Académique Documents
Professionnel Documents
Culture Documents
Rida T. Farouki
a,
, Carlotta Giannelli
b
, Maria Lucia Sampoli
c
,
Alessandra Sestini
d
a
Department of Mechanical and Aerospace Engineering, University of California, Davis, CA 95616, USA
b
Institut fr Angewandte Geometrie, Johannes Kepler Universitt Linz, Altenberger Strasse 69, A-4040 Linz, Austria
c
Dipartimento di Ingegneria dellInformazione e Scienze Matematiche, Universit di Siena, Pian dei Mantellini 44, 53100 Siena, Italy
d
Dipartimento di Matematica e Informatica, Universit degli Studi di Firenze, Viale Morgagni 67a, 50134 Firenze, Italy
a r t i c l e i n f o a b s t r a c t
Article history:
Received 1 February 2013
Received in revised form 5 September 2013
Accepted 26 November 2013
Available online 4 December 2013
Keywords:
Rotation-minimizing frames
Pythagorean-hodograph curves
Quaternions
Angular velocity
Rigid body motion
Ruled surfaces
An orthonormal frame (f
1
, f
2
, f
3
) is rotation-minimizing with respect to f
i
if its angular
velocity satises f
i
0 or, equivalently, the derivatives of f
j
and f
k
are both parallel
to f
i
. The Frenet frame (t, p, b) along a space curve is rotation-minimizing with respect to
the principal normal p, and in recent years adapted frames that are rotation-minimizing
with respect to the tangent t have attracted much interest. This study is concerned
with rotation-minimizing osculating frames (f, g, b) incorporating the binormal b, and
osculating-plane vectors f, g that have no rotation about b. These frame vectors may be
dened through a rotation of t, p by an angle equal to minus the integral of curvature
with respect to arc length. In aeronautical terms, the rotation-minimizing osculating frame
(RMOF) species yaw-free rigid-body motion along a curved path. For polynomial space
curves possessing rational Frenet frames, the existence of rational RMOFs is investigated,
and it is found that they must be of degree 7 at least. The RMOF is also employed to
construct a novel type of ruled surface, with the property that its tangent planes coincide
with the osculating planes of a given space curve, and its rulings exhibit the least possible
rate of rotation consistent with this constraint.
2013 Elsevier B.V. All rights reserved.
1. Introduction
A general spatial motion of a rigid body is specied by describing its position and orientation as functions of time. A par-
ticular point of the body (e.g., the center of mass) is usually chosen to describe position. To describe orientation, the varia-
tion of an orthonormal frame (e
1
, e
2
, e
3
) embedded within the body may be specied. In general, position and orientation
vary independently, but in certain motion problems they may be correlated. This study is concerned with constrained spatial
motions, in which the instantaneous angular velocity of a rigid body is related to the geometry of its center-of-mass path.
The Frenet frame is the most familiar orthonormal frame on a space curve, comprising the tangent t, principal normal p,
and binormal b =t p. When the Frenet frame is used to orient a body along a path, its angular velocity satises p 0
i.e., it has no component in the principal normal direction. This means that the body exhibits no instantaneous rotation
about the principal normal vector p from point to point along the path.
In aerodynamics, an embedded frame is used (Cook, 1997) to characterize variations in the attitude of an aircraft, in
terms of roll, pitch, and yaw axes through its center of mass the roll (or longitudinal) axis is aligned with the fuselage;
()
=ds/d,
where s is the cumulative arc length of r(), measured from some xed point. The curve r() is regular if its parametric
speed satises () = 0 for all . The variation of an orthonormal frame (e
1
(), e
2
(), e
3
()) dened along r() may be
specied by its angular velocity () through the relations
e
1
= e
1
, e
2
= e
2
, e
3
= e
3
. (1)
Since (e
1
, e
2
, e
3
) comprise a basis for R
3
we can write
=
1
e
1
+
2
e
2
+
3
e
3
, (2)
and hence the relations (1) become
e
1
=(
3
e
2
2
e
3
), e
2
=(
1
e
3
3
e
1
), e
3
=(
2
e
1
1
e
2
). (3)
For a given reference direction, specied by a unit vector eld c() along r(), one may characterize frames
(e
1
(), e
2
(), e
3
()) that are rotation-minimizing with respect to c() as follows.
Denition 1. The frame (e
1
, e
2
, e
3
) is rotation-minimizing with respect to c if its angular velocity has no component in
the direction of c, i.e., c 0.
Now if the frame vector e
1
() is chosen as the reference direction, the rotation-minimizing frame (RMF) satises
1
0
in (2). Eqs. (3) then yield an alternative characterization, in terms of the derivatives of e
2
() and e
3
().
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 29
Corollary 1. The frame (e
1
, e
2
, e
3
) is rotation-minimizing with respect to e
1
when e
2
or e
3
is always parallel to e
1
along the
curve r(). An analogous characterization holds when e
2
or e
3
is chosen as the reference direction.
For a frame that is rotation-minimizing with respect to e
1
, the vectors e
2
, e
3
exhibit no instantaneous rotation about e
1
.
They vary only because e
1
varies along r(), and they must remain orthogonal to it. This is equivalent to the observation in
Corollary 1 that the derivatives of e
2
, e
3
are parallel to e
1
.
The Frenet frame (t, p, b) on a space curve r() is dened (Kreyszig, 1959) by
t =
r
|r
|
, p =
r
|r
|
t, b =
r
|r
|
. (4)
The tangent t is the instantaneous direction of motion along the curve, the principal normal p points to the center of curvature,
and the binormal b =t p completes the frame. The orthogonal planes spanned by (t, p), (p, b), and (b, t) are the osculating,
normal, and rectifying planes, respectively.
The angular velocity of the frame (4) is given (Kreyszig, 1959) by the Darboux vector
=b +t, (5)
where the curvature () and torsion () of r() are dened by
=
p t
=
|r
|
|r
|
3
and =
p b
=
(r
) r
|r
|
2
.
