Академический Документы
Профессиональный Документы
Культура Документы
5, November 2016
THE
COLLEGE
MATHEMATICS
JOURNAL
In this Issue:
The sine of one degree via geometry, algebra, and complex analysis
Knowing whether you can win a football confidence pools is provably hard
Pythagorean theorem: a new long proof and questioning Leonardo da Vincis
EDITORIAL POLICY
The College Mathematics Journal serves all who are interested in college mathematics. CMJ seeks lively, well-motivated articles that will enrich undergraduate instruction
and enhance classroom learning, as well as expository
papers that stimulate the thinking and broaden the perspectives of those who teach undergraduate-level mathematics. Articles involving all aspects of mathematics are
welcome, including history, philosophy, problem solving,
pedagogy, applications, computation, and so on. Emphasis
is placed on topics taught in the first two years of college.
Contributions from teachers in high schools, two-year colleges, four-year colleges, and universities are welcome.
In addition to articles, CMJ publishes short pieces containing material suitable for immediate classroom use, problems, solutions to published problems, student research
projects, media reviews, and all sorts of mathematical
ephemera: proofs without words, arresting examples
of fallacious proofs, mistakes and/or other mathematical anomalies, quotations, poetry, humor, cartoons, and
doodles.
SUBMITTING AN ARTICLE
For more information about writing for CMJ, please visit
the web site at www.maa.org/cmj and follow the links to
Information for Authors and Electronic Production Information. Submissions of articles and Classroom Capsules
are required via the CMJs Editorial Manager System. The
authors name(s) should not appear in the file. Initial
submissions in pdf or LATEX form can be made at www.
editorialmanager.com/collmathj/. The Editorial Manager
System will cue the author for all required information
concerning the paper. Questions concerning submission
of papers can be addressed to the editor at cmj@maa.
org. Authors are strongly encouraged to use LATEX and the
CMJ style file, available on the website and EM system,
without any additional packages or custom formatting.
Contributions to Problems and Solutions, and Media Highlights should be submitted to the appropriate section editor at the address listed inside the journal. Letters to the
Editor on any topic are invited. Comments, criticisms, and
suggestions for making CMJ more lively, entertaining, and
informative are welcome and should be forwarded to the
editor at cmj@maa.org.
Microfilmed issues may be obtained from University Microfilms International, Serials Bid Coordinator, 300 North
Zeeb Road, Ann Arbor MI 48106.
Advertising correspondence should be addressed to
MAA Advertising
1529 Eighteenth St. NW
Washington DC 20036
Phone: (202) 319-8461
E-mail: advertising@maa.org
Further advertising information can be found online at
www.maa.org.
Permission to make copies of individual articles, in paper
or electronic form, including posting on personal and
class web pages, for educational and scientific use is
granted without fee provided that copies are not made or
distributed for profit or commercial advantage and that
copies bear the following copyright notice: Copyright 2016
Mathematical Association of America. All rights reserved.
Abstracting with credit is permitted. To copy otherwise,
or to republish, requires specific permission of the MAAs
Director of Publications and possibly a fee.
Periodicals postage paid at Washington DC, and additional
mailing offices.
Postmaster: Send address changes to Membership/Subscription Department, Mathematical Association of America, 1529 Eighteenth Street, NW, Washington DC 20036-1385.
Printed in the United States of America
THE
COLLEGE
MATHEMATICS
JOURNAL
Editor
Brian Hopkins, Saint Peters University, Jersey City, NJ
Classroom Capsules Editors
Ricardo Alfaro, University of MichiganFlint, Flint, MI
Lixing Han, University of MichiganFlint, Flint, MI
Kenneth Schilling, University of MichiganFlint, Flint, MI
Problems & Solutions Editors
Curtis Cooper, University of Central Missouri, Warrensburg, MO
Charles N. Curtis, Missouri Southern State University, Joplin, MO
Media Highlights Editors
Tanya Leise, Amherst College, Amherst, MA
Warren Page, New Rochelle, NY
Philip Straffin, Longmont, CO
Associate Editors
sarah-marie belcastro, Mathematical Staircase, Hadley, MA, and Smith College, Northampton, MA
Arthur Benjamin, Harvey Mudd College, Claremont, CA
Susan Goldstine, St. Marys College of Maryland, St. Marys City, MD
Heather Hulett, University of WisconsinLa Crosse, La Crosse, WI
Daniel King, Sarah Lawrence College, Bronxville, NY
Dominic Klyve, Central Washington University, Ellensburg, WA
Michael Maltenfort, Northwestern University, Evanston, IL
Vincent J. Matsko, University of San Francisco, San Francisco, CA
Peter Mercer, State University College at Buffalo, Buffalo, NY
Nancy Ann Neudauer, Pacific University, Forest Grove, OR
Brigitte Servatius, Worcester Polytechnic Institute, Worcester, MA
Jeff Suzuki, Brooklyn College of the City University of New York, Brooklyn, NY
Todd Will, University of WisconsinLa Crosse, La Crosse, WI
Robin Wilson, California State Polytechnic University, Pomona, CA
Beverly Ruedi, Managing Editor
Bonnie Ponce, Assistant Managing Editor
Though the radian is the undisputed king of angular measurement in calculus, required
as it is for the familiar trigonometric derivative and integration formulas to hold true,
it is the degree to which most of us are first introduced. The humble degreeone
three-hundred-sixtieth part of a complete rotationis one of the oldest methods used
to measure angles. Its use dates to antiquity and it can be found in ancient Babylonian,
Greek, Indian, and Mayan mathematical traditions [1, 4]. Many theories have been
suggested for its ubiquity, ranging from the astronomical (a solar year has approximately 360 days) [2, p. 37] to the practical (360 has lots of divisors) [8, pp. 1011],
but the fact remains that the degree is, if not a mathematically ideal way to measure
angles, a particularly human way to do so.
In a typical trigonometry class we devote time learning those special angles that
admit exact trigonometric ratios involving roots of integers, such as 30 and 45 . A
question that might interest students of trigonometry is whether this is also true of the
archetypal measure 1 ? That is, is its sine expressible as a ratio of some combination
of radicals and integers? Can we compute the sine of a single degree exactly?
The answer to this question is yes, but the path to computing it meanders through
classic geometry, polynomial algebra, and complex numbers. The purpose of this paper
is to follow this path to value of sin 1 and to see some beautiful mathematics along
the way.
1 sin2 .
http://dx.doi.org/10.4169/college.math.j.47.5.322
MSC: 01A99
322
The sine of an angle can also be computed via isosceles triangles. Consider an
isosceles triangle with vertex angle 2 and congruent sides of length 1. Since the altitude of an isosceles triangle bisects both the vertex angle and the base, we have
sin =
1
base of this isosceles triangle.
2
(1)
The two most familiar isosceles triangles are the equilateral (606060 degree) triangle
and the right-isosceles (454590)
triangle, from which the familiar sine formulas
36
36
x
1
108
72
x
36
72
72
72
36
Assume the congruent sides are length 1 and denote the length of the base by x. An
angle bisector constructed through one of the base angles divides the golden triangle
into the two smaller isosceles triangles depicted in Figure 1: a short, squat 3636108
triangle with sides of lengths x, x, and 1; and a tall, skinny 367272 triangle with
sides of length x, x, and 1 x. This latter triangle is another golden triangle, so it is
similar to the larger one. Comparing ratios of side lengths gives
1
x
= ,
1x
x
from which we conclude x = ( 5 1)/2. Greeks recognized this as the golden mean
and is the reason the triangle earned its name. Applying (1), we conclude
51
sin 18 =
.
(2)
4
Can we deduce the sine of yet smaller angles? If we can construct two angles > ,
then it is also possible to construct their difference . Moreover, if we know the
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
323
sin sin
sin
cos
cos sin
sin cos
sin ( )
cos ( )
Using this process, how small an angle can we obtain? Given that we know the sine
and cosine at 45 and 30 , we can compute
2
31
sin 15 = sin(45 30 ) =
,
(3)
4
2
3+1
.
cos 15 = cos(45 30 ) =
4
Combining these two values with the trigonometric values at = 18 from (2), we
can work down to 3 :
2
3+1
51 2
31
5+ 5
sin 3 =
.
(4)
16
This is just a stones throw away from sin 1 , but sadly, that is as close as we are
going get using only the Greek geometers tools of straightedge and compass. None
of the Greek geometersor indeed any of the geometers to come aftercould find
a straightedge-and-compass construction that would yield 1 . It would not be until
centuries later that such a construction was shown to be impossible, a consequence
of the powerful algebraic theorems devised by mathematicians in the 18th and 19th
centuries [14]. Perhaps the most relevant result to us is the following: An angle of
integer degree measure can be trisected if and only if it is a multiple of 9 . Since this is
not true of 3 , we cannot construct 1 . Interested readers can find a wonderful survey
of the algebraic results regarding the constructibility of angles in the Polya Award
winning article [7].
324
(5)
which algebraically relates the sine of an angle to the sine of thrice the angle. Setting
= 15 in the triple-angle formula (5) gives
3+1
2 3
,
(6)
sin 45 = 3 sin 15 4 sin3 15 =
2
using the exact value of sin 15 from (3). After a moments
pause (or panic?), we see
that (6) is equal to a more familiar value sin 45 = 2/2 by squaring both sides.
Since we know the exact value of sine at 3 , setting = 1 in (5) gives
sin 3 = 3 sin 1 4 sin3 1 ,
showing that x = sin 1 is a solution to the polynomial equation sin 3 = 3x 4x 3 .
Isolating the x 3 term and using the exact expression for sin 3 from (4), this becomes
3
2 ( 3 + 1)( 5 1) 2 ( 3 1) 5 + 5
3
.
(7)
x = x
4
64
Figure 3 shows a plot of the curves y = x 3 and y = (3x sin 3 )/4, which clearly
intersect three times, so the solution closest to the origin must be the exact value of
sin 1 , but how do we find it?
1
0.2
0.5
0.1
0.5
0.5
1 0.2
0.1
0.1
0.5
0.1
0.2
0.2
325
The solution to this equation, or rather, the more general depressed cubic equation
x 3 = Ax + B,
(8)
(9)
What makes this formula such a wonder is that Tartaglia and Cardano derived it
entirely geometrically as a statement about the volume of a large cube with a smaller
cube removed from one corner! A wonderful description of its discovery, and the feud
that eventually resulted between the two mathematicians, can be found in [5, pp. 133
154]. Comparing (8) and (9), if we can find values of t and u that solve the pair of
equations
A = 3tu,
B = t 3 u3,
(10)
then x = t u will solve the original depressed cubic (8); so let us solve (10) for t
and u.
Solving both equations simultaneously for u 3 yields
A3
= t 3 B,
27t 3
A
= 0.
27
This is a quadratic in t 3 , so applying the quadratic formula and taking the positive
radical (as Cardano would have) gives
A3
B2
B
3
.
t = +
2
4
27
Taking the cube root of this yields t, which we can then use to solve for u to obtain
2
3
3 B
3
A
A3
B
B
B2
+
,
x =t u =
2
4
27
2
4
27
as a solution of the depressed cubic.
Now we apply this to the problem at hand. If we set A = 3/4 and B = (sin 3 )/4,
then we recover (7). Using the TartagliaCardano technique, it follows that one of the
solutions must be
2
3
3
1
1
sin
3
sin
3
sin
3
sin2 3
+
+
.
(11)
x=
8
64
64
8
64
64
326
Before we declare victory and call this value sin 1 , recall that (7) has three real
solutions, so this x might be one of the other values. Unfortunately for us, numerically
evaluating the expression in (11) on a calculator gives
x 0.009 + 0.015 i
where i denotes the so-called imaginary number, i.e., i = 1. That is, the x given by
(11) is not even a real number, much less one of the solutions to our depressed cubic
equation, and much much less our elusive sin 1 !
Where did we go wrong? With a little algebra, we can rewrite (11) as
x=
sin 3
3
=
sin 3
3
2
+ sin 3 1 sin 3 + sin2 3 1
+i
3
sin 3 + i cos 3
2
cos 3
(12)
Of course, it was not the additive properties of complex numbers that befuddled
mathematicians; it was their multiplicative habit of squaring to negative numbers.
Remarkably, viewing complex numbers as vectors gives a simple geometric description of complex multiplication as well. To see it, imagine for a moment we have a
complex number z, which we represent as a vector. If z has length r and direction
angle (measured from the positive x-axis), it can be written as
z = r cos , r sin = r cos + i r sin
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
327
using (modern) trigonometry. Consider a second complex number (or plane vector,
whichever you prefer) w = s cos + i s sin and compute the product zw by expanding the terms and recalling that i 2 = 1:
zw = (r cos + i r sin )(r cos + i r sin )
= r s(cos cos sin sin ) + i r s(sin cos + cos sin ).
That may not look like an improvement, but the expressions in parentheses are
precisely the angle addition formulas for sine and cosine! That means
zw = r s cos( + ) + i r s sin( + ),
that is, zw is a vector of length r s that makes a + angle with the positive x-axis.
Said differently, to multiply two plane vectors as complex numbers, we multiply their
lengths and add their direction angles.
This simple geometric description gives a nice explanation of why the product of
two negative real numbers is a positive one: Every negative number makes an angle
of 180 with the positive x-axis, so when they are multiplied together their angles are
added to obtain 360 , yielding a vector pointing in the direction of the positive x-axis.
(Can you now see why it also shows that i 2 = 1?)
Perhaps more germane to our problem, it also explains how complex cube roots
work. If z is a complex number with vector magnitude r and direction angle , then z 3
is a vector with magnitude r 3 and direction angle 3. Hence, to find a complex number
that cubes to 8i, we need only a vector whose length cubes to 8 (i.e., the length of 8i)
and whose angle triples to be coterminal with 90 (i.e., the angle of 8i). Obviously, a
vector of length 2 and angle 30 should work, so that
2 cos 30 + i 2 sin 30 = 3 + i
is a cube root of 8i. The reader is encouraged to cube this by algebraic expansion to
see that it does indeed equal 8i.
However, unlike the case for real numbers, this is not the only cube root. For
example, a vector of length 2 and angle 150 would cube to a vector of length 8 and
angle 450 which is coterminal to 90 . Thus,
algebraic cube root of z. Of the three, the root with the minimumdirection angle (in
absolute value) is called the principal cube root and is denoted 3 z. In the example
above, we would have
(8i)1/3 =
3 + i, 3 + i, 2i
whereas
3
8i = 3 + i.
