Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
MONTHLY
VOLUME 120, NO. 10 DECEMBER 2013
NOTES
REVIEWS
Member Plus
$169/Year
$249/Year
Individual Member Benefits
$35/Year
Member
Student
Electronic Subscriptions to:
The American Mathematical Monthly (10 per year) X X X
The College Mathematics Journal (5 per year) X X X
Mathematics Magazine (5 per year) X X X
Math Horizons (4 per year) X X X
MAA FOCUS (6 per year) X X X
MONTHLY
Volume 120, No. 10 December 2013
EDITOR
Scott T. Chapman
Sam Houston State University
ASSOCIATE EDITORS
William Adkins Ulrich Krause
Louisiana State University Universitat Bremen
David Aldous Jeffrey Lawson
University of California, Berkeley Western Carolina University
Elizabeth Allman C. Dwight Lahr
University of Alaska, Fairbanks Dartmouth College
Jonathan M. Borwein Susan Loepp
University of Newcastle Williams College
Jason Boynton Irina Mitrea
North Dakota State University Temple University
Edward B. Burger Bruce P. Palka
Williams College National Science Foundation
Minerva Cordero-Epperson Vadim Ponomarenko
University of Texas, Arlington San Diego State University
Beverly Diamond Catherine A. Roberts
College of Charleston College of the Holy Cross
Allan Donsig Rachel Roberts
University of Nebraska, Lincoln Washington University, St. Louis
Michael Dorff Ivelisse M. Rubio
Brigham Young University Universidad de Puerto Rico, Rio Piedras
Daniela Ferrero Adriana Salerno
Texas State University Bates College
Luis David Garcia-Puente Edward Scheinerman
Sam Houston State University Johns Hopkins University
Sidney Graham Susan G. Staples
Central Michigan University Texas Christian University
Tara Holm Dennis Stowe
Cornell University Idaho State University
Roger A. Horn Daniel Ullman
University of Utah George Washington University
Lea Jenkins Daniel Velleman
Clemson University Amherst College
Daniel Krashen
University of Georgia
EDITORIAL ASSISTANT
Bonnie K. Ponce
NOTICE TO AUTHORS Proposed problems or solutions should be sent to:
The MONTHLY publishes articles, as well as notes and DOUG HENSLEY, MONTHLY Problems
other features, about mathematics and the profes- Department of Mathematics
sion. Its readers span a broad spectrum of math- Texas A&M University
ematical interests, and include professional mathe- 3368 TAMU
maticians as well as students of mathematics at all College Station, TX 77843-3368
collegiate levels. Authors are invited to submit arti-
cles and notes that bring interesting mathematical
In lieu of duplicate hardcopy, authors may submit
ideas to a wide audience of MONTHLY readers.
pdfs to monthlyproblems@math.tamu.edu.
The MONTHLYs readers expect a high standard of ex-
position; they expect articles to inform, stimulate,
challenge, enlighten, and even entertain. MONTHLY Advertising Correspondence:
articles are meant to be read, enjoyed, and dis- MAA Advertising
cussed, rather than just archived. Articles may be 1529 Eighteenth St. NW
expositions of old or new results, historical or bio- Washington DC 20036
graphical essays, speculations or definitive treat-
ments, broad developments, or explorations of a Phone: (877) 622-2373
single application. Novelty and generality are far E-mail: tmarmor@maa.org
less important than clarity of exposition and broad
appeal. Appropriate figures, diagrams, and photo- Further advertising information can be found online
graphs are encouraged. at www.maa.org
Notes are short, sharply focused, and possibly infor- Change of address, missing issue inquiries, and
mal. They are often gems that provide a new proof other subscription correspondence:
of an old theorem, a novel presentation of a familiar MAA Service Center, maahq@maa.org
theme, or a lively discussion of a single issue.
All at the address:
Submission of articles, notes, and filler pieces is re-
quired via the MONTHLYs Editorial Manager System. The Mathematical Association of America
Initial submissions in pdf or LATEX form can be sent 1529 Eighteenth Street, N.W.
to the Editor Scott Chapman at Washington, DC 20036
Abstract. How might we determine in practice whether 23/67 equals 33/97? Is there a quick
alternative to cross-multiplying?
How about reducing? Cross-multiplying checks equality of products, whereas reducing is
about the opposite, factoring and cancelling. Do these very different approaches to equality of
fractions always reach the same conclusion? In fact, they wouldnt, but for a critical prime-free
property of the natural numbers more basic than, but essentially equivalent to, uniqueness of
prime factorization.
This property has ancient, though very recently upturned, origins, and was key to number
theory even through Eulers work. We contrast three prime-free arguments for the property,
which remedy a method of Euclid, use similarities of circles, or follow a clever proof in the
style of Euclid, as in Barry Mazurs essay [22].
1 My unscientific surveys have shown that people, including mathematicians, display a remarkably diverse
collection of knowledge and approaches to the question of determining equality. Surprisingly, even children
from the same home and school may take quite different tacks, some expanding and some reducing. And to
my astonishment, not all mathematicians even think of the possibility of reducing.
2 I have often wondered what intuition the average person has about whether prime factorization of natural
numbers is unique or not, since it is so key to the nature of basic mathematics. Somewhat disappointingly, my
informal surveys suggest that most people have no intuition one way or the other on this matter.
3 Note that the questions of efficacy and efficiency are quite different. Reduction may be efficacious, always
distinguishing unequal fractions, but whether it is more efficient than cross-multiplying is another matter. We
might at first think that reduction is comparatively inefficient, requiring factorization before reduction, but in
fact it does not, since the irreducible reduction of a fraction can be obtained by calculating the greatest common
divisor of numerator and denominator using the Euclidean algorithm, which is highly efficient.
868
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120
only certain categories of natural numbers forming acceptable multiplicative systems
(specifically, commutative semigroups with cancellation).
1. The even integers. Here 2/6 = 6/18, but both are irreducible (6 has no proper
factorization in the even numbers), so the mathematicians method will not work:
Different irreducible fractions can be equal.
2. The H-integers (after David Hilberts use of them [5, 15]), consisting of the
numbers one more than a multiple of three. Here 4/10 = 10/25; again, different
irreducible fractions can be equal.
In both cases cross-multiplying displays a failure of unique factorization.
We are thus appropriately admonished to recall that the natural numbers are per-
fectly built, so that every fraction has a unique irreducible reduction. And indeed, this
powerful property of fractions is essentially unique factorization, even though primes
are not mentioned. An amazing amount of beautiful and important mathematics hinges
thereon.
In fact, we should view it as almost miraculous that the natural numbers is a world
where reduction is as effective as expansion, i.e., where reduction and expansion of
fractions detect the same equivalence relation. After all, in any commutative semigroup
with cancellation, although equality of fractions is precisely the equivalence relation
have an identical expansion, there is no reason why the relation have an identical
reduction should even be transitive. For instance, in the H-integers above, 4/10 and
40/100 have an identical reduction, and so do 40/100 and 10/25, but 4/10 and 10/25
do not, so transitivity fails.
Returning to ordinary fractions, can we answer our reduction question without
having first to introduce primes, then prove uniqueness of prime factorization, and
then build another proof upon that? We ask: Is there a straightforward way to deduce
uniqueness of lowest terms for a rational number without invoking primes?
Identical reduction of equal fractions. Given two equal fractions, there is some frac-
tion to which they both reduce.
Reduction tells all for fractions. Given two equal fractions, they have identical irre-
ducible reductions.
Thus, proving any of these three equivalent properties of fractions will answer
Does reduction tell all? affirmatively.
By his parts language, Euclid means that there exist natural numbers x, y, m, n
such that a = mx, b = nx, c = my, and d = ny. In the language of fractions, this
simply says that a/b and c/d both reduce to m/n.
The only if is trivial, and the if has been called the four numbers theorem.
This was first named by Laszlo Kalmar as the Vierzahlensatz [16], and proven by
him by mathematical induction. It appears that Kalmar did not notice that this is Eu-
clids VII.19.