When (e
1
, e
2
, e
3
) =(t, p, b) and is specied by (5), the relations (1) are equivalent (Kreyszig, 1959) to the well-known
FrenetSerret equations
_
_
t
_
_
=
_
_
0 0
0
0 0
_
_
_
_
t
p
b
_
_
. (6)
Remark 1. From Denition 1, we observe that the Frenet frame is rotation-minimizing with respect to the principal nor-
mal p, but not with respect to the tangent t or binormal b. For a plane curve, it is also rotation-minimizing with respect to
the tangent t.
Although the Frenet frame is not rotation-minimizing with respect to t, one can easily derive such a rotation-minimizing
frame from it. New normal-plane vectors (u, v) are specied through a rotation of (p, b) according to
u =cos p +sinb, v =sinp +cos b, (7)
where (Guggenheimer, 1989) we dene
() =
0
_
0
d, (8)
0
being an integration constant note that the minus sign is missing from (8) in Guggenheimer (1989). It is assumed in
(7) that r() is free of inections, so p and b are dened at each point. The frame (t, u, v) has angular velocity
=b =(sinu +cos v), (9)
satisfying t 0. A frame incorporating the tangent t is an adapted frame, and when it is rotation-minimizing with respect
to t it is called a rotation-minimizing adapted frame (RMAF). Such frames describe natural spatial motions of a rigid body that
maintains one principal axis aligned with the path tangent, while the other principal axes have no instantaneous rotation
about the tangent. These motions are useful in computer animation, the use of sweep operations in geometric design, robot
path planning, programming of CNC machines with rotary axes, and related applications.
By modication of the Frenet frame, one can also dene a frame that is rotation-minimizing with respect to the binor-
mal b. New osculating-plane vectors (f, g) are specied through a rotation of (t, p) according to
f =cos t +sinp, g =sint +cos p, (10)
where, in this case, we dene
() =
0
_
0
d. (11)
30 R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742
Fig. 1. Comparison of (left) the Frenet frame vectors p, b and (right) the rotation-minimizing adapted frame vectors u, v on the circular helix (13).
Fig. 2. Comparison of (left) the Frenet frame vectors t, p and (right) the rotation-minimizing osculating frame vectors f, g on the circular helix (13).
Using
= f, g
= g, b
R
2
+k
2
and b =k/
R
2
+k
2
as
t =(a sin, a cos , b), p =(cos , sin, 0), b =(b sin, b cos , a)
and the parametric speed, curvature, and torsion are =
R
2
+k
2
, =a/, = b/. The angular deviation (8) of the
RMAF normal-plane vectors (u, v) from the Frenet frame vectors (p, b) is thus =
0
b. Hence, for
0
= 0, the RMAF
basis vectors (7) are
u =(cos cos b b sin sinb, sin cos b +b cos sinb, a sinb),
v =(cos sinb +b sin cos b, sin sinb b cos cos b, a cos b).
Fig. 1 compares the Frenet frame and RMAF normal-plane vectors along the helix (13) with R =2.5 and k =0.6. The angular
deviation (11) of the RMOF osculating-plane vectors (f, g) from the Frenet frame vectors (t, p) is =
0
a, and hence for
0
=0 the RMOF basis vectors (10) are
f =(a sin cosa +cos sina, a cos cosa +sin sina, b cosa),
g =(a sin sina cos cosa, a cos sina sin cosa, b sina).
Fig. 2 compares the Frenet frame and RMOF osculating-plane vectors.
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 31
For the Frenet frame (t, p, b) we have (
1
,
2
,
3
) = (, 0, ) from (5), so a body aligned with this frame along
the path r() exhibits pitch-free motion. For the rotation-minimizing adapted frame (t, u, v) we have (
1
,
2
,
3
) =
(0, sin, cos ) from (9), so a body aligned with the RMAF exhibits roll-free motion. Finally, for the rotation-minimizing
osculating frame (f, g, b) we have angular velocity components (
1
,
2
,
3
) =( cos , sin, 0) from (12), and hence a
body aligned with the RMOF exhibits yaw-free motion.
Rotation-minimizing frames can also offer intuitive characterizations for certain special curves on smooth surfaces. The
Darboux frame (n, t, h) along a surface curve comprises the surface normal n, the curve tangent t, and the tangent normal
h =n t i.e., a unit vector in the surface tangent plane, orthogonal to the curve tangent. The variation of the Darboux
frame is specied (Kreyszig, 1959; Struik, 1961) in terms of its angular velocity
=
g
n
g
t
n
h
by the relations n
= n, t
= t, h
= h, where
g
=
h t
,
g
=
h n
,
n
=
n t
,
are the geodesic curvature, geodesic torsion, and normal curvature along the surface curve. In terms of these quantities, the
geodesics, line of curvatures, and asymptotic lines on a smooth surface may be characterized (Kreyszig, 1959; Struik, 1961) as
loci along which
g
0,
g
0, and
n
0, respectively. From Denition 1, we obtain the following alternative succinct
characterizations, which are useful in the construction of surface patches bounded by geodesics (Farouki et al., 2009c;
Paluszny, 2008; Snchez-Reyes and Dorado, 2008); lines of curvature (Biard et al., 2010; Li et al., 2011; Martin, 1983); and
asymptotic lines (Bayram et al., 2012).
Proposition 1. A surface curve is (i) a geodesic; (ii) a line of curvature; or (iii) an asymptotic line, if and only if the Darboux frame
(n, t, h) is rotation-minimizing with respect to (i) the surface normal n; (ii) the curve tangent t; or (iii) the tangent normal h.
3. Rational rotation-minimizing frames
Exact rational frames are preferable to approximations when possible, since they provide a precise and ecient means
of orienting a rigid body along a given path. After briey reviewing the theory of rational RMAFs below, we consider its
adaptation to the problem of identifying rational RMOFs.
