Roller Coasters without Differential Equations – A Newtonian

Approach to Constrained Motion

Rainer Müller
Technische Universität Braunschweig, Physikdidaktik,
Pockelsstr. 11, D-38106 Braunschweig, Germany∗
(Dated: August 23, 2009)

Abstract
Within the context of Newton’s equation, we develop a simple approach to the constrained motion
of a body forced to move along a specified trajectory. Because the formalism uses a local frame of
reference it is simpler than previous methods, making more complicated geometries accessible. No
Lagrangian multipliers are necessary to determine the constraining forces. Although the method
is able to deal with friction, it becomes particularly simple for conservative systems. We give an
analytic expression for the constraining force for any two-dimensional frictionless trajectory that
can be written in the form y = f (x). The approach is illustrated with examples from roller coaster
physics, e. g. the circular loop or the clothoid loop. We will find analytic expressions for the
constraining force in both cases.

1
I. INTRODUCTION

When a body slides along a guiding track its motion is said to be constrained. Constrained
motion is not easily described by Newton’s equation of motion. In order to make the body
follow a curved track, a force must act between the track and the body. This force is called
constraining force. The trouble is that it is not a priori known. Thus, in Newton’s equation
F~ = m · ~a, neither the total force F~ nor the acceleration ~a are known at any given time. It
seems that Newton’s equation does not suffice to solve the problem.
Of course, problems with constrained motion can be solved. Holonomous constraints10
are routinely treated in textbooks of theoretical physics. Usually, one uses an approach
where the governing equations are not Newton’s but Lagrange’s (see e. g. [1]). Generalized
coordinates are chosen so that the number of dynamic equations is reduced to the number
of independent degrees of freedom
Although the Lagrangian approach is perfectly valid in principle, there may be problems
in practice. The guiding track may have a complicated shape. Then it is difficult to find
a set of coordinates in which the constraint equations become simple (technically: one or
more of the coordinates have to become ”cyclic”, i. e. they must not explicitly appear in
the Lagrangian). Usually, one does not see the Lagrange formalism applied to a guiding
track more complicated than a circle or a parabola.
Can problems with constrained motion be solved with Newton’s equation? It is indeed
possible to include the constraining force into Newton’s law [2–4]. For a single particle forced
to move under a constraint f (~r, t) = 0, Newton’s equation takes the form

m · ~r¨ = F~other + λ∇f,

~ (1)

where the last term on the right hand side represents the constraining force while F~other
stands for other (external) forces. The gradient fixes the direction of the constraining force
because it is orthogonal to the surfaces of constant f . The unknown constant λ enters
because we do not know the magnitude of the constraining force. It is called a Lagrange
multiplier.
The three equations (1) together with the constraint equation f (~r, t) = 0 are sufficient
to solve for the four unknowns ~r(t) and λ. Thus, problems with constrained motion can in
principle be solved with the Lagrange multiplier method. As with the approach described

2
earlier, problems arise in practical calculations because in all but the simplest problems it is
virtually impossible to find a coordinate system in which the constraint equation becomes
tractable.
In the present paper we present an approach in which the practical difficulties that render
the traditional methods almost useless are avoided. We consider the case of a particularly
simple holonomic constraint: a body is forced to move on a fixed trajectory ~r(t). The key
observation that characterizes our approach is the following: In order to formulate Newton’s
law for a body moving along a given trajectory we do not need a global set of coordinates
which is adapted to the problem. It is sufficient to define coordinates in the vicinity of
the given trajectory. In the language of General Relativity: We only need a local frame of
reference. Elementary differential geometry provides the mathematical tools needed.
The paper is organized as follows. In Sec. II we review the basics of elementary differential
geometry needed to formulate our approach. Secs. III–IV deal with velocity and acceleration
for constrained motion and introduce the notion of constraining forces. Newton’s law for
constrained motion is formulated in Sec V. From Sec. VI on we will talk about conservative
systems (without friction). In this case, not even a differental equation has to be solved
to obtain the constraining force. We will be able to derive a closed expression for the
constraining force for a fairly general class of trajectories (Sec. VII).
Roller coasters are an interesting physics application of the type of constrained motion
considered here. In order to make the presentation more vivid we will talk about roller
coasters from now on (a more detailed discussion of roller coaster physics can be found in
[5, 6]). In example problems we will discuss some well-known (as well as some not-so-well-
known) results from roller coaster physics. To illustrate the power of the formalism even in
nontrivial situations we will treat the clothoid loop in Sec. IX.

