [0, 1] in Bernstein form, with coefficients {𝑎0 , 𝑎1 }. This operation is tantamount
to introducing a base point (Farin, 1999), the root 𝑢𝑎 = 𝑎0 ∕(𝑎0 − 𝑎1 ) of 𝑎(𝑢). Let us summarize this result formally, rewriting in a more
precise way the algebraic characterization of cubic conics found in (Boehm and Prautzsch, 1994):

Lemma 2. All cubic representations of a nondegenerate conic have one base point, i.e., a homogeneous form 𝐜(𝑢) that can be factorized as
the product (4).

All nondegenerate conics in quadratic form 𝐛(𝑢) are properly reparameterized and hence related by a Moebius reparametrization,
a property that the introduction of a base point (4) transfers to the cubic form:

Corollary 3. All cubic representations of a nondegenerate conic relate by a Moebius reparametrization.

3
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 3. Obtaining a cubic polygon 𝐜 from a quadratic conic 𝐛(𝑢) via generalized degree-elevation: Cases where 𝐜 forms a strictly convex quadrilateral (𝐜).

Denker and Herron (1997) employ the term generalized degree elevation to denote the insertion of a base point as it encompasses
standard degree elevation (𝑎0 = 𝑎1 ), as the limit case of a base point 𝑢𝑎 at infinity. They provide formulas in the general case 𝑛 − 1 → 𝑛
of a final degree 𝑛, so we have only to particularize them for 𝑛 = 3. In terms of the homogeneous Bézier points 𝐛𝑖 of 𝐛(𝑢), 𝑢 ∈ [0, 1],
the resulting cubic points 𝐜𝑖 take the form:
{ }3 { 1 }
𝐜𝑖 𝑖=0 = 𝑎0 𝐛0 , 𝑎 𝐛
3 1 0
+ 23 𝑎0 𝐛1 , 2
𝑎 𝐛
3 1 1
+ 13 𝑎0 𝐛2 , 𝑎1 𝐛2 . (5)
In affine space, the cubic endpoints 𝐜0 = 𝐛0 , 𝐜3 = 𝐛2 coincide with those of the conic, whereas 𝐜1 , 𝐜2 lie on their respective legs
𝐛0 𝐛1 , 𝐛1 𝐛2 , with ratios depending on the base point and the original quadratic weights 𝑧𝑖 :
𝑧1 2 𝑧1 𝑎0
𝜆1 = ratio(𝐛0 , 𝐜1 , 𝐛1 ) = 2𝛼 , 𝜆2 = ratio(𝐛2 , 𝐜2 , 𝐛1 ) = , 𝛼= . (6)
𝑧0 𝛼 𝑧2 𝑎1
Denker and Herron (1997) do not consider the case of an inner point 𝐛1 at infinity (𝑧1 = 0), i.e., an homogeneous point 𝐛1 =
{𝑥1 , 𝑦1 , 0} mapping to a pure vector ⃗𝐛1 = {𝑥1 , 𝑦1 }. However, this case requires a particularized analysis because it furnishes parallel
legs 𝐜0 𝐜1 ‖𝐜3 𝐜2 , as Fig. 3(b) shows, so the resulting ratios (6) 𝜆1 = 𝜆2 = 0 become useless to locate the inner points 𝐜1 , 𝐜2 along their
legs. Luckily, projecting 𝐜1 , 𝐜2 (5) leads to simple formulas for the affine positions of 𝐜1 , 𝐜2 in terms of their signed distances 𝑑1 , 𝑑2
(along ⃗𝐛1 ) to 𝐜0 , 𝐜3 :
2𝛼 ⃗ 2 ⃗
𝑑1 = ‖𝐛 ‖, 𝑑2 = ‖𝐛 ‖. (7)
𝑧0 1 𝛼𝑧2 1
A nondegenerate conic guarantees end weights 𝑧0 , 𝑧2 ≠ 0, but expressions (6), (7) are still ill-defined if 𝑎0 = 0 or 𝑎1 = 0, that is, if
the base point 𝑢𝑎 lies at either of the endpoints of the domain [0, 1]. This location implies a vanishing ratio (6) or distance (7), hence
coinciding points 𝐜0 = 𝐜1 or 𝐜3 = 𝐜2 . We avoid this degeneracy by assuming, as in Section 2, a nondegenerate 𝐜 with no three points
collinear, but we place no other restriction on 𝐜. We consider general signed ratios, that is, 𝐜1 , 𝐜2 not confined to their respective
legs 𝐛0 𝐛1 , 𝐛2 𝐛1 (𝜆1 , 𝜆2 > 0), so they can lie outside them (𝜆1 < 0 or 𝜆2 < 0), on the lines 1 , 2 these segments define. A simple
analysis reveals that only the three cases (a-c) depicted in Fig. 3 generate a strictly convex (𝐜), but a nonconvex (𝐜), i.e., concave
or crossed (Fig. 2), is also admissible.
At first sight, it might seem that a Moebius reparametrization of 𝐛(𝑢) could provide extra freedom to generalized degree-elevation,
in addition to the scalar 𝛼 (6). However, reparametrization means replacing the original weights 𝑧𝑖 with {𝑧0 , 𝛽𝑧1 , 𝛽 2 𝑧2 }, 𝛽 ≠ 0, which
amounts to employing 𝛼 ̂ = 𝛼𝛽 instead of 𝛼 in the ratios (6) or distances (7). For instance, Fig. 3 plots the segment contained in
(𝐜), but its complementary one, obtained with 𝛽 = −1, would also generate 𝐜 with 𝛼 ̂ = −𝛼 . We conclude that a given Bézier conic 
generates a one-parameter family of cubic polygons 𝐜(, 𝛼), as there exists just one degree of freedom, namely 𝛼 . This non-uniqueness
of the cubic Bézier representation of conics, already observed by Farin and Worsey (1991), comes as no surprise: The Bézier polygon
is unique only for properly parameterized curves with no base points (Berry and Patterson, 1997).