In terms of divisibility, this is nicely recast as the following.
Shifting our consideration only briefly to contrast with prime factorizations, from
any of the above we can easily deduce what we often think of today as underpinning
uniqueness of prime factorization.
Proof from the viewpoint of fractions. If p | ad, then ad = pl, so a/ p = l/d. Now
either p | a, or else a/ p is irreducible, in which case by identical reduction of equal
fractions, l/d reduces to a/ p, and therefore p | d.
870
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120
Proof from the viewpoint of divisors of a product. If p | ad, then by divisors of a
product, p is a product of two numbers, necessarily 1 and p, dividing a and d, and
thus p itself divides either a or d.
Thus, it should not surprise that many results proven today using unique factoriza-
tion were at least as easily proven in the past without primes, e.g., directly from VII.19,
also known as the four numbers theorem.
Remarkably, uniqueness of prime factorization as we know it today was not formu-
lated until barely two centuries ago, in Carl Gausss 1801 Disquisitiones Arithmeticae.
Prior to Gauss, the focus was on knowing how to list all divisors of products of num-
bers. This is nicely accomplished via the four numbers theorem, which ensures that the
collection of divisors of a product of two numbers comprises precisely, and crucially
no more than, all the pairwise products of individual divisors of the two numbers. Of
course, from knowing all the divisors of a product we can easily obtain uniqueness of
prime factorization, but the precise notion of unique factorization was neither explic-
itly considered by, nor of apparent interest or necessity to, those before Gauss. Mathe-
maticians like al-Fars (c. 12671319) [2, 3], Jean Prestet (16481690) [12; 24, Ch. 6,
p. 146ff], and Leonhard Euler (17071783) [9, Book 1, Abschrift 1, 41] dealt very
successfully with divisibility using results such as VII.19 and VII.30 [1, 4, 18, 20].
For instance, Euler made an occasional nod to Euclid as justification where today
we appeal to unique factorization. Two illustrations from Eulers works [18, pp. 186
187], easily proven from the four numbers theorem, are his use of the following.
If a product of two relatively prime numbers is a nth power, then so are the individ-
ual factors.
Euler explicitly says that, thanks to Euclid, it would be superfluous for him to
provide a proof [7, Lemma 1, p. 126; 18, p. 186; 20; 21]. He used this result in
proving Fermats Last Theorem for exponent four [17].
Any divisor of a product of two relatively prime numbers is uniquely a product of a
divisor of each of them.
Euler needed this result in the development of his function, which counts the
natural numbers up to n that are relatively prime to n [10, Ch. 4, p. 16; 20; 21], and
in his study of amicable numbers, in order to know that the sum of proper divisors
of n function (n) is, for example, multiplicative on relatively prime products mn
[8, Lemma 1, p. 24; 10, Ch. 3; 18, pp. 186187; 20]. At one point, Euler makes
explicit use of the four numbers property [8, Prob. 2, p. 61; 20].
In sum, prime factorizations are quite unnecessary for obtaining reduction tells all
for fractions, and also for most of the number theory achieved by Euler; all that is
needed is the four numbers theorem, Euclids VII.19.4
Returning to resolving the reduction question for fractions without mentioning
primes, it appears at first blush that Euclid did this nicely for us with his VII.19,
interpreted as identical reduction of equal fractions. However, we are about to see that
there are very serious problems with this tidy conclusion.
four numbers theorem and unique factorization in the setting of ring theory is explored in [18, 19, 20].
Greatest common divisor for unique irreducible reduction. We prove unique irre-
ducible reduction in a manner close to our fix [23] for Euclids proof of VII.19.
Proof. Given a/b, let g = gcd(a, b) be the greatest common divisor of a and b, and
write a = ge, b = g f , so that e/ f is an irreducible reduction of a/b. Now, consider
an arbitrary reduction c/d of a/b, so a = hc, b = hd.
We need a crucial fact about the natural numbers, that every common divisor of
two numbers must divide their greatest common divisor (gcd). This is stated by Euclid
as his Porism (corollary) to the Euclidean algorithm, and follows by examining the
algorithms iterative process both upward and downward.7 It is an algebraic property
of the gcd.
Thus, h divides g, so writing g = mh we have a = mhe and b = mh f . Thus, c =
me and d = m f , hence c/d reduces to e/ f . Then, since every reduction of a/b reduces
to e/ f , the irreducible reduction is unique.
Circle similarity for the four numbers theorem. Janos Suranyi [25, 26] gives a
proof of the four numbers theorem using translation congruence properties of an in-
teger lattice and similar triangles, that simultaneously automatically selects two gcds
from the geometry. Klaus Hartig and Suranyi [5, 13, 14] make a geometric circle proof.
It appears that Hartig and Suranyi, like Kalmar, were unaware that the four number
5 The serious gap in Euclids proof of VII.19, not identified until 1970 [23, 27], leaves us wondering when
Euclids key number theoretic results, such as Euclids Lemma (VII.30) relying on VII.19, were actually validly
proven by anyone. Since al-Fars relied directly on Euclids results such as VII.30 [3], it seems that Jean Prestet
in 1689, who developed his results independently of Euclid, might have been the first [12]. Prestet uses the
Euclidean algorithm to prove his key result, from which he derives results like VII.30 [24, Ch. 6, p. 146ff].
Interestingly, quite unlike Euclid, the key result Prestet first derives from a rather elaborate argument with the
Euclidean algorithm is not about divisors of a product, but rather about least common multiples. He proves
that the least number divisible by two relatively prime numbers is their product. All else follows from this.
6 In [23], we analyzed Euclids path to Euclids Lemma (VII.30) via VII.19, including the major lacuna in
his proof of VII.19, explained below after our first proof of unique irreducible reduction.
7 Today we tend first to derive the much more recent Bezout equation, that gcd(a, b) is an integer linear
combination of a and b, from the upward process, and then derive Euclids Porism therefrom. But Euclid
obtains the Porism directly from the algorithm, without needing the Bezout equation.
8 Euclid does not work with fractions per se, but his analysis involves the equivalent of this.
872
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120
theorem is Euclids VII.19. Of course, they were also unaware that Euclids proof is
flawed.
Here we present a different geometric proof from theirs, with a modular flavor,
using similarity of partitioning of circles, invented because I knew of the existence
of a Hartig-Suranyi circle proof, but couldnt immediately obtain a copy of it. To my
surprise, the following proof is remarkably different from their circle proof, using
geometric similarity of three circles rather than a counting analysis on a single circle.9
Four numbers theorem. If ad = bc, then there exist x, y, m, and n such that a =
mx, b = nx, c = my, d = ny.
a b = nx
a = mx
x = ra + sb
x
b
Now, scale the partitioned circle up by a factor of d, to a similar circle with cir-
cumference bd and initial arclength da(= bc), and arclength between adjacent points
d x = d(ra + sb) = r bc + sbd = b(r c + sd) = by with y = r c + sd also an integer.
d x = d(ra + sb)
da = bc
= bcr + bsd
= b(r c + sd)
= by
d x = by
db
Then, scale this circle down by a factor of b, yielding a circle with circumference
d, initial arclength c, and arclength y between adjacent points.
9 The arguments do not really rely specifically on the geometry of circles, and could equally be phrased
with polygons.
y
d
The two geometric similarities have rescaled the lengths of arcs, converting b to d,
a to c, and x to y, while the multiples encoding the relationships between arcs remain
the same, and so the original a = mx and b = nx rescale to c = my and d = ny.
It is geometrically obvious that x = gcd(a, b). Or, note that since x | a, b and x =
ra + sb (the Bezout equation), we have x = gcd(a, b). Likewise, y = gcd(c, d).
The rotational symmetry relied on in combination with similarity leads directly to
both gcds occurring simultaneously with the same multiplicities, thereby avoiding the
need for Euclids Porism on divisibility properties of greatest common divisors.