3.1. Rational RMAF on PH curves
The existence of rational RMAFs on polynomial space curves has recently been studied (Farouki, 2010). A polynomial
space curve with a rational RMAF must be a Pythagorean-hodograph (PH) curve, since only the PH curves have rational unit
tangents (Farouki, 2008). The distinctive property of a polynomial PH curve r() = (x(), y(), z()) is that its derivative
components satisfy (Farouki, 2008) the condition
()
2
=x
2
() + y
2
() +z
2
() =
2
() (14)
for some polynomial (). A spatial PH curve may be generated (Choi et al., 2002) from a quaternion polynomial
A() =u() + v()i + p()j +q()k (15)
and its conjugate A
() =A()iA
() =
_
u
2
() + v
2
() p
2
() q
2
()
_
i +2
_
u()q() + v()p()
_
j
+2
_
v()q() u()p()
_
k. (16)
Alternatively, one may use complex polynomials
() =u() +iv(), () =q() +ip(), (17)
in the Hopf map expression
r
() =
_
()
()
2
, 2Re
_
()()
_
, 2Im
_
()()
__
. (18)
Lemma 1. If the complex polynomials in (18) are of the form () = f (), () =e
i
g() for real f (), g() the curve dened by
integrating (18) is planar. Furthermore, if g() =kf () for real k, it is just a straight line.
32 R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742
Proof. When () = f () and () =e
i
g() for constant , we have
x
() = f
2
() g
2
(), y
() =2 f ()g() cos , z
() =2 f ()g() sin.
Since siny
() +cos z
() = 0, we have siny
() +cos z
() = 0 and siny
() +cos z
() r
()
_
r
() =
() y
() z
()
x
() y
() z
()
x
() y
() z
()
are linearly dependent, so the torsion vanishes and the curve is planar. Also, note that expression (18) reduces to r
() =
f
2
()(1 k
2
, 2k cos , 2k sin) if g() = kf (), which denes a straight line. Lemma 1 also holds in the case () =
e
i
f (), () = g(). 2
On the PH curve dened by (15) and (16), a rational orthonormal adapted frame the EulerRodrigues frame (ERF) is
dened (Choi and Han, 2002) by
e
1
() =
A()iA
()
|A()|
2
, e
2
() =
A()jA
()
|A()|
2
, e
3
() =
A()kA
()
|A()|
2
.
Here e
1
is the curve tangent while e
2
, e
3
span the normal plane. The variation of the ERF is characterized by its angular
velocity through (1), where
() =
()
=u
2
() + v
2
() + p
2
() +q
2
()
is the parametric speed of the PH curve r(). Han (2008) showed that the PH curve dened by (15)(16) possesses a rational
RMAF if and only if relatively prime polynomials a(), b() exist, such that u(), v(), p(), q() satisfy
uv
v pq
+ p
q
u
2
+ v
2
+ p
2
+q
2
=
ab
b
a
2
+b
2
. (19)
This is equivalent to the existence of a rational normal-plane rotation that maps the ERF vectors e
2
(), e
3
() onto the RMAF
vectors u(), v() so the ERF angular velocity component in the e
1
direction is exactly cancelled. For PH quintics, specied
by quadratic quaternion polynomials with Bernstein coecients A
0
, A
1
, A
2
, the condition for a rational RMAF may be
expressed (Farouki, 2010) as a single constraint on these coecients, namely
A
2
iA
0
+A
0
iA
2
=2A
1
iA
1
.
In the Hopf map model, this is equivalent (Farouki, 2010) to the constraints
Re(
0
2
) =|
1
|
2
|
1
|
2
,
0
2
+
2
0
=2
1
1
on the Bernstein coecients
0
,
1
,
2
and
0
,
1
,
2
of (t) and (t).
3.2. Rational RMOF on DPH curves
The existence of rational RMOFs (f, g, b) on polynomial space curves is a more dicult problem. Whereas all PH curves
have rational tangents t(), their principal normals n() and binormals b() do not, in general, depend rationally on the
curve parameter, because of the |r
() r
2
=
_
y
_
2
+
_
z
_
2
+
_
x
_
2
must be the perfect square of a polynomial in . Now it can be shown (Farouki et al., 2004, 2009b) that every PH curve
satises the relation
() r
()
2
=4
2
()(),
where the polynomial can be expressed in terms of u, v, p, q and u
, v
, p
, q
p + vq
q
_
2
+
_
uq
q vp
+ v
p
_
2
. (20)
Hence, the binormal vector is rational if and only if we have
() =
2
() (21)
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 33
for some polynomial (), i.e.,
2
=
_
y
_
2
+
_
z
_
2
+
_
x
_
2
=4( )
2
. (22)
Polynomial curves satisfying (14) and (22) are double PH (DPH) curves (Beltran and Monterde, 2007; Farouki et al., 2009b).
They have rational Frenet frames, curvature, and torsion, dened by
t =
r
, p =
r
2
, b =
r
2
, =
2
2
, =
(r
) r
4
2
2
. (23)
Remark 2. Because of their rational Frenet frames, curvature, and torsion, we shall focus here on rational RMOFs on DPH
curves, and do not address at present the existence of non-DPH curves that possess rational binormals and admit rational
rotation-minimizing osculating-plane basis vectors f, g.
A (general) helix is characterized by the fact that its tangent t maintains a constant inclination (the pitch angle) with
respect to a xed direction a (the axis of the helix) and the curvature/torsion ratio is constant (Kreyszig, 1959; Struik, 1961).
A polynomial helix must be a PH curve (Farouki et al., 2004) in fact, a DPH curve (Farouki et al., 2009b). All PH cubics
and all helical PH quintics are DPH curves, but non-helical DPH curves of degree 7 exist. Complete details may be found
in Beltran and Monterde (2007), Farouki et al. (2009b), Monterde (2009).
Remark 3. For any DPH curve, the Frenet frame species a rational pitch-free motion, since the Frenet frame is rotation-
minimizing with respect to the principal normal, and it is also rational for double PH curves.