4. Finding the unique conic generating a given cubic polygon 𝐜

In the preceding section, we obtained the polygons 𝐜(, 𝛼) that a given Bézier conic  generates. Now, we solve the reverse
problem and find what conic can generate a given 𝐜, which turns out to be unique and must hence coincide with the conic (𝐜)
associated with 𝐜 (Section 2).
From Lemma 2, we derived the ratios 𝜆1 , 𝜆2 (6) for any cubic conic. By multiplying them, the factor 𝛼 cancels out, and we readily
get the conic shape factor identifying (𝐜):

4
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 4. Particular cases of parallel end legs: Semicircle (𝜎∞ = 1) and segments of an equilateral hyperbola (𝜎∞ = −1).

𝑧21 𝜆1 𝜆2
𝜎= = . (8)
𝑧0 𝑧2 4
Once again, parallel end legs require a particularized analysis as always 𝜎 = 0, which conveys no shape information. We propound
an alternative invariant 𝜎∞ to Moebius reparameterizations, resulting from the product of distances (7):

‖⃗𝐛1 ‖2 𝑑1 𝑑2
𝜎∞ = 4 = , 𝑑 = ‖𝐛0 𝐛2 ‖, (9)
𝑧0 𝑧2 𝑑2 𝑑2
where the normalizing factor 𝑑 2 in the denominator adds invariance to similarities. This bespoke factor indicates shape as follows:

• 𝜎∞ > 0: ellipse. The segment is an affine version of a semicircle, with factor 𝜎∞ = 1 because its standard quadratic form has
weights {𝑧𝑖 }2𝑖=0 = {1, 0, 1} and a vector of length ‖⃗𝐛1 ‖ = 𝑑∕2. Standard degree-elevation of this semicircle (Fig. 4, left) yields the
cubic form employed in Section 2, with distances (7) 𝑑1 = 𝑑2 = 𝑑 defining a square (𝐜).
• 𝜎∞ < 0: hyperbola (two disconnected segments). Segments on an equilateral hyperbola (𝜎∞ = −1) have the quadratic form
described in (Sánchez-Reyes and Fernández-Jambrina, 2008), with ‖⃗𝐛1 ‖ = 𝑑∕2 and weights {1, 0, −1}. Standard degree elevation
(Fig. 4, right) results in distances 𝑑1 = −𝑑2 = 𝑑 defining a cross-trapezoid (𝐜).