Proof (as in B. Mazurs essay [22]). Given a/b, let e/ f be the fraction with smallest
numerator that is equal to a/b, i.e., for which a f = be. Our goal is to show that a is a
multiple ke, for then it will follow that b = k f , so a/b must reduce to e/ f . Then, e/ f
must be the unique irreducible fraction equal to a/b, since all fractions equal to a/b
will then reduce to the one with smallest numerator.
If a = e, we are done. If not, let ne be the largest multiple of e still less than a. Then,
check that n f will be less than b. Now verify the equality (a ne)/(b n f ) = e/ f
by cross-multiplying.12 Since e/ f has smallest numerator among all fractions equal to
it, we must have a ne e, in addition to a ne e from the choice of n. Hence
a ne = e, so a is a multiple of e as desired.
The proof that Mazur gives of unique irreducible reduction requires no primes, no
Euclidean algorithm, no algebraic divisibility of greatest common divisor, no geome-
try, no mathematical induction, and no contrapositive or contradiction. It relies on so
little, i.e., repeated but bounded subtraction of smaller from larger, and a unit to allow
repeated addition to transform to integer multiples.13 Its simplicity is enabled by the
clever choice of a new third equal fraction based on the given different but equal ones.
874
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120
7. WHAT TO CONCLUDE? We have examined three prime-free ways of showing
that reduction tells all for fractions (the mathematicians secret), equivalent to unique
irreducible reduction of fractions and to identical reduction of equal fractions:
A la Euclid (repaired via Euclidean algorithm),
Geometric circle similarity (modularity),
Mazurs [22] (mixing equal proportions).
The beautiful proof that Mazur gives of unique irreducible reduction, based solely
on a single clever combination of proportionalities, is remarkably simple and satisfy-
ing, and one Euclid could have been very happy with. Is it perhaps also the shortest
and most elementary path to uniqueness of prime factorization, since Euclids Lemma
follows immediately?
ACKNOWLEDGMENTS. Many thanks for valuable comments received from Pat Baggett, Andrzej Ehren-
feucht, Bill Julian, Reinhard Laubenbacher, Franz Lemmermeyer, Barry Mazur, Fred Richman, Virginia
Warfield, Robert Wisner, and the referees.
REFERENCES
1. A. G. Agargun, The fundamental theorem of arithmetic dissected, Math. Gaz. 81 (1997) 5357, available
at http://dx.doi.org/10.2307/3618768.
2. , Kamal al-Din al-Fars and the fundamental theorem of arithmetic, in Science in Islamic civili-
sation (Istanbul, 1991/1994), Stud. Sources Hist. Sci. Ser. 9, Res. Cent. Islam. Hist. Art Cult. (IRCICA),
Istanbul, 2000. 185192.
3. A. G. Agargun, C. R. Fletcher, al-Fars and the fundamental theorem of arithmetic, Hist. Math. 21 (1994)
162173, available at http://dx.doi.org/10.1006/hmat.1994.1015.
4. A. G. Agargun, E.M. Ozkan, A historical survey of the fundamental theorem of arithmetic, Hist. Math.
28 (2001) 207214, available at http://dx.doi.org/10.1006/hmat.2001.2318.
5. P. Erdos, J. Suranyi, Topics in the Theory of Numbers. Springer, New York, 2003.
6. Euclid, The Thirteen Books of Euclids Elements. Translated by T. L. Heath. Dover, New York, 1956,
available at http://aleph0.clarku.edu/~djoyce/java/elements/Euclid.html.
7. L. Euler, Theorematum quorumdam arithmeticorum demonstrationes, E98, Comm. Acad. Sci. Petrop. 10
(1738/1747) 125146, available at http://www.eulerarchive.org/.
8. , De numeris amicabilibus, E152, Opusc. Varii Argumenti 2 (1750) 23107, available at http:
//www.eulerarchive.org/.
9. , Vollstandige Anleitung zur Algebra (Petersburg 1770), E387, Reclam, Leipzig, 1920, available
at http://www.eulerarchive.org/.
10. , Tractatus de numerorum doctrina capita sedecim quae supersunt, E792, Comment. Arithm. 2
(1849) 503575, available at http://www.eulerarchive.org/.
11. A. Genocchi, Demonstration de quelques propositions darithmetique dapres Euclide, Nouv. Ann.
Math 13 (1854) 426428, available at http://archive.numdam.org/ARCHIVE/NAM/NAM_1854_1_
13_/NAM_1854_1_13__426_0/NAM_1854_1_13__426_0.pdf
12. C. Goldstein, On a seventeenth century version of the fundamental theorem of arithmetic, Hist. Math.
19 (1992) 177187, available at http://dx.doi.org/10.1016/0315-0860(92)90075-M.
13. K. Hartig, J. Suranyi, Combinatorial and geometric investigations in elementary number theory, Mat.
Lapok 23 (1972) 2529. (Hungarian)
14. , Combinatorial and geometric investigations in elementary number theory, Periodica Mathemat-
ica Hungarica 6 no. 3 (1975) 235240.
15. D. Hilbert, Zahlentheorie, Lecture Notes 18971898. Edited by E. Maus. Univ. Gottingen, Gottingen,
1990.
16. L. Kalmar, A szamelmelet alaptetelerol (Uber den Fundamentalsatz der Zahlentheorie), Mat. Fiz. Lapok
43 (1936) 2745. (Hungarian; German summary)
17. R. Laubenbacher, D. Pengelley, Mathematical Expeditions: Chronicles by the Explorers, Springer, New
York, 1999.
18. F. Lemmermeyer, Zur Zahlentheorie der Griechen: Teil I. Euklids Fundamentalsatz der Arithmetik,
Math. Semesterber. 55 (2008) 181195, available at http://www.rzuser.uni-heidelberg.de/
~hb3/publ/euk-1.pdf.
DAVID PENGELLEY is professor emeritus at New Mexico State University. His research is in algebraic
topology and history of mathematics. He develops the pedagogies of teaching with student projects and with
primary historical sources, and created a graduate course on the role of history in teaching mathematics. He
relies on student reading, writing, and mathematical preparation before class to enable active student work to
replace lecture. He has received the MAAs Haimo teaching award, loves backpacking and wilderness, is active
on environmental issues, and has become a fanatical badminton player.
Department of Mathematical Sciences, New Mexico State University, Las Cruces, NM 88003
davidp@nmsu.edu
876
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120
Linear Algebra via Complex Analysis
Alexander P. Campbell and Daniel Daners
7 (I A)1
1
( a)1 =
a
1
=
( 0 ) + (0 a)
(0 a)1
=
1 + ( 0 )(0 a)1
(1)k (0 a)(k+1) ( 0 )k
X
= (1.1)
k=0
http://dx.doi.org/10.4169/amer.math.monthly.120.10.877
MSC: Primary 15A21, Secondary 47A10; 47A56
For a matrix A, (A) is the set of eigenvalues because the rank-nullity theorem implies
that ker(I A) = {0} if and only if I A is invertible. Replacing a by I A
in (1.1) and absolute value by operator norm (see (2.1)), we might expect for 0 %(A)
that
(1)k (0 I A)(k+1) ( 0 )k
X
(I A)1 = (1.4)
k=0
if
1
| 0 | < . (1.5)
k(0 I A)1 k
That is, if 0 %(A), then (I A)1 can be expanded in a power series about 0
with radius of convergence at least 1/k(0 I A)1 k. Hence, (I A)1 is an analytic
(holomorphic) function of %(A), with Taylor series (1.4) at 0 %(A), and %(A)
is open. In Section 2 we make these arguments rigorous.