II. THE GEOMETRY OF CONSTRAINED MOTION

We consider the type of motion illustrated in Fig. 1. A body is forced to move along
a given trajectory in three dimensions. Before we start a physical analysis we will give a
purely mathematical description of the motion. Elementary differential geometry provides
a neat way for doing this. A curve in three-dimensional space is described by a set of three
mutually orthogonal unit vectors at any point of the curve, the well-known Frénet vectors.

3
 y

®
eT ®
eB

®
®
r eN

®
r (t)
z

FIG. 1: A body is forced to move along a given trajectory. On each point of the trajectory, a set
of three orthogonal unit vectors can be defined.

y local circle
of curvature

®
r eN
®
eT

FIG. 2: The normal vector is directed to the center of the local circle of curvature.

Although the topic is covered in many mathematical texts [7] as well as some physics
books [8] we find it useful to review some properties of the Frénet vectors. At any point of
the curve, the tangent vector ~eT is directed tangential to the curve (Fig. 1). Its direction
thus changes from point to point, whereas is length remains fixed to 1 by definition.
The direction of the normal vector ~eN is chosen so that it points to the local center of
curvature (Fig. 2). It is orthogonal to the tangent vector and has unit length. Finally, the
third of the Frénet vectors, the binormal vector ~eB is defined to be orthogonal to both ~eB
and ~eN . It has unit length, too.
In principle, the curve can be parametrized by an arbitrary variable t. A particularly
advantageous parametrization uses the arc length s, defined by
Z t0
s= v(t) dt, (2)
0

4
 FIG. 3: Circular trajectory

where
p
v(t) = ẋ2 (t) + ẏ 2(t) + ż 2 (t). (3)

Differentiation of Eq. (2) leads to the relation

ds
v= = ṡ. (4)
dt

Explicit expressions for the tangent vector and the normal vector can be obtained as
follows. For a given curve ~r(s), the tangent vector is calculated by differentiation:

d~r
~eT = . (5)
ds

If s denotes the arc length, this vector is normalized to unity automatically. The normal
vector is defined by
d~eT
~eN = ρ · . (6)
ds
The direction of ~eN is fixed so that the normalizing factor ρ becomes positive. ρ can be
interpreted geometrically as the radius of the local circle of curvature (Fig. 2). Often, the
reciprocal of ρ is used: the local curvature κ = 1/ρ.

Problem 1: Circular trajectory

Determine tangent vector, normal vector, and radius of curvature for the circular trajectory
shown in Fig. 3:

x(t) = R sin ωt, y(t) = −R cos ωt, z(t) = 0. (7)

5
Solution:
In Eq. (7) the trajectory is parametrized by the time t. To use the arc length instead, we
calculate with Eq. (3):
p
v= ẋ2 (t) + ẏ 2(t) + ż 2 (t) = ωR, (8)

where the relation sin2 ωt + cos2 ωt = 1 was used. s und t are therefore related by
Z t
s= ωR dt = ωRt. (9)
0

Thus, when parametrized by the arc length, Eq. (7) becomes

~r(s) = x(s), y(s), z(s)
 s s 
= R sin , −R cos , 0 . (10)
R R
The tangent vector is obtained by differentiation of Eq. (10):
d~r   s  s 
~eT = = cos , sin , 0 . (11)
ds R R
As expected, the length of this vector is unity. Similarly, the normal vector is calculated
according to Eq. (6):

d~eT 1  s s 

= · − sin , cos , 0 . (12)
ds R |
|{z} R {z R }
= 1/ρ =~
eN

It is directed to the center of the circle. Not unexpectedly, the local radius of curvature is
constant along the trajectory and is equal to the radius of the circle R.

Problem 2: General trajectory of the form y = f (x)

Determine tangent vector, normal vector, and radius of curvature for any trajectory in the
x-y plane that can be written in the form y = f (x).