Let us summarize our results in the following Theorem, subsuming Theorem 3 in (Wang and Wang, 1992) and completing it with
the conic shape factor for parallel end legs:

Theorem 4. Given a planar, nondegenerate cubic polygon 𝐜 = {𝐜𝑖 }3𝑖=0 , the only conic generating 𝐜 is the conic (𝐜) associated with 𝐜. In
quadratic Bézier form, (𝐜) has endpoints 𝐛0 = 𝐜0 , 𝐛2 = 𝐜3 , inner point 𝐛1 where the end legs of 𝐜 intersect (at infinity if 𝐜0 𝐜1 ‖𝐜3 𝐜2 ), and
conic shape factor (8) or (9).

5. Characterizing conics in cubic Bézier form

Lemma 2 algebraically characterizes a conic in homogeneous cubic form 𝐜(𝑢) as having a common root in all its three components.
This root could be detected with specific methods (Denker and Herron, 1997; Farin and Worsey, 1991) that involve finding the
roots of cubic polynomials, or (Sederberg, 1984) by computing their GCD (Greatest Common Divisor), which can be done without
abandoning the Bézier form (Bourne et al., 2020; Busé and Goldman, 2007).
Our previous results lead to simpler ways to identify Bézier conics in cubic form. Theorem 4 states that the conic (𝐜) associated
with a given polygon 𝐜 is unique. Since by Corollary 3 all cubic versions of (𝐜) relate by a Moebius reparametrization, which does
not alter the cubic shape factors, we conclude that the values (3) characterize the only possible conic with given 𝐜. Also, we can
rewrite this property with a more geometric flavor by expressing (Farin, 1999) the oriented curvatures 𝜅0 , 𝜅3 at the endpoints 𝐜0 , 𝐜3
in terms of 𝜎1 , 𝜎2 :

4 𝐴3 4 𝐴0
𝜅0 = , 𝜅3 = .
3𝜎1 ‖𝐜0 𝐜1 ‖3 3𝜎2 ‖𝐜3 𝐜2 ‖3
For cubics with same 𝐜, sharing 𝜎1 , 𝜎2 is clearly tantamount to sharing 𝜅0 , 𝜅3 . Let us put together these results:

Theorem 5. Given a rational Bézier cubic with a planar, nondegenerate polygon 𝐜, the following properties are equivalent:

5
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 5. Extended range (c) over quadratics (a) of the rho-value 𝜌 offered by cubic conics with positive weights.

(i) The cubic is a nondegenerate conic.

(ii) The cubic has the shape factors 𝜎1 , 𝜎2 (3) of the conic (𝐜) associated with 𝐜.
(iii) At the endpoints 𝐜0 , 𝐜3 , it has the oriented curvatures of (𝐜).

Some remarks about these characterizations:

(ii) It extends Theorem 2 by Wang and Wang (1992), which characterizes cubic conics restricted to positive weights and derives the
convexity of 𝐜 as a condition. Indeed, positive weights are tantamount to a strictly convex , which implies a convex 𝐜, but this
is a different issue. We neatly separate these two problems: detection of cubic conics and the geometry of 𝐜 for positive weights
(Lemma 1).
(iii) It borrows from Farin (1999) the interpretation of (𝐜) as a degree-reduced version of a given cubic. In principle, (𝐜) provides
only 𝐺1 Hermite approximation, but if the approximation order improves to 𝐺2 , then it coincides with the cubic.

6. Additional expressive power of the cubic Bézier form: extended range of the rho-value

As observed in the Introduction, the cubic Bézier representation of conics deserves attention, among other reasons, for its higher
expressive power (sticking to positive weights). Indeed, Lemma 1 states that positive weights imply a strictly convex (𝐜), which
translates into only three possible ranges of ratios/distances (Fig. 3), and by Theorem 4 in the following ranges of shape factors:

(a) 𝜆1 , 𝜆2 ∈ (0, ∞) ⇒ 𝜎 > 0.