A matrix has only finitely many eigenvalues, so they are isolated singularities of
the resolvent. Hence, it is natural to use Laurent expansions about the eigenvalues to
analyze the structure of the resolvent. If 1 , . . . , q are the distinct eigenvalues of A,
then the expansion turns out to be
m j 1
N kj Pj
(1)k B k+1 ( j )k ,
X X
(I A)1 = + + j (1.6)
k=1
( j )k+1 j k=0
1
Z
Pj = (I A)1 d, (1.7)
2i Cj
1 (I A)1
Z
Nj = d. (1.8)
2i Cj j
The Laurent expansion (1.6) is the core of our exposition and is discussed in Section 3.
n j := min{k 1 : N kj = 0} m j . (1.9)
Hence, A is diagonalizable if and only if all of its eigenvalues are simple poles. The
order of j as a pole of the resolvent is, therefore, a measure for how far an operator is
from being diagonalizable. We show in Section 5 that p() = kj=1 ( j )n j is the
Q
minimal polynomial of A, and we prove the CayleyHamilton theorem.
This number is finite for every linear operator on finite dimensional spaces. Note that
every finite dimensional space has a norm induced by the Euclidean norm on Cn and
some isomorphism from V to Cn . As all norms on finite dimensional vector spaces are
equivalent, it does not matter which one we use; see [8, Sections II.13].
The expansion (1.4) was motivated by a geometric series. The counterpart of the
geometric series in operator theory is the Neumann series.
Then k=0 B k converges if r < 1 and diverges if r > 1. Moreover, r kBk. If the
P
series converges, then (I B)1 exists and
Bk .
X
(I B) 1
= (2.2)
k=0
Proof. The root test for the absolute convergence of series implies that nk=0P B k con-
P
verges if r < 1 and diverges if r > 1; see [1, Theorem 8.5]. The partial sum nk=0 B k
satisfies the identity
n n n n
!
k k
Bk B k+1 = I B n+1 .
X X X X
(I B) B = B (I B) = (2.3)
k=0 k=0 k=0 k=0
!
k k
X X
(I B) B = B (I B) = I.
k=0 k=0
Theorem 2.2 (analyticity of resolvent). The resolvent set %(A) is open and the map
7 (I A)1 is analytic on %(A). If 0 %(A), then the power series expansion
(1.4) is valid whenever satisfies (1.5).
Proof. We use a calculation similar to (1.1) with a replaced by A. The difficulty is that
we need to show that (I A) is invertible for close to 0 , so we cannot start with
(I A)1 . In the spirit of (1.1), we write
and then show that we can invert. The first term in (2.4) is of the form I B with
B := ( 0 )(0 I A)1 . By Proposition 2.1, I B is invertible if
We next prove that (A) 6 = by giving an operator theory version of a simple proof
of the fundamental theorem of algebra from [7]. The proof relies only on the Cauchy
integral formula and a decay estimate for (I A)1 . Having proved that (A) 6 = ,
it makes sense to define the spectral radius
of A.
1 (I A)1
Z
A =
1
d
2i ||=R
2 R k(I A)1 k 1
kA1 k
2 R R kAk
for all R > kAk. Letting R , we see that kA1 k = 0, which is impossible. Hence,
(A) 6= , as claimed.
Theorem 3.1. Let 0 (A). Then there exist operators P0 , N0 , and B0 so that for
in a neighbourhood of 0 ,
N0k P0
(1)k B0k+1 ( 0 )k .
X X
(I A)1 = k+1
+ + (3.1)
k=1
( 0 ) 0 k=0
We defer the proof of the theorem to Section 6 and now discuss some consequences.
We show that N0 is nilpotent and deduce that every eigenvalue of A is a pole of the
resolvent.
Next we discuss the structure of the regular and singular parts of the Laurent expan-
sion.
V = im(P0 ) ker(P0 ).
Choose bases of im(P0 ) and ker(P0 ) to form a basis of V . With respect to that basis,
P0 can be written as a block matrix
I 0
P0 = .
0 0
Similarly, with respect to the basis introduced, (ii) and (v) of the theorem imply that
N0 and A are block matrices of the form
N 0 I+N 0
N0 = and A = 0 .
0 0 0 A0
0 0
B0 = .
0 (0 I A0 )1
In particular, the regular part of the Laurent expansion (3.4) is consistent with (1.4),
and coincides with the power series expansion of the resolvent (I A0 )1 about 0 .
Further note that the singular part of the Laurent expansion is trivial on ker(P0 ), and
the regular part is trivial on im(P0 ), so the singular and regular parts live on comple-
mentary subspaces.
is the singular part of the Laurent expansion (3.1), and is valid for all C \ {0 }.
Moreover,
(1)k B0k+1 ( 0 )k
X
(I P0 )(I A)1 = (I A)1 (I P0 ) = (3.4)
k=0
In Theorem 3.1(v) we already see how the Jordan decomposition arises from the
Laurent expansion about 0 , since 0 P0 is diagonalizable on im(P0 ) and N0 is nilpo-
tent. The following proposition is useful to prove uniqueness of the Jordan decompo-
sition.
I A = I D N = (I D) I (I D)1 N . (3.7)
for all n N. Therefore, if K %(D) is compact, then there exists M > 0 such that
for all n N and K ,
As spr(N ) = 0, for every > 0 there exists n 0 N such that, if n > n 0 and K ,
then
N (I D)1 n
1/n MkN n k1/n < . (3.8)
dk
(I D)1 = k!(I D)(k+1) ,
dk
so (I D)(k+1) has primitive (I D)k on %(A) for all k 1. Hence, all integrals
in (3.9) vanish except for the first one, and (3.9) reduces to (3.6).
1
Z
I = P1 + + Pq = (I A)1 d, (4.1)
2i C R
Proof. As R > spr(A), the circle C R encloses all eigenvalues. By the residue theorem
and the Laurent expansion (2.5), we get
q q
1
X X Z
Pj = (I A)1 d
j=1 j=1
2i Cj
1
Z
= (I A)1 d
2i C R
Z
1 X Ak
= d
2i k=0 C R k+1
I 1
Z
= d = I
2i C R
as all terms in the series are zero except the one with k = 0. We next show that P j is a
projection parallel to Pk if k 6 = j. We have
Z Z
2
(2i) P j Pk = (I A) d
1
(I A)1 d
Cj Ck
Z Z
= (I A)1 (I A)1 d d.
Cj Ck
(I A)1 (I A)1
Z Z
(2i)2 P j Pk = d d
C j Ck
1 1
Z Z Z Z
= (I A) 1
d d (I A)1
d d
Cj Ck Ck Cj
= 0,
since the circle C j is outside Ck and vice versa. This completes the proof of (4.2). The
fact that the direct sum reduces A follows from Corollary 3.4.
m
To prove (4.3), note that Theorem 3.1(v) implies that (A j I )m j P j = N j j = 0
and so im(P j ) ker( j I A)m j . By Corollary 3.4, (A j I )m j (I P j ) is injective
on ker(P j ), so ker( j I A)m j im(P j ), proving (4.3).
From Corollary 3.4 we know that A : im(P j ) im(P j ) has j as its only eigen-
value. This motivates the following definition.
Definition 4.2. We call im(P j ) the generalized eigenspace associated with the eigen-
value j and m j = dim(im(P j )) the algebraic multiplicity of j .
The identity (4.3) ensures that Definition 4.2 agrees with the usual definition of the
generalized eigenspace. We now derive a formula for the resolvent in terms of N j and
P j similar to a partial fraction decomposition of a rational function.
Proof. Using Corollary 3.4 and Proposition 4.1, for every %(A)
q
X
(I A)1 = (I A)1 Pj
j=1
q
X
= (I A)1 P j
j=1
q m j 1
X P j
X N kj
= + k+1
j=1
j k=1
( j )
as claimed.
where P j and N j are defined as before. In particular, D and N are uniquely determined
by A. Finally, A is diagonalizable if and only if all eigenvalues of A are simple poles
of (I A)1 .
Pk N j = Pk P j N j = k j N j = N j Pk ,
so the direct sum (4.2) reduces N . In particular, N is nilpotent, since each N j is nilpo-
tent, and also DN = ND, since this is the case on im(P j ).
