Roads and Wheels

Leon Hall; Stan Wagon

Mathematics Magazine, Vol. 65, No. 5. (Dec., 1992), pp. 283-301.

Stable URL:
http://links.jstor.org/sici?sici=0025-570X%28199212%2965%3A5%3C283%3ARAW%3E2.0.CO%3B2-4

Mathematics Magazine is currently published by Mathematical Association of America.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/maa.html.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact [email protected].

http://www.jstor.org
Tue Mar 11 09:35:33 2008
 ARTICLES

Roads and Wheels

LEON HALL
University of Missouri-Rolla
Rolla, M O 65401

STAN WAGON
Macalester College
Saint Paul, M N 55105

Introduction

San Francisco's Exploratorium contains an intriguing exhibit of a square wheel that

rolls smoothly on a road made up of linked, inverted catenaries (see FIGURE 4). That
exhibit inspired us to generate a computer animation of a rolling square and further
explore the relationship between the shapes of wheels and roads on which they roll.
In a sense, we are bringing up to date the paper by G. Robison [4], showing how
much more can be done, both numerically and graphically, with modern computer
hardware and software. The problem of the square wheel has been rediscovered and
solved several times; see [S,71.
All the diagrams and animations were prepared in Muthematica. Our package that
generated the diagrams and the associated animations (see Section 5) can be obtained
by sending a Macintosh disk to one of the authors. It is noteworthy that some of the
results of this paper, in particular the discovery of a cycloidal locus generated by a
noncircular wheel, were discovered only after viewing certain graphics. Muthematica
was also used to do all the symbolic integrations that occur. For further applications
of symbolic and graphic computation to wheel/road problems, in particular, a
complete discussion of the cycloid, see [6, Chapter 21.
The paper is organized as follows. Section 1 discusses the theory and the funda-
mental differential equation. Section 2 contains many closed-fonn examples. Section 3
shows how numerically approximating the solution to the differential equation is an
excellent approach to diverse examples, even those solvable in closed form. Section 4
squares the circle by considering Fourier approximations to the catenary. And Section
5 discusses the Muthemuticu package that we built.

1. Building a wheel

Suppose we are given a road in the form of a rectifiable curve in the lower half-plane
parametrized by f ( t ) = (x(t), y(t)), where x(t) is increasing, x(O) = 0, y(t) 5 0. By
the wheel corresponding to the road we mean a curve that will roll smoothly on the
road. More precisely, a wheel will be a curve given by a polar function r = 4 0 ) such
that the axle of the wheel, which initially is at (0, O), stays on the x-axis directly above
the wheel-road contact point as the curve rolls along the road. The wheel's axle may
or may not coincide with the wheel's geometric center. The road is assumed to
284 MATHEMATICS MAGAZINE

provide enough friction so that there is never any slipping of the wheel. The rolling
motion can be described by a function 8 = B(t) that describes the amount of angular
rotation for the wheel to roll from f(O) to f(t). These functions must satisfy the
following conditions (see FIGURE 1):
1. Initial condition. The initial contact point is at f(O), directly under the origin,
whence e(0) = - ~ / 2 .
2. Rolling condition. The amount unravelled on the wheel matches the distance
travelled on the road: For any t, the arc length of f between f(0) and f ( t ) equals
the arc length of the polar curve between O(0) and B(t).
3. Radius condition. The radius of the wheel matches the depth of the road at the
corresponding point: For any t, r(e(t)) = - y(t).
FIGURE 1 illustrates the formation of the wheel in the case when the road is given
as y =f(x), with f ( x ) nonpositive, in which case the conditions simplify accordingly
(that is, x can be used as the parameter, and so 0 becomes a function of x).

FIGURE 1
If a road is given by y =f(x), then the relationship between 0 and x is obtained from the
equality of the arc lengths AB and AC and of the radius vector OC and the depth of the road
+
(dashed lines). The road illustrated is given by y = - fi cos x , where fi has been
chosen so that the wheel closes up on itself (see Remark 5).

The key to getting a wheel is finding the function e(t), since the radius condition
will then yield r(B). The first two conditions become a simple differential equation,
which can lead to either a closed-form description of 0 or a numerical approximation.
The rolling condition is:

Differentiating both sides with respect to t and squaring yields:

d r d e dy
Now substitute - - =
de a t - --a t (obtained by differentiating the radius condition) to
get:
V O L . 65. NO. 5. DECEMBER 1 9 9 2 285
which simplifies to:

Because dO/dt is to be positive, the differential equation we seek is:

with initial condition O(O) = - ~ / 2 .