(b) 𝑑1 𝑑2 > 0 ⇒ 𝜎∞ > 0 (semiellipse).
(c) 𝜆1 , 𝜆2 ∈ (−1, 0) ⇒ 𝜎 ∈ (0, 1∕4) (elliptical segment).

Whereas case (a) does not extend the range over quadratics, cases (b),(c) do because a quadratic would require zero or negative
weights.
In case (c), this added expressive power boils down to extending the admissible locations for the shoulder point 𝐬, the point on
the conic characterized by a tangent parallel to the chord 𝐛0 𝐛2 . In a quadratic, the user employs 𝜎 as shape handle to displace 𝐬,
1
but positive weights confine 𝐬 to the median 𝐌𝐛1 of the triangle 𝐛0 𝐛1 𝐛2 , where 𝐌 = (𝐛0 + 𝐛2 ). In contrast, with cubics 𝐬 can move
2
beyond 𝐌, along the ray  extending the median. Let us investigate what happens to the conic rho-value 𝜌 (Lee, 1987; Wolters,
2002), the barycentric coordinate of 𝐬 on the median:
√
± 𝜎
𝐬 = (1 − 𝜌)𝐌 + 𝜌𝐛1 , 𝜌= √ . (10)
1± 𝜎
The ± option signals that, in fact, the conic comprises two complementary segments, each one with its shoulder point. In quadratics,
positive weights (plus sign) restrict the rho-value to 𝜌 ∈ (0, 1), whereas case (c) (minus sign) adds the range 𝜌 ∈ (−1, 0), as Fig. 5
illustrates.
Regarding how to deal with the degree of freedom in the cubic polygon 𝐜 representing a given conic, the analysis in Section 3
provides a neat way to control it: Choose either 𝜆1 or 𝜆2 (𝑑1 or 𝑑2 ) and compute the other ratio or distance from equalities (8) or
(9). This procedure is way simpler and more intuitive than other options in the literature, such as selecting the free parameter that
Theorem 1 by Hang and Chou (2013) involves. A sensible choice for the above cases (a-c) is:
√ √
(a),(c): 𝜆 = 𝜆1 = 𝜆2 = ±2 𝜎, (b): 𝑑1 = 𝑑2 = 𝑑 𝜎∞ , (11)

6
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 6. Cubic polygons resulting from standard degree-elevation of a quadratic conic 𝐛(𝑢).

which corresponds to standard degree-elevation (𝛼 = 1) of a quadratic 𝐛(𝑢) in standard form (𝑧0 = 𝑧2 = 1). As Fig. 6 shows, the
resulting middle leg 𝐜1 𝐜2 is parallel to 𝐜0 𝐜3 and defines a trapezoid, the quadrilateral (𝐜) with the tightest bounding box, in
1
particular, a parallelogram in case (b). For cases (a),(c), the midpoint 𝐦 = (𝐜1 + 𝐜2 ) lies on the line  at a position determined by
2
𝜆 = ratio(𝐌, 𝐦, 𝐛1 ). Since ratio(𝐌, 𝐬, 𝐛1 ) = 𝜆∕2 by combining (10), (11), points 𝐌, 𝐬, 𝐦, 𝐛1 on  are harmonic (Farin, 1999), i.e., they
have cross ratio 1/2. Observe that this property trivially holds for the semicircle 𝐂(𝑢) in Fig. 1, and perspective maps preserve cross
ratios.

7. Cubic circles

In this section, we particularize our investigation to the remarkable case of cubic circles. Their characterization comes from
combining Theorems 4, 5, in a compact version of the main result by Wang (1991):

Corollary 6. A rational Bézier cubic with a planar, nondegenerate polygon 𝐜 is a circular arc iff:

(i) The conic (𝐜) associated with 𝐜 is a circle, and

(ii) The cubic has the shape factors 𝜎1 , 𝜎2 (3) of (𝐜).