Solution:
This problem encompasses a large class of trajectories. Its solution will therefore open up
a broad range of applications. In a roller coaster context, for example, arbitrarily shaped
curves, hills, and valleys are covered. Loops and spirals, however, are not, because they
cannot be represented by a single-valued function f (x).
The position vector reads


~r(t) = x(t), f (x(t)), 0 . (13)

6
Using the chain rule we write

df dx
ẏ(t) = · = f ′ (x) · ẋ, (14)
dx dt

so that
p p
v= ẋ2 (t) + ẏ 2 (t) + ż 2 (t) = ẋ2 (t) (1 + f ′2 (x)) (15)

and Z tp Z x(t) p
dx
s= 1 + f (x) ·
′2 dt = 1 + f ′2 (x) dx. (16)
0 dt x(0)

To calculate the tangent vector, we use the chain rule again:

d~r d~r dx
~eT = = · . (17)
ds dx ds

By differentiation of Eq. (16) we obtain

dx 1
=p . (18)
ds 1 + f ′2 (x)

With ~r from Eq. (13) and d~r/dx = (1, f ′ (x), 0), this yields for the tangent vector:

dx 1
~eT = (1, f ′ (x), 0) · = p · (1, f ′ (x), 0) . (19)
ds 1+ f ′2 (x)

Finally we calculate the normal vector, again using the chain rule:

d~eT d~eT dx
= ·
ds dx ds
|f ′′ (x)| sign(f ′′ (x))
= 3 · p · (−f ′ (x), 1, 0) . (20)
(1 + f (x)) 2
′2 1 + f (x)
′2
| {z } | {z }
= 1/ρ =~
eN

The local radius of curvature for a trajectory of the form y = f (x) is thus given by:
3
(1 + f ′2 (x)) 2
ρ= . (21)
|f ′′ (x)|

This formula (which is a standard result in differential geometry) completes the problem.

III. VELOCITY AND ACCELERATION

Up to now, everything has been mathematics. No physics was involved, even if we have
given suggestive names to some quantities (like v). Let us now interpret ~r(t) as the physical

7
 ®
FC

®
FG

FIG. 4: The constraining force is directed normal to the track

trajectory of a moving body. The kinematical variables of its motion can be expressed with
the help of the unit vectors considered above. The velocity, for example, can be written
d~r d~r ds
~v = = · . (22)
dt |{z}
ds |{z}
dt
= ~eT =v

so that
~v = v · ~eT with v = ṡ. (23)

Likewise, the acceleration can be obtained by differentiating Eq. (23):

d~v d
~a = = (v · ~eT ) . (24)
dt dt
With the product rule we get:
d~eT
~a = v̇ · ~eT + v ·
dt
d~eT ds
= v̇ · ~eT + v · · . (25)
ds |{z}
|{z} dt
= ~eN /ρ =v

The acceleration of a moving body can thus be expressed solely in terms of the normal vector
and the tangent vector (cf. e. g. [8]:
v2
~a = v̇ · ~eT + · ~eN . (26)
ρ
By definition of the Frénet vectors, there is no component of the acceleration in the direction
of the binormal vector:

IV. CONSTRAINING FORCES

In a roller coaster, the car is guided along a specified track (Fig. 4). If the track is curved,
forces must act between track and car in order to change the direction of motion. The forces
that accomplish these direction changes are called constraining forces.

8
 The motion of the car is described by Newton’s law F~ = m · ~a. Some effort is required,
however, to apply this law to constrained motion. First, we formally separate the constrain-
ing forces F~C (which are a priori unknown) from all other forces (which we assume to be
known):
F~ = F~C + F~other . (27)

We assume that the constraining forces have no component in tangential direction. They
do not change the speed but only the direction of the car. Strictly speaking, this is not
an assumption but a definition. In general, the force between track and car will have a
component in tangential direction. However, we usually prefer to call these force components
friction forces and assign them to the ”other forces” mentioned above. The physics is
unaffected by this formal separation; our formalism will be able to deal with friction forces.
The components of the constraining forces in the direction of ~eN and ~eB will be denoted
by FC and FC′ , respectively. Newton’s law now reads

m · ~a = F~other + FC · ~eN + FC′ · ~eB . (28)

It cannot be solved in the usual manner because the constraining force components FC
and FC′ are still unknown. In the usual Newtonian approach, they are treated as Lagrange
multipliers. We will find a simpler way to deal with them.