The ERF is of no use in constructing a rational RMOF, since it does not incorporate the binormal b and two rational
vectors spanning the osculating plane, to which a rotation can be applied to cancel the angular velocity component in the
direction of b. Instead, we use the rational Frenet frame (23) for a DPH curve, and seek a rational osculating-plane rotation
_
f()
g()
_
=
1
a
2
() +b
2
()
_
a
2
() b
2
() 2a()b()
2a()b() a
2
() b
2
()
_
_
t()
p()
_
(24)
that maps the Frenet frame vectors t(), p() to the RMOF vectors f(), g() and cancels the b component of the Frenet
frame angular velocity (5).
Now (10) and (24) give = 2tan
1
b/a and
= 2(a
b ab
)/(a
2
+b
2
), and if (24) denes an RMOF, we must have
= from (11). Thus, we deduce from (23) that a DPH curve satisfying (14) and (22) has a rational RMOF if and only
if polynomials a(), b() exist, such that
u
2
+ v
2
+ p
2
+q
2
=
ab
b
a
2
+b
2
. (25)
3.3. Rational RMOF on DPH cubics
Every spatial PH cubic is a helical curve (Farouki and Sakkalis, 1994), and also a DPH curve (Farouki et al., 2009b). Such
curves admit a closed-form (but not rational) expression for the RMOF. A cubic space curve with control points p
0
, p
1
, p
2
,
p
3
and control polygon leg vectors L
0
=p
1
p
0
, L
1
=p
2
p
1
, L
2
=p
3
p
2
is a PH curve (Farouki and Sakkalis, 1994) if and
only if L
0
, L
2
lie on a right-circular cone of half-angle
1
2
about L
1
as axis, and their azimuthal separation on this
cone is given by
cos =1
2L
2
1
L
0
L
2
, (26)
where L
i
= |L
i
|. Adopting suitable coordinates, one may assume that L
1
= L
1
(0, 0, 1), L
0
= L
0
(sin, 0, cos ), L
2
=
L
2
(sin cos , sin sin, cos ), with satisfying (26) and L
1
L
0
L
2
. The parametric speed is then
() =
()
=3
_
L
0
(1 )
2
+ L
1
cos 2(1 ) + L
2
2
_
, (27)
where = L
0
L
2
L
2
1
cos
2
0, so that () 0 for all . Since |r
| =6L
1
sin() and (r
) r
=108(L
0
L
1
) L
2
,
the curvature and torsion can be expressed (Farouki and Sakkalis, 1994) in terms of (27) as
() =
6L
1
sin
2
()
, () =
3L
0
L
2
sin
L
1
2
()
. (28)
Hence, the angle function (11) that species that RMOF vectors f, g in terms of the Frenet frame vectors t, p is given by
34 R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742
() =
2L
1
sin
tan
1
(L
1
cos L
0
)(1 ) +(L
2
L
1
cos )
,
where
0
=
0
+(2L
1
sin/
) tan
1
((L
1
cos L
0
)/
).
Proposition 2. The condition (25) for a rational RMOF can be satised by PH cubics only in the degenerate case of planar curves.
Proof. Since |r
|
2
= 4( )
2
and |r
| = 6L
1
sin we see that () = 3L
1
sin is a constant. Substituting this
together with (27) for the denominator on the left in (25), and assuming a() =a
0
(1 ) +a
1
, b() =b
0
(1 ) +b
1
on
the right, satisfaction of (25) requires that
a
0
b
1
a
1
b
0
=3L
1
sin, a
2
0
+b
2
0
=3L
0
, a
0
a
1
+b
0
b
1
=3L
1
cos , a
2
1
+b
2
1
=3L
2
.
With c
0
=a
0
+ib
0
, c
1
=a
1
+ib
1
these equations are equivalent to |c
0
|
2
=3L
0
, c
0
c
1
=3L
1
e
i
, |c
1
|
2
=3L
2
, and we must have
|c
0
| =
3L
0
, |c
0
||c
1
| =3L
1
, |c
1
| =
3L
2
. From (26) these conditions imply that cos =1, sin =0. It then follows from
(28) that the torsion vanishes, so the curve is planar. 2
3.4. Rational RMOF on DPH quintics
One is thus led to consider satisfaction of (25) in the case of DPH quintics, which are equivalent to the quintic helical
polynomial space curves: for such curves, () is a quadratic polynomial. The characterization of curves that have rational
RMOFs, as solutions to (25), is more dicult than the problem of identifying curves with rational RMAFs, satisfying (19).
Unlike (19), the numerator polynomial on the left in (25) is not known explicitly in terms of u, v, p, q instead, must
be determined from conditions (20)(21).
Now for the polynomials (17) dening the Hopf map form (18), we have
=
_
uq
q vp
+ v
p
_
+i
_
up
p + vq
q
_
,
and hence (20) can be written as = |
|
2
. The condition (21) for to be a perfect square is thus equivalent
(Farouki et al., 2009b; Monterde, 2009) to the requirement that
()
()
()() =h()w
2
() (29)
for a real polynomial h() and complex polynomial w() =r() +is() with gcd(r(), s()) =1, where deg(h) +2deg(w) =
2max(deg(), deg()) 2. The polynomial () dened by (21) is then given by () =h()|w()|
2
.
For DPH quintics with max(deg(), deg()) = 2, the solutions to (29) may be categorized according to whether
(a) deg(h) =0 and deg(w) =1; or (b) deg(h) =2 and deg(w) =0. These correspond (Beltran and Monterde, 2007; Farouki et
al., 2004, 2009b; Monterde, 2009) to the monotone helical and general helical PH quintics, respectively the former maintain
a xed sense of rotation of the tangent t about the axis a, while the latter may exhibit a reversal of the sense of rotation.