To show the uniqueness of the decomposition, assume that A = D + N with D di-
agonalizable, N nilpotent, and DN = ND. Proposition 3.5 implies that %( D) = %(A)
and that the spectral projections are equal. Hence, 1 , . . . , q are the eigenvalues of D.
As D is diagonalizable, DP j = j P j and NP j = AP j j P j = N j for j = 1, . . . , q.
Hence, D = D and N = N , as claimed.
The last assertion of the theorem follows, since N = 0 if and only if N j = 0 in
the Laurent expansion (1.6) for all j = 1, . . . , q, which means that all eigenvalues are
simple poles.
To obtain the Jordan canonical form for matrices, it is sufficient to construct a basis
of im(P j ) such that the matrix representation of AP j consists of Jordan blocks; see,
e.g., [2, Theorem 8.47]. For many purposes, the full Jordan canonical form is not
needed, as examples in [6] show.
k
` k
Z
k` (I A)`1 d
X
= (1)
`=0
` C R
Z
= k (I A)1 d
CR
as required.
and therefore
k! p()
Z
p ( j ) =
(k)
d.
2i Cj ( j )k+1
The monic polynomial p of smallest degree such that p(A) = 0 is called the min-
imal polynomial of A. According to Theorem 5.2, it is the polynomial p of smallest
degree with
q n j 1
X p (k) ( j )
N kj = 0,
X
p(A) = p( j )P j + (5.3)
j=1 k=1
k!
where n j is the order of j as a pole of the resolvent given by (1.9). By Remark 3.2
and Proposition 4.1, the set of operators
{P j : 1 j q} {N kj : 1 k n j , 1 j q}
Theorem 5.4 (minimal polynomial). Let A be a matrix over C with distinct eigen-
values 1 , . . . , q . Then the minimal polynomial of A is given by
q
( j )n j ,
Y
p() =
j=1
The preceding theorem shows that the minimal polynomial determines the order of
the poles of the resolvent and vice versa.
and similarly,
(I A) (I A)1 (I A)1 (I A) = (I A) (I A)
= ( )I.
Applying (I A)1 from the left and (I A)1 from the right, we get (ii). Finally,
(iii) follows from (ii) by interchanging the roles of and .
Property (ii) is often referred to as the resolvent identity. It corresponds to the partial
fraction decomposition
1 1
= .
( a)( a) a a
where
1 (I A)1
Z
Bn = d (6.2)
2i Cr ( 0 )n+1
and Cr is a positively oriented circle of radius r centered at 0 , not enclosing any other
eigenvalue of A; see [3, Theorem 8.3.1] or [4, Section 3.11]. We next prove some
recursion relations between the Bn . The aim is to be able to express all Bn in terms of
B2 , B1 , and B0 .
Moreover,
(
Bn1 + 0 Bn if n 6 = 0,
ABn = Bn A = (6.4)
Bn1 + 0 Bn I if n = 0.
In the spirit of a partial fraction decomposition, we use the resolvent identity from
Proposition 6.1(ii) to get
(I A)1 (I A)1
Z Z
(2i)2 Bn Bm = d d
Cr C s ( )n+1 m+1
(I A)1 1
Z Z
= m+1 n+1 ( )
d d
Cs Cr
(I A)1 1
Z Z
n+1 m+1
d d. (6.5)
Cr Cs ( )
and
m
1 1 X 1 1
m+1
= m+1 + (6.7)
( ) ( ) k=0 mk+1 k+1
to evaluate the inner integrals. Note that Cs is outside the circle Cr . Using (6.6) if
n 0, and the Cauchy integral theorem if n < 0, we get
1
1 1 if n 0,
Z
d = n+1
2i Cr n+1 ( )
0 if n < 0.
For the other integral, note that if m 0, then both = 0 and = are singularities
enclosed by Cs . Hence, using (6.7) and the residue theorem, we obtain
1 1 1 1
Z
d = = 0.
2i Cs m+1 ( ) m+1 m+1
1 1 1
Z
d = m+1 .
2i Cs m+1 ( )
1 (I A)1
Z
Bn Bm = d = Bn+m+1 .
2i Cs m+n+2
If m, n < 0, then the first of the inner integrals on the right-hand side of (6.5) is zero,
and the other is 2i(m+1) and therefore
1 (I A)1
Z
Bn Bm = d = Bn+m+1 .
2i Cr m+n+2
1 A(I A)1
Z
ABn = d
2i Cr n+1
1 (I A)1 I
Z
= d
2i Cr n+1
1 (I A)1 I 1
Z Z
= d d
2i Cr n 2i Cr n+1
I 1
Z
= Bn1 d.
2i Cr n+1
This completes the proof of the lemma, since the last integral is zero if n 6 = 0, and is
2i if n = 0.
P0 := B1 and N0 := B2 .
P02 = B1 B1 = B11+1 = B1 = P0 ,
which proves (i). Similarly, applying (6.3) again we get (ii), since
P0 N0 = B1 B2 = B12+1 = B2 = N0 ,
P0 B0 = B1 B0 = 0 = B0 B1 = B0 P0 .
We next use induction to show that Bn = (1)n B0n+1 . This is obvious for n = 0, so
assume that n 1. Then by (6.3) and the induction assumption,
as claimed. Hence, (3.1) follows. To prove (iv), note that from (6.2),
for some constant K > 0 depending on r . The constant K is finite, since the circle
|| = r is compact and the resolvent is continuous. Hence,
REFERENCES
DANIEL DANERS received his Ph.D. from the University of Zurich, Switzerland, in 1992. He currently
teaches at the University of Sydney, where, given the opportunity, he likes to show students interesting mathe-
matics outside the usual curriculum. He likes to spend his free time with his family and children.
School of Mathematics and Statistics, The University of Sydney, NSW 2006, Australia
daniel.daners@sydney.edu.au
Abstract. Flip a coin repeatedly, and stop whenever you want. Your payoff is the proportion
of heads, and you wish to maximize this payoff in expectation. This so-called ChowRobbins
game is amenable to computer analysis, but while simple-minded number crunching can show
that it is best to continue in a given position, establishing rigorously that stopping is optimal
seems at first sight to require backward induction from infinity.
We establish a simple upper bound on the expected payoff in a given position, allowing ef-
ficient and rigorous computer analysis of positions early in the game. In particular, we confirm
that with 5 heads and 3 tails, stopping is optimal.
1 2 1 4 1 1 101
+ + = 0.601,
2 3 4 7 4 2 168
whereas stopping at 3-2 would have given us 3/5 = 0.6. The proposed strategy is not
itself optimal, but it shows that with 3 heads and 2 tails, continuing must be the correct
decision. More elaborate calculations will show that for a number of other positions
too, continuing is better than stopping. On the other hand, no similar calculation will
ever determine when stopping is optimal.
Chow and Robbins [1] established a number of basic properties of the game, like
the nontrivial fact that there is an optimal strategy that stops with probability 1. Precise
asymptotical results were obtained by Aryeh Dvoretzky [2] and Larry Shepp [4]. In
particular, Shepp showed that for the optimal strategy, the proportion of heads required
for stopping after n tosses is asymptotically equal to
1 0.41996 . . .
+ ,
2 n
http://dx.doi.org/10.4169/amer.math.monthly.120.10.893
MSC: Primary 60C05, Secondary 60J22
2. LOWER BOUND ON V (a, n). In position (a, n), we can guarantee payoff a/n
by stopping. Moreover, if a/n < 1/2, then by the recurrence of simple random walk
on Z, we can wait until the proportion of heads is at least 1/2. Therefore,
a 1
V (a, n) max , . (2)
n 2
We can recursively establish better lower bounds by starting from the inequality (2) at a
given horizon, and then working our way backward using (1). An obvious approach
is to let the horizon consist of all positions with n = N for some fixed N . In practice,
it is more efficient
to use (1) only for positions where in addition a n/2, say when
|a n/2| c N for some suitable constant c, and to resort to (2) outside that range.