Remarks.
1. If the road is given by y =f ( x ) , the differential equation for 0 becomes simply
dO/dx = - l / f ( x ) . In this case

2. If the function 0 ( t ) can be inverted to t ( 0 ) then the wheel is given by the polar
equation r = - y(t(0)).
3. An alternative approach to characterizing O(t) proceeds by matching slopes instead
of arc lengths. The rolling condition then becomes: For any t , the slope of the road
at f ( t ) equals the slope of the tangent to the polar curve at O(t), rotated clockwise
through O(t) + n-/2 radians. This leads to the same differential equation.
4. The inverse problem starts with r(O), a polar representation of a wheel, and seeks
the appropriate road. The preceding discussion implies that the road is given by
y ( x ) = -r(O(x)), where 0 satisfies 0(0) = - ~ / 2and dO/dx = l / r ( O ) . One can
also deal with the case that the wheel is given parametrically by ( x ( t ) ,y(t)); see
Case 4 in Sectioil 3.
5 . Suppose the road y =f ( x ) is periodic with period a. Then the corresponding
wheel does not necessarily close up on itself to form a topological disk (see l & c u ~ e
2). The condition for such closure-the closed-wlaeel condition-is that there
exists a rational number r so that 2n-r = O(a) - O(0) = 1:- l / f ( x ) dx. If r = l / n ,
n a positive integer, then the wheel rolls over n periods of the road during each
complete revolution. As an example, consider the road given by y = d cos x, +

FIGURE 2
+
The wheel for a cosine road given by - 3.5 cos x winds around endlessly without closing up,
because 3.5 is not one of the special values d s . This wheel was generated by the
numerical technique discussed in Section 3.
286 MATHEMATICS MAGAZINE

where d I - 1. For each positive integer n, there is a unique value of cl, which
turns out to be - \ll+n2, for which the wheel closes up into one that covers n
periods per revolution. (See FICURE1 for the n = 4 case and FIGURE13 for the
n = 1 case.) To see that the values of d in the cosine case are as claimed, observe
+
that /,:" - l / ( d cos x) cE?c = 2n-/ Jd"-l. It is easiest to integrate from O to rr
+
and then double. More generally, the road y = d 17 cos(cx) yields a closed wheel
that covers n periods when d = - d w . Cycloid roads and inverted
cycloid roads provide two more examples in which the closed-wheel condition
integral can be evaluated (see FIGURE9(c) for the rz = O case, and FIGURE19(d)
for a fractional example, viz., n = 1/21, The closed-wheel conditioil has a coin-
pletely analogous form in the case of a parametrically defined road as well; see the
cycloid example in Case 3 of Section 3.

2. Closed-form solutions

This section discusses several examples for which the differential equation can be
solved in closed form. It is remarkable that, although the arc length of familiar curves
is generally not solvable in closed form, the wheel-road problem is solvable for a
wide variety of functions.

Polygonal wheels The roads corresponding to polygonal wheels are derived from
the case of a wheel that is nothing more than a straight line. Consider the polar
equation r = -csc 0, -n- < < < 0, whose graph is a horizontal line one unit below
the x-axis. The results of Section 1 show that the road on which this polar line will
roll is given by y =f (x) = - r(O(x)) where O(0) = - n-/2 and

The solution to this initial-value problem is x = -log(- tan(O/2)), or 0 =

- 2 arctan e x . Hence the road is given by:

whose graph is an inverted catenary. This means that the polar line will roll on the
catenary so that the polar origin, which we imagine as attached to the line, stays on
the x-axis.
Modifying the straight-line example yields roads for wheels that are regular
polygons. Consider the square wheel. By simply truncating the catenaiy where its
slope is kl-this occurs at x = karcsinhl-and forming a periodic road by
translating copies of the truncated catenary, the angle at the junctions will be 90"
Hence a square will smoothly pass over the junction. FIGURE4 shows several images
from an animation of a rolling square on such a road, along with the locus of a vertex,
which is related to the involute of the catenary.
The road appropriate for a regular n-gon may also be obtained from the catenary
y = - cosh x. If the catenary is truncated at x = k arcsinh[tan(rr/n)] then the angle
VOL. 6 5 . NO. 5 , DECEMBER 1 9 9 2 287
at the road's cusp matches the interior angle of the n-gon, and the amount of rotation
+
to get the wheel into the cusp is O(x) ~ / 2 T . his works out to be exactly 2 ~ / n(the
details, which involve a horrendous-looking identity involving tan, arcsinh, and arctan,
are left to the reader). So the wheel corresponding to the road made of pieces of
inverted catenaries closes up exactly into a regular n-gon. The case of a triangle is
noteworthy in that the rolling cannot happen physically: Because the cusp angle is
less than 90°, the triangle will crash into the road before the vertex gets into the cusp
(FIGURE
5).

FIGURE 3
Some stills from an animation showing the polar line r = - csc 8, - T < 8 < 0, rolling over the
catenary y = - cosh x. The x-values and 8-values correspond to the x-coordinate of the point
of tangency and the 8-value of the point of tangency viewed as a point on the polar line. Adding
~ / to2 the @-valueyields the amount of rotation of the horizontal line.

FIGURE 4
A road made up of pieces of an inverted catenary allows a square to roll smoothly. The dots are
the locus of a vertex during the rolling. Note that the slope of the tangent to the locus has a
discontinuity at the cusp.