In case (a) we may set h() =1 and w() =w
0
(1 ) +w
1
without loss of generality. Satisfaction of (29) by quadratic
polynomials (t) and (t) with Bernstein coecients
0
,
1
,
2
and
0
,
1
,
2
then implies that
(
0
0
,
0
0
,
1
1
) =
_
1
2
w
2
0
, w
0
w
1
,
1
2
w
2
1
_
(30)
which is equivalent to (
0
1
1
0
)(
1
2
2
1
) =
1
4
(
0
2
2
0
)
2
. If this constraint is satised and w
0
, w
1
are
complex values determined from (30), the polynomial () in the rational RMOF condition (25) has the form
() =|w
0
|
2
(1 )
2
+Re(w
0
w
1
)2(1 ) +|w
1
|
2
2
.
In case (b) we set h() =h
0
(1 )
2
+h
1
2(1 ) +h
2
2
and w() =e
i
without loss of generality. Then satisfaction of
(29) implies that
(
0
0
,
0
0
,
1
1
) =
_
1
2
h
0
e
i2
, h
1
e
i2
,
1
2
h
2
e
i
_
(31)
i.e.,
0
0
,
0
0
,
1
1
are just real multiples of each other. In this case, with the coecients of (t)
and (t) satisfying (31), the polynomial () in (25) has the form () =h
0
(1 )
2
+h
1
2(1 ) +h
2
2
.
Eqs. (30) and (31) differ only in the complex values on the right. We now consider their solutions for general values
(z
0
, z
1
, z
2
) = (0, 0, 0) on the right, focusing on the generic case with z
1
= 0 and z
0
2
+ z
2
0
= 0 (the cases z
1
= 0 or
z
0
2
+z
2
0
=0 also yield only linear or planar curves: since the arguments for these cases are rather technical, we omit
them).
Remark 4. If (z
0
, z
1
, z
2
) = (0, 0, 0) the left-hand side of (25) vanishes, and it is satised with a(), b() constant, i.e., the
Frenet frame is an RMOF. But this occurs only if the curvature vanishes, so the curve is just a straight line.
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 35
Proposition 3. For z
1
=0, the solutions to the system of equations
0
=z
0
,
0
0
=z
1
,
1
1
=z
2
, (32)
where z
0
2
+z
2
0
=0 may be dened be expressing
1
,
0
,
2
in terms of
0
,
2
,
1
as
1
=
z
0
2
+z
2
0
z
1
,
0
=
(
0
1
z
0
)z
1
z
0
2
+z
2
0
,
2
=
(
2
1
+z
2
)z
1
z
0
2
+z
2
0
. (33)
Proof. Writing (32) as a linear system for
0
,
1
,
2
in the form
_
_
1
0
0
2
0
0
0
2
1
_
_
_
_
2
_
_
=
_
_
z
0
z
1
z
2
_
_
, (34)
the matrix on the left is seen to be of rank 2. Solutions exist if and only if the 34 augmented matrix, dened by appending
the vector on the right in (34) as a fourth column to the matrix on the left, is also of rank 2. The determinants of the three
3 3 matrices dened by combining two of the rst three columns of the augmented matrix with the last column are
2
(z
0
2
+z
2
0
z
1
1
),
1
(z
0
2
+z
2
0
z
1
1
),
0
(z
0
2
+z
2
0
z
1
1
).
Now (
0
,
1
,
2
) =(0, 0, 0) cannot satisfy (32) if z
1
=0, so we must have
z
0
2
+z
2
0
z
1
1
=0.
This condition yields the expression for
1
in (33) and if it holds, Eqs. (34) admit solutions in which one of
0
,
1
,
2
may be treated as a parameter. Choosing
1
as the free parameter, and substituting for
1
into the rst and third equations
of (32), we obtain the expressions for
0
,
2
in (33). 2
Hence, when Eqs. (32) are satised with z
1
=0 and z
0
2
+z
2
0
=0, we can express (t), (t) in terms of
0
,
2
,
1
and z
0
, z
1
, z
2
as
() =
0
(1 )
2
+
z
0
2
+z
2
0
z
1
2(1 ) +
2
2
,
() =
(
0
1
z
0
)z
1
z
0
2
+z
2
0
(1 )
2
+
1
2(1 ) +
(
2
1
+z
2
)z
1
z
0
2
+z
2
2
.
Considerable simplication is achieved by considering canonical-form curves, with (
0
,
0
) =(1, 0). Since canonical form is
achieved by a scaling/rotation, that maps r
(0) to the unit vector i, it does not alter the PH or DPH nature of a curve, or the
existence of rational rotation-minimizing frames on it (Farouki, 2010).
Setting
0
=1 and
1
=z
0
gives the canonical-form polynomials
() =(1 )
2
+
z
0
2
+z
2
z
1
2(1 ) +
2
2
,
() =z
0
2(1 ) +z
1
2
. (35)
These polynomials dene a monotone or general helical DPH quintic, when (z
0
, z
1
, z
2
) are of the form (
1
2
w
2
0
, w
0
w
1
,
1
2
w
2
1
)
or e
i2
(
1
2
h
0
, h
1
,
1
2
h
2
).
Now suppose that condition (25) is satised with quadratic polynomials a(), b(). Combining them into a complex
polynomial c() = a() + ib() with Bernstein coecients c
0
, c
1
, c
2
, we see that the right-hand side of (25) is equal to
Im(cc
)/|c|
2
. Since this expression is unchanged on dividing c() by any complex constant, we may choose c
0
=1 without
loss of generality.