That complex numbers have multiple cube roots explains precisely the problem we
faced in our algebraic determination of sin 1 . When a modern calculator computes
a complex cube root, it determines only the principal root. This is why the algebraic
solution in (11) failedthe cube roots we demand of the solution are not necessarily
the principal cube roots. If we return to our discussion of the solution to the depressed
cubic equation (7), we realize that instead of (12), a more accurate description of the
TartagliaCardano solution is
x
(13)
Unfortunately, this no longer defines a single value, but rather nine different values,
three for each cube root. How do we sift among them to find the value sin 1 ?
One approach is to graph the nine complex numbers described in (13). For convenience, define
z = sin 3 + i cos 3
and
w = sin 3 + i cos 3
cos 31 + i sin 31 ,
cos 151 + i sin 151 , .
=
cos 271 + i sin 271
cos 87 + i sin 87
cos 31 + i sin 31
329
2x
2 sin 1
2 sin 1
w1/3
cos 29 + i sin 29 ,
cos 149 + i sin 149 , .
=
cos 269 + i sin 269
These vectors appear in Figure 4 right in red/bold, together with the three cube roots
of z for comparison.
On the one hand, the expression z 1/3 w1/3 can represent any of the vectors formed
as a vector difference of one of the three vectors representing z 1/3 with one of the vectors representing w1/3 or, equivalently, any vector connecting a w1/3 vector with a z 1/3
one. Of the nine such arrows, exactly three of them are horizontal, shown in Figure 5
left. These vectors represent three real numbers, two positive and one negative, consistent with the graphical information we found in Figure 3. In fact, the bottom-most real
vector, formed by connecting the two cube roots located exactly 1 on either side of
the negative y-axis, is the base of an isosceles triangle with equal sides length 1, thus
the base length is 2 sin 1 by (1).
On the other hand, the expression 3 z 3 w, which equals twice our original erroneous TartagliaCardano solution x from (11), represents exactly one vector: the difference between the two principal cube roots. As these two roots are both
located in
the firstquadrant, it follows that 2x is the small vector connecting the tip of 3 w to the
tip of 3 z, illustrated in Figure 5 right. Since each cube root of z (or w) is a rotated
copy of the principal cube root, we need only rotate 2x clockwise 120 to make it
horizontal, and hence match 2 sin 1 exactly.
Using the geometry of complex multiplication, this means we need only multiply the incorrect x in (12) by cos(120 ) + i sin(120 ) to obtain the desired
value:
1
3 x
sin 1 = i
2
2
2
1+i 3
3
3
sin 3 + i cos 3 sin 3 + i cos 3 .
=
4
330
Using the explicit formula for sin 3 given in (4) and thePythagorean identity, we can
finally express the exact value of sin 1 . Replacing i by 1, we can then write sin 1
in terms of radicals and integers as
( 5 1) 2 + 3 2 5 + 5 2 3
1 + 3
3
2
8
( 5 1) 2 + 3 2 5 + 5 2 3 64
+
(1 5) 2 + 3 + 2 5 + 5 2 3
3
2
( 5 1) 2 + 3 2 5 + 5 2 3 64.
+
(14)
Epilogue
The expression (14) is an eyeful, but the complex approach can suggest other radical
expressions for sin 1 . It is clear that
(cos 1 + i sin 1 ) (cos 1 i sin 1 ) = 2i sin 1 .
However, using the properties of complex roots, we can also conclude that
n
cos 1 + i sin 1 = cos n + i sin n
for any n with 1 n 90 where we are using the principal nth root. Combining these
two observations, it follows that
1
n
n
cos n + i sin n cos n i sin n
sin 1 =
2i
for any n with 1 n 90. This provides an unusual way to compute sin 1 in terms
of radicals of trigonometric functions at angles we may already know.
For example, n = 3 gives (14) in slightly rearranged form and n = 30 yields
30
30
1
3
+
1
3
,
sin 1 =
2
2
2 1
which is at least prettier to look at. Using n = 45 gives the similarly concise
1 45 1 + 1
1
45 1
.
sin 1 =
2 1
2
2
Perhaps the most exotic description for sin 1 comes from the extreme case n = 90,
for which the cosine term vanishes and the sine term becomes 1:
90
90
1 1
sin 1 =
,
2 1
which, in addition to being delightfully bizarre and strangely beautiful, has the added
bonus of being true.
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
331
Acknowledgment. The author would like to thank the professors and students of the South
Dakota School of Mines and Technology for both their comments and suggestions to improve
the paper and their patience for listening to him present it over and over again.
Summary. Ostensibly a derivation of an algebraically exact formula for the value of the sine
of 1 degree, we present this calculation as a historical romp looking at the problem through
the tools of geometry, then algebra, and finally complex analysis. Each one of these approaches
gets the reader nearer to the correct value, but also serves to frame a vignette of surprising or
beautiful mathematics.
References
1. C. B. Boyer, A History of Mathematics. Second ed. Wiley, New York, 1991.
2. M. Cantor, Vorlesungen u ber Geschichte der Mathematik. Vol. 1. Teubner, Leipzig, Germany, 1907.
3. J. W. Brown, R. V. Churchill, Complex Variables and Applications. Eighth ed. McGrawHill, New York,
2009.
4. Dirghatamas, Rig Veda, 1.164.48.
5. W. Dunham, Journey through Genius: The Great Theorems of Mathematics. Penguin, New York, 1990.
6. Euclid, Elements, Book IV, Proposition 10.
7. R. L. Francis, A note on angle construction, College Math. J. 9 (1978) 7580, http://dx.doi.org/10.
2307/3026605.
8. W. Hopkins, Elements of Trigonometry. Baldwin and Craddock, London, 1833.
9. G. G. Joseph, Crest of the Peacock: Non-European Roots of Mathematics. Third ed. Princeton Univ. Press,
Princeton, 2011.
10. B Mazur, Imagining Numbers: (particularly the square root of minus fifteen). Picador, New York, 2003.
11. T. Needham, Visual Complex Analysis. Oxford Univ. Press, Oxford, 1997.
12. R. B. Nelsen, Proofs Without Words II: More Exercises in Visual Thinking. Mathematical Association of
America, Washington, DC, 2000.
13. F. Swetz, J. Fauvel, O. Bekken, B. Johansson, V. Katz, Learn From the Masters. Mathematical Association
of America, Washington, DC, 1995.
14. P. Wanzel, Recherches sur les moyens de reconnatre si un probl`eme de Geometrie peut se resoudre avec la
r`egle et le compas, J. Math. Pures Appl. 2 (1837) 366372.
332
Form
Sarah Blake (sarah@sarahblakepoetry.com), Philadelphia, PA
Im well versed in theories. My favorite is that one infinity can be smaller than another
(e.g., there are fewer natural numbers than there are real numbers just between zero
and one).
I have learned patterns of thought. I catch myself picking out all the contradictions and
contrapositives before holding a thing in my hands.
For sixteen weeks I learned how to say that one knot is not the same as another. Similarly, a list of disparities can define me. And the power of discord has me sleepless.
At my best, Ive learned that when I close my mouth Im topologically equivalent to
the Great Pyramid.
When I open it, I resemble a fish.
This poem first appeared in Sentence: A Journal of Prose Poetics. Sarah Blake has
an undergraduate mathematics degree from The College of New Jersey and was a
National Endowment for the Arts 2013 Literature Fellow. Her Mr. West, an unauthorized lyric biography of Kanye West, was published by Wesleyan University Press in
2015, and an e-chapbook Named After Death is recently available from Banango
Editions with an illustrated companion workbook.
http://dx.doi.org/10.4169/college.math.j.47.5.333
333
A function is said to be convex if the interior of any line segment drawn between points
on its graph lies above the graph itself. (In calculus, these functions are often referred to
as concave up.) Convex functions have several well-known properties: Any tangent
line to a convex function lies below its graph; if a convex function is differentiable,
then its derivative is increasing; if a convex function is twice differentiable, then its
second derivative is nonnegative, and vice versa [1].
Maple has a feature that allows one to gather up frames of an animation. An
animation of tangent lines sliding along the graph of f (x) = x 2 produces an image
like Figure 1.
The absence of tangent lines above the graph is clear enough, but this illustration
raises another question: Is there a tangent line through every point below the graph?
We investigate this question first.
334
4
3
y 2
1
1
2
Figures 1 and 2 (and the difference between them) are the primary motivation for
this investigation. One can imagine the (very powerful and narrowly focused) headlights of a car driving along a curved road. The figures show us places that no light
would reach. There appears to be a region in Figure 2 below the graph through which
no tangent line passes. We begin by confirming this impression.
Lemma 1. If b |a|, then there is no tangent line to f that passes through the
point (a, b).
Proof. Observe that the graph of f lies entirely above the graph y = |x|. If (x, y)
is on the graph of f and b |a|, then line through (x, y) and (a, b) has slope
y
1
x 2 + 1.
335
b y |a| + |x| |a| + |x|
|a| + |x|
>
a x a x = |a x| |a| + |x| = 1.
But 1 < f (x) < 1 for all x, so the line through (x, y) and (a, b) cannot be a tangent
line.
The following lemma helps focus several of our results.
Lemma 2. Suppose f is continuously differentiable. If there is a tangent line to f
passing above the point (a, b) and a tangent line passing below (a, b), then there is a
tangent line passing through (a, b).
Proof. Since f is continuously differentiable, the tangent linethought of as a
function of two variables, T (c, x) = f (c) + f (c)(x c)is continuous in each variable separately. If T (c, a) (as a function of c) is less than b at some point and greater
than b at some other, it must be equal to b somewhere in between by the intermediate
value theorem.
We observe that, as long as (a, b) is not on the graph of f , one of the two conditions
on tangent lines happens automatically. This helps us prove the following theorem.
Theorem 1. Let f be a continuously differentiable convex function. Every point
below the graph of f lies on a tangent line to f if and only if | f (x)| is unbounded.
Proof. Suppose that | f (x)| is unbounded. Note again that any point (a, b) that
is below the graph of f also lies below a tangent line to f , in particular the one at
(a, f (a)). Since the derivative of f is increasing and unbounded, it must either be the
case that limx f (x) = or limx f (x) = . We will assume the former;
the proof for the other case is similar.
Consider a point (a, b) with b < f (a). Let L be a line through (a, b) and consider
a point on the graph (x, f (x)) where x > a. Suppose L intersects the graph just once.
It might be tangent to the graph, in which case there is a tangent line passing through
(a, b), and we are done. If L is not tangent to the graph, then it must lie above the
graph for any point to the right of x (since L intersects the graph just once). If there
is a tangent line whose slope is greater than the slope of L, then that tangent line will
overtake L as x and L will then be below a tangent line, consequently below
the graph, and so must intersect the graph twice. Thus the slope of L must be greater
than the slope of any tangent line. But this cannot be the case because f (x) .
Suppose instead that L intersects the graph twice and consider the rightmost of
the two points of intersection, say (c, f (c)). The slope of the line through (a, b)
and (c, f (c)) is ( f (c) b)/(c a). Since f (x) , there is a point p > c with
f ( p) > ( f (c) b)/(c a). The tangent line to the graph through ( p, f ( p)) passes
below (a, b). A similar argument holds if x < a. This is illustrated in Figure 3.
Now suppose that f (x) is bounded. Then the graph of f approaches asymptotes
as x . No point situated below both asymptotes can be on a tangent line to f .
This can be seen in Figure 2, and the proof is that of Lemma 1.
We make two remarks on the proof. It is a property of convex functions that the
slope of any tangent line is greater than the slope of any secant line whose two points
of intersection with the graph lie to the left of the point of tangency. Consequently, the
point (c, f (c)) would serve the purpose of ( p, f ( p)) in the proof [1]. Also, note that
it is not necessary for f (x) to be unbounded in both directions. The result holds, for
instance, for f (x) = e x , where the derivative is bounded as x and unbounded
as x .
336
( p, f ( p))
(c, f (c))
(a, b)
Figure 4 is the Maple image of the function f (x) = ex sin(x) (which is not convex
and is difficult to see in the picture) and a collection of its tangent lines. Though this
image is more complex than the others, one thing seems clear: There appear to be two
regions in the plane through which no tangent line passes. In this section we examine
the structure of this excluded region (abbreviated XR).
This requires additional definitions. A subset S of the plane is convex if, whenever
P1 and P2 are points in S, the entire line segment connecting them is also contained in
S. (Notice that the region above the graph of a convex function is a convex set, but not
every convex set arises in this way. The interior of a circle, for instance, is convex.)
A subset S of the plane is path connected if, whenever P1 and P2 are points in S,
there is a path in S from P1 to P2 (that is, a continuous function p : [0, 1] S with
4
3
2
1
8
1
2
3
4
337
p(0) = P1 and p(1) = P2 ). We will say that a set S is a component of a set T with a
certain property if S has the property, S T , and no subset of T larger than S has that
property.
Theorem 2. If f is a continuously differentiable function and R is a path connected
component of XR, then R is convex.
Proof. Suppose that P1 and P2 are points in R and that p is a path connecting them
such as shown in Figure 5. Suppose further that the line segment connecting P1 and
P2 is not entirely contained in R. Then there is a point on the line segment through
which a tangent line passes (the intersection of the two lines in Figure 5). None of the
points on this tangent line are in XR and the tangent line cannot pass through either
P1 or P2 , since they are in XR. Because p is continuous, this tangent line and p must
intersect. But the point of intersection would both be in XR (since it is on p) and not
in XR (since it is on the tangent line).
P1
p(t)
P2
We observe another property common to all the excluded regions we have seen,
although it is perhaps less obvious.
Theorem 3. If f is a continuously differentiable function and R consists of a component of XR that is unbounded above, then R is also unbounded both to the right and
to the left.
That is, the collection of x-coordinates of the elements of a component of XR is
unbounded both above and below. The result also holds for a component of XR that is
unbounded below.
Proof. Suppose a component of XR is unbounded above yet bounded on the right.
This means there is a vertical line consisting entirely of points through which tangent
lines pass, and to the right of any point P XR, as shown in Figure 6. By Lemma 2,
these tangent lines must pass below P. As the points on the vertical line go to , the
slopes of these tangent lines must also go to . Thus f (x) would be unbounded on
some interval around the vertical line. This cannot be the case, since f is continuously
differentiable.
We can say more about the the possibilities for XR.
Theorem 4. If f is a continuously differentiable function, then its excluded region
consists of at most two convex sets, one of which opens upward and one downward.
338
Proof. We continue with the notation of the previous proof. If P is above the graph
of f and P XR, then every point above P is also in XR (again, by Lemma 2, no
tangent line can pass above P). Likewise, if P XR is below the graph, then every
point below P is also in XR. Thus XR might consist of a component that is unbounded
above or one that is unbounded below, or possibly one of each.