Condition (i) is tantamount to a quadratic form 𝐛(𝑢), specified by Theorem 4, corresponding to a circular arc. If 𝐛1 does not lie
at infinity, the required geometry is well-known (Wolters, 2002): 𝐛0 𝐛1 𝐛2 must form an isosceles triangle, and the conic shape factor
(8) take the value 𝜎 = cos2 𝜃 , where 𝜃 ∈ (0, 𝜋∕2) denotes the base angle of the triangle. In the limit case 𝜃 = 𝜋∕2 (𝐛1 at infinity), a
factor 𝜎∞ = 1 indicates a semicircle, as noted in Section 4. Therefore, condition (i) spells out as:

𝜆1 𝜆2 = 4 cos2 𝜃 if 𝜃 ∈ (0, 𝜋∕2)

∠𝐛2 𝐛0 𝐛1 = ∠𝐛0 𝐛2 𝐛1 = 𝜃, .
𝑑1 𝑑2 = 𝑑 2 if 𝜃 = 𝜋∕2
As a result, the three cases (a-c) with positive weights listed in Section 6 furnish arcs (Fig. 7) spanning the following angles Δ:

(a) 𝜆1 , 𝜆2 ∈ (0, ∞) ⇒ 𝜃 ∈ (0, 𝜋∕2) ⇒ Δ = 2𝜃 ∈ (0, 𝜋).

(b) 𝑑1 𝑑2 > 0 ⇒ Δ = 2𝜃 = 𝜋 .
(c) 𝜆1 , 𝜆2 ∈ (−1, 0) ⇒ 𝜃 ∈ (𝜋∕3, 𝜋∕2) ⇒ Δ = 2(𝜋 − 𝜃) ∈ (𝜋, 4𝜋∕3).

Thus, we readily reach the restriction (Chou, 1995; Wang, 1991) that, in cubic Bézier form, only arcs spanning Δ ∈ (0, 4𝜋∕3) admit
positive weights, which corresponds to the range 𝜌 ∈ (−1, 1).
Finally, the choice (11) yields a symmetric trapezoid (𝐜) with 𝜆 = ±2 cos 𝜃 in cases (a),(c), or a square (𝐜) for the semicircle
(Fig. 4, left) of case (b).

8. Conclusions

We have shed light on the cubic Bézier representation of conics by appealing to geometry rather than algebraic manipulation.
Previous results are expanded, reinterpreted, and combined in a simplified and more logical fashion in terms of shape invariants,
ratios, and the concept of conic (𝐜) associated with a cubic control polygon 𝐜. This unique conic comes from a cubic semicircle,
whose control polygon forms a square, by mapping this square to 𝐜 via a perspective transformation. Detecting whether a given cubic
is a conic amounts to comparing the cubic with (𝐜): They coincide if they share shape factors or, equivalently, oriented curvatures
at the endpoints so that (𝐜) provides 𝐺2 interpolation. We pose no restriction on 𝐜, save that it must not have three of its points
collinear to guarantee a nondegenerate conic, albeit positive cubic weights are restricted to polygons 𝐜 forming a strictly convex

7
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Fig. 7. Cubic circles with positive weights, spanning an angle Δ: (a) Δ ∈ (0, 𝜋), (b) Δ = 𝜋 , (c) Δ ∈ (𝜋, 4𝜋∕3).

quadrilateral (𝐜). We thus separate the geometry required for positive weights from the problem of detecting cubic conics. Also, we
rigorously address the case resulting from a quadratic with the inner point at infinity by introducing a bespoke conic shape factor.
In the customary quadratic Bézier representation of conics, positive weights restrict rho-values to 𝜌 ∈ (0, 1). The cubic form offers
added expressive power by duplicating the allowable range (with positive weights) to 𝜌 ∈ (−1, 1), allowing elliptical segments with
𝜌 ∈ (−1, 0) and semiellipses. For circles, this extended range means arcs spanning an angle up to 4𝜋∕3. Recall that positive weights
are required to guarantee containment in the convex hull or to compute bounds on the magnitude of the derivatives, such as those
proposed in (Deng and Li, 2013) and related works.
In contrast, cubics do not add any additional freedom regarding possible parametrizations (shared with quadratics), so the optimal
one (closer to arc length) coincides with that found by Cattiaux-Huillard et al. (2009) for quadratics. To improve it, we should resort
to improper parametrizations, which means at least quartics (Hu and Wang, 2007).
Cubic conics can be characterized in other rational bases, as done by Hu and Wang (2011) in the rational Ball basis (Qiulin
and Davies, 1987), or Hu and Wang (2010) in the rational version (Dejdumrong, 2006) of the basis put forward by Delgado and
Peña (2003). However, only the rational Bézier form yields simple formulas in terms of shape factors, invariant with Moebius
reparameterizations, because such factors are exclusive to the Bézier form (Sánchez-Reyes, 2023). In the rational form associated
with any other polynomial basis, reparameterization changes not only the weights but also the control polygon (Sánchez-Reyes,
2022), complicating the analysis. This more straightforward treatment of cubic conics provides yet another reason, besides those
already known (Farouki, 2012), to adhere to the Bernstein-Bézier representation for CAGD applications.