With c
0
= 1, if
0
,
1
,
2
are the Bernstein coecients of (), satisfaction of (25) means that for some real number
=0 we must have
0
=2Im(c
1
),
1
=Im(c
2
),
2
=2Im(c
1
c
2
), (36)
|
0
|
2
+|
0
|
2
=,
Re(
0
1
+
0
1
) =Re(c
1
),
Re(
0
2
+
0
2
) +2
_
|
1
|
2
+|
1
|
2
_
=
_
Re(c
2
) +2|c
1
|
2
_
,
Re(
1
2
+
1
2
) =Re(c
1
c
2
),
|
2
|
2
+|
2
|
2
=|c
2
|
2
. (37)
36 R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742
3.4.1. Monotone helical quintics
For the DPH quintic specied by (35) with (z
0
, z
1
, z
2
) =(
1
2
w
2
0
, w
0
w
1
,
1
2
w
2
1
) and (
0
,
1
,
2
) =(|w
0
|
2
, Re(w
0
w
1
), |w
1
|
2
),
we obtain =1 from the rst equation in (37). Eqs. (36) and the other equations in (37) are then
|w
0
|
2
=2Im(c
1
), Re(w
0
w
1
) =Im(c
2
), |w
1
|
2
=2Im(c
1
c
2
),
Re(
1
) =Re(c
1
),
Re(
2
) +2|
1
|
2
+
1
2
|w
0
|
4
=Re(c
2
) +2|c
1
|
2
,
Re
_
2
+
1
2
|w
0
|
2
w
0
w
1
_
=Re(c
1
c
2
),
|
2
|
2
+|w
0
|
2
|w
1
|
2
=|c
2
|
2
,
where
1
=
1
2
(w
1
/w
0
) +
1
2
(w
0
/w
1
)
2
,
2
is free, and
1
=
1
2
w
2
0
,
2
= w
0
w
1
. Setting c
1
=a
1
+ ib
1
, c
2
=a
2
+ ib
2
, w
0
=
w
0
e
i
0
, w
1
= w
1
e
i
1
,
2
=u
2
+iv
2
, c =cos(
1
0
), s =sin(
1
0
) then gives
w
2
0
=2b
1
, w
0
w
1
c =b
2
, w
2
1
=2(a
1
b
2
a
2
b
1
), (38)
w
0
w
1
(cu
2
+sv
2
) +
w
1
w
0
c =2a
1
,
4c(cu
2
+sv
2
) +
w
2
0
w
2
1
_
u
2
2
+ v
2
2
_
+
w
2
1
w
2
0
+ w
4
0
=2a
2
+4
_
a
2
1
+b
2
1
_
,
w
1
w
0
(cu
2
+sv
2
) +
w
0
w
1
c
_
u
2
2
+ v
2
2
_
+ w
3
0
w
1
c =2(a
1
a
2
+b
1
b
2
),
u
2
2
+ v
2
2
+ w
2
0
w
2
1
=a
2
2
+b
2
2
. (39)
Lemma 2. The solutions to (38)(39) generate straight lines if b
1
=0, and planar curves if v
2
=0 and
1
=
0
+k for integer k, i.e.,
(c, s) =(1, 0).
Proof. If b
1
= 0, Eqs. (38) imply that w
0
= w
1
= 0, and thus w() 0. Consequently, from (29) we have ()
()
()() 0, and since the curvature of r() can be expressed (Farouki et al., 2009b) as
() =2
|()
()
()()|
(|()|
2
+|()|
2
)
2
,
this implies that () 0 i.e., the curve degenerates to a straight line. If
1
=
0
+k and v
2
=0 (i.e.,
2
is real), we
have
(
0
,
1
,
2
) =
_
1,
1
2
_
(w
1
/w
0
) +(w
0
/w
1
)u
2
_
, u
2
_
,
(
0
,
1
,
2
) =e
i2
0
_
0,
1
2
w
2
0
, w
0
w
1
_
,
and (), () are thus of the form stated in Lemma 1 for a planar curve. 2
Proposition 4. Monotone helical DPH quintics admit rational RMOFs only in the degenerate cases of straight lines or planar curves.
Proof. For Eqs. (38) to dene real w
0
, w
1
values we must have b
1
> 0 and a
1
b
2
a
2
b
1
0, where we invoke Lemma 2 to
discount the case b
1
=0. Moreover, consistency of these equations implies that
c
2
=
b
2
2
4(a
1
b
2
a
2
b
1
)b
1
, (40)
and hence we must also have 4(a
1
b
2
a
2
b
1
)b
1
b
2
2
. Assuming these conditions hold, substituting (38) and (40) into (39)
gives
4
b
1
b
2
c(cu
2
+sv
2
) +
b
2
b
1
=4a
1
,
4c(cu
2
+sv
2
) +
b
1
a
1
b
2
a
2
b
1
_
u
2
2
+ v
2
2
_
+
a
1
b
2
a
2
b
1
b
1
=2a
2
+4a
2
1
,
4
a
1
b
2
a
2
b
1
b
2
c(cu
2
+sv
2
) +
b
2
a
1
b
2
a
2
b
1
_
u
2
2
+ v
2
2
_
=4a
1
a
2
,
u
2
2
+ v
2
2
+4(a
1
b
2
a
2
b
1
)b
1
=a
2
2
+b
2
2
.
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 37
From the rst and fourth of these equations we obtain
c(cu
2
+sv
2
) =
b
2
b
1
_
a
1
b
2
4b
1
_
, u
2
2
+ v
2
2
=a
2
2
+b
2
2
4(a
1
b
2
a
2
b
1
)b
1
, (41)
and thus, to ensure that
2
= u
2
+iv
2
exists, we have the further inequality a
2
2
+b
2
2
4(a
1
b
2
a
2
b
1
)b
1
. On substituting
(41) into the second and third equations, we arrive at two conditions involving only a
1
, b
1
, a
2
, b
2
namely
_
a
2
1
+b
2
1
a
1
b
2
a
2
b
1
1
b
1
__
b
2
2
b
1
4(a
1
b
2
a
2
b
1
)
_
=0, (42)
_
a
1
a
2
+b
1
b
2
a
1
b
2
a
2
b
1
a
1
b
1
__
b
2
2
b
1
4(a
1
b
2
a
2
b
1
)
_
=0. (43)
Consider rst the case where the common second factor in (42)(43) vanishes. Then (c, s) =(1, 0) from (40), and Eqs. (41)
reduce to
u
2
=
b
2
b
1
_
a
1
b
2
4b
1
_
, u
2
2
+ v
2
2
=a
2
2
.