There cannot be a single component of XR that is unbounded both above and below,
else there would be no tangent lines at all. Nor can there be two distinct upwardpointing components of XR, as the union of two such sets would either be convex
(and thus one of the components itself) or not convex (and so not a component of an
XR); likewise two distinct downward-pointing components.
Further questions
We conclude with some possible extensions of these ideas.
What about functions that are not continuously differentiable? Many excluded
regions seem to have boundaries that consist of lines connected by corners. Perhaps
the notion of tangent line could be altered to accommodate such curves.
The upward-pointing and downward-pointing components of XR are clearly not
independent, at least in their positions. Can we describe the relationship between the
two components?
Finally, the inverse question: Given a suitable region, can we construct a function
for which the region is its XR? Such a function is almost certainly not unique, but to
what extent can such functions be characterized?
Summary. We consider the set consisting of the union of all tangent lines to continuously
differentiable functions. In particular, the complement of this set has a predictable structure.
Reference
1. J. van Tiel, Convex Analysis: An Introductory Text. Wiley, New York, 1984.
339
340
Explorations
In the Fold Four Boxes activity, students turn crease patterns such as those shown in
miniature in Figure 2 into boxes such as those shown in Figure 1. This builds the
necessary intuition to answer the following questions, which are also posed in a more
guided way in the subsequent activity sheets.
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
341
Question 1. How do you construct a crease pattern for an n-gonal-based twist box?
There are a number of aspects of the crease pattern students need to analyze here.
They should think about which parts of the crease pattern correspond to the sides of the
box and which to the bottom of the box, how many rectangles they need in the crease
pattern, which rectangles in the crease pattern could be changed in size, and how the
crease pattern forms angles that correspond to interior angles of a regular polygon. As
a follow-up, students could use GeoGebra to make crease patterns for several other
values of n (say n = 710, although n = 8 will be the next easiest to fold).
The first row of rectanglesthe one with the diagonal creasesforms the polygonal base of the box. The other two rows of rectangles form the sides of the box.
Because the formation of the box requires overlapping the two rectangles on the ends,
the crease pattern must have n + 1 rectangles per row. The third-row rectangles are
folded behind the second-row rectangles; therefore the third-row rectangles should be
a bit shorter than the second-row rectangles. The diagonal-crease folds result in
being half of the interior angle of the regular n-gon base, so = (/2)((n 2)/n).
A bonus observation is that because the diagonal creases each bisect the interior angle
of the polygon, the diagonal creases pass through the center of the polygon.
Question 2. While constructing the four boxes, sometimes the corners of the twist
line up with the corners of the box, and sometimes they do not. How does this relate to
n? If the corners of the twist do not line up with the corners of the box, where do they
land?
Students may want to look at the insides of two different boxes where one lines
up nicely and another does not, as in Figure 3, and unfold these boxes to see what
is happening in the crease pattern. They should see that for n even, n-gonal-based
twist boxes have clean bottoms, while for n odd, these boxes have messy bottoms.
In a messy box, it appears that the corners of the twist land halfway between the base
vertices; it is as though a copy of the entire base has rotated by /n.
Examining the folding process on the diagonally creased rectangles reveals that
each emanates inward from an edge of the base. (Figure 5(a) shows this.)
When n is even, the corners of the rectangle then align with the corners of the base
polygon, and when n is odd, the corners of the rectangle do not align. In fact, when n
is even a regular n-gon is invariant under rotation by ; when n is odd, a regular n-gon
has no such invariance.
You might now wonder how the height of the rectangle was determined so that it
would be exactly the diameter of the base polygon in the even-n case. Try using a bit
of plane geometry (similar triangles) and perhaps some trigonometry to figure it out,
and then consult [1] for the answer.
342
Question 3. The rectangles in the crease pattern for which the lengths of the sides
have strict limitations are the rectangles with the diagonal creases. A picture of one
such rectangle is shown in Figure 4 below. How does h relate to s and ? How does h
relate to s and n?
Students should have noted from exploring Question 1 that s corresponds to a sidelength of the regular polygon base and the labeled angle is half of the interior angle
of the base n-gon, or = (n 2)/(2n). This question is not challenging (it mainly
requires simple trigonometry) but answering it is important for the next question.
Question 4. What happens to h in Figure 4 as n increases?
There are several ways to think about this. The easiest way is to set s = 1 and
see what happens as one varies h. Students should quickly see that this limit goes to
infinity, so as n increases, larger and larger sheets of paper would be needed to make
boxes with fixed s. A more origami-based approach is to hold the size of the paper
constant and see what happens to h as n increases. This is roughly equivalent to setting
each boxs perimeter to 1 instead of holding s constant; the limits are the same, and
because the calculations are slightly simpler for box perimeter 1, we take this approach
in the activity sheets. The behavior in the limit is now much more interesting! This is
a good place to make some numerical calculations and come up with a conjecture.
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
343
Calculating the limit using lHopitals rule is much easier after solving for h in terms
of instead of considering h in terms of n. In this case,
lim h = lim
1
( 2) sin()
2 sin() + ( 2) cos()
= lim
= .
2
2 cos()
2 sin()
(b)
s
d
a
Figure 5. One rectangle of the interior twist for a pentagonal-base twist box.
Using Figure 5, we see that we want to know how much longer d is than the altitude
of the n-gon that forms the base of the box. The altitude of the n-gon is r + a where
r is the radius of the circumscribed circle and a is the altitude of one of the isoceles triangles composing the n-gon. This introduces two new variables, so we need to
describe them in terms of our already existing variables s, , d, h. We can accomplish
this goal in many ways, including by using the Pythagorean theorem, trigonometry, or
plane geometry.
For simplicity here, we will use similar right triangles. The larger triangle is the
basic triangle shown in Figure 4 and outlined in Figure 5(b), and the smaller triangle is one-half of one of the isoceles triangles composing the n-gon. Comparing
the hypotenuse and the side adjacent to the angle , we have d/s = r/(s/2), so
d = 2r . Thus, our desired length x = d (r + a) = r a. We may then observe
that sin() = a/r , so a = r sin() and x = r r sin() = r (1 sin()).
As n increases, the distance between each side of an n-gon and its circumscribed
circle becomes smaller, so we would expect x to decrease as n increases. Specifically,
n2
= 0.
= 1 sin
lim x = lim (1 sin()) = 1 lim sin
n
n
n
2
n
2
Ah! The troublesome interior twist will become less troublesome for larger values
of n. Indeed, when n becomes large enough, one can simply do a bit of squishing of
the box bottom instead of tucking the excess paper away.
344
Conclusion
In our experience, students have been excited to see the interesting (and unexpected)
connection between accessible mathematics and origami. In addition, there are a number of extensions to this material that students could explore to deepen this experience.
For example:
References
1. s.-m. belcastro, T. Veenstra, Constructing regular n-gonal twist boxes, in Origami4 . Ed. R. J. Lang. A K Peters,
Natick, MA, 2009. 4149.
2. S. Fujimoto, M. Nishiwaki, Invitation to Creative Origami Playing (in Japanese). Asahi Culture Center, Tokyo,
1982.
3. T. Fuse, Boxes in One Piece (in Japanese). Chikuma Shobo, Tokyo, 1992.
4. , Origami Gift Boxes (in Japanese). Chikuma Shobo, Tokyo, 2000.
5. T. C. Hull, Project Origami. A K Peters, Natick, MA, 2006.
345
s
A = s2
2R
= Rs
2s
R
2R
2R
Theorem. The lateral surface area A of a frustum of a right circular cone with
base radii r and R and slant height s is A = (r + R)s.
Proof.
s
2r
R
2R
2r
2R
2R
2r
Summary. We present visual proofs for the lateral surface area of a frustum of a right
circular cone by relating the unwrapped surface to a rectangle.
http://dx.doi.org/10.4169/college.math.j.47.5.346
MSC: 97G30
346
As the vibrantly colorful fall season slowly fades to the more austere and grim winter
season, there is fortunately a third season that warms the hearts of manythe college
football bowl season. Every year, numerous college football programs have their seasons extended by one game, as they meet on the gridiron in such storied contests as
the Raycom Media Camellia Bowl and the Famous Idaho Potato Bowl (along with the
Rose Bowl, Orange Bowl, and a host of others). Besides giving teams one final chance
for pigskin glory, these bowl games also afford the opportunity for a different type of
season for those not blessed with the ability to actually play the game: the college bowl
betting season.
For the past several years, I have run a type of pool known as a confidence pool (for
intellectual purposes only, of course). It works as follows: For each game in the pool,
you not only pick a winner but also assign a confidence value to your pick. Each game
must have a unique confidence value in the range 1, . . . , n where n is the total number
of games (in the 201516 season, n = 40). A high value indicates strong confidence
in your pick while a low value indicates weak confidence. The game where you are
sure that the winner is an absolute lock will be assigned the confidence value n, the
next most confident n 1, and so on, until you reach the so-called pick-em games
to which you assign low confidence values like 2 and 1. For each game in which you
correctly pick the winner, you get points equal to the confidence value for that game,
and the player who amasses the most of these points is declared the winner. Ideally,
you could score as many as n(n + 1)/2 points, though typical scores are much lower
given the propensity for players to pick with their heart instead of their head (e.g., I
always pick Purdue no matter what) and the almost 100% chance for upsets (do not
get me started on this one).
CoPoW
An interesting situation arises from this type of pool: Suppose at half-way through
the bowl season that you have won a few and lost a few of your confidence bets. Do
http://dx.doi.org/10.4169/college.math.j.47.5.347
MSC: 03D99
347
you still have a shot at winning the whole thing, or have you been mathematically
eliminated from the winners circle? Note that even if you have a big lead, you are
not automatically assured a chance to winyou may have already won most of your
high-confidence picks, while those behind you may not have. Table 1 shows one such
scenario. You are in a pool with four other losers . . . I mean, players, and are currently winning with 50 points with two games left to play. You have only your 4 and
5 confidence valued games left, while each of the others have their 10 and 11 games.
Note that no matter what the outcome of the two games, you will always come in
second.
Table 1. Confidence pool scenario with five players and two remaining games.
player
current
points
you
Red
Ned
Ted
Ed
50
40
35
35
30
game 1
One State Two State
4
10
10
game 2
Red State Blue State
5
11
11
10
10
11
11
Throughout the rest of this paper we will use the convention that if a confidence
value c is positive, then it represents a confidence value c assigned to the home team
for the game, while if c is negative, them it represents a value c assigned to the
visiting team. For example, if we assume in Table 1 that One State and Red State
are the two home teams, then Neds confidence picks would be 10 and 11, while
Eds would be 10 and 11. Given this convention, we can now state the problem as
follows.
Problem (Confidence pool winning, CoPoW). Given a set of p players with
current scores s1 , s2 , . . . , s p , a set of n remaining games, and a set of confidence picks
ci, j , the confidence value of player i on game j for 1 i p and 1 j n, determine whether it is possible for Player 1 to win the pool.
Note that CoPoW (pronounced ka-POW!) is a more general form of the problem stated
in the introduction. Here the current scores can be any value (including negative) and
there is no constraint on the values of the confidence picks (they could all be the same,
some could be zero, etc.). Later, we will return to the original version where the scores
are nonnegative and the confidence picks are positive and unique.
How hard is it to solve CoPoW? For small numbers of games and/or contestants it
is relatively simple, but what if we were in a large pool with thousands of contestants
betting on hundreds of games (which given recent trends may not be that far fetched)?
We will show that CoPoW falls into a category of problems known as NP-complete
problems. This means that, with very high probability, there is no easy or quick way
to solve this problem. Algorithms can be constructed to solve the problem (you can
probably think of a few given a little time), but when the parameters start getting large
(thousands of contestants, hundreds of games) the time to solve the problem becomes
so large that the algorithm is essentially useless.
348
349
CoPoW is NP-complete
To show that CoPoW is NP-complete, we follow the steps outlined above: Show that
CoPoW is in NP and then find some existing NP-complete problem which can be
reduced to CoPoW in polynomial time. One of the most commonly used NP-complete
problems for this task is the 3cnf satisfiability problem (3-SAT for short): Given a
Boolean expression in 3 conjunctive normal form (3cnf), determine if there is a truth
assignment to the Boolean variables which makes the entire formula true. An expression is in 3cnf if it consists of a conjunction of clauses where each clause is a disjunction of three literals (where a literal is either a Boolean variable or its negation). For
example, the formula
(x1 x2 x3 ) (x1 x1 x2 ) (x2 x3 x4 )
is in 3cnf and is satisfiable with the assignment x1 = x2 = x3 = x4 = TRUE (among
many other satisfiable assignments). The formula
(x1 x1 x1 ) (x1 x1 x1 )
is also in 3cnf but is not satisfiable. 3-SAT was one of the earliest problems shown to
be NP-complete [3] and we now use it to show that CoPoW is NP-complete.
Theorem. CoPoW is NP-complete.
Proof. Clearly CoPoW is in NP: Given a solution to CoPoW, we can verify it in
polynomial time by simply taking the n outcomes, determining every players score,
and seeing if Player 1 is the winner.
The work is reducing 3-SAT to CoPoW by a polynomial reduction. Assume the 3SAT expression e we wish to satisfy consists of m clauses. We construct an instance
of CoPoW which consists of 3m games, one for each literal in e. We call each group
of games corresponding to three literals in one clause a triad. Next we construct three
types of playersPlayer 1, clause players, and consistency playersdefined below by
their current score and their set of 3m confidence picks.
(1) Player 1 is the subject of our problem: Is she mathematically eliminated from
the pool? We define her by
current score 3m + 5 and
for each game corresponding to a nonnegated variable we assign a confidence
value 1 (i.e., betting on the home team) and for each negated variable we assign a
value 1 (a bet on the away team).
We note two things concerning Player 1: Her maximum possible score is 6m + 5.
Also, if she gets at least one game correct in each triad, then she scores at least 4m + 5
points. We will call each of these correct games an anchor game for the triad.
350
It may seem that when this latter case is true, we can convert anchor games to a satisfying assignment of truth values for e as follows: If the anchor game is a home game,
then we assign TRUE to the corresponding variable; otherwise we assign FALSE to
the variable. In each case, the assignment would make the corresponding clause TRUE
and e would be satisfied. However, there is one problem: It is possible that one anchor
game may assign TRUE to some variable xi while another may assign FALSE to the
same variable. Our object then is to construct the remaining players so that Player 1
wins if and only if
she has at least one anchor game in each triad, and
the assignment of truth values to the variables is consistent.