V. COMPONENTS OF NEWTON’S LAW

The problem of the unknown constraining force components can be solved by separating
the tangential, normal, and binormal component of Newton’s law. We insert ~a from Eq. (26)
to get
mv 2
m · v̇ · ~eT + · ~eN = F~other + FC · ~eN + FC′ · ~eB . (29)
ρ
We multiply this equation by ~eT to obtain its tangential component. Using the orthonor-
mality of the Frénet vectors we get

m · v̇ = F~other · ~eT . (30)

This equation can be regarded as Newton’s law along a specified track. It is remarkable that
the unknown constraining force components do not appear. The motion along the track is

9
governed only by the tangential component of the other forces F~other ·~eT . Eq. (30) is a scalar
equation for the change of the ”speed” v along the specified track. As it stands, it can be
solved for v(t) and s(t) by any of the standard methods.
The binormal component of Newton’s law is particularly simple. By definition, there is
no binormal component of the acceleration. Therefore the constraining force FC′ just cancels
any component of the other forces F~other acting in this direction:

FC′ = −F~other · ~eB . (31)

Finally, we obtain the normal component of Newton’s law by multiplying Eq. (29) with
~eN :
mv 2
= F~other · ~eN + FC . (32)
ρ
The interpretation of this equation is straightforward. The left hand side denotes the local
centripetal force. In order to guide the car along the track, the total force acting on the car
in normal direction must have the magnitude mv 2 /ρ (it must act as centripetal force). The
constraining force FC adjusts itself so that this equality is fulfilled.
Once we have solved Eq. (30), we can insert v(t) into the equation above to determine
the constraining force component FC :
mv 2
FC = − F~other · ~eN . (33)
ρ
Our problem is solved. In addition to the kinematical variables s(t) and v(t), we are able
to calculate the constraining force components FC and FC′ that guide the car through the
track.

VI. CONSTRAINED MOTION WITHOUT FRICTION

With the approach described above, problems with constrained motion along a given tra-
jectory can be treated in a familiar manner, i. e. by solving ordinary differential equations.
However, we can do even better for frictionless problems. If friction can be neglected, the
total energy of the car is conserved. We assume that only kinetic and potential energy are
relevant so that Etot = Ekin + Epot = const. In this case, we need not solve the differential
equation Eq. (30) for v(t). It can be obtained much easier from energy conservation:
r
2
v(t) = (Etot − Epot (t)). (34)
m

10
 FIG. 5: A swingboat as an example of a pendulum (Photo courtesy of Heidepark, Germany)

R
x
®
FC f

®
FG y

FIG. 6: Geometry of the pendulum

For roller coasters, the potential energy is m · g · y. The total energy is conveniently specified
by the height of the lifthill y0 : Etot = m · g · y0 . The top of the lifthill is the highest point of
the track where the car is disengaged from the lifting cable. We assume the kinetic energy
to be negligible there.
If we use Eq. (34) together with Eq. (33), we can determine the constraining forces
without ever solving a differential equation:
2 (Etot − Epot ) ~
FC = − Fother · ~eN . (35)
ρ
This fact is a particularly appealing feature of the approach. In the following problem,
we will apply the formalism to determine the constraining force acting in the rope of a
pendulum. In keeping with the roller coaster spirit we will consider a swingboat as a specific
example.

Problem 3: Pendulum
The swingboat shown in Fig. 5 has a total height of 15 m and and a pendulum length of

11
 y/m x/m

0.26 g acceleration caused

by constraining
8g force
0 .9

g
6
1.

g
2 .1

2.4 g

2.5 g
FIG. 7: Constraining acceleration (i. e. FC /m) for the swingboat pendulum

12 m. It swings up to a maximum angle of 75◦ . Assume that it can be modeled as a point

mass and determine the constraining force as well as the total force acting through the
”pendulum rope” onto the body of the boat.