FIGURE 5
A vertex of a rolling triangle crashes into the road just before the vertex arrives at the cusp. The
rightmost diagram is a close-up of the collision.
288 MATHEMATICS MAGAZINE

As we shall see several times in this paper, the depth of a road plays a crucial role
in deteimining the shape of the wheel. If the catenaiy road y = -cosh x is raised or
lowered, the shape of the wheel changes. Consider the family of roads k - cosh x
where k < 1. If k = 0, then the wheel is a straight line, but other values of k yield
radically different wheels, as shown in FIGURE6. The closed-form solution, obtained
with the help of A4athernnticu's integrator, is given by 0(x) = +(x) - 4(O) rr/2 +
where

roads: y = - 1- cosh x 0.9 - cosh N 0.7 - cosh x 0.676454 - cosh x

FIGURE 6
Raising or lowering a catena17 road via k - cosh x leads to a variety of wheel shapes. Only the
choice k = 0 leads to a straight-line wheel. The wheel in the last case has vertical asymptotes.

Tilted roads Let f ( x ) = - 1 - x, x 2 0, define a downward-sloping road. The solu-

+
tion to the initial-value problem for 0 is then 0(x) = log(x 1) - rr/2. Therefore the
+
wheel for this oblique line is the equiangular spiral r(0) = 1 exp(0 rr/2) - 1 = +
+
exp(O rr/2), which is shown in FIGURE 7. In the general case that the road is given
+
by - 1 - nzx, the wheel is the spiral given by r = exp[(0 rr/2)/m].
As with the catenary, we can turn the 45" road into a periodic function, in this case
a sawtooth. I11 order to get a smoothly rolling wheel that covers four teeth of the road
in one revolution, we need to find x such that 0(x) = -rr/4: 0(x) = -rr/4 if and
+
only if log(x 1) - rr/2 = -rr/4 if and only if x = e"/" 1 = 1.19... . Now we can
cut off the straight line at this point, generate a sawtooth road, and paste appropri-
ately truncated pieces of the spiral together at right angles to get the sawtooth wheel
shown in FIGURE8. The pasting yields 90" angles because the tangent to the
equiangular spiral makes a constant 45" angle with the radius vector; these angles
plus their mirror images yield the right angles. A wheel that covers more teeth during
each revolution can be obtained by truncating the line at a value of x that satisfies
O(x) = - T / ~ Z and pasting together more and shorter pieces of the spiral.
V O L . 6 5 . NO. 5. DECEMBER 1 9 9 2 289

FIGURE 7
An exponenti:tl, or eclui:tngular, spir~il1.01ling along a tilted line

FIGURE 8
Pieces of ecluiangular spiral can 1)e pasted together to get a wheel that rolls o n a sa\vtooth
road. The examples shown cover four and five teeth per rc~volution,respectively.

Cycloidal roads The cycloid is tlle famous curve tllat is the path of a point on a
traditional round wheel rolling on a straigllt road; iilverti~~g the cycloid leads to the
parametric curve f ( t ) = ( t - sill t,cos t - 1). The wheel tllat rolls on all inverted
cycloid call be found by solving the differential ecluation clO/dt = ( - 1 cos t)/ +
+
(cos t - 1) = 1, so O(t) is simply t - ~ / 2 Hence . the wlleel is given in polar form
I)y r = - ,j(t(O))
i = 1 - cos(6' + ~ / 2= +
) 1 sill 0, the polar fonn of a cardioid (FIC:LKE
9(a)). The cusp of the cardioid rolls o\ler the cusp of the cycloid, at least in theor),. In
practice, there is a crash between the road and the cardioid, siinilar to tlle one that
happens wit11 a rolling triangle (F~c:rrti~< 5). As pointed out hy Rol)ison, a physical
inodel can I)e built so as to avoid the cusp problem by introducing pieces of a
catenary into the road and a straigllt line segment illto the wheel so as to Iqlpass the
cusps. See [4, F I G U K41E for details.
The locus of the top point of the cardioid seems to have the saine general shape as
a cycloid. As an exercise, the reader can verify that, indeed, this locus is a cycloid
stretched vertically by a fhctor of 2. As further exercise, the reader can investigate the
clover-like wlleels that arise f o m lo~veringthe cycloid so that tlle closed-wheel
cor~clitionis met. The case of ( t - sin t, - 13,"s - cos t ) is illustrated ill Frc:r;rtl.: 9(11).
One can also consider the unin\~ertedcycloitl that is tangent to the x-axis from below
- ( t - sill t , - 1 - cos t ) - for \vhich the wheel is derived froin the function:

e(x) = -x + 12si11x
+ cos x - -
7i-
2
This leads to a spiral wheel that requires infinitely many re\~olutionsto pass over tlle
cycloid's lligll point (FIC:LKI'9(c)). Cycloidal roads are discussed furtller in Casc 3 of
Section 3.
290 MATtiEMATICS MAGAZINE

FIGURE 9
(a). A cardioid rolls 011 an inverted cycloid. and the locus of the point opposite the c;lrdiod's
cusp is a vertically scaled c).cloid. The close-up vien- sl~o\\-sthat, in actualit!,, a c;lrdioid-shaped
tire would be puncturecl by the cycloid's cusp. (11) Lowering the cycloid leiids to the
clover-shaped n.heels. (c) A right-side-up cycloid touching the x-axis yields a \\heel that takes
infinitely mi111y r e \ ~ ~ l u t i o ntos pass over the c!,cloid's high point.