We enforce these two conditions with the remaining two types of players.
(2) Clause players ensure that Player 1 wins if and only if there is at least one anchor
game in each triad. There will be a total of m clause players, one for each clause in e.
We construct clause player i by
current score 0 and
for the games in the ith triad we assign a confidence value (2m + 2) if the
corresponding variable in e is nonnegated and assign a confidence value 2m + 2 if the
corresponding variable is negated. (Note that for this triad, clause player i is betting
on the opposite teams from Player 1.) All other games have confidence value 0.
If there is no anchor game in the ith triad, then clause player i will amass 6m + 6
points and will beat Player 1 no matter the outcome of the other games. However, if
there is at least one anchor game in each triad, Player 1 scores at least 4m + 5 while
the most any clause player can score is 4m + 4, so in this case Player 1 will beat all
the clause players.
(3) Consistency players come in pairs, for each pair of occurances of the same
variable in e, whether negated or not. They have
current score 0 and
for the two games corresponding to the pair of variables, set the confidence values for the first consistency player to 3m + 3 and (3m + 3) and confidence values
(3m + 3) and 3m + 3 for the second consistency player. (Note that these values are
independent to whether the variable is or is not negated.) All other games have confidence value 0.
Assume that Player 1 has two anchor games associated with the same variable, one
a home win and the other an away win. This would result in assigning both TRUE and
FALSE to the same variable. In this case, one of the two consistency players associated
with this variable pair will amass 6m + 6 points and will beat Player 1, no matter the
outcome of the other games. However, if the same two anchor games are both home
or both away wins, then each consistency player will only score 3m + 3 points and
Player 1 will beat each of them as long as there is one anchor in each triad.
Table 2 shows an example of this reduction applied to the logical expression
e = (x1 x2 x3 ) (x1 x1 x2 )
which has m = 2 clauses with p = 4 pairs of identical variables (from the three x1
or x1 terms taken pairwise and then two x2 or x2 terms). In this case there are a
total of six games and 11 players. In general, the number of games is 3m and the
number of players is 1 + m + 2 p where p is the number of pairs of identical variables
in the expression. The maximum value for p occurs when every literal in e is for the
same variable; in this case p = 3m(3m 1)/2. So the maximum number of players is
9m 2 2m + 1 and the maximum number of confidence picks is 27m 3 6m 2 + 3m,
i.e., O(m 3 ). This shows that the reduction of 3-SAT to CoPoW is polynomial.
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
351
current
points
Player 1
games
g1,3 g2,1
g2,2
g2,3
g1,1
g1,2
11
clause player 1
clause player 2
0
0
6
0
6
0
6
0
0
6
0
6
0
6
consistency player 1
consistency player 2
consistency player 3
consistency player 4
consistency player 5
consistency player 6
consistency player 7
consistency player 8
0
0
0
0
0
0
0
0
9
9
9
9
0
0
0
0
0
0
0
0
0
0
9
9
0
0
0
0
0
0
0
0
9
9
0
0
9
9
0
0
0
0
9
9
9
9
0
0
0
0
0
0
0
0
9
9
We now show that the 3-SAT clause e is satisfiable if and only if there is a possible
winning outcome for Player 1. We treat each implication in turn.
If e is satisfiable, then Player 1 can win: Let A be the assignment of truth values
that make e true. For any variable set to TRUE in A we select the home team to win
for any game associated with that variable; otherwise, we select the away team. In
each clause in e either one nonnegated variable must be set to TRUE or one negated
variable must be set to FALSE; in either case Player 1 will win the corresponding game
in the triad associated with that clause and will score at least 4m + 5 points, while the
most any clause player will be able to score is 4m + 4. Furthermore, since all games
associated with any variable are either all home wins or all away wins, the most any
consistency player can score is 3m + 3, so the selection of wins based on A gives a
winning scenario for Player 1.
If Player 1 can win, then e is satisfiable: Since Player 1 has a winning strategy, there
must be an anchor game in each triad, else one of the clause players would win. For
every anchor game which is a home win, assign TRUE to the corresponding variable;
for every anchor game which is an away win, assign FALSE. Furthermore, since Player
1 has beaten all of the consistency players, we know that no variable has been assigned
both TRUE and FALSE in this process. Since home game selections correspond to
nonnegated variables, setting those variables to TRUE makes the clause containing
that variable TRUE. Similarly, away game selections correspond to negated variables,
so setting them to FALSE makes the clauses containing those variables TRUE as well.
Since every triad has an anchor game, every clause in e is set to TRUE and we have a
satisfying assignment of truth values to a set of variables in e. Any variable not already
set can be randomly set TRUE or FALSE without affecting the satisfiability of e.
Therefore, since CoPoW is in NP and 3-SAT can be polynomially reduced to it,
CoPoW is NP-complete.
352
game
corresponding variable is
nonnegated
negated
3i 2
3i 1
3i
(20m 3 1)
20m 3
(20m 3 + 1)
20m 3 1
20m 3
20m 3 + 1
consistency player 1
consistency player 2
(30m 1)
30m 3 + 1
30m 3 + 1
(30m 3 1)
An example of a reduction using the scheme above is shown in Table 5. Note that
the confidence values will be unique and can always be considered to be taken from
an original set of 30m 3 + 1 games, most of which have been played already (and
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
353
current
points
Player 1
400
clause player 1
clause player 2
consistency player 1
consistency player 2
consistency player 3
consistency player 4
consistency player 5
consistency player 6
consistency player 7
consistency player 8
games
g1,1
g1,2
g1,3
g2,1
g2,2
g2,3
0
0
159
1
160
2
161
3
4
159
5
160
6
161
0
0
0
0
0
0
0
0
241
239
241
239
1
1
1
1
2
2
2
2
2
2
241
239
3
3
3
3
3
3
3
3
239
241
4
4
241
239
4
4
5
5
239
241
239
241
5
5
6
6
6
6
6
6
239
241
for which all of the clause and consistency players have bet incorrectly, since they
each have a current score of 0). The choices of 50m 3 , 20m 3 1, 30m 3 1, etc., are
relatively arbitrary: Any large number relative to the sum of the remaining confidence
values for Player 1 (which equals 3m(3m + 1)/2) will suffice. Using these values,
the results of the games involving confidence values in the range 1, . . . , 3m have a
very small impact on the final score for each player relative to the games involving
the much larger confidence values. Because of this, the clause and consistency players
will play the same role in the constrained version of CoPoW as they did in the previous
example. Namely, if there is no anchor game in some triad, there will be a clause
player scoring at least 60m 3 points, which will beat Player 1s top possible score of
50m 3 + 3m(3m + 1)/2. Likewise, a consistency player will score at least 60m 3 points
if there is a home win and an away win for games associated with the same variable
in e. Using these facts, the proof that 3-SAT has a satisfiable assignment if and only if
Player 1 has a winning scenario is almost identical to the proof for the unconstrained
version of CoPoW.
This paper highlights one of the most interesting aspects of mathematics; how it
has something to say about normal, everyday occurrences. I had run the confidence
pool for many years and this question had always been in the back of my mind, but
it was not until I had time to truly ponder it that I realized the actual complexity of
the problem. Will it have major repercussions on my approach to the next confidence
pool? Probably notI will keep on losing regardless. But it is always satisfying to
come across a little problem and solve it. That is a nice win.
Summary. This article address the question of whether or not it is possible to know if you are
mathematically eliminated from a type of betting pool known as a confidence pool. We show
that this problem falls in the category of NP-complete problems, meaning that there is almost
surely no quick method to determine the answer.
354
References
1. S. A. Cook, The complexity of theorem-proving procedures, in Proceedings of the Third Annual ACM Symposium on the Theory of Computing. Eds. M. A. Harrison, R. B. Banerji, J. D. Ullman. Association for Computing
Machinery, New York, 1971. 151158.
2. M. Garey, D. S. Johnson, Computers and Intractability: A Guide to the Theory of NP-Completeness. Freeman,
San Francisco, 1979.
3. R. M. Karp, Reducibility among combinatorial problems, in Complexity of Computer Computations. Eds.
R. E. Miller, J. W. Thatcher. Plenum, New York, 1972. 85103.
r
y
x
2R
Moreover, since s > r + 2R for acute triangles and s < r + 2R for obtuse
triangles, s = r + 2R characterizes right triangles [1].
Summary. We use a figure to relate the semiperimeter, inradius, and circumradius of a
right triangle.
Reference
1. W. J. Blundon, On certain polynomials associated with the triangle, Math. Mag. 36 (1963) 247248,
http://dx.doi.org/10.2307/2687913.
http://dx.doi.org/10.4169/college.math.j.47.5.355
MSC: 51N05
355
356
To set out the terminology, the Pythagorean theorem says that for any right-angled
triangle, in which the hypotenuse has a length of r and the other two sides lengths of
a and d, it must be that r 2 = a 2 + d 2 . We prove the theorem using two lemmas, stated
and proved in the next two sections.
D
B
d
E
R2
d
B
Figure 1.
Lemma (Isosceles). In the situation described, the sum of the areas of rectangles
R1 and R2 is equal to the area of the square on the third side, BC. In other words,
r 2 = 2dc.
Proof. Continuing with the triangle shown in Figure 1 left, drop a perpendicular
from A to the side BC and let E be the point of intersection.
Clearly E bisects BC, hence BE and EC both have length r/2. It is easy to see that
triangles AEB and CDB are similar. That is, their only (possible) difference is the size.
Hence, in terms of lengths, DB is to BC what BE is to AB. That is,
r/2
d
=
r
c
from which we conclude r 2 = 2dc.
Corollary. In the special case in which the isosceles triangle happens to be rightangled, the isosceles lemma implies the Pythagorean theorem.
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
357
R4
F f G
c
B f
R4
Figure 2.
Lemma (Right-angled). In the situation described, the sum of the areas of the
rectangles R3 and R4 is equal to the square on the side FB. That is, a 2 = b f + c f .
Proof. As shown in Figure 3, draw a line parallel to BG through the vertex F and
let H be the point of intersection with the line determined by A and B.
Since FH is parallel to GB and ABG is isosceles (by construction), AHF must be
isosceles as well. Hence, the length of HB is f . Now draw HG and let J be the point
where it intersects FB. Let n be the length of FJ and m the length of JB. Clearly, it
follows n is also the length of HJ and m is also the length of JG. Therefore, n + m = a.
358
F
n
f
m
G
c
Figure 3.
It is easy to see GFJ and BFA are similar triangles, hence n/m = c/b. Combining
this with n + m = a gives n = a m = a bn/c which is equivalent to
n=
ac
.
b+c
b
B
Figure 4.
359
This establishes the Pythagorean theorem. There are no surprises in the end, not
after 2,600 years. But it is hoped that the journey, especially the encounter en route
with the isosceles lemma, is of some intrinsic interest to both students and geometers.
Acknowledgment. While the present paper is an exercise in pure geometry, it is the outcome
of a long-standing interest of the author in using geometry to understand economics [2]. I owe
initial interest in this particular topic to a conversation with Arunava Sen. I am grateful to
Tito Cordella, Tapan Mitra, and S. Subramanian for reading through the paper and the proofs,
and for giving many valuable comments; to an anonymous referee of the journal for useful
suggestions; and to Bintao Wang for excellent technical support.
Summary. This paper provides a new, long proof of the Pythagorean theorem. The two lemmas used should be of some intrinsic interest, especially one on isosceles triangles.
References
1. K. Arrow, Social Choice and Individual Values. Wiley, New York, 1951.
2. K. Basu, A geometry of non-Walrasian general equilibrium theory, J. Macroecon. 14 (1992) 87103,
http://dx.doi.org/10.1016/0164-0704(92)90019-5.
3. C. Cavafy, Collected Poems. Trans. D. Mendelsohn. Knopf, New York, 2009.
4. T. L. Heath, The Thirteen Books of Euclids Elements. Dover, Mineola, NY, 1956.
5. J. Geanakoplos, Three brief proofs of Arrows impossibility theorem, Econ. Theory 26 (2005) 211215,
http://dx.doi.org/10.1007/s00199-004-0556-7.
6. J. A. Garfield, Pons asinorum, New Engl. J. Educ. 3 (1876) 161.
7. A. Sen, Collective Choice and Social Welfare. Holden-Day, San Francisco, 1970.
8. G. Willans, R. Searle, Down with Skool!: A Guide to School Life for Tiny Pupils and Their Parents. Parrish,
London, 1953.
360
While collecting various proofs of the Pythagorean theorem for presenting them in my
class [8], I discovered a beautiful proof credited to Leonardo da Vinci. It is based on
the diagram below; I leave the pleasure of reconstructing the simple proof from this
diagram to the reader.
D
E
E
C
F
B
F
B
Z
L
Z
M
M
N
Since I had decided to give correct references to as many results as possible (if only
to set an example), I started looking at the many presentations of Leonardo da Vincis
proof on the Internet to find out where he had published it. Although this is a very
well-known proof, none of the many sources were able to point me to a book with a
sound reference, let alone to one of Leonardo da Vincis original publications. For this
reason, I gave up and simply remarked, As a rule, one has to be very careful in such
situations, meaning that claims that cannot be verified often turn out to be wrong.
One can easily find dozens of books that credit the proof above to Leonardo da
Vinci, including Maor [10, pp. 104106] and Ostermann & Wanner [14, p. 26].
http://dx.doi.org/10.4169/college.math.j.47.5.361
MSC: 01A55
361
363
something like the telltale figure above. If Leonardo da Vinci did develop this proof,
then I hope that a scholar of his work will provide the details.
It is my belief, however, that the legend of Leonardo da Vincis proof was given
birth by Max Simons 1906 report, from whence it was copied in turn by Heath, Boon,
and Loomis, and then was spread throughout the mathematical literaturenowadays,
the claim that Johann Tobias Mayers proof was found by Leonardo da Vinci can be
found in dozens of books, hundreds of articles, and thousands of web pages.
Acknowledgments. I thank the editor for several suggestions and for providing me with many
details concerning Boons book.
Summary. We present evidence suggesting that the proof of the Pythagorean theorem widely
attributed to Leonardo da Vinci is actually due to J. T. Mayer in the late 18th century.
References
1. F. C. Boon, A Companion to Elementary School Mathematics. Longmans, London, 1924.
2. E. G. Forbes, Tobias Mayers claim for the longitude prize: A study in 18th century Anglo-German relations,
J. Navig. 28 (1975) 7790, http://dx.doi.org/10.1017/S0373463300025674.
3. T. L. Heath, The Thirteen Books of Euclids Elements. Three volumes. Univ. Press, Cambridge, 1908.
4. J. J. I. Hoffmann, Der Pythagorische Lehrsatz mit zwey und dreysig theils bekannten, theils neuen Beweisen.
Kupferberg, Mainz, Germany, 1819. Second ed., 1821.