Solution:
The geometry of the problem is shown in Fig. 6. The constraining force FC is calculated
with Eq. (35). In Problem 1, we have already determined ~eN for a circular trajectory so that
we can write immediately (using φ = s/R):

F~G · ~eN = −m · g · cos φ. (36)

If we further use
y = −R · cos φ (37)

to simplify the potential energy, we can write down the following expression for the con-
straining force for the pendulum
2Etot
FC = + 3 mg cos φ. (38)
R
Note that no small-angle approximation was required to obtain this result.
For the swingboat under consideration, the constraining force calculated with Eq. (38)
is shown in Fig. 7 for various points of the trajectory. The constraining force is largest at
the lowest point of the trajectory where the velocity has its maximum value (and gravity
has to be cancelled, too). Because the passengers move together with the boat, our analysis
applies also to the force acting from the seats onto the passengers. We see from Fig. 7 that
they have to withstand a maximum acceleration of 2.5 g for the Heidepark swingboat.
The total force, shown in Fig. 8, is the vector sum of the constraining force and the
gravitational force (i. e. the right-hand side of Eq. (28)). It is remarkable that for a simple

12
 y/m x/m

0.97 g total
acceleration
0.99 g

1.1 g
g
3
1.

1.5 g
1 .4
FIG. 8: Total acceleration for the swingboat pendulum

system like a pendulum the magnitude as well as the direction of the total force change in
a relatively complicated way along the trajectory.

VII. THE ROLLER COASTER FORMULA

We now will derive a closed formula for the constraining force for any trajectory that
can be represented in the form y = f (x). We will assume y to be the vertical, hence the
gravitational force has the components F~G = (0, −m · g, 0). In Problem 2 we have already
calculated the normal vector for this type of motion (Eq. (20)) so that we can write

sign(f ′′ (x))
F~G · ~eN = − p · m · g. (39)
1 + f ′2 (x)

We have also determined the local radius of curvature in Eq. (21). As usual v(t) is obtained
from energy conservation: r
2
v= Etot − 2gf (x). (40)
m
By inserting these expressions into Eq. (35) we obtain a formula for the constraining force
under these very general circumstances. We will call it the roller coaster formula:

2 |f ′′ (x)| sign(f ′′ (x))

FC = 3 (Etot − m · g · f (x)) + p · m · g. (41)
(1 + f ′2 (x)) 2 1 + f ′2 (x)

Problem 4: Parabola-shaped hill

The parabola-shaped hill in Fig. 9 can be represented by the function y = −c · x2 (with
c > 0). With the roller coaster formula, calculate the constraining force acting on the
passengers in a car moving over the hill. What happens for FC > 0? How can a roller
coaster constructor achieve ”airtime” (i. e. weightlessness)?

13
 FC > 0

®
eT
®
eN

FIG. 9: Parabola-shaped hill

Solution:
We need the derivatives f ′ (x) = −2cx and f ′′ (x) = −2c. Hence, sign(f ′′ (x)) = −1 every-
where. From Eq. (41), we obtain:

4c
mg
FC = 3 Etot + mgcx2 − √ . (42)
(1 + 4c2 x2 ) 2 1 + 4c2 x2

At the top of the hill, at x = 0, this reduces to

FC (top) = 4c · Etot − m · g. (43)

Energy conservation tells us, that the car does not reach the top if Etot < 0. If the total
energy is positive and smaller than mg/(4c), the constraining force ist negative at the top of
the hill. It is directed upward, i. e. opposite to ~eN . According to Eq. (43), it is smaller than
the gravitational force. Thus, the magnitude of the force that the seat exerts on a passenger
is reduced from its usual value mg. The passengers feel somewhat lighter as they roll over
the hill.
If Etot > mg/(4c), the constraining force becomes positive at the top of the hill, i. e. it
will be directed downward. An ordinary rail track cannot exert a downward force on the
car. The car will tend to leave the track and follow the dashed line in Fig. 9. To avoid
this unwanted behavior, the car must have additional counter wheels below the track. In
addition the passengers must be tied to their seats by a safety harness.
”Airtime” is a particular attraction in which the passengers experience weightlessness for
a short time. We expect this to be the case if the car and the passengers follow a free-
fall trajectory. The track must have the shape of a free-fall parabola. Fig. 10 shows an
impressive example of parabola-shaped hills. If v0 denotes the velocity at the top of the hill,

14
 FIG. 10: Parabola-shaped hills can give airtime. Photo courtesy of Europapark, Germany

the free-fall trajectory has the form

g
y=− · x2 . (44)
2v02
We thus have to insert into Eq. (42)
g 1
c= and Etot = mv02 , (45)
2v02 2
which leads to
FC (x) = 0 for all x. (46)

The constraining force is zero everywhere. As expected, the car (as well as the passengers)
follow the track without the necessity of any constraining force, resulting in weightlessness.