Declaration of competing interest

The author declares that he has no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

No data was used for the research described in the article.

Acknowledgements

This work was supported by grant PID2019-104586RB-I00 funded by MCIN/AEI/10.13039/501100011033; by grant PID2022-
137598NB-I00 funded by MCIN/AEI/10.13039/501100011033/FEDER,UE; and grant SBPLY/19/180501/000247 by Consejería de
Educación Cultura y Deportes (Junta de Comunidades de Castilla-La Mancha) and grant 2022-GRIN-34077 by Universidad de Castilla-La
Mancha, co-financed by the European Union (ERDF, European Regional Development Fund).

References

Barnhill, R.E., 1992. Geometry Processing for Design and Manufacturing. SIAM, Philadelphia.
Barrowclough, O.J.D., 2014. A basis for the implicit representation of planar rational cubic Bézier curves. Comput. Aided Geom. Des. 31, 148–167. https://doi.org/
10.1016/j.cagd.2014.02.007.

8
J. Sánchez-Reyes Computer Aided Geometric Design 108 (2024) 102266

Berry, T.G., Patterson, R.R., 1997. The uniqueness of Bézier control points. Comput. Aided Geom. Des. 14, 877–879. https://doi.org/10.1016/S0167-8396(97)00016-2.
Boehm, W., Prautzsch, H., 1994. Geometric Concepts for Geometric Design. AK Peters.
Bourne, M., Winkler, J.B., Su, Y., 2020. The computation of the degree of the greatest common divisor of three Bernstein basis polynomials. J. Comput. Appl.
Math. 373, 112373. https://doi.org/10.1016/j.cam.2019.112373.
Busé, L., Goldman, R., 2007. Division algorithms for Bernstein polynomials. Comput. Aided Geom. Des. 25 (9), 850–865. https://doi.org/10.1016/j.cagd.2007.10.003.
Cattiaux-Huillard, I., Albrecht, G., Hernández-Mederos, V., 2009. Optimal parameterization of rational quadratic curves. Comput. Aided Geom. Des. 26 (7), 25–732.
https://doi.org/10.1016/j.cagd.2009.03.008.
Chou, J.J., 1995. Higher order Bézier circles. Comput. Aided Des. 27 (4), 303–309. https://doi.org/10.1016/0010-4485(95)91140-G.
Dejdumrong, N., 2006. Rational DP-Ball curves. In: Proc. Int. Conf. on Computer Graphics, Imaging and Visualization, pp. 478–483.
Delgado, J., Peña, J.M., 2003. A shape preserving representation with an evaluation algorithm of linear complexity. Comput. Aided Geom. Des. 20, 1–20. https://
doi.org/10.1016/S0167-8396(02)00190-5.
Deng, C., Li, Y., 2013. A new bound on the magnitude of the derivative of rational Bézier curve. Comput. Aided Geom. Des. 30, 175–180. https://doi.org/10.1016/j.
cagd.2012.11.002.
Denker, W.A., Herron, G.J., 1997. Generalizing rational degree elevation. Comput. Aided Geom. Des. 30, 399–406. https://doi.org/10.1016/S0167-8396(96)00036-2.
Farin, G., 1999. NURBS: from Projective Geometry to Practical Use. AK Peters.
Farin, G., Hansford, D., 2000. The Essentials of CAGD. AK Peters.
Farin, G., Worsey, A., 1991. Reparameterization and degree elevation for rational Bézier curves. In: Farin, G. (Ed.), NURBS for Curve and Surface Design. SIAM,
Philadelphia, pp. 47–58.
Farouki, R.T., 2012. The Bernstein polynomial basis: a centennial retrospective. Comput. Aided Geom. Des. 29, 379–419. https://doi.org/10.1016/j.cagd.2012.03.001.