5. , Die geometrischen Bucher der Elemente des Euclides; als Leitfaden zum Unterricht in der
Elementar-Geometrie, mit vielen Anmerkungen. Kupferberg, Mainz, Germany, 1829.
6. , Beitrage zur Elementar-Geometrie. Wailandt, Aschaffenburg, Germany, 1848.
7. F. C. Jetze, Dissertatio inauguralis philosophico-mathematica sistens theorematis Pythagorici demonstrationes plures. Grunert, HalleMagdeburg, Germany, 1752.
8. F. Lemmermeyer, Von Ahmes bis Cardano. Ein Loblied auf die binomischen Formeln. In progress.
9. E. S. Loomis, The Pythagorean Proposition. Mohler, Bera, OH, 1927. Second ed., Edwards, Ann Arbor, MI,
1940. Reissued, National Council of Teachers of Mathematics, Reston, VA, 1968. Available at
http://files.eric.ed.gov/fulltext/ED037335.pdf.
10. E. Maor, The Pythagorean Theorem: A 4000-Year History. Princeton Univ. Press, Princeton, 2007.
11. F. G. Mehler, Hauptsatze der Elementar-Mathematik zum Gebrauche an Gymnasien und Realschulen.
Reimer, Berlin, 1859.
12. J. W. Muller, Mathematische und historische Beytrage und Erganzungen zu Johann Joseph Ignaz Hoffmanns
neuester Schrift: der Pythagorische Lehrsatz mit zwey und dreyig Beweisen. Lechner, Nurnberg, Germany,
1819.
13. , Neue Beitrage zu der Parallelen-Theorie, den Beweisen des Pythagoraischen Lehrsatzes und den
Berechnungsarten der Pythagoraischen Zahlendreiecke. Jenisch and Stage, Augsburg and Leipzig, 1826.
14. A. Ostermann, G. Wanner, Geometry by Its History. Springer, Berlin, 2012.
364
CLASSROOM CAPSULES
EDITORS
Ricardo Alfaro
Lixing Han
Kenneth Schilling
University of MichiganFlint
ralfaro@umflint.edu
University of MichiganFlint
lxhan@umflint.edu
University of MichiganFlint
ksch@umflint.edu
Classroom Capsules are short (13 page) notes that contain new mathematical insights on a topic
from undergraduate mathematics, preferably something that can be directly introduced into a college
classroom as an effective teaching strategy or tool. Classroom Capsules should be prepared according
to the guidelines on the inside front cover and submitted through Editorial Manager.
and
cos x 1
= 0.
x0
x
lim
(1)
These typically arise as necessary ingredients in the most common proofs of the rules
for differentiating the basic trigonometric functions,
(sin x) = cos x
and
(2)
The usual proof of the first derivative in (2) using the limit definition of the derivative
[1] involves expanding sin(x + h) using an angle sum formula, factoring the resulting
expression, and applying both of the limits in (1). Students are usually required to
demonstrate their understanding of these special limits by computing related limits
such as
sin (3 sin (2 sin x))
.
x0
x
lim
Nevertheless, the importance of such limits is easily misplaced; their role is purely
auxiliary and indeed (1) is nothing more than the special case of (2) when x = 0.
Direct proofs of (1) take about a page or more and are usually banished to an appendix.
Moreover, the insight to be gained from (1)that x must be measured in radians for
(1) and (2) to be correct as statedis easily buried in the details.
The purpose of this note is to identify a particularly simple proof of (2) that does
not require the limits (1). As an added bonus, the argument makes clear why x must
be measured in radians.
To begin, we recall the definition of the two basic trigonometric functions: cos t and
sin t are, respectively, the x- and y-coordinates of the point on the unit circle with angle
t measured from the positive x-axis. Thus, the curve (t) = (cos t, sin t) traverses the
unit circle counterclockwise. Recall that the length of an arc on the unit circle is equal
to the angle it subtends (when measured in radians). Hence, when t is measured in
radians, (t) moves at unit speed.
http://dx.doi.org/10.4169/college.math.j.47.5.365
MSC: 97I40
365
t
O
Figure 1. The point P = (cos t, sin t). The position triangle OPA and velocity triangle
PCB are congruent when t is measured in radians.
Now fix t R. For simplicity, we may assume t [0, /2]. Let O = (0, 0),
P = (t), and A = (cos t, 0). Define points B and C so that PC is the velocity vector
Since PC is tangent to the circle at P, OPC = /2. Since AP and PB are parallel,
AOP = BPC and OPA = PCB. Hence, triangles OPA and PCB are similar.
If t is measured in radians, then |PC| = 1. Since |OP| = 1, triangles OPA and PCB
are actually congruent.
and
Combining this with the congruence of the position and velocity triangles, we have
and
Reference
1. J. Stewart, Calculus. Eighth ed. Cengage, Boston, 2016.
366
AB
AB
AB
AB
Figure 1. Two-event Venn diagrams showing the state spaces and corresponding probabilities.
p+
q 1 and r + s 1.
http://dx.doi.org/10.4169/college.math.j.47.5.367
MSC: 97K50
367
Moreover,
p+
q = 1 if and only if r = s,
and likewise,
r + s = 1 if and only if p = q.
Proof. For necessity, from the conditions r s = pq and p + q + r + s = 1, we have
the following chain of equivalent inequalities:
0 ( r s)2 = r + s 2 r s,
2 r s r + s,
2 r s 1 p q,
p + 2 pq + q 1,
( p + q)2 1,
p + q 1.
Notice that, when r = s, the first inequality becomes an equality, hence the last
inequality becomes anequality
as well.
The inequality for r + s follows similarly, with the analogous equalities resulting when p = q.
It would be interesting to apply the same reasoning and derive similar relations for
three (and more) independent events. Of special interest would be the extension of the
above formulas and inequalities to random variables.
Acknowledgment. The authors wish to thank the editor for help simplifying the proof of
Theorem 1 and for comments that improved the presentation of this paper.
Summary. In this short note, we present an equivalent statement for two independent events
and inequalities related to their probabilities.
368
Curtis Cooper
Charles N. Curtis
CMJ Problems
Department of Mathematics and
Computer Science
University of Central Missouri
Warrensburg, MO 64093
cmjproblems@maa.org
CMJ Solutions
Mathematics Department
Missouri Southern State University
3950 Newman Road
Joplin, MO 64801
cmjsolutions@maa.org
This section contains problems intended to challenge students and teachers of college mathematics.
We urge you to participate actively both by submitting solutions and by proposing problems that are
new and interesting. To promote variety, the editors welcome problem proposals that span the entire
undergraduate curriculum.
Proposed problems should be sent to Curtis Cooper, either by email (preferred) as a pdf, TEX, or
Word attachment or by mail to the address provided above. Whenever possible, a proposed problem
should be accompanied by a solution, appropriate references, and any other material that would be
helpful to the editors. Proposers should submit problems only if the proposed problem is not under
consideration by another journal.
Solutions to the problems in this issue should be sent to Chip Curtis, either by email as a pdf,
TEX, or Word attachment (preferred) or by mail to the address provided above, no later than February
15, 2017.
PROBLEMS
1086. Proposed by Michel Bataille, Rouen, France.
Let AB be a given line segment with length c > 0. Construct a right-angled triangle
whose legs a, b satisfy a 2/3 + b2/3 = c2/3 using a ruler and square set (a drafting tool
with a right angle, also known as a set square).
1087. Proposed by Ian Cavey (student), Boise State University, Boise, ID.
Let R(n) denote the regular n-gon with unit sides. Define Center(n) to be the set of
interior points of R(n) whose distance to each of the n vertices is greater than 1.
(a) Show that Center(n) is empty for n 6.
(b) Determine the area of Center(n) for n > 6.
1088. Proposed by Eugen J. Ionascu, Columbus State University, Columbus, GA.
Let f be a differentiable function such that, for some a, b satisfying 0 < a < b < 1,
we have
1
a
f (x) d x
f (x) d x
0
+ b
= 0.
a(1 a)
b(1 b)
Let K =
a
0 f (x) d x
a(1a)
1
0
ba
sup | f (x) + 2K |.
f (x) d x
2
x(0,1)
http://dx.doi.org/10.4169/college.math.j.47.5.369
369
1089. Proposed by Angel
Plaza, University of Las Palmas de Gran Canaria, Spain.
Prove that if x, y, z 1, then
1
1
1
1
1
3
1
+
+
+
+
1+x
1+y
1+z
1 + xy
1 + yz
1 + zx
1 + 3 x yz
1090. Proposed by Ovidiu Furdui and Alina Sntamarian, Technical University of ClujNapoca, Romania.
Find all nonconstant, differentiable functions f : R R that satisfy the functional
equation f (x + y) f (x y) = 2 f (x) f (y) for all x, y R.
SOLUTIONS
a m + bm
a m1 + bm1
n+1
a n+1 + bn+1
.
2
ap + bp
,
p1
a
+ b p1
M p (a, b) =
ap + bp
2
1/ p
.
Here, we will prove the stronger result that, if p and q are real numbers such that
, then
p 1 and q p+1
2
L q (a, b) M p (a, b).
Note that the original inequality follows from this result by letting p = n + 1, q = m,
.
and showing m n 2 implies p 1 and q p+1
2
For the proof, note that L q (a, b) is increasing in q. This can be proved by logarithmic differentiation:
a q log a + bq log b a q1 log a + bq1 log b
d
log L q (a, b) =
dq
a q + bq
a q1 + bq1
=
370
Assume without loss of generality that a b and let x = b/a. After dividing both
sides of the inequality by a, it becomes L p+1 (1, x) M p (1, x) with x 1. For x = 1,
2
the inequality becomes equality, so all we need to prove is that L p+1 (1, x) grows faster
2
than M p (1, x) as x 1 increases. To that end, we compute
p+1 p1
p1 p3
x 2
x 2
d
x p1
2
2
p3
2
Also solved by ROBERT AGNEW, Buffalo Grove, IL, and Palm Coast, FL; M ICHEL BATAILLE, Rouen, France;
B RIAN B RADIE, Christopher Newport U.; H ONGWEI C HEN, Christopher Newport U.; J OHN C HRISTOPHER,
California State U. Sacramento; E UGENE H ERMAN, Grinnell C.; PANAGIOTIS K RASOPOULOS, Athens, Greece;
E LIAS L AMPAKIS, Kiparissia, Greece; K EE -WAI L AU, Hong Kong, China; PAOLO P ERFETTI, U. Rome Tor
Vergata, Italy; and the proposer. One incomplete solution was received.
371
b6
c6
a6
+
+
>
4
3S.
(a)
(a 2 b2 )(a 2 c2 ) (b2 a 2 )(b2 c2 ) (c2 a 2 )(c2 b2 )
b6
c6
a6
+
+
>
4
3S.
(b)
(a b)2 (a c)2
(b a)2 (b c)2
(c a)2 (c b)2
This correction to Problem 1062 appeared in the March 2016 issue of The College
Mathematics Journal. Its solution will be published in the March 2017 issue.
3
c8
>
144
3r .
(a + b)(c a)2 (c b)2
1
a8
a4
or equivalently
cyc
a8
1
2
2
(b + c)(a b) (a c)
4s
cyc
a4
(a b)(a c)
2
.
cyc
a4
= a 2 + b2 + c2 + ab + ac + bc,
(a b)(a c)
(a 2 + b2 + c2 + ab + ac + bc)2
a8
.
Note that
a 2 + b2 + c2 + ab + ac + bc
372
2
8s 2
(a + b + c)2 =
.
3
3
Indeed,
3(a 2 + b2 + c2 + ab + ac + bc) 2(a + b + c)2 = a 2 + b2 + c2 ab bc ca
is nonnegative. The expression is zero if and only if a = b = c, but triangle ABC is
nonisosceles. Therefore,
cyc
8
a
>
(b + c)(a b)2 (a c)2
8s 2
3
4s
2
=
16s 3
.
9
16(3 3r )3
a8
16s 3
= 144 3r 3 .
>
2
2
(b + c)(a b) (a c)
9
9
n
k
1 22n
2n
< 1.
cos2n+1
2n + 1
n
k=1
This correction to Problem 1064 appeared in the March 2016 issue of The College
Mathematics Journal. Its solution will be published in the March 2017 issue.
n
n
1
ak bk
|ak |2 + ||2
|bk |2 .
(a) Re
2
k=0
k=0
k=0
n
n
(2n
+
1)(2n
+
3)
1
ak bk
|ak |2 +
|bk |2 .
(b) Re
2(n
+
1)
3
k=0
k=0
k=0
Solution by Eugene Herman, Grinnell College, Grinnell, IA.
We prove stronger versions of both (a) and (b):
n
n
n
1
ak bk
|ak |2 + ||2
|bk |2 .
(a )
2
k=0
k=0
k=0
n
n
n
1
(b )
ak bk
|ak |2 + L
|bk |2 , whenever 0 < K and K 2 L.
2K
k=0
k=0
k=0
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
373
Note that the coefficients in (b ) do not depend on n and that (n + 1)2 (2n+1)(2n+3)
3
for all real numbers n 0.
For (a ), let x =
a0 , a1 , . . . , an and y =
b0 , b1 , . . . , bn . Then, using the
CauchySchwarz inequality and 2ab a 2 + b2 ,
n
n
n
2
ak bk = 2|
x, y| 2x y x2 + y2 =
|ak |2 + ||2
|bk |2
k=0
k=0
k=0
n
n
1
1
(x2 + Ly2 ) =
|ak |2 + L
|bk |2
K
K k=0
k=0
Also solved by A DNAN A LI, Mumbai, India (part (b)); M ICHEL BATAILLE, Rouen, France; B ILL D UNN,
Montgomery C.; D MITRY F LEISCHMAN, Santa Monica, CA (part (b)); E LIAS L AMPAKIS, Kiparissia, Greece;
W ILLIAM S EAMAN, Bethlehem, PA; and the proposer.
Please note that we now have MAA email addresses for submitting problems and
solutions. Rather than using the editors university email addresses, please use
CMJproblems@maa.org and CMJsolutions@maa.org, respectively.
Editors Note. Wei-Kai Lai of the University of South Carolina Salkehatchie was omitted from the list of solvers of problem 1045 in the January 2016 issue.