Substituting the former into the latter gives
v
2
2
=
[b
2
2
4(a
1
b
2
a
2
b
1
)b
1
][4(a
1
b
2
+a
2
b
1
)b
1
b
2
2
]
16b
4
1
,
and since b
2
2
4(a
1
b
2
a
2
b
1
)b
1
=0, we obtain v
2
=0. The conditions v
2
=0 and (c, s) =(1, 0) of Lemma 2 for a planar
curve are thus satised.
Now consider that case where the rst factors in Eqs. (42)(43) vanish simultaneously. This occurs (provided that a
2
1
a
2
)
when
b
1
=
_
a
2
1
a
2
and b
2
=2a
1
_
a
2
1
a
2
. (44)
From (40) we then have c
2
= a
2
1
/(2a
2
1
a
2
), and using this together with s
2
= 1 c
2
in (41) gives v
2
2
= a
2
1
(2a
2
1
a
2
u
2
)
2
/(a
2
1
a
2
). Substituting into the second equation of (41) and simplifying yields the quadratic equation
u
2
2
2a
2
1
u
2
+4a
4
1
6a
2
1
a
2
+3a
2
2
=0
in u
2
. Since this has discriminant 12(a
2
1
a
2
)
2
, there are no real solutions for u
2
unless a
2
1
=a
2
, in which case we have
b
1
=0 from (44) and Lemma 2 implies that the curve degenerates into a straight line. 2
3.4.2. General helical quintics
For a DPH quintic specied by (35) with (z
0
, z
1
, z
2
) = (
1
2
h
0
e
i2
, h
1
e
i2
,
1
2
h
2
e
i2
) and (
0
,
1
,
2
) = (h
0
, h
1
, h
2
), we have
=1 again from the rst equation in (37), so (36) and the other equations in (37) become
h
0
=2Im(c
1
), h
1
=Im(c
2
), h
2
=2Im(c
1
c
2
),
Re(
1
) =Re(c
1
),
Re(
2
) +2
_
|
1
|
2
+|
1
|
2
_
=Re(c
2
) +2|c
1
|
2
,
Re(
1
2
+
1
2
) =Re(c
1
c
2
),
|
2
|
2
+|
2
|
2
=|c
2
|
2
,
with
1
=
1
2
(h
0
2
+h
2
)/h
1
,
2
free, and
1
=
1
2
h
0
e
i2
,
2
=h
1
e
i2
. Setting c
1
=a
1
+ib
1
, c
2
=a
2
+ib
2
,
2
=u
2
+iv
2
then
gives
h
0
=2b
1
, h
1
=b
2
, h
2
=2(a
1
b
2
a
2
b
1
), (45)
h
0
u
2
+h
2
=2h
1
a
1
,
2
_
h
2
1
+h
0
h
2
_
u
2
+h
2
0
_
u
2
2
+ v
2
2
_
+h
2
0
h
2
1
+h
2
2
=2h
2
1
_
a
2
+2
_
a
2
1
+b
2
1
__
,
h
0
_
u
2
2
+ v
2
2
_
+h
2
u
2
+h
0
h
2
1
=2h
1
(a
1
a
2
+b
1
b
2
),
u
2
2
+ v
2
2
+h
2
1
=a
2
2
+b
2
2
. (46)
38 R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742
Proposition 5. General helical DPH quintics admit rational RMOFs only in the degenerate case of planar curves.
Proof. Substituting (45) into (46) and simplifying gives
u
2
=a
2
,
_
b
2
2
+4(a
1
b
2
a
2
b
1
)b
1
_
u
2
+2b
2
1
_
u
2
2
+ v
2
2
_
=4a
1
a
2
b
1
b
2
+a
2
b
2
2
2a
2
2
b
2
1
,
b
1
_
u
2
2
+ v
2
2
_
+(a
1
b
2
a
2
b
1
)u
2
=a
1
a
2
b
2
,
u
2
2
+ v
2
2
=a
2
2
.
From the rst and fourth equations we have (u
2
, v
2
) =(a
2
, 0) and these values also satisfy the other equations. Hence,
2
=
u
2
and
1
=
1
2
(h
0
u
2
+h
2
)/h
1
, and (), () have the form identied in Lemma 1 with planar curves, since (
0
,
1
,
2
) =
(1,
1
2
(h
0
u
2
+h
2
)/h
1
, u
2
) and (
0
,
1
,
2
) =e
i2
(0,
1
2
h
0
, h
1
). 2
4. Ruled surfaces
Surfaces generated by the motion of a straight line through space are called ruled surfaces. Each instance of the line is
a ruling or generator of the surface. A special class of ruled surfaces, the developable surfaces, is characterized by a constant
surface normal along each ruling. A surface is developable if and only if it is the envelope of a one-parameter family of
planes. A developable surface, regarded as a thin material sheet, can be attened (or developed) onto a plane, without
stretching or compressing it (Hilbert and Cohn-Vossen, 1952; Kreyszig, 1959; Struik, 1961).
Several important ruled surfaces are associated with a space curve. The surface generated by the curve tangent lines is a
developable surface, known as the tangent developable it can also be interpreted as the envelope of the curve osculating
planes. However, for space curves, the surfaces generated by lines along the principal normal and binormal directions at
each point are not developable (Kreyszig, 1959). The envelopes of the normal planes and the rectifying planes to a space
curve are called the polar developable and rectifying developable.
If the tangent developable is developed onto a plane, the space curve maps to a plane curve, the generators of the
surface becoming its tangents. The two curves have equal curvatures at corresponding points, and equal arc lengths of
corresponding segments (Hilbert and Cohn-Vossen, 1952). When the rectifying developable is developed onto a plane, the
curve maps to a straight line, corresponding segments of the curve and the line having equal lengths. The polar developable
can be regarded as the analog, for a space curve, of the evolute of a plane curve (Struik, 1961).
Consider the construction of a ruled surface from a space curve r(), such that the surface tangent plane is coincident
with the curve osculating plane. Expressing the surface in terms of a unit vector d() dened along r() as
s(, ) =r() +d(), (47)
it has the desired property if d() is of the form
d() =cos () t() +sin()p() (48)
since we than have
s
=sinb +
_
sin(cos p sint)
_
+
_
b
_
,
and thus when =0 i.e., along the curve r() the surface normal is
n =
s
|s
|
=sign(sin)b. (49)
Coincidence of the osculating plane of r() with the tangent plane of s(, ) identies the curve as an asymptotic line of the
surface (47).