VIII. CIRCULAR LOOP

Why do we never see circular loops in roller coasters? Real loops always have an inverted-
teardrop shape (the form of the clothoid loop we will encounter in the next Section). There
is a physical reason that can be discussed with our approach. The constraining force for
circular motion has already been derived in Problem 3 (Eq. (38)):
2Etot
FC = + 3 mg cos φ. (47)
R
From Problem 4 we know that at any point of the track FC > 0 must hold if there are no
counter wheels. Otherwise, the car will fall out of the track. Applied to Eq. (47), this gives
a condition for Etot :
3
Etot > − mgR cos φ. (48)
2

15
If the car is to make it through the loop, this condition must be fulfilled everywhere, espe-
cially at the top of the loop where φ = π. Thus, there is a minimum total energy the car
must have at the beginning of the loop:
3
Etot > mgR. (49)
2
Conveniently, this expression is written as a condition for the minimum height of the lifthill:

y0 > 2.5 R. (50)

The lifthill must be higher than the loop by at least half a loop radius.
We can now understand the problem of the circular loop. If the condition Eq. (49) is
fulfilled, the constraining forces at the beginning and the end of the loop become forbiddingly
large. If we insert Etot from Eq. (49) into Eq. (47), we obtain:

FC = 3 mg + 3 mg cos φ. (51)

At the loop entry and exit (φ = 0 bzw. φ = 2π), this becomes

FC,max = 6 mg. (52)

As displayed in Fig. 11 the passengers suddenly have to stand six times their own weight at
the entry of the loop. This is unacceptable from a medical point of view. Note that there is
no design parameter to resolve the situation because the value of FC,max is independent of
the loop radius.

IX. CLOTHOID LOOP

Up to now, we have considered relatively simple geometries in our concrete examples. To

show the power of the formalism we will turn to something more difficult now: the clothoid
loop. The clothoid is a particularly challenging geometry because it can be mathematically
described only by a parametric representation. Of course, the problem is also physically
interesting because real loops have a clothoid shape.
The parametric representation of the clothoid (Fig.12) reads:

x(t) = R · C(t),
y(t) = R · S(t). (53)

16
FIG. 11: Constraining acceleration in a circular loop. At the entry, the acceleration suddenly rises
from 1 g to 6 g. Only half of the loop is shown in the diagram on the right.

FIG. 12: The clothoid

The two functions C(t) and S(t) are known as Fresnel integrals. They cannot be represented
by elementary functions. Instead they are defined by definite integrals (cf. [9]):
Z t π 
C(t) = cos u2 du,
2
Z0 t π 
S(t) = sin u2 du. (54)
0 2

To calculate the arc length for the clothoid we need

d π 
C(t) = cos t2 ,
dt 2
d π 
S(t) = sin t2 . (55)
dt 2

17
FIG. 13: The characteristic shape of a roller coaster loop. Photograph: istockphoto/Marcio Silva

According to Eq. (2):

Z tp
s= ẋ2 (u) + ẏ 2 (u) du
Z0 t  π   π  12
= R · cos2 u2 + sin2 u2 du
0 | 2 {z 2 }
=1

= R · t. (56)

The arc length and the parameter t are related via t = s/R.
An actual roller coaster loop is shown in Fig. 13. As a comparison with Fig. 14 shows, this
shape can be constructed from two mirror-symmetrical sections of the clothoid, extending
√
from s/R = 0 to s/R = 2. Because of the mirror symmetry, we will consider only the first
half of the loop in the following.
In the parametric representation (53), the position vector along the track reads
 s  s 
~r(s) = R · C , R·S . (57)
R R
The tangent vector is obtained by differentiation:
d~r  ′  s  ′
 s 
~eT = = C , S . (58)
ds R R

18
 FIG. 14: A roller coaster loop is constructed from two clothoid section.

With Eq. (55) we get     

π s2 π s2
~eT = cos , sin . (59)
2 R2 2 R2
Another differentiation leads to the normal vector and to the local radius of curvature:
    