374
BOOK REVIEW
My Search for Ramanujan: How I Learned to Count by Ken Ono and Amir D. Aczel,
Springer International Publishing, Cham, Switzerland, 2016, xvi + 238 pp., ISBN 9783-319-25566-8, $29.99 (hardback), 978-3-319-25568-2, $19.99 (electronic).
Reviewed by Brian Hopkins (bhopkins@saintpeters.edu), Saint Peters University,
Jersey City, NJ
While in graduate school, we organized a program for the Preparing Future Faculty initiative where mathematicians discussed their career paths. Fortunately, we had pizza
and beer to keep the graduate students engaged because all but one faculty account
came down to I was always good at math, so I thought I should become a math professor, and I did. The exception was a professor whose initial ambition to work assembly
at Boeing was foiled by a badly timed broken leg; he ended up in a calculus class for
premedical students that awakened a love of mathematics (a story to remember when
we teach service classes). Still, each account, once on the mathematical track, went
from success to success to professorship.
Such a program today at Emory University, where Ken Ono is on the faculty, would
have at least one very different and riveting account. Fortunately, that story has been
written by Ono and Amir Aczel for a wide audience. This assisted autobiography
illustrates in close detail the personal difficulties Ono went through before his current success. As the title suggests, Srinivasa Ramanujan plays an important role, not
just mathematically.
In fact, there is not much mathematical content in the book, although that restraint
gives way in the last 11 pages. The central story is Onos life through the year 2000,
but the interconnected accounts span four countries and more than 100 years. Ono and
Aczel have written a very engaging and readable book about the lives of several mathematicians focusing on themes of isolation, generosity, and finding oneself, portrayed
with many interesting connections and parallels. There are moving accounts of caring
and effective mentors along with some cautionary tales of debilitating discouragement.
Tiger cage
The young Ken Ono was good at math by almost any standard, but he did not meet
his immigrant parents expectations. His father, noted mathematician Takashi Ono,
escaped the isolation and deprivations of postwar Japan through hard work and the
generous support of Andre Weil, who attended a legendary 1955 TokyoNikko conference on algebraic number theory. Takashi and his wife Sachiko first came to Princeton and then settled near Baltimore, following the mathematicians appointments at
the Institute for Advanced Study and Johns Hopkins University. They were the only
Asian-American family in their Maryland neighborhood. Between the lingering American postwar animosity toward Japanese and their dream of returning to Japan, the Ono
family took on something of the kaikin policy isolation enforced during Japans Tokugawa period.
Borrowing the terminology popularized by Amy Chua, Ken and his two brothers
grew up in a tiger cage in stark contrast to the lives of their classmates. Each son had
a prestigious life planned out by his parents, and there was no room for activities that
http://dx.doi.org/10.4169/college.math.j.47.5.375
375
did not contribute to those goals. The book opens with a young Ken stealthily watching
Gilligans Island while his parents are away, ready with ice water and a washcloth
to cool the television set in case they return unexpectedly. Around the same time,
when he scored in the 98th percentile in mathematics and 97th percentile in reading
comprehension on a standardized test given to third grade students, his parents rebuke
informed a negative voice that would plague him for years. It spoke against his selfesteem in a pidgin English.
Ken-chan, one in fifty people did better in math. Even more better in English.
Thats fifty thousand kids in country. You arent even among best fifty thousand.
Harvard only accepts few thousand. . . We sacrifice everything for you three boys,
and this is how you thank us?
Kens designated path was to become a second generation prestigious mathematician (and excel at violin). His often distant father connected with Ken around
mathematics, such as telling stories about great mathematicians and playing a prized
recording of a David Hilbert lecture. There is a warm account of Takashi teaching a
grade-school Ken about Fermats little theorem in the guise of a trick.
Ken-chan, pick a number that isnt a multiple of 7. Id pick a number, such as
13. Now raise it to the sixth power and subtract 1. Id run around the corner of
his giant steel desk, to my little kiddie desk, and Id work it out by hand. After a
minute or so, Id arrive at the answer: 4,826,808. I bet thats a multiple of 7. Ill
take that bet, I thought. He has only a 1 in 7 chance of being right, and Ill take
those odds any day. And Id run back to my kiddie desk, and I would divide 7
into this large number, and I found that he was right: 4,826,808 = 7 689,544.
Id say something like, You got lucky. Lets do it again.
Later, Takashi would have Ken write programs on a brand new Apple II to compute
class numbers of quadratic forms for the Hopkins graduate students.
But even with Kens proficiency, the pressure became too much. He began to stand
up for himself, taking up cycling in seventh grade, racing in distant contests by high
school, quitting violin and the Peabody Preparatory Orchestra in tenth grade. Soon
376
after, he dropped out of high school and left home, a course of events surprisingly
related to his first knowledge of Ramanujan.
Takashi Ono was one of some 100 mathematicians who contributed to an effort,
coordinated by Richard Askey, to have a bust of Ramanujan designed, cast, and
installed in India. Janaki Ammal, Ramanujans widow, sent thank you notes to the
contributors. Kens father was deeply moved by the letter and told his son about
Ramanujan. Ken focused on one detail: Ramanujan had dropped out of college yet
went on to mathematical greatness. With that example, he eventually got his parents
permission to leave home without finishing high school.
377
Figure 2. Ken Ono and Basil Gordon in 1992 (photo by Keith Kendig).
From Basil Gordon I learned what it means to do mathematics. . . . Doing mathematics is a mental voyage in which clarity of thought and openness to insight
make it possible to see the deeper beauty of a mathematical structure, to enter
a world where triumph over a problem depends less on conscious effort than on
confidence, creativity, determination, and intellectual rigor.
Andrew Wiless work on Fermats last theorem came at an opportune time for Ono,
highlighting his small corner of mathematics that was an ingredient in the proof.
Ken Onos path continued to a visiting assistant professorship at the University of
Illinois and a two-year membership at the Institute for Advanced Study. Ken was able
to host his parents there, almost 40 years after their time living in Princeton. They had
a sweet reconciliation, finally silencing the negative voices that had long plagued the
son, initiating a warm adult relationship.
Multiple endings
The detailed chronology continues through Ken Onos time at Pennsylvania State
University, where he and George Andrews hosted a conference in honor of Basil Gordon. Ono also proved a remarkable result, the most exciting and significant work
on congruences for the ordinary partition function since the inaugural conjectures of
Ramanujan in 1919 [2], circling back to the mathematician of the books title. This
section ends with another meeting with Takashi Ono, on the occasion of Bill Clinton presenting the younger Ono a Presidential Early Career Award for outstanding
378
contributions to number theory and for unusual talent for fostering mathematical abilities in students at different levels [5]. The heartwarming details of the interaction
between father and son are left for the readers enjoyment.
Stylistically, that would have been a strong ending for the book, but there are four
more sections. One summarizes a motivational TEDxEmory talk that highlights some
the biographical details given above [4]. The last official chapter explains the spiritual
lives of both generations of Onos in contrast to the agnosticism of Kens childhood.
An epilogue of 17 pages and an afterword of 16 pages are longer the chapters (which
average just under six pages each) and cover a few highlights in contrast to the thorough biographical details through the year 2000. The epilogue covers three events. In
2005, Ono spoke at a conference in India and visited Ramanujans childhood home
and the Sarangapani Temple where his mother sang and worshippers etched special
numbers on the walls; perhaps one from Ramanujan himself? At the end of 2014, Ono
began an association with the film The Man Who Knew Infinity that quickly went from
assisting with props to being in charge of the mathematics, working with Jeremy Irons
and Dev Patel (playing Hardy and Ramanujan, respectively) on their dialogue. Finally,
there is an account of recent research extending various of Ramanujans results with
several collaborators.
The afterword includes an assessment of Ramanujan as a great anticipator of mathematical developments and shows a throughline from topics discussed at the 1955
conference attended by Takashi Ono to the proof of Fermats last theorem. And, as if
the authors could no longer hold back, the final pages are dedicated to mathematical
gems of Ramaujans mathematics, including the famous story of the number 1729
taxi, highly composite numbers, and his tau function.
379
[1], still his most popular book. Several more popular books on mathematical and
scientific topics followed, ranging from the ancient Khmer origins of zero to Nicolas
Bourbaki (along with updates of the statistics textbooks).
The book under review is a moving account of overcoming debilitating expectations. The pervasive self-doubt, given voice over a dozen times, can wear on the reader,
but perhaps that is intentional to help us feel its tenacious grip. The authors effectively
communicate the excitement and beauty of advanced mathematics without requiring
technical background. The collective hero of the interconnected stories is the generous mentor who recognizes talent and provides opportunities for growth and success.
We need more accounts of these important relationships: Hardy and Ramanujan are a
well-known example; the chapter here documenting Basil Gordons time as Ken Onos
thesis advisor is an especially warm addition.
Rarely are we given an inside view of the personal struggles of an accomplished
mathematician, and tributes to mentors are often reserved for banquet dinners at conferences tied to the honorees 65th, 70th, etc. birthday. Ono and Aczel have given a
highly readable treatment of these issues that will be appreciated by several readers:
graduate students searching for models for their future, faculty wanting motivation to
be effective mentors, and all readers inspired by a lively account of overcoming difficulties to achieve great success in the beautiful world of mathematics.
References
1. A. Aczel, Fermats Last Theorem: Unlocking the Secret of an Ancient Mathematical Problem. Reprint of the
1996 original. Basic, New York, 2007.
2. B. C. Berndt, Review of distribution of the partition function modulo m by K. Ono. Math. Rev. M1745012
(2000).
3. W. Grimes, Amir Aczel, 65; Wrote scientific cliffhanger, New York Times, December 9, 2015,
B15, http://www.nytimes.com/2015/12/08/science/amir-aczel-author-of-scientific-cliff
hanger-dies-at-65.html.
4. K. Ono. Live mathematically, but not by the numbers. TEDxEmory, 2015, http://tedxtalks.ted.com/
video/Live-Mathematically-But-Not-by.
5. President honors top junior faculty in science and engineering, National Science Foundation Office of Legislative and Public Affairs, April 2000, https://www.nsf.gov/od/lpa/news/press/00/pr0022.htm.
6. R. P. Schneider, Uncovering Ramanujans lost notebook: An oral history, Ramanujan J. 29 (2012) 324,
http://dx.doi.org/10.1007/s11139-012-9445-z.
380
FILM REVIEW
The Man Who Knew Infinity, directed by Matthew Brown, 1 hour 48 minutes, Edward
R. Pressman Film, 2015.
Reviewed by Jennifer Wilson (wilsonj@newschool.edu), Eugene Lang College, The
New School, New York, NY
As most readers of this J OURNAL are probably aware, the recent film The Man Who
Knew Infinity is a biopic about the life of Ramanujan and his collaboration with
G. H. Hardy. Arriving on the heels of 2014s banner year for math-related films,
including The Imitation Game and The Theory of Everything, The Man Who Knew
Infinity is based on the book of the same title by Robert Kanigel [5]. Prior depictions
of Ramanujan include Complicites theatrical production A Disappearing Number
and the 2014 Indian film Ramanujan directed by Gnana Rajasekaran and produced in
Ramanujans native Tamil and English. Number theorist and lost notebook discoverer George Andrews has recently written a review-cum-reminiscence of the current
film for the Notices of the AMS [1]; Alexandra Brancombe discusses it in the MAA
Focus [2].
Although Ramanujans history and accomplishments are familiar to mathematicians, he is less well-known to the general public. Screenwriter and director Matthew
Brown aims to bring his story to a wider audience. Filmed in Madras and Cambridge
(among other locations), the story unfolds amid colorful scenes rich with historical
detail. Opening images of Ramanujan covering the temple floor with chalked equations give way to scenes of him skirting the hallowed grass of the Trinity College
courtyard where the iconic figures of Bertrand Russell and John Littlewood populate
the common rooms. Unlike similar-feeling Merchant Ivory movies based on the novels
of E. M. Forster (a friend of Hardys), however, the films true subject is mathematics.
Previous theatrical stories of mathematicians (such as Proof and Good Will Hunting)
direct their attention to the characters. The Man Who Knew Infinity instead uses the
relationship between Ramanujan and Hardy as a lens through which to understand
why mathematicians are drawn to their work and to explore the nature of mathematical
knowledge and collaboration. In this, the filmmakers were assisted by mathematicians
Manjul Barghava and Ken Ono; the latters recent book My Search for Ramanujan
(coauthored with Amir Aczel) [6] is reviewed in this issue (p. 375). An interesting
and accessible discussion of Ramanujans work and the way the film does and does
not capture the process of mathematical creation can be found in James Propps blog
Mathematical Enchantments [7].
The film
The narrative focuses on the five-year period between 1914 and 1919 when Ramanujan
(played by Dev Patel) and Hardy (Jeremy Irons) worked together at Trinity College,
Cambridge. We learn little of Ramanujans background or education; we are introduced to him as a young clerk at the Madras Port Trust Office supporting his wife
Janaki (Devika Bhise) and mother Komalatammal (Arundathi Nag) and working on
mathematics at night. Encouraged by a sympathetic boss, he writes letters to several
prominent mathematicians asking for their opinion and assistance with publication.
http://dx.doi.org/10.4169/college.math.j.47.5.381
381
G. H. Hardy is the only one who takes him seriously. About the letter, Hardy famously
said,
I had never seen anything like [these results] before. . . they must be true because,
if they were not true, no one would have had the imagination to invent them. [3]
After consultation with Littlewood, Hardy arranges for Ramanujan to travel to
Cambridge where their collaboration begins.
Much of the films drama lies in the contrasting experience, motivations, and mathematical styles of Ramanujan and Hardy. Ramanujan is self-taught, a mathematician
with deep intuition but little formal training. Hardy, by contrast, was already wellknown for bringing rigor to analytic methods and is determined to teach or corral
Ramanujan into providing formal reasoning for his results. Despite warnings from
Littlewood and Russell (played by Toby Jones and Jeremy Northam, respectively) not
to rein in Ramanujans talents, Hardy insists Ramanujan complete his proofs before
publishing.
Its not that I cant see what youve claimed, its that Im not sure you know how
you got there. Or indeed that your claims are correct. There are subtleties that
. . . intuition is not enough; it has to be held accountable.
Figure 1. Jeremy Irons as Hardy (photo by Sam Pressman and Richard Blanshard).
Ramanujan initially resists, partly because he has so many results to publish, but
mostly because he is already convinced of their correctness. Eventually, Ramanujan
accedes; it is never clear whether he actually agrees. As their collaboration progresses,
we see Hardy and Ramanujan working together on his prime number theorem and later
applying Hardys circle method to studying the asymptotic behavior of the partition
function, Ramanujans insights sharpened and made rigorous through Hardys analytic
techniques.