The tangent developable and the principal normal surface correspond to choosing t() and p() for d(). The angular
velocity describing the changing orientation of the rulings in these cases is the Darboux vector, =b +t. However, the
component b of this angular velocity denes an orientational change within the osculating plane, that is unnecessary
to satisfaction of the constraints imposed on s(, ). A ruled surface satisfying the prescribed interpolation conditions, with
the least orientational variation of its rulings, can be constructed by choosing one of the RMOF vectors f(), g() dened by
(10)(11) or any vector xed relative to them for d().
4.1. Tangent developable
The family of tangent lines to a curve r() dene the ruled surface
s(, ) =r() + t(), (50)
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 39
with derivatives s
=(t+p) and s
=t satisfying s
|s
|
=sign()b.
This is the singular case () 0 of (49). The tangent developable consists of two sheets corresponding to < 0 and > 0
that meet at =0, i.e., along the curve r(), where n is indeterminate (it is also indeterminate along rulings corresponding
to points at which =0 i.e., the inections of r()). Since n exhibits a sudden reversal on passing through =0 along
a ruling of the surface (50), the curve r() is called (Kreyszig, 1959) the cuspidal edge or edge of regression of the tangent
surface s(, ). It may also be regarded as the locus of intersection points of neighboring generators, at parameter values
and +, as 0. Since n is independent of (except at =0), each sheet of the surface (50) is developable, and can
be generated by folding a piece of paper. In applications, one considers portions of the surface (50) with < 0 or > 0, to
avoid inclusion of the singular locus dened by =0.
4.2. Principal normal surface
The principal normal surface has rulings along the principal normal vectors to the curve r(), and thus has the parame-
terization
s(, ) =r() +p(). (51)
Since s
|s
|
=
(1 )b t
_
(1 )
2
+()
2
=
b
|b |
,
where is the Darboux vector dened by (5). Note that n is non-singular at = 0. Since n depends explicitly on , the
surface (51) is not developable.
If the curve has vanishing torsion at =
) of the
corresponding ruling this identies the center of curvature for the curve point r(
() d
()
|d
()|
2
d(). (52)
For the tangent surface, choosing the tangent t() for d() gives q() = r() i.e., the line of striction is the edge of
regression. However, for the principal normal surface, choosing p() for d() gives the line of striction
q() =r() +
()
2
() +
2
()
p().
The pitch of a non-developable ruled surface is a measure of the variation of the surface tangent plane along each ruling. If
and h are the angle and distance between rulings at and +, the pitch p() is (Hilbert and Cohn-Vossen, 1952)
the limit of the ratio /h, as 0. For the surface (47), it is specied (Pottmann and Wallner, 2001) by
p() =
|d
()|
2
r
() [d() d
()]
. (53)
One can verify that, for the principal normal surface, p =(
2
+
2
)/.
4.3. Rotation-minimizing ruled surface
For the tangent surface (50) and principal normal surface (51), the angular velocity of the rulings is dened by the
Darboux vector (5), in which the term b species an unnecessary rotation in the osculating plane i.e., it is not required
for the tangent plane of s(, ) to be coincident with the osculating plane of r() at each point. A rotation-minimizing ruled
surface, interpolating a given curve and its osculating planes, can be dened by using one of the basis vectors (10)(11) to
specify the rulings, so as to eliminate the term b in the Darboux vector. For example, the rulings of the surface
s(, ) =r() +f(), (54)
40 R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742
Fig. 3. Comparison of tangent developable (left), principal normal surface (center) and rotation-minimizing ruled surface (right) for the circular helix (13)
with R =
1
2
and k =
1
2
on the parameter domain [0, ].
have the angular velocity (12). The surface derivatives are s
= t + f
and s
=
sinb. Hence, we have s
|s
|
=sign(sin)
b +g
1 +
2
2
. (55)
Because the surface normal varies with , the surface (54) is not developable.
The surface normal (55) is singular along each ruling at a parameter value for which the angle (11) is an integer
multiple of i.e., at points where the ruling direction f is parallel or anti-parallel to the tangent. Since the curvature is
non-negative, the angle (11) is monotone-increasing with , and such singular rulings occur at points on r() determined
by its curvature. In practice, one would restrict the surface (54) to intervals between such points: the integration constant
0
in (11) can be used to adjust such intervals.
Proposition 6. The line of striction on the ruled surface (54) coincides with the curve r(), as for the tangent surface. The pitch of this
surface is equal to the torsion of the curve r(), and is the smallest possible for the ruled surfaces dened by (47)(48), whose tangent
planes coincide with the osculating plane of r(). These results also hold if the frame vector f is replaced by g in (54).
Proof. Choosing d() =f() in (52), and recalling that f
= sinb, we obtain r
=
2
sint b =0, and hence q() =
r(). Similarly, with d() =f() in (53), we have |f
|
2
=
2
2
sin
2
, and
r
_
f f
_
=t
_
(cos t +sinp) sinb
_
=
2
sin
2
,
and hence p =. Moreover, substituting (48) into (53) and simplifying gives
p = +
(
+)
2
2
sin
2
.
This expression is of least magnitude when
2(4
3
4
))/8 and f = 3
2/8 for [
1
4
,
3
4
]. These curves have a common end-point r(
1
2
) = s(
1
4
) =
R.T. Farouki et al. / Computer Aided Geometric Design 31 (2014) 2742 41
Fig. 4. The principal normal surface (left) and rotation-minimizing ruled surface (right) for a composite curve, consisting of segments of a PH cubic (red
curve) and a circular helix (blue curve). The rotation-minimizing ruled surface is tangent-plane continuous, but the principal normal surface is not.
1
4
(7, 1, 2
2), curvature =2