π s2 π s2
~eN = − sin , cos (60)
2 R2 2 R2
and
R2
ρ= . (61)
πs
This formula illustrates a defining feature of the clothoid: its curvature (the inverse of
ρ) increases linearly with s. As usual, the velocity of the car is calculated from energy
conservation:
q
v(s) = v02 − 2g · y(s)
r s
= v02 − 2gR · S , (62)
R
where v0 is the velocity at the loop entry.
We have now all ingredients needed to calculate the constraining force from Eq. (33):
s  2
2πs 2πs πs
FC = 2 Etot − mg S + mg cos . (63)
R R R 2 R2
This is our result: an analytic expression for the constraining force in a clothoid loop. It will
allow us to discuss the reasons why a clothoid loop is less harmful to the passengers than a
circular loop.
As with the circular loop we first have to calculate the minimum velocity at the entry of
the loop. Again, it is determined by the condition FC > 0 everywhere. The most critical

19
FIG. 15: Constraining acceleration for a clothoid loop (height 14 m) with minimal entry velocity.

√
point is the top of the loop: s/R = 2. To evaluate the constraining force there, we insert
√
the numerical value of the Fresnel integral S( 2) = 0.71. We obtain the condition

Etot > 0.83 mgR, (64)

or, refering to the top of the lifthill:

y0 > 0.83 R. (65)

√
Since the total height of the loop is R · S( 2) = 0.71R (cf. Eq. (53)), the lifthill has to be
only 16% higher than the top of the loop. Compare this to 25% for the circular loop. Here
is a first advantage of the clothoid loop: a lower entry speed is required, corresponding to
lower centripetal forces.
The main advantage of the clothoid loop can be understood with the help of Fig. 15. It
shows the constraining acceleration for a loop with R = 20 m (corresponding to a height
of 14 m) with the minimum entry velocity specified by Eq. (64). We see that there is no
hard onset of the constraining force at the entry of the loop. Instead, the force gradually
increases from 1 g to a maximum value of 3.6 g and then decreases again near the top. The
linear increase is the reason that of all geometrical shapes a clothoid is chosen. As we have
seen above, its characteristic feature is the linear increase of curvature along the track. The
clothoid makes loops in roller coasters feasible. Accelerations of 3–4 g are not problematic
for healthy passengers.

X. DISCUSSION

We have developed a Newtonian approach to constrained motion along a specified trajec-

tory that is simpler than the traditional methods. Because it uses only local coordinates it

20
makes complicated geometries easier accessible. The approach is based on elementary differ-
ential geometry, especially the Frénet vectors. The approach becomes especially simple for
conservative motion. We have been able to give an analytic expression for the constraining
force for any two-dimensional trajectory that can be written in the form y = f (x).
As a physically interesting application of the approach we considered various roller coaster
tracks. Especially, we calculated the constraining forces for a circular loop and a clothoid
loop. It was possible to find an analytic expression for the constraining force in both cases.

∗ Electronic address: [email protected]

1 H. Goldstein, C. Poole, J. Safko, Classical Mechanics (3rd Edition) (Addison Wesley, San Fran-
cisco, 2001).
2 A. Budo, Theoretische Mechanik, 12th ed. (Wiley-VCH, Weinheim, 1990).
3 S.-P. Liu, Physical approach to constrained motion, Am. J. Phys. 49, 750–753 (1981).
4 W. R. Davies, Constraints in particle mechanics and geometry, Am. J. Phys. 35, 916–920 (1967).
5 R. Müller, Klassische Mechanik – vom Weitsprung zum Marsflug (de Gruyter, Berlin, 2009).
6 A. M. Pedrill, Acceleration in one, two, and three dimensions in launched roller coasters, Phys.
Ed. 43, 483–491 (2008).
7 E. Kreyszig, Differential Geometry (Dover, New York, 1991).
8 W. Greiner, Classical mechanics. Point particles and relativity (Springer, New York, 2004).
9 M. Abramowitz, I. A. Stegun, Handbook of Mathematical Functions with Formulas, Graphs,
and Mathematical Tables, (Dover, New York, 1964).
10 A holonomous constraint can be described by an equation of the form f (~r1 , ~r2 , . . . , t) = 0

21