Fiorot, J.-C., Jeannin, P., Taleb, S., 1995. Rational cubic curves as BR-curves. Comput. Aided Geom. Des. 12, 67–88. https://doi.org/10.1016/0167-8396(93)E0058-L.
Hang, H., Chou, R., 2013. Investigation of practical representation and parameterization of the rational cubic conic sections. Telkomnika 111 (1), 9–18. https://
doi.org/10.12928/telkomnika.v11i1.
Hoschek, J., Lasser, D., 1993. Fundamentals of Computer Aided Geometric Design. AK Peters.
Hu, Q.-Q., Wang, G.-J., 2007. Necessary and sufficient conditions for rational quartic representation of conic sections. J. Comput. Appl. Math. 203, 190–208. https://
doi.org/10.1016/j.cam.2006.03.024.
Hu, Q.-Q., Wang, G.-J., 2010. Representing conics by low degree rational DP curves. J. Zhejiang Univ. Sci. C 11, 278–289. https://doi.org/10.1631/jzus.C0910148.
Hu, Q.-Q., Wang, G.-J., 2011. Rational cubic/quartic Said-Ball conics. Appl. Math. J. Chin. Univ. 26, 198–212.
Lee, E., 1987. The rational Bézier representation for conics. In: Farin, G. (Ed.), Geometric Modeling: Algorithms and New Trends. SIAM, Philadelphia, pp. 3–20.
Li, Y., Deng, C., Ma, W., 2014. C-shaped 𝐺2 Hermite interpolation by rational cubic Bézier curve with conic precision. Comput. Aided Geom. Des. 31, 258–264.
https://doi.org/10.1016/j.cagd.2014.03.002.
Piegl, L., Tiller, W., 1997. The NURBS Book, 2nd ed. Springer.
Qiulin, D., Davies, B.J., 1987. Surface Engineering Geometry for Computer-Aided Design and Manufacture. Ellis Horwood, Chichester, UK.
Sánchez-Reyes, J., Fernández-Jambrina, L., 2008. Curves with rational chord-length parametrization. Comput. Aided Geom. Des. 25, 205–213. https://doi.org/10.
1016/j.cagd.2007.11.003.
Sánchez-Reyes, J., 2022. The uniqueness of the rational Bézier polygon is unique. Comput. Aided Geom. Des. 96, 102118. https://doi.org/10.1016/j.cagd.2022.
102118.
Sánchez-Reyes, J., 2023. Shape factors and shoulder points for shape control of rational Bézier curves. Comput. Aided Des. 157, 103477. https://doi.org/10.1016/j.
cad.2023.103477.
Sederberg, T.W., 1984. Degenerate parametric curves. Comput. Aided Geom. Des. 1, 301–307. https://doi.org/10.1016/0167-8396(84)90018-9.
Sederberg, T.W., 1986. Improperly parametrized rational curves. Comput. Aided Geom. Des. 3, 67–75. https://doi.org/10.1016/0167-8396(86)90025-7.
Sederberg, T.W., Zheng, J., 2015. Birational quadrilateral maps. Comput. Aided Geom. Des. 32, 1–4. https://doi.org/10.1016/j.cagd.2014.11.001.
Wang, G.J., 1991. Rational cubic arcs and their applications in CAD. Comput. Ind. 16, 283–288. https://doi.org/10.1016/0166-3615(91)90066-I.
Wang, G.J., Wang, G.-Z., 1992. The rational cubic representation of conics. Comput. Aided Geom. Des. 9 (6), 447–455. https://doi.org/10.1016/0167-8396(92)90043-
O.
Wolberg, G., 1990. Digital Image Warping. IEEE Computer Society Press.
Wolters, H.J., 2002. Rational techniques. In: Farin, G., Hoschek, J., Kim, M.-S. (Eds.), Handbook of Computer Aided Geometric Design. Elsevier, Amsterdam,
pp. 111–140.