It turns out to be worth considering illore general cycloidal roads, such as

trochoidal roads represented by ( t - (1 sin t , - 1 + a cos t ) . The usual conlputations
+
show that the wheel is given by r = 1 (1 sin 0, a l i m a ~ o nthat, wlleil (1 = 1, is the just
discussed cardioid. This liinason intersects the positive y-axis at (0, cr 11, points
+
whose loci during the rolling are ( t i sill t , ( u 1)cos t ) . Tllesc loci are cycloids
stretched in the y-direction. Yote that there are t ~ v ocases in which thc stretching
constant is 1-that is, the locus is an exact cycloid: The classical case (I = 0, with

FIGURE 10

Rolling a 1imac;on along a trochoid yields an exact c!,cloidal locus

VOL. 65. NO. 5. DECEMBER 1 9 9 2 291

straight road and round wheel, and the case when a = 2 (or -521, where the road is a
trochoid, the wheel is a lima~on,and the locus of the top of the inner loop is an exact
inverted cycloid (FIGURE10). Are these the only two examples of road-wheel
combinations for which a point on the wheel traces out an exact cycloid?

A road that is its own wheel Can there be a road for which the corresponding
wheel is congruent to the road and for which the rolling motion matches points that
correspond under the congruence? Consider the road that is given by the parabola
= - T .2 - 1/4. The differential equation yielding 0(x) is then d8/dx = l / ( x %
1/4), for which the solution is O(x) = 2 arctan(52x) - ~ / 2 ,or x = 1/52 t a n ( ~ / 4+
8/52). The polar wheel then has the form r(8) = -y(x(8)), which simplifies to
r = 1/(2 - 2 sin 8). This last is the graph of the parabola y = x" 1/4, which is the
reflection of the road in the line y = - 1/4. FIGURE11 shows this singular situation
of a wheel rolling on itself. We leave the verification that corresponding points touch
as an exercise. Robison [4] showed that this parabola is the only curve that has this
property.
Note that raising or lowering the parabola changes the shape of the wheei
dramatically (see FIGURE11).

road: y = - x 2 - i; road: y = -x2 - road: y = -x2 - I

wheel: y =x2 -

FIGURE 1 1
The wheel corresponding to the parabolic road given by - x 2 - f is simply a reflection of the
road itself. But the wheels for other parabolic roads are not parabolic.

Round wheels can roll on round roads A well-known puzzle can be interpreted as
follows: What is the wheel corresponding to a road that is an upward-opening
semicircle whose highest points' are on the x-axis? Such a road is given by f ( t ) =
(cos t, sin t), n- I t 5 2n-. The differential equation for 0(t) is simply d0/dt = 1, so
0(t) = t - n-/2 and ~ ( 0 =) - s i d e +
n-/2) = - cos 0, n-/2 5 0 I 3 ~ 1 2 .Thus the
wheel is a polar circle with geometrical center at ( - 1/2,0). The aforementioned
puzzle is the one that asks for the locus of a point on the circumference of a circle that
is rolling in a way tangent to the interior of a circle twice as large. The locus is a
straight line, which shows itself in FIGURE12 as the vertical lines in the arch-shaped
locus.
292 MATHEMATICS M A G A Z I N E

FIGURE 12
A circle rolls on inverted semicircle. The dotted paths are the loci of two points in the wheel's
circumference.

Off-centered elliptical wheels Consider the ellipse given in polar form by r = ka/
(1 - a sin 8), where 0 < a < 1 andk > 0; the origin, which corresponds to the axle of
the wheel as it rolls, is one focus of the ellipse, the other focus is on the positive
y-axis, a is the eccentricity, and k is the distance from the origin to the corresponding
directrix. Such an elliptical wheel rolls on the road y = -(ka/u"(l - a cos(cr)),
where c = u/ka and a = fi?. The derivation of this is slightly complex. Here's a
sketch:
1. Solve the initial-value problem to get

) + arctan ( 1 a ) .
ax
- = arctan
tan(8/2)
-
2ka -a

2. Take the tangent of the relationship in (1) and use some trig forlnulas to get

1+ E 1 - cosS(cx) -
-
1 sin 8 +
(1+cos(cx))2 l S i n 8

3. Solve the preceding for sin 8, substitute into

and simplify to get the desired representation of the road as (ka/a"(l - a cos(cx)).