This allows a greater value of N at given computational resources.
If in this way, we find that V (a, n) > a/n, then in position (a, n), continuing is
better than stopping. For instance, it is straightforward to check (see the discussion
in [3]) that V (2, 3) > 2/3, from which it follows that with 2 heads versus 1 tails, we
should continue.
The third column of Table 1 in Section 4 shows positions where we have determined
that continuing is better than stopping. These results are based on a calculation with a
horizon stretching out to n = 107 . They agree with [3, Section 5] with one exception;
Medina and Zeilberger conjectured, based on calculations with a horizon of 50000
that, in the notation of [2, 3, 4], 127 = 9, meaning that the difference (number of
heads minus number of tails) required in order to stop after 127 flips, is 9. Accord-
ingly, they suggested stopping with 6859, but our computation shows that continuing
is slightly better.
On the other hand, in order to conclude that stopping is ever optimal, we need a
nontrivial upper bound on V (a, n). Clearly, such an upper bound cannot come from
(1) alone, since that equation is satisfied by V (a, n) 1.
3. UPPER BOUND ON V (a, n). We let V (a, n) be the expected payoff from po-
sition (a, n) under infinite clairvoyance, in other words, assuming we have complete
knowledge of the results of the future coin flips and stop when we reach the maxi-
mum proportion of heads. Obviously, V (a, n) V (a, n), so that any upper bound on
V (a, n) is also an upper bound on V (a, n).
1
r
1
.
4 N 2h
That is, h (2/ ) N . In our computations leading to the results of Table 1
(with N = 107 ), we have taken h = 3568. The second column of Table 1 lists positions
where we have determined that stopping is optimal. This includes 5 heads to 3 tails,
a position discussed in [3] and for which computational evidence [3, 6] strongly sug-
Table 1. Opening theory for the first 1000 steps of the ChowRobbins game. If the difference (number of
heads number of tails) is nonpositive, we always continue. If the difference is 27 or more and the total
number of flips is at most 1000, stopping is optimal. For differences from 1 to 26, stopping is optimal up to and
including the position in column 2, while continuing is optimal from the position in column 3 and so on. There
are seven positions in this range for which we havent determined the optimal decision: 116104, 200184,
278259, 337316, 436412, 472447, and 509483.
Difference Stop with But go with Difference Stop with But go with
1 10 21 14 155141 156142
2 53 64 15 176161 177162
3 96 107 16 199183 201185
4 1612 1713 17 224207 225208
5 2318 2419 18 250232 251233
6 3226 3327 19 277258 279260
7 4235 4336 20 306286 307287
8 5446 5547 21 336315 338317
9 6758 6859 22 368346 369347
10 8272 8373 23 401378 402379
11 9887 9988 24 435411 437413
12 115103 117105 25 471446 473448
13 134121 135122 26 508482 510484
5. PROOF OF THE MAIN THEOREM. For a and n as before, and p [0, 1], let
P(a, n, p) denote the probability that, starting from position (a, n), at some point now
or in the future the total proportion of heads will strictly exceed p. In other words,
P(a, n, p) is the probability of success starting from (a, n) if, instead of trying to
maximize expected payoff, we try to achieve a proportion of heads exceeding p, and
continue as long as this has not been achieved. When p is rational, P(a, n, p) is al-
gebraic and can in principle be calculated with the method of [5]; but we need an
inequality that can be analyzed averaging over p.
Key Lemma. Suppose that in position (a, n), the nonnegative integer k is such that
at least k more coin flips will be required in order to obtain a proportion of heads
exceeding p. Then
1
P(a, n, p) . (4)
(2 p)k
Proof. We can assume that p > max(a/n, 1/2), since otherwise the statement is triv-
ial. One way of proving that an event is unlikely is to show that conditioning on it
causes strange things to happen. From position (a, n), we condition on the event that
the total proportion of heads will at some later point exceed p. We want to show that
this considerably increases the probability of an immediate run of k consecutive heads.
Our next task is to use the Key Lemma to estimate V (a, n). We have
Z 1 Z 1
a 1
V (a, n) = P(a, n, p) dp = max , + P(a, n, p) d p. (5)
0 n 2 max(a/n,1/2)
If p > max(a/n, 1/2), then the requirement that at least k more coin flips are needed
to obtain a proportion of heads exceeding p is equivalent to
a+k1
p,
n+k1
which we rearrange as
np a
k 1+ .
1 p
np a np a
k 1+ ,
1 p 1 p
we conclude using (4) that for p in the range max(a/n, 1/2) < p < 1 of integration
in (5),
1
P(a, n, p) npa .
(2 p) 1 p
log(1 + t)
t,
1t
we obtain
Z 1
a 1 1 dt
V (a, n) max , + (1+t)n2a
n 2 2 (1 + t) 1t
max 2an
n ,0
(1 + t)n 2a
Z 1
a 1 1
= max , + exp log(1 + t) dt
n 2 2 max 2an 1t
n ,0
1 1
Z
a 1
max , + exp ((1 + t)tn + 2at) dt. (6)
n 2 2 max 2an
n ,0
By putting u = t n and replacing the upper bound of integration by infinity, we arrive
at
Z
a 1 1 2a n
V (a, n) max , + exp u +
2
u du. (7)
n 2 2 n max 2an n
,0
n
Now, notice that by the substitution w = u (2a n)/ n,
Z Z
2a n 2a n
exp u 2 + u du = exp w 2 w dw. (8)
2an
n 0 n
n
The bound (9) can be used directly in computations by first tabulating values of the
integral, but we have chosen to simplify the error term further. We can discard either
of the two terms inside the exponential in (9). On the one hand, the error term is at
most
1
Z r
1 2
.
exp u du =
2 n 0 4 n
suggests that there might be a more direct proof. There might be a simple function
F(a, n) for which we can verify that
a F(a, n + 1) + F(a + 1, n + 1)
F(a, n) max , . (10)
n 2
It would then follow by backward induction that V (a, n) F(a, n). Unfortunately,
(10) fails for small a with F(a, n) = max(a/n, 1/2) + 1/(2 |2a n|), as well as with
F(a, n) equal to the right-hand side of (3). Numerical evidence suggests that (10)
holds for all a and n if we set F(a, n) equal to the right-hand side of (9), but we have
been unable to verify this.
The Main Theorem allows us to calculate V (a, n) to any desired precision. This is
because (1) has the property that if V (a, n + 1) and V (a + 1, n + 1) are both known
with an error of at most , then the same is true of V (a, n). To obtain the desired level
of precision, we therefore only need to start our calculation from a horizon where the
error term in (3) is sufficiently small.
On the other hand, it is difficult to say in advance how far we have to take our
computations in order to find the optimal decision in a given position, as the expected
payoff on continuing may be very close to the payoff a/n on stopping. For instance,
we have no idea how hard it will be to find the optimal decision in the position 116
104 (the first one whose status we havent determined). For all we know, the question
of whether stopping is optimal in this position might be undecidable by our method,
although that would require the expected payoff on continuing to be, miraculously,
exactly equal to the payoff on stopping.
REFERENCES
1. Y.-S. Chow, H. Robbins, On optimal stopping rules for sn /n, Ill. J. Math. 9 (1965) 444454.
2. A. Dvoretzky, Existence and properties of certain optimal stopping rules, Proceedings of the Fifth Berkeley
Symposium on Mathematical Statistics and Probability, 1:441452. Univ. of California Press, Berkeley,
CA, 1967.
3. L. A. Medina, D. Zeilberger, An Experimental Mathematics Perspective on the Old, and still Open, Ques-
tion of When To Stop?, in Gems in Experimental Mathematics, Contemporary Mathematics series v. 517
(AMS), Edited by T. Amdeberhan, L. Medina, and V. Moll, 265274, also available at arXiv:0907.0032v2
[math.PR].