382
Figure 2. Dev Patel (left) as Ramanujan and Kevin MacNally (right) as MacMahon (photo by
Sam Pressman and Richard Blanshard).
Toward the end of the film, Ramanujan falls ill with tuberculosis and decides to
complete his work with Hardy so that he can return to his family. The final events of
the story follow quickly: the completion of his partition theorem, his acceptance as
Fellow of the Royal Society (and subsequently at Trinity), and his leaving for home.
His death, one year later, is shown in flashbacks interspersed with scenes of Hardys
paying tribute to him at his Cambridge memorial.
Themes
Recurring motifs help to maintain the films tight focus. Mathematics is depicted
largely through the act of writing: Ramanujan using chalk on the floor of the temple
(reputedly because paper was scarce); cramming his notebooks with tersely worded
statements of his results; reveling in Cambridge at the large stack of blank paper in his
rooms; working with Hardy at a chalk board; hallucinating, when ill, that equations
are writhing on his arms.
A sense of history-making is elicited through repeated images of walking feet. The
actual Ramanujan was famously uncomfortable in Western clothes. In the film, we
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
383
Omissions
While hewing closely to the truth, the film necessarily leaves out many details of
Ramanujans life, including no more than a passing reference to his Brahmin status
and his abbreviated university career. Ramanujan left few reflections behind about his
thought processes and feelings regarding his time in Cambridge. So we are left with
the words of Hardy and his contemporaries to flesh out the details of their relationship. Several statements made by Hardy in the film come directly from C. P. Snows
introduction to Hardys A Mathematicians Apology [4]. Snow also describes Hardys
discomfort in social situations, his dedication to logic and formal reasoning, his love
of cricket, and his adamant atheism. Like the more vocal Bertrand Russell, Hardy was
against Britains participation in what we now know as World War I. His involvement
in college protests did not cause him to be forced out of Trinity, as did Russells, but
provides additional fodder in the film for the tension between him and his more conservative colleagues. Hardy refers ruefully to his own social diffidence and deafness
to the ordinary niceties of friendship, but the film does not touch upon Hardys homosexuality or any romantic leanings he might have harbored toward Ramanujan. (The
film does, however, open with Hardys statement, My association with him is the one
romantic incident of my life. [3])
The film leaves other things unspoken. Tellingly, the film does not make clear
exactly what prompts Hardy to respond to Ramanujans letter, aside from a disinterested appreciation of mathematical talent. After an evening with Littlewood spent
assessing the letters claims, Hardy says flatly much better let him rot away in his
office in Madras. Yet the next scene shows Ramanujan preparing to make the journey to England. C. P. Snow remarked that Hardy was comfortable with people of all
class and education and had little patience for elitism. His radicalism was impersonal,
but real, and in later years he was involved in the labor movement, briefly serving as
President of the Association of Scientific Workers.
The issue of race appears frequently if not always directly. Fellows at Trinity speak
dismissively of Ramanujan, and, in one war-time incident, he is beaten by soldiers and
left lying on the street. When Hardy asks him why he thinks his colleagues want him
to fail, Ramanujan replies it is because he is Indian. Hardy answers matter of factly
well there is that, but then goes on to suggest what was for him the more relevant
factor of professional jealousy.
The central tension in the film, however, remains Ramanujan and Hardys disagreement about the roles of intuition and proof in mathematics. In a climatic sense at the
hospital, Ramanujan tells Hardy that his ideas come from the goddess Namagiri who
causes them to appear fully formed in his mind. An equation has no meaning to me
unless it expresses the thought of God. Hardy, while revering Ramanujans talents,
384
cannot accept this. A letter from Littlewood to Hardy, sent during Littlewoods time
at the front, offers a more subtle look at this conundrum. In a few short sentences, he
notes that he has disproved Ramanujans prime number theorem but then states that
Ramanujan is a genius of the order of Euler or Jacobi whose inspiration and talents
must be fostered and treasured. In this way, the film hints at the more complex truths
lying behind this apparent dichotomy: that intuition is necessary but can be wrong,
that mathematical ideas arise in many different ways, and that intuition and rigor play
integral and interrelated roles in the act of mathematical creation.
Moving forward
The film, which was ten years in the making, was a labor of love for director Matthew
Brown and producer Edward Pressman. Inspired by the idea that, as Littlewood says,
great knowledge often comes from the humblest of origins, the filmmakers have
teamed with Expii, an online education platform, to create the Spirit of Ramanujan Talent Initiative (https://www.expii.com/ramanujan). Sponsored by the Templeton
Foundation, the mathematics competition is open to all students; this inaugural round
closes December 1. Finalists will receive support for summer research projects and for
attending the 2017 Joint Mathematics Meeting in Atlanta. Perhaps you will see them
there.
References
1. G. Andrews, The Man Who Knew Infinity: A report on the movie, Notices Amer. Math. Soc. 63 (2016)
178180, http://dx.doi.org/10.1090/noti1349.
2. A. Branscombe, Ramanujan goes to Hollywood, MAA Focus 36(3) (June/July 2016) 67.
3. G. H. Hardy, Ramanujan. Cambridge Univ. Press, London, 1940.
4. , A Mathematicians Apology. Cambridge Univ. Press, London, 1940.
5. R. Kanigel, The Man Who Knew Infinity. Washington Square, New York, 1991.
6. K. Ono, A. Aczel, My Search for Ramanujan: How I Learned to Count. Springer, New York, 2016.
7. J. Propp, The Man Who Knew Infinity: What the film will teach you (and what it wont),
Mathematical Enchantments, July 17, 2016, https://mathenchant.wordpress.com/2016/07/17/
the-man-who-knew-infinity-what-the-film-will-teach-you-and-what-it-wont/.
385
MEDIA HIGHLIGHTS
Warren Page, Section Editor
Tanya Leise
Philip Straffin
New Rochelle, NY
wxpny@aol.com
Amherst College
Amherst, MA
Longmont, CO
Media Highlights are short, approximately half-page, reviews intended to help CMJ readers monitor
a broad spectrum of publications, web materials, professional activities, and instructional resources.
Readers are encouraged to submit items that will be of interest to colleagues in the mathematical
community. Media Highlights should be sent to Warren Page at wxpny@aol.com.
386
`
calculated 205 places in 1844. But Jean-Etienne
Montucla reported in his 18th-century
Histoire des mathematiques that one Baron Zach saw in the 1780s a manuscript in
Oxford containing a calculation of to 154 places. Benjamin Wardhaugh was able,
using the now-digitalized manuscripts catalogue of the Bodleian Library, to track down
this manuscript, which turned out to be a handwritten book containing, in addition to
the calculation of , several hundred pages of tables of logarithms and sines calculated
to 30 decimal places. In the unsigned manuscript, the unknown author calculated the
154 digits in 1721, using not an efficient formula like Machins, but the much cruder
1
1
1
1
= 6 arctan
+
+ .
= 12 1
3 3 5 32
7 33
3
This involved calculating 12 to 154 decimal places and then calculating and adding
an amazing 314 terms of the series. After having carefully checked the work twice, the
author wrote out the final clean copy in 1759, in which 152 places were correct. Wardhaugh comments, Its a strange story and perhaps a melancholy one. The thought of
a provincial schoolmastersaybeavering away at mathematical tables and calculations for half a lifetime and never seeing them printed is a sobering one, even if it was
done as self-improvement and not with the hope of publishing. There is one more
remarkable fact: This work was done in Philadelphia. Thus, for a period of 123 years,
the most accurate known evaluation of was done in the remote British colony of
Pennsylvania. PDS
Painting New Lines: Maximizing Color Difference in Metro Maps, S. F. Griffioen and A. V. Kiselev. The Mathematical Intelligencer 38:1 (Spring 2016) 2531,
http://dx.doi.org/10.1007/s00283-015-9597-y.
The Moscow metro has 14 different lines and on its stylized map each line is colored
a different color. In the near future, new lines are to be added. This article describes a
model to determine what color a new line should be in order to maximize its perceptual
difference from the colors already in use. In order to do this, we need to define a space
of colors and a metric on that space which correlates well with perceived color differ
ences. In 1931, the Commission Internationale de lEclairage,
based on measurements
on the vision of hundreds of subjects, recommended using a three-dimensional space
with coordinates L (lightness), a(red-green scale), and b (yellow-blue scale) with the
Euclidean metric d(C1 , C2 ) = (L 1 L 2 )2 + (a1 a2 )2 + (b1 b2 )2 . Within this
space, now known as CIELAB space, the set of colors that can be represented by a
given printing system is a compact set, called the gamut of the system. The authors
illustrate and use the gamut of Adobe 98 and model it as a polyhedron. Now the
problem becomes: Given a finite set of points in a polyhedron P, find the point whose
minimal distance from all those points is as large as possible. One computationally
efficient method for doing this is to construct the Voronoi polyhedral decomposition
of P corresponding to the given points and check its vertex points to find the one with
the largest minimal distance. The authors use a variation of this method to determine
that the next five new lines in the Moscow metro should be colored, in order, cyan,
bright green, blue, pale lavender, and rose pink. At the end of the article, they note
that the next step in modeling would be to take into account that not all lines intersect
and maximize the color differences between lines which do intersect. The article is
beautifully illustrated with maps of the Moscow metro and its planned extension, the
London metrowhich we are invited to analyzeand a stereographic image (bring
your eyes in close) of the Moscow metro colors in CIELAB space. PDS
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
387
389
and he was embittered by Pascals demeaning comments on it. Many years later, in
1685, Wallis published a treatise on algebra that included, according to the authors,
a fairly biased review of the algebraic works of his contemporaries. While he praised
his fellow countrymens publications on algebra, he criticized the work done by great
mathematicians such as Rene Descartes and others from outside of England. For more
information on Walliss disputes see Jacqueline Stedalls article John Wallis and the
French: His Quarrels with Fermat, Pascal, Dulaurens, and Descartes, Historia Mathematica 39:3 (August 2012) 265279. In the 1690s, Wallis published his theological
thoughts on the doctrine of The Holy Trinity. These used a geometrical model of a
cube to relate its length, width, and height to the Father, Son, and Holy Ghost! The
authors observe that Walliss material on the Trinity was very popular throughout the
various religious sects. PR
Deepwater Horizon: Locating Submerged Oil in Ocean Water Using the
k-means Clustering Technique, T. Dey, W. Hickey, B. Parakkal, and W. Woyczynski. Chance 29:2 (April 2016) 4652, http://dx.doi.org/10.1080/09332480.
2016.1181965.
In 2010, the Deepwater Horizon oil well exploded and released about 5,000,000 barrels
of crude oil before it was capped. For comparison, the intentional Gulf War oil spill
in Kuwait is estimated to be slightly larger than this and the Exxon Valdez discharged
about a tenth as much in Prince William Sound, Alaska. In spite of the damage caused
in the Gulf of Mexico, the rate of oil drilling permits issued by the United States
today is about the same as it was right before 2010. Consequently, developing better
techniques to estimate the amount of oil in a body of water from samples taken is still
relevant and the authors describe a method using k-means, a data mining algorithm, to
do this.
The United States government has made available its measurements from the Gulf
of Mexico through the National Oceanic and Atmospheric Administration (NOAA).
The authors used 16 weeks of temperature, depth (1, 2, 3, 4, and 5 meters), and conductivity (from which salinity can be estimated) data. Estimates of fluid density can
be made from these parameters, leading to an estimate of the proportion of oil in each
sample. The k-means algorithm is a clustering technique that finds mean vectors for
each cluster that minimize the sum of squared distances between these and the data
vectors of each cluster. The authors found four clusters defined by ranges of mean
density in kg/m3 ; the one with an upper bound of 1,020 kg/m3 is strongly correlated
with regions of high levels of oil. This makes sense because ocean water has a density
around 1,025 kg/m3 , while the oil at that location is only 840 kg/m3 , so contamination results in lower mean density. Finally, the authors method is cheap because it is
collected by unmanned, underwater drones instead of expensive-to-run hydrographic
survey ships, which is yet another example in statistics of exchanging some accuracy
in order to obtain lower costs. RB
Higher-Order Organization of Complex Networks, A. Benson, D. Gleich, and
J. Leskovec. Science 353:6295 (8 July 2016) 163166, http://dx.doi.org/10.
1126/science.aad9029.
Network science has rapidly become one of the fundamental interdisciplinary tools
for understanding the world, with applications in a variety of areas including gene
regulation networks, transportation systems, and neural structures. Much of network
analysis is based on examining local properties of nodes, which are then averaged in
some way over the larger network. Global analysis of a network is often built on spectral analysis of some appropriate matrix, such as the adjacency matrix, representing the
390
network. What has been lacking, until now, are computationally tractable methods for
global analysis of large networks to identify how recurring subgraphs drive network
structure. The authors accomplish this with a clever generalization of the conductance
metric from spectral graph theory.
For a particular subgraph M called a motif, the motif conductance
M (S) =
cut M (S, S)
is a measure of how much of a given network contains instances of M. Here cut M (S, S)
refers to the number of instances of the motif M that are cut by the partitioning and
vol M (S) is the number of nodes in S that are found in instances of M. Essentially, one
tests all partitions of the graph into a set S and its complement S until a minimum
of M (S) is found. If one then forms a motif adjacency matrix W M by counting the
number of times two nodes co-occur in the motif, then an eigenvector of the Laplacian
of W M provides an ordering of the nodes of the network. Looking at the nested sets
of nodes of increasing size, Sr = {1 , 2 , . . . , r }, one can then identify the r value
for which the motif conductance is minimal. That set provides, as the authors prove
in the supplemental material online, a near-optimal higher order cluster. Applying this
to the c. elegans neuronal network, they identify a 20-node higher-order cluster of a
four-node bi-fan motif (where nodes 1 and 2 each point to nodes 3 and 4, expressing
two neurons passing information separately to the next layer of neurons) that governs
communication. The supplementary material provides additional proofs and details
on how this analysis generalizes to directed, undirected, weighted, and signed networks, and an analysis of the computational efficiency of the technique, which scales
as
(edges1.2 ) for triangular motifs in the worst case scenario. KHG
Novel Fibonacci and Non-Fibonacci Structure in the Sunflower: Results of a
Citizen Science Experiment, J. Swinton, E. Ochu, and The MSI Turings Sunflower
Consortium. Royal Society Open Science 3 (2016) 3: 160091, 17 pp., http://dx.
doi.org/10.1098/rsos.160091.