+
If we set k = 1 and s = 1/ fi then the road is just y = - fi cos r (see FIGURE
13); this is a special case of one of the proper depths to lower the cosine so as to get a
closed wheel (see Remark 5 in 51).

FIGURE 13
An ellipse rolls on a cosine curve. In the example shown the road is given by y = - fi + cos x
and the ellipse has the polar form r = l/(& - sin 0).
V O L . 65, N O . 5, DECEMBER 1 9 9 2 293
Centered elliptical wheels The preceding example involved an off-center elliptical
wheel; that is, the axle is not at the ellipse's center. Let's now find the road on which
an ellipse with centered axle will roll. The ellipse (x/a)" ( ~ / b=) ~1 has the polar
representation r = b/ J,- where rn abbreviates 1 - b2/a'; in this repre-
sentation the polar center-the axle of the wheel-coincides with the center of the
ellipse. We can find the appropriate road by first finding the relationship between x
and 0; the differential equation for 0 separates to $0 = ( l / b ) dx and
the initial condition then leads to:

Substituting 9 = 4 + n-/2 yields sin @ = cos 0 and

Now this is an incomplete elliptic integral of the first kind [l, 17.2.2, 17.2.171.
+
Therefore r / b = F(O .rr/2lrn), which can be inverted by using what is known as
+
the Jacobian elliptic function sn: sin(0 n-/2) = sn(r/b, m). The road is therefore
given by y = -r(e(x)) = -b/ dl
- rn s n 2 ( r / b , rn) . The sn function is built into

Mathernatica, and so it is a simple matter to generate a diagram of the road and

wheel. But some complications arise when one tries to use the closed form to
generate an animation; they can be worked around by adding multiples of n-/2 as x
increases through the quarter-period value. But life is much simpler if we use the
numerical approach of the next section to generate the animation. It is easiest to start
with the ellipse and use the Case 2 discussion, as that avoids the computation of
values of sn. The result of one such computation is shown in FIGURE14 ( a = 1/2,
b = 1, eccentricity = 0.87). Ellipses with eccentricities greater than about 0.97 lead to
a road-wheel crash similar to the one for the triangle (FIGURE 5).

FIGURE 14
Four frames from an animation showing an elliptical wheel (a = i,b = 1) with its axle at its
center rolling on a road defined using the elliptic sine function.

An elliptical road We can consider bounded roads as well as roads that protrude
above the x-axis. As one example, consider the ellipse given parametrically by
( a sin t, - b cos t), where a and b are positive. The closed-wheel integral is 2n-a/b,
and integer values of b/a lead to closed wheels that are familiar curves. The usual
+
calculation shows that t(0) is just (b/a)(O n-121, whence the wheel's polar form is
r(0) = - y(t) = b cos[(b/a)(O +
n-1211, a polar rose (FIGURE15(a)). The locus here
was a surprise to us as it turns out to be a piriform, which we had considered earlier
in another context (93, Case 4). If b/a is rational then the wheel is a rosette, as
defined by Hall [2].
294 MATHEMATICS MAGAZINE

11

FIGURE 15
(a). A four-leafed rose rolls inside the ellipse (isin t , -cos t ) , and the locus of the tip of a petal
is a piriform. (b) The wheels corresponding to lnore general ellipses [shown here is
( 3 sin t , - 2 cos t ) ] are rosettes.

Vertical scaling We have seen in some of the preceding examples that raising or
lowering the road usually changes the wheel significantly, and may destroy the
closed-wheel condition. Another way to change the road is by scaling the y-coordi-
nate: y =f(x) becomes y = kf(x), and x =f,(t), y =f,(t) becomes N =fl(t), y =
kf,(t). The closed-wheel condition is affected as follows.
1. If the a-periodic road y =f(x) has a closed wheel, then so does y = kf(x) for any
positive rational scaling factor k.
2. If the a-periodic road y =f(x) does not have a closed wheel, the scaled road
y = kf(x) does have a closed wheel whenever k is a rational multiple of

Similar results hold for roads defined parametrically.

As an example of scaling, consider the road y = (cos x) - b, b > 1. The scale factor
1/ J E produces a closed wheel. Note that the scale factor goes to zero as b goes
to infinity, and that the scaled road approaches y = - 1. Since the wheels for these
roads are ellipses, larger values of b correspond to wheels with smaller eccentricity.

FIGURE 16

A scaled cycloidal road has a cuspitate rosette as its wheel.

VOL. 6 5 , NO. 5, DECEMBER 1 9 9 2 295

Another case where scaling gives interesting road-wheel pairs is an inverted cycloid.
+
r = t sin t , y = k(- l - cos t). Assume k is a positive rational. Then the wheel is
+ +
given by r(8) = k ( l cos[k8 kn-/2]). For k a positive integer these wheels are
cuspitate rosettes, curves similar in appearance to standard polar roses, but having
cusps at their center. For more about cuspitate rosettes see [2] and the references
therein.