4. L. A. Shepp, Explicit solutions to some problems of optimal stopping, The Annals of Mathematical Statis-
tics 40 (1969) 9931010, available at http://dx.doi.org/10.1214/aoms/1177697604.
5. W. Stadje, The maximum average gain in a sequence of Bernoulli games, Amer. Math. Monthly, December
(2008) 902910.
6. J. D. A. Wiseman, The Chow & Robbins Problem: Stop at h = 5 t = 3, available at http://www.
jdawiseman.com/papers/easymath/chow_robbins.html.
OLLE HAGGSTROM received his Ph.D. in 1994 from Chalmers University of Technology, Gothenburg,
where he is currently a professor. He has received numerous prizes and awards for his work on percolation
theory, but his interests also include philosophy and climate science. To the general public, he is best known
for his books and articles vigorously defending secular humanism and the scientific method against superstition
and pseudoscience.
Department of Mathematical Sciences, Chalmers University of Technology, 412 96 Gothenburg, Sweden
olleh@chalmers.se
Proof. Let (xn ) be any sequence, xn 6 = c, such that xn c. Then, either there
exists an n 0 such that for all n n 0 , g(xn ) 6 = g(c), or there exists a subsequence
(xn k ) such that g(xn k ) = g(c) for all k. The former case implies that for n n 0 ,
f (g(xn k )) f (g(c))
0 = f 0 (g(c))g 0 (c).
xnk c
f (g(x)) f (g(c))
This implies that limxc xc
= f 0 (g(c))g 0 (c).
REFERENCES
1. P. F. McLoughlin, A simple proof of the chain rule, Amer. Math. Monthly, 120 (2013) p. 94.
Abstract. We prove a generalization of both Pascals Theorem and its converse, the
BraikenridgeMaclaurin Theorem: If two sets of k lines meet in k 2 distinct points, and if
dk of those points lie on an irreducible curve C of degree d, then the remaining k(k d)
points lie on a unique curve S of degree k d. If S is a curve of degree k d produced in
this manner using a curve C of degree d, we say that S is d-constructible. For fixed degree
d, we show that almost every curve of high degree is not d-constructible. In contrast, almost
all curves of degree 3 or less are d-constructible. The proof of this last result uses the group
structure on an elliptic curve and is inspired by a construction due to Mobius. The exposition
is embellished with several exercises designed to amuse the reader.
A a
B
b
C c
Figure 1. Two syzygies: the centroid (left) and Pappuss configuration (right)
Pappuss Theorem, which dates from the fourth century A . D ., describes another
syzygy. It is one of the inspirations of modern projective geometry.
http://dx.doi.org/10.4169/amer.math.monthly.120.10.901
MSC: Primary 14H50, Secondary 14H52
Pappuss Theorem has inspired a lot of amazing mathematics. The first chapter
of a fascinating new book by Richter-Gebert [18] describes the connections between
Pappuss Theorem and many areas of mathematics, including cross-ratios and the
GrassmannPlucker relations among determinants.
Pappuss Theorem appears in his text Synagogue [17], a collection of classical
Greek geometry with insightful commentary. David Hilbert observed that Pappuss
Theorem is equivalent to the claim that the multiplication of lengths is commutative
(see, e.g., Coxeter [3, p. 152]). Thomas Heath believed that Pappuss intention was
to revive the geometry of the Hellenic period [11, p. 355], but it wasnt until 1639
that the sixteen-year-old Blaise Pascal generalized Pappuss Theorem [4, Section 3.8],
replacing the two lines with a more general conic section.
Pascals Theorem has an interesting converse named after the British mathemati-
cians William Braikenridge and Colin Maclaurin. Braikenridge and Maclaurin seem
to have arrived at the result independently, though they knew each other and their cor-
respondence includes a dispute over priority.
(x, y, z) (x, y, z) 6 = 0,
then each equivalence class corresponds to a line in R3 through the origin. We denote
the equivalence class of points on the line through (x, y, z) by [x : y : z]. This is a
sensible notation, since the ratios between the coordinates determine the direction of
the line. Returning to our earlier model of P2 , the points with z 6 = 0 correspond to
points in our usual copy of R2 , while the points with z = 0 correspond to points at
infinity.
If points in P2 correspond to lines through the origin, then what do lines in P2
look like? If we once again identify the plane z = 1 with R2 , we see that the points
The polynomial needs to be homogeneous (all terms in the polynomial have the same
degree) in order for the curve to be well-defined (see Exercise 4.2 below). It is tradi-
tional to call degree-d homogeneous polynomials degree-d forms. The curve C is said
to be a degree-d curve when F(x, y, z) is a degree-d polynomial. We say that C is an
irreducible curve when F(x, y, z) is an irreducible polynomial. When F(x, y, z) fac-
tors, then the set C is actually the union of several component curves, each determined
by the vanishing of one of the irreducible factors of F(x, y, z).
Exercise 4. If this is the first time youve met projective space, you might try these
exercises to get a feel for projective space.
1. Show that the line ax + by + cz = 0 in P2 consists of all the points of the form
[x : y : 1] such that ax + by + c = 0, together with a single point at infinity (the
point [b : a : 0]). We say that the line ax + by + cz = 0 is the projectivization
of the line ax + by + c = 0. Now show that the projectivizations of two parallel
lines ax + by + c = 0 and ax + by + d = 0 in R2 meet at a point at infinity.
2. The projectivization of the hyperbola x y = 1 in R2 is the set of points in P2
that satisfy x y z 2 = 0. Show that whether a point [x : y : z] lies on the pro-
jectivization of the hyperbola or not is a well-defined property (i.e., the answer
doesnt depend on which representative of the equivalence class [x : y : z] we
use). Where does the projectivization meet the line at infinity?
3. Show that if a1 x + b1 y + c1 z = 0 and a2 x + b2 y + c2 z = 0 are two distinct lines
in P2 , then they meet in a point P = [a3 : b3 : c3 ] whose coordinates are given
by the cross product,
Projective space P2 enjoys many nice properties that Euclidean space R2 lacks.
Many results are easier to state and more elegant in projective space than in Euclidean
space. For instance, in Euclidean space two distinct lines meet in either one point or in
no points (in the case where the two lines are parallel). By adding points at infinity to
Euclidean space, weve ensured that any two distinct lines meet in a point. This is just
the first of a whole sequence of results encapsulated in Bezouts Theorem.
Bezouts Theorem requires that we work in complex projective space; in P2R , two
curves may not meet at all. For instance, the line y 2z = 0 misses the circle x 2 +
y 2 z 2 = 0 in P2R ; the points of intersection have complex coordinates. In the rest
of the paper, well work in complex projective space (denoted P2 ) so that we can
take advantage of Bezouts Theorem. The points of P2 correspond to one-dimensional
subspaces of C3 .
To say what it means to count points appropriately, requires a discussion of in-
tersection multiplicity. This can be defined in terms of the length of certain modules
[8], but an intuitive description will be sufficient for our purposes. When two curves
meet transversally at a point P (there is no containment relation between their tangent
spaces), then P counts as 1 point in Bezouts Theorem. If the curves are tangent at
P or if one curve has several branches passing through P, then P counts as a point
with multiplicity. One way to determine the multiplicity of P is to look at well-chosen
families of curves C1 (t) and C2 (t) so that C1 (0) = C1 and C2 (0) = C2 , and to count
how many points in C1 (t) C2 (t) approach P as t goes to 0. For instance, the line
y = 0 meets the parabola yz = x 2 in one point P = [0 : 0 : 1]. Letting C1 (t) be the
family of curves y t 2 z = 0 and letting C2 (t) be the family consisting only of the
parabola, we find that if t 6 = 0, then C1 (t) C2 (t) = {[t : t 2 : 1], [t : t 2 : 1]}; so two
points converge to P as t goes to 0. In this case, P counts as two points. The reader
interested in testing their understanding could check that the two concentric circles
x 2 + y 2 z 2 = 0 and x 2 + y 2 4z 2 = 0 meet in two points, each of multiplicity two.