The spiral arrangement of seeds in sunflower heads has been a favorite example of
Fibonacci numbers occurring in nature, with some anecdotal evidence and small studies supporting the notion. But how consistently does that really occur? In this citizen
science experiment, volunteers grew sunflowers, took photos of the flower heads, then
marked the spirals and submitted them, yielding data on 657 sunflower heads which
were independently reviewed to ensure reliable results. Typically, both clockwise and
counterclockwise spirals, called parastichies, can be discerned, and the numbers in
each direction form a parastichy pair. These parastichy pairs have often been observed
to be adjacent Fibonacci numbers like (34, 21), or adjacent double Fibonacci numbers
like (68, 42). Among the parastichy pairs observed here, 80% had Fibonacci structure
and a further 10% were one-off from Fibonacci (with one less than Fibonacci much
more frequent than one more). The authors systematically examine the exceptions to
Fibonacci-like organization and categorize the deviations that can occur. Across most
of the samples, one spiral direction tends to be more prominent than the other direction, and in a few of the non-Fibonacci examples, only one spiral direction could be
discerned. In some other problematic samples, regions of disorder occurred, preventing reliable counting. There were also instances of multiple spiral patterns occurring
rather than only two clear ones. This large data set could facilitate the testing of different models of how such parastichy pairs may arise, such as Turings geometrical
phyllotaxis model, which could make a great undergraduate research project. TL
VOL. 47, NO. 5, NOVEMBER 2016 THE COLLEGE MATHEMATICS JOURNAL
391
estimates produced by the other group members regarding the estimated fairness of the
spinner in relation to the number of spins completed. Comparisons among groups naturally give rise to the ideas behind type I errors (wrongly convicting a noncheater), type
II errors (letting the suspect go when cheating has occurred), and power, which teachers can introduce as the discussion evolves. Groups produce graphs that demonstrate
the variability of the estimate of the spinners fairness as a function of the number
of spins and guided questions are provided to enhance discussion. The second class
period can be devoted to a simulation activity that allows the students to collect more
data over a wider range of conditions. Using TinkerPlots, the groups explore how the
fairness of the spinner, the number of spins, and the significance level required to convict all influence the rates and types of errors produced. Guided questions and selected
by-hand calculations help students sort through the data and draw conclusions about
the ways in which sample size and effect size increase the power of a test. While
designed for a high school Advanced Placement statistics course and the Common
Core, this pair of activities meets the expectations laid out in the Guidelines for Assessment and Instruction in Statistics Education College Report for helping students at all
levels develop experience with these concepts. These activities would also be useful in
many college level introductory statistics or liberal arts mathematics courses. KHG
AMS President Robert Bryant: His Early Life and His Views on Mathematics,
A. Jackson. Notices of the AMS 63:6 (June/July 2016) 614620, http://dx.doi.
org/10.1090/noti1393.
Robert Bryant grew up in a small farming community in North Carolina. He said he
hated arithmetic as a kid, but after he found his uncles college calculus textbook
he was fascinated by the pictures and word problems. He taught himself college algebra by the eighth grade and was inspired by a professor to major in mathematics in
his senior year. He attended North Carolina State University and originally planned
to go to Harvard for graduate study. He was not accepted and he had not applied to
other schools because he did not want to spend money on application fees. Because
of these hardships, Bryant supports the AMSs involvement in the Math Alliance to
support underserved students. Bryant is most fond of his proof of the existence of Riemannian manifolds with special holonomy. He was surprised by physicists interest
in his work, which turned out to be connected to supersymmetry, string theory, and
M-theory. When asked about how his personal life has interacted with professional
life, Bryant said that as a gay man, he faced some challenges. To bring equality, he
was involved in persuading Duke University to recognize same-sex domestic partners
(the first university in the South to do so). He is proud of the mathematics community for moving the 1995 Joint Mathematics Meetings (JMM) from Denver to San
Francisco after Colorado passed a discriminatory amendment. Bryant helped set up
the first reception for LGBT mathematicians at the 1995 JMM and 21 years later the
tradition of this reception still continues. FW
Media Correspondents
RB Roger Bilisoly; KHG Kris H. Green; RNG Raymond N. Greenwell; TL Tanya
Leise; RM Roland Minton; PR Peter Ross; A&JS Annie & John Selden; PDS Philip
D. Straffin; FW Frank Wang.
393
George Polya
Awards for 2016
The George Polya Awards, named for the distinguished Stanford University mathematician and author, best known for his book How to Solve It, recognizes the authors
of exceptional pieces from The College Mathematics Journal. For the year 2016, there
are two awards. The following citations were written by the Committee on George
Polya Awards, chaired by Natacha C. Fontes-Merz.
G. Hamilton, K. S. Kedlaya, H. Picciotto, Square-sum pair partitions, The College
Mathematics Journal 46:4 (2015) 264269.
Mathematicians love a good puzzle. But what does good mean? While answers may
vary, the article Square-Sum Pair Partitions delivers a puzzle to warm many of our
hearts. The basic version is this: Arrange a given set of whole numbers into pairs so
that the sum of the numbers in each pair is a perfect square. For example, the whole
numbers from 1 to 8 can be paired into {1, 8}, {2, 7}, {3, 6}, and {4, 5}. Sets of the form
{1, 2, 3, . . . , 2n} follow, prompting readers to question when a solution exists, when a
solution is unique, and how many solutions are possible in the case of non-uniqueness.
Generalizations abound: sets need not start at one, and need not increase by one, with
arithmetic progressions { p, p + d, p + 2d, . . . , p + (n 1)d} suggested as another
type of set to consider. Alternate constraints, such as summing to powers of 2, and
many other extensions, turn the initial puzzle into a wealth of ideas.
Returning to the idea of a good puzzle, most notably, the puzzle in this article is
easy to state; elementary grade students can begin is tackle it. With regard to educational goals, a priority for many puzzle seekers, the article speaks directly to mathematics teachers, suggesting ways to implement these challenges in class and providing
ample resources for teachers and puzzle hobbyists. Another good puzzle quality is a
compelling visual representation and the puzzles in this article have a beautiful rainbow pairing shown in several colorful illustrations. It is also true that a good puzzle
often has layers, with each solution prompting further questions and extensions, something demonstrated repeatedly in this article. The extensions and suggestions about
proving results provide solid material for the college level and beyond. Finally, and
of particular importance for the George Polya Award, the article is written expertly,
making it easy and fun for readers to enjoy the many good puzzles within.
3 4
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Figure 1. A partial square-sum pair partition from the article. Can you complete it?
http://dx.doi.org/10.4169/college.math.j.47.5.394
394
(1)
x2 + y2 = 1
(2)
in the unit square [0, 1] [0, 1] looks awfully circular itself. Thanks to the symmetry
of a circle and the reciprocal nature of these equations, we might want to take a close
look before deciding that (2) yields some other kind of curve.
Starting from this straightforward premise, Boualem and Brouzet take us on a tour
of important techniques from algebra, geometry, and calculus, each step illustrated by
a new proof that (2) is not, in fact, an arc of a circle. The chain of fourteen proofs,
many framed as exercises, culminates in an application of the formula for curvature.
But the fact that (1) and (2) share a structural bond prompts the authors to linger
on the topic. Surely, the two curves, though not identical, are members of a family.
Indeed they are: after proving that (2) is an arc of a parabola, the authors demonstrate
that these two curves are the only conics (aside from a degenerate case) that satisfy
x r + y r = 1 for r > 0 in the region [0, 1] [0, 1]. This discussion not only revisits
curvature but also relies on series expansions. The authors conclude their tour with a
glimpse at Lames curves, a larger family that contain both (1) and (2).
If we compare the article to a mountain trail, and think of the fourteen proofs as base
camps, then the article is something that a teacher can give to a student who is equipped
with the tools of elementary calculus, and the student can begin to climb. Certain
passes may require equipment that the student can acquire on the fly, and some of the
base camps may turn out to be (for now) inaccessible. At all times, however, a view
of the destination is visible from the trail, drawing the student onwards. This article
inspires persistence and rewards the hard work that we have all put into understanding
what we must in order to make this journey, and thus it shares much with Polya award
winners of the past.
3
2
1
Figure 2. Curves (1) and (2) plotted together beyond [0, 1] [0, 1].
395
NOTES ON VOLUME 46
In these pages, we share corrections and notes for articles published in volume 46
(2015) of The College Mathematics Journal. Unfortunately, these do not address all
of the published mistakes, only the ones we are aware of that are more problematic
than typographic or similar errors. Readers are encouraged to write cmj@maa.org with
notes on the current volume.
Proof without words: The maximum sum of inradii, David Richeson, The College
Mathematics Journal 46 (2015) 23, http://dx.doi.org/10.4169/college.
math.j.46.1.23.
This page generated an interesting discussion about the intent and requirements for
the popular Proofs Without Words series. Based on comments by a referee, the claim
preceding the final figure reads The triangulation shown below provides a way to
see this convergence rather than claiming a proof, but Michael Maltenfort expected
more: While admitting that the format prevents including all details, he explains, As
a reader, I want to be able to come away from a PWW and have the ability to write up
a complete proof. The author agrees that There are definitely nontrivial details to be
addressed. We maintain that the triangulation shown provides valuable insight as to
why the sum in question equals the diameter. Perhaps this raises the need for a revised
title that includes a broader category of mathematically elucidating figures.
Elvis lives: Mathematical surprises inspired by Elvis, the Welsh corgi, Steve J. Bacinski, Mark J. Panaggio, Timothy J. Pennings, The College Mathematics Journal 46
(2015) 8291, http://dx.doi.org/10.4169/college.math.j.46.2.82.
Due to a production error, an uncorrected formula was published on page 87. It
should be as follows.
d L
=
d x y
y
1 + (y )2
3
v(x, y)
The fastest path between two points, with a symmetric obstacle, Kathleen Bell, Shania
Polson, Tom Richmond, The College Mathematics Journal 46 (2015) 9297, http://
dx.doi.org/10.4169/college.math.j.46.2.92.
Michael Maltenfort points out that in the section the other diagonal, the authors
lose track of their convention that the axes are not part of the pool and should consider
the purely on-land path of running to the edge of the pool and then up to the corner. The
authors agree that their application of Snells law does not address this path, which
will be optimal if running is much faster than swimming. They go on to point out
that much faster would be difficult to quantify, depending on a quartic with three
parameters.
On combining and convolving fractals, Nicholas Cotton, Cam McLeman, Daniel
Pinchock, The College Mathematics Journal 46 (2015) 99108, http://dx.doi.
org/10.4169/college.math.j.46.2.99.
The authors are careful to explain that their convolution of fractals is commutative, but Michael Maltenfort points out that they fail to account for this in two places.
In Theorem 6, assuming dF dG does lose generality; a stronger statement, with
essentially the same proof, would say that if dF = dG , then the dimension of F G is
http://dx.doi.org/10.4169/college.math.j.47.5.396
396
strictly between dF and dG . Also, the order of the two fractals on the left-hand side
of the graphic equation in Figure 10 should be reversed to give the convolved fractal
shown on the right-hand side.
Sequences of power lines, Ricardo Alfaro, The College Mathematics Journal 46 (2015)
113120, http://dx.doi.org/10.4169/college.math.j.46.2.113.
Michael Maltenfort reported some strict inequalities that should allow for equality
and nonstandard uses of notation along with a more serious issue:
In the paragraph beginning In the second case, the author is considering the
case in which limn (bn /an ) = L 1. The author claims that there exists
n 0 such that for all n > n 0 we have bn > an . An elementary calculus problem we
might give our students is to invent examples in which bn < an for all n and yet
limn (bn /an ) = 1. Happily, it turns out that this error
can berepaired. The
conclusion that such an n 0 exists is unnecessary because nn bn /an < nn /an
follows from an and bn being positive.
Healso pointed out that the graphs shown in Figure 5 should all cross the x-axis at
(1 + 5)/2. The author acknowledges submitting the wrong graphics file and appreciates the correction to Theorem 1 which, as noted, does not change the conclusion.
Groupoid cardinality and Egyptian fractions, Julia E. Bergner, Christopher D. Walker,
The College Mathematics Journal 46 (2015) 122129, http://dx.doi.org/10.
4169/college.math.j.46.2.122.
Multiple readers were confused by the authors visual representation of groupoids.
They explain, The dot and arrow diagrams we drew were meant to be schematic,
not rigorous descriptions of groupoids. While it was admittedly ambiguous how many
arrows were meant by the double arrow from the one dot to itself, the picture was
supposed to show an identification of the row of dots by folding itthe middle dot
gets identified with itself and so has a (single) nonidentity arrow which is its own
inverse.
There were also responses that the greedy algorithm cannot be applied to find Egyptian fraction decompositions for m/n for integers m > n. This is true, and alternative
methods are available, but as the authors stated, they were only considering the case
m < n.
On the shrinking volume of the hypersphere, Michael H. Peters, The College Mathematics Journal 46 (2015) 178180, http://dx.doi.org/10.4169/college.
math.j.46.3.178.
Lawrence Glasser mentioned to the author a related problem he posed in the
Monthly in 1961: Show that the volume of any sphere in Hilbert space is zero. A solution by D. J. Newman and comments by J. V. Ryff appeared in Amer. Math. Monthly
69 (1962) 240241 http://dx.doi.org/10.2307/2311073.
Square-sum pair partitions, Gordon Hamilton, Kiran S. Kedlaya, Henri Picciotto, The
College Mathematics Journal 46 (2015) 264269, http://dx.doi.org/10.4169/
college.math.j.46.4.264.
After publication, the authors found another relevant reference which includes
determining the n for which {1, 2, . . . , 2n} admits a partition into square-sum pairs:
J. Schwartzman, H. S. Shultz, Square-dance numbers, Math. Teacher 85 (1989)
380382.
397
REFEREES IN 2016
Listed here are the mathematicians who, in addition to members of the Editorial Board,
reviewed articles for CMJ over the past year (May 2015 to August 2016). The College
Mathematics Journal is only produced thanks to the labor of these volunteers.
http://dx.doi.org/10.4169/college.math.j.47.5.398
398
399
www.maa.org/ebooks
CONTENTS
322364
ARTICLES
322
333
334
340
346
347
355
356
361
365368
CLASSROOM CAPSULES
365
367
369374
375380
BOOK REVIEW
375
381385
381
386393
394
FILM REVIEW
The Man Who Knew Infinity, directed by Matthew Brown
Reviewed by Jennifer Wilson
MEDIA HIGHLIGHTS
GEORGE PLYA AWARDS FOR 2016
396-397
NOTES ON VOLUME 46
398-399
REFEREES IN 2016