Summary of some road-wheel relationships Here is a summary of road-wheel

relationships including some not in closed form (12,13) and some whose derivations
are left as an exercise for the reader (14,15,16).

\$'heel Form of Road Locus

1. Straight line Catenary

2. Regular polygons Piecewise catenaries
3. Circle Horizontal line Cycloid/Trochoid
4. Circle, axle on circumference Circle, radius doubled Circular arch
5. Eq~~iangular spiral Oblique line
6. Piecewise equiangular spiral Sawtooth
7. Ellipse, axle at focus Cosine
8. Ellipse, axle at center 1/d1 - (elliptic sine)'
9. Parahola ( r 2 - 1/41 Parabola ( -"2 1/41
10. Cardioid; see F I G U K9(a)
E Inverted cycloid Scaled cycloid
11. Spirals, clovers; see FIC,I.HE9(b, c) Lowered inverted cycloids
12. Pointed wheels; see F ~ c u ~ 19
rr Lowered cycloids
13. Piriform See F I C U ~20~ E
14. Hippopede See FIGIIHE 18
15. Roses, rosettes Ellipses Piriform
16. Lilna~on Trochoid Cycloid
17. +
Hyperbolic spiral [ r = 1/(0 T ) ] Exponential [ = -(2/.ir)e-'1
18. Cuspitate rosettes Scaled cycloids

3. Generating solutions numerically

As we demonstrated in the last section, surprisingly many road-wheel problems are

solvable in closed form. When one deals with arc length, however, functions that are
not integrable in terms of elementary functions eventually show up. In this section we
describe how to generate plots of roads and wheels numerically. Our procedures will
be described for Mathernatica, but the methods, with the possible exception of the
animations, could be adapted to other software with graphics capabilities. What is
needed is the ability to apply a numerical differential equation algorithm and plot
points and lines.
The key step, as we have seen, is to solve the initial-value problein relating the
road parameter and the polar angle. The differential equation involved is separable,
but the resulting integrations can be daunting, even when feasible. Furthermore, we
have often found this closed-form approach too slow to generate plots of the roads and
wheels. A inore efficient method is to solve the initial-value problem nu~nerically
using a standard Runge-Kutta algorithm; then a set of points defining the wheel or
road can be generated with the help of the radius condition. The points are joined by
lines to get the final image. The relationship between the road parameter and the
polar angle is almost always nonlinear (an exception: a circular wheel rolling on a
straight road), which means that constant velocity and constant angular velocity
296 MATHEMATICS M A G A Z I N E

cannot be achieved simultaneously. The use of a fixed step-size in the numerical

method can guarantee that, in an animation, one of these velocities is constant.
When animating one of these wheels rolling along its road, the relation between
the polar angle 8 and the horizontal coordinate x must be known. The rolling can be
broken down into two parts: a rotation, given by the cllange in 8 , and a translation,
given by tlle corresponding change in x.
There are four cases to consider. Some stripped-down code for Case 1, the simplest
case, is given in Section 5.
Case 1. Suppose the road is given by y =f ( x ) and we want to generate tlle wheel.
The initial-value problem is dO/dx = - l / f ( x ) , 8(0) = - ~ r / 2 . The Runge-Kutta
method generates a table of ordered pairs ( x , 8). This table and the radius condition
are used to produce the wheel.
+
Example. Let y = - 1.887365 - (2/3)cos x sin x - ( 1 / 2 ) sin 2 x , where the con-
stant term has been chosen so that the closed-wheel condition holds for one
revolution per ~ e r i o d The
. wheel (FIGURE
17) is generated by using a Runge-Kutta
method to solve the initial-value problem, taking the independent variable to be x.
Thus, when animated, this wheel's axle moves with constant linear velocity. This
example shows that wheels need not have an axis of symmetry.

FIGURE 1 7
This example shows that the angular speed during rolling can vary a lot. The wheel rotates
quickly when it is above the high bumps in the road, as illustrated by the steepness in the 0 vs.
I plot.

Case 2. Suppose the wheel is given by r = g ( 6 ) and we want to generate the road.
The initial-value problem is d x / d 8 = g(B), x ( - ~ / 2 )= 0. This time the numerical
method gives pairs ( 8 , x), and the road can be generated using the radius condition.

Example. Let r = 445 - 4 sin" , a polar curve called a hippopede (see [3]). Again,
Runge-Kutta is used to solve the initial-value problem, but this time it is convenient
to take 8 as the independent variable. Thus, when animated, the wheel (FIGURE 18)
would exhibit constant angular velocity.

FIGURE 18
The example of a hippopedal wheel illustrates the case that the wheel is given in polar form
and the road is found numerically.
 VOL. 65, NO. 5, DECEMBER 1 9 9 2 297
Case 3. Suppose the road is given parametrically by x =fl(t), y =f2(t), where
fl(0) = 0. Now the initial-value problem relates the parameter t and the polar angle
0, and reduces to dO/dt = -f;(t)/f,(t), O(0) = - 7 ~ / 2 . This time the numerical
solution gives ordered pairs (t, O), and the radius condition can be used to produce
the wheel, very much as in Case 1.
Example. Let the road be a cycloid, lowered sufficiently so the closed-wheel
condition (with an integer number, n, of periods per revolution) holds, and translated
+
so the y-axis bisects one arch. The parametric equations are x = t sin t , y =
cos(t) - d,. For this cycloid, the closed-wheel condition is

7r +
1 cos t 7T
c o s t dt= -
n

+
for which the positive solutions are d, = 1 2n"(2n +
1). (One could also consider
negative solutions: dl = - 1 yields a road with a cardioid wheel similar to that in
FIGURE9.) Using the table of (t, 0) values obtained with the Runge-Kutta algorithm,
and the radius condition, we get the wheels corresponding to different periods (see
FIGURE19). In this case both x and 0 are given as functions of the parameter t, so in
an animation neither will increase linearly. FIGURE19 also shows the case of a
period-l/2 roller; that is, n = 1/2, d = 1: and the wheel rotates twice for each
period of the cycloid. For the n = 0 case (infinitely many revolutions) see FIGURE
9(c>.
Case 4. Suppose the wheel is given parametrically by x = gl(t), y = g,(t). In order
to get a simple closed wheel, assume g l and g, are periodic with the same period. In
terms of g l and g 2 , the initial-value problem relating the x-coordinate of the road
with the parameter t is:

The numerical method produces pairs (t, x), and if care is taken to use the same
t-values, the corresponding coordinates of the road are found froin the radius
+
condition to be given by: y(t) = - J g 1 ( t l 2 g 2 ( t l P .
To generate an animation of this case, we must also have the polar angle 0 in terms
of t. Unfortunately, there are complications involving branches of the arctangent
function that prevent the direct use of 0 = arctan[g,(t)/g,(t)], so we generate
another table of values, this time (t, 8), by applying Runge-Kutta to:

again being careful to use the same t-values as were used to generate the (t, x) pairs.
Together, the (t, x) pairs and the (t, 0) pairs define a set of (x, 0) pairs that can be
used to animate the wheel.
+
Example. Let the wheel be defined by x = - sin t (1/2) sin2t, y = - cos t. This
shape (FIGURE 20) is known as a pirz$orm (again, see [3]). As in the example for Case
3, neither linear velocity nor angular velocity will be constant in an animation.
298 MATHEMATICS M A G A Z I N E

FIGURE 19
Some wheels corresponding to a cycloidal road. These wheels can be given in closed form, but
when generating images or animations it is much more convenient to ignore the closed form
and just use the numerical approach. The top wheel covers one cycloid period during each
revolution; the next covers two; the next covers four. The bottom wheel covers a half-period of
the road during each revolution and corresponds to n = in the closed-wheel condition.

FIGURE 20
The piriform is an example of a parametrically given wheel. Its road is found, as in the other
cases, by numerically solving the appropriate initial-value problem.
VOL. 65, NO. 5, DECEMBER 1 9 9 2 299
Remark. In all four cases, the choice of independent variable in the Runge-Kutta
step is arbitrary. If an animation is planned, then this choice call be made to cause a
particular quantity, usually x or 6 , to change linearly in the animation.

4. Squaring the circle with Fourier series

A natural thought when dealing with periodic functions, such as the roads for closed
wheels, is to look at the Fourier series. For wheels with an axis of symmetry through
the axle, we can make the periodic function even, thus yielding a cosine series. The
question then arises as to what shape the wheels for the various Fourier approxima-
tions will have. Clearly, for the approximation using only the constant term, the wheel
will be a circle and, as the Fourier series more closely approximates the road, the
wheels for the Fourier approximations will more closely approximate the original
wheel. Thus if we begin with the road for a rolling square, the wheel for the Fourier
approximations will "square the circle." Obviously, we are not restricting ourselves to
Euclidean tools! There is one problem with this process, however: The Fourier
approximatioils fail to satisfy the closed-wheel condition.
For example, suppose p(x) = cos x - 6 ,which is a finite Fourier series already
with period 2 ~ satisfying
, the closed-wheel condition. The constant term or 0th
Fourier approximation, is p,(x) = - 6, 6
and so
is the radius of the corresponding
wheel (a circle). It was shown earlier that the wheel for p(x) traverses two periods of
the road in one revolution, so the circumference of the circular wheel must be 4m-
(taking the period of the constant function the same as that of p(x)). But this makes
the radius of the wheel equal to 2, a contradiction.
The case when p(x) is the road for a square wheel is similar. Here, the circular

-
wheel that rolls on the 0th Fourier approximatioil has circumference 7.12866 instead
of 8 arcsinh 1 7.05099. And the closed-wheel condition fails here by 0.00138 for the
two-term Fourier approximation road. Of course, as the trigonometric polynomials
more closely approximate the original road, they will also come closer to satisfying the
closed-wheel condition, but the condition fails nevertheless for each Fourier approxi-
mation.
What is needed, then, is a sequence of approximations, each satisfying the same
closed-wheel condition as the road, which converges to the road. Such a sequence can
be constructed by exploiting the orthogonality of the cosine functions in a Fourier
series, along with the fact that the closed-wheel condition involves the reciprocal of
the road function.

Road approximations having closed wheels Let p(x) be a continuous, even,

negative, ~ e r i o d i cfunction with ~ e r i o dT. Then q(x) = l / p ( x ) is also continuous,
even, negative, and T-periodic, and so can be expanded in a Fourier cosine series.
Denote the partial sums of the Fourier series for q(x) by q,(x), n = 0 , 1 , 2 , .. . . The
q,'s converge to 9, are continuous, even, T-periodic, and, for large enough n ,
negative. If, in addition, we assume that the derivative of q. is piecewise continuous,
-

the convergence is uniform. Finally, we shall assume that q is "nice enough" so that
all the q,'s are negative, which is the case in all our examples. We now form the
sequence (p,(x)}, where p,(x) = l/q,(x), which converges uniformly to p(x), and
we shall use this sequence to approximate the road. The p,'s have the same period as
p and satisfy the same closed-wheel condition because
300 MATHEMATICS MAGAZINE

and, because the definite integrals of the cosines in the Fourier series vanish,

Approximations to the square wheel Recall that the road for a square wheel with
side 2 is the periodic extension of y = - cosh x, for - arcsinh 1 I x 5 arcsinh 1. The
first seven Fourier coefficients for the reciprocal are: -0.891107, -0.12537,
0.0230828, - 0.0100959, 0.0056363, - 0.00359458, 0.00249146. These were found
using the standard formulas and integrating numerically. The first three approxima-
tions to the road are:

For the wheel corresponding to the approximation P ~ ( X )the, ~ o l a rangle as a

function of x is described by the initial-value problem dO/dx = - l/p,(x), 0(0) =
-7r/2. This can be integrated in closed form since - l / p k ( x ) is a trigonometric
polynomial, but the ilumerical method described in Section 3 is more efficient,
especially for produciilg animations.
FIGURE21 shows two correspoildiilg positioils of the road-wheel pairs for p,(x),
p,(x), p,(x), and p6(x). Note how the circle becomes square-like very quickly.

F I G U R E 21
Approximating a catenary road with partial sums of its Fourier series yields wheels that
transform a circle to a square.

5. A Mathernatica wheel-building package

We have written a complete Mathernatica package (version 1.2) that generates roads
from wheels and vice versa. The package has options by which the user can generate
stills or animations, with or without a locus, spokes, shading, and so on. The notebook
can be obtained by sending a Macintosh disk to one of the authors. We include here a
bare-bones version of the routine to give some idea of how such a program is written.
This routine takes a function defining a road and displays the road and the wheel with
its axle at the origin.
VOL. 65, NO. 5, DECEMBER 1 9 9 2 301
Even when a closed-form solution is available, it is often simpler to generate a
diagram or animation by taking a numerical approach. Nevertheless, sometimes the
closed form must be used. For example, rolling polygons are best generated directly,
without numerical approximations. Some code for doing so can be found in the
Appendix to [6].

RoadMovieCfunc-, Cx-, xmax-11 := NoduleCCthetal, theta),

thetalCx-I = thetaCx3 /.

F i r s t @NDSoLveC

C t h e t a ' C x l == - l / f u n c , t h e t a C 0 1 == - P i / 2 1 ,
t h e t a C x l , Cx, 0, xmax), PrecisionGoaL->21;
ParametricPLotCC-func CCosCthetalCxll, SinCthetalCxll), Cx, func)),
Cx, 0, xmax), PLotRange->ALL, AspectRatio->Automatic] 1

For example, R o a d M o v i e C C o s C x I - S q r t C l O l , C x , 6 P i ) ] will generate

the period-3 closed wheel for a cosine road (similar to FIGURE2, but with n = 3).

REFERENCES

1. M. Abramowitz and I. Stegun, Handbook of Mathematical Functions, Dover Publications, Inc., New
York, 1965.
2. L. Hall, Trochoids, roses and thorns-beyond the Spirograph, College Math. J. 23(19921, 20-35.
3. J. D. Lawrence, A Catalog of Special Plane Curves, Dover Publications, Inc., New York, 1972.
4. G. B. Robison, Rockers and rollers, this MAGAZINE33 (19601, 139-144.
5. P. Said, Problem E1809, Amer. Math. Monthly 74 (19671, 87.
6. S. Wagon, Mathematicn in Action, W . H . Freeman & Co., New York, 1991.
7. D. 6. Wilson, Problem E1668, Amer. Math. Monthly 72 (19651, 82-83.

If you ask mathematicians what they do, you always

get the same answer.. They think. They think about
difficult and unusual problems. They do not think
about ordinary problems: they just write down the
answers.
M. Egrafov (translatedfrom Russian),
contributed by the late R.P. Boas, Jr.