More details can be found in Fultons lecture notes [8, Chapter 1].
a0 x 2 + a1 x y + a2 x z + a3 y 2 + a4 yz + a5 z 2 = 0. (1)
Multiplying the formula by a nonzero constant gives the same curve, so the curve C
can be identified with the point [a0 : a1 : a2 : a3 : a4 : a5 ] in P5 . More generally, letting
Rd be the vector space of degree-d homogeneous polynomials in three variables, the
degree-d curves in P2 are identified with points in the projective space P(Rd ), where
we identify polynomials if they are nonzero scalar multiples of one another. A basis of
Rd is given by the D = d+2 2
monomials of degree d in three variables, so the degree-d
curves in P2 are identified with points in the projective space P(Rd ) = P D1 .
2
Returning to the case of degree-2 curves in P , if we require C to pass through
a given point, then the coefficients a0 , . . . , a5 of C must satisfy the linear equation
produced by substituting the coordinates of the point into (1). Now, if we require C to
pass through five points in P2 , then the coefficients must satisfy a homogeneous system
of five linear equations in six variables. The Rank-Nullity Theorem shows that such a
system always has a non-trivial solution: There is a conic through any five points in
P2 . If the points are in general position (so that the resulting system has full rank), then
the system has a one-dimensional solution space and so there is a unique conic passing
through all five points (see [1] for details).
The Zariski topology is the coarsest topology that makes polynomial maps from
Pm to Pn continuous. More concretely, every homogeneous polynomial F in n + 1
variables determines a closed set in Pn
and every closed set is built by taking finite unions and arbitrary intersections of such
sets. Closed sets in the Zariski topology are called varieties. The nonempty open sets
in this topology are dense; such a set is not contained in a proper subset of the form
V(F). Well say that a property holds for almost every point in Pn if it holds on a
dense Zariski-open set in Pn . More details on the Zariski topology can be found in
Shafarevich [19, Section 4.1].
1 Both the MathSciNet and Zentralblatt reviews of the English translation [2] are entertaining. The assess-
ment in the MathSciNet review is atypically colorful: This is not a book to be taken to the office, but to be
left at home, and to be read on weekends, as a romance, while the review in Zentralblatt Math calls it an
immortal evergreen of astonishing actual relevance.
Theorem 6. Let F(x, y, z) = 0 and G(x, y, z) = 0 define two curves of degree k and
assume that these curves meet in a set 0 of k 2 distinct points. If H (x, y, z) = 0 defines
an irreducible curve C of degree d > 0 that passes through kd of the points in 0, then
there is a curve S = 0 of degree k d that passes through the remaining k(k d)
points in 0 \ C. Moreover, S = 0 is the unique curve of degree k d containing 0 \ C,
if G factors into distinct linear forms G = G 1 G k .
The first part of the following proof is due to Kirwan [13, Theorem 3.14]. Katz [12,
Theorem 3.1] gives the proof of the case of Max Noethers Theorem discussed in the
second paragraph.
The degree-k curve X meets the degree-d curve C in at least kd + 1 points, namely the
kd points of 0 C and the point [a : b : c], so by Bezouts Theorem, C and X must
share a common component. Since C is irreducible, M(x, y, z) = H (x, y, z)S(x, y, z)
for some degree k d form S. Since M vanishes on 0, the curve defined by S = 0
must contain all the k(k d) points of 0 off C.
Now, we assume that G has k distinct linear factors G 1 , . . . , G k , and show that
any degree-k form N defining a curve that contains 0 must satisfy N = a F + bG
for suitable constants a and b. This is a special case of a result that Max Noether
called the Fundamental Theorem of Algebraic Functions, though today it is known by
a more technical name, the AF + BG Theorem. Both N and F have the same zeros
when restricted to G 1 = 0, so for some constant a, N a F vanishes identically on
the first line G 1 = 0, and N a F = G 1 Q k1 for some degree k 1 form Q k1 . Now,
Q k1 vanishes at all points of 0 off the first line G 1 = 0. In particular, for each of
the remaining lines G i = 0, Q k1 vanishes at k points on the line, hence G i divides
Q k1 . Since the forms G 2 , . . . , G k are relatively prime, Q k1 = bG 2 G k for some
constant b and N = a F + bG.
Now, if S1 (x, y, z) and S2 (x, y, z) are two forms of degree k d vanishing on
0 \ C, then H S1 = a1 F + b1 G and H S2 = a2 F + b2 G for constants a1 , a2 , b1 , and b2 .
Then H must divide both b2 H S1 b1 H S2 = (b2 a1 b1 a2 )F and a2 H S1 a1 H S2 =
(a2 b1 a1 b2 )G. Now if a2 b1 a1 b2 6 = 0, then the curve C given by H = 0 must be
contained in the set of points 0 = (F = 0) (G = 0), a contradiction. So a2 b1
a1 b2 = 0 and the forms H S1 and H S2 are scalar multiples of one another. It follows
that the curves defined by S1 = 0 and S2 = 0 are identical.
In the rest of the paper we will be interested in the case where both of the forms
F and G factor completely into linear forms, in which case their zero-sets determine
collections of k blue and k red lines, respectively. If the polynomial HS in Theorem 6
also factors completely into linear forms, then the resulting arrangement of linesin
which each point of intersection lies on a line from HS = 0, a red line, and a blue line
is called a multinet, an intriguing combinatorial and geometric object in the theory of
hyperplane arrangements (see Falk and Yuzvinsky [7] for details on the connection
between multinets and resonance varieties).
Theorem 7. Suppose that k red lines and k blue lines meet in a set 0 of k 2 distinct
points, with 2k points of 0 lying on a conic Q. If a line L contains k 1 of the k 2 2k
points of 0 off Q, then L contains one other point of 0 off Q as well.
To see the connection with Mobiuss result, suppose that a polygon with 2k =
4n + 2 sides is inscribed in an irreducible conic. Working around the perimeter of the
polygon, color the edges alternately red and blue. Since there are 4n + 2 sides, oppo-
site sides have opposite colors. Extending the edges to lines, consider the k = 2n + 1
points of intersection of the pairs of opposite sides. If k 1 = 2n of these points lie
on a line L, then Theorem 7 shows that another of the points in 0 lies on L as well.
Since the points of 0 are distinct, the only possibility is that the remaining pair of
corresponding edges intersect on the line L.
Mobius also proved a result involving two polygons inscribed in a conic. Consider
two polygons P1 and P2 , each with 2k edges, inscribed in a conic, and associate one
edge from P1 with one edge from P2 . Working counterclockwise in each polygon, as-
sociate the other edges of P1 with the edges of P2 . Extending these edges to lines,
Mobius proved that if 2k 1 of the intersections of pairs of corresponding edges lie
on a line, then the last pair of corresponding edges also meet in a point on this line.
Assuming that the points of intersection of the lines are distinct, this result also fol-
lows from Theorem 6. A similar construction using a pair of inscribed polygons will
reappear when we consider constructible cubics in the next section.
The two curves S and C are said to be directly linked via the set 0. The notion of
linkage has important applications in the study of curves. Indeed, special properties of
one curve are reflected in special properties of the other curve. This point of view leads
to the beautiful subject of liaison theory. The last chapter of Eisenbud [5] introduces
this advanced topic in commutative algebra; more details can be found in Migliore and
Nagels notes [14].
Well restrict our attention in the rest of the paper to the question of which curves
are d-constructible. A simple dimension count shows that most curves of high degree
are not d-constructible, so the d-construction is not dense in high degrees.
The curves
Proof.
t+2 2
of degree t are parameterized by a projective space of dimension
2
1 = (t + 3t)/2. Lets try to parameterize the set of d-constructible curves of
It is easy to see that all lines are d-constructible. For instance, if L is a line, then
choose any set of d + 1 points on L and pick red and blue lines that pass through
these points and meet in (d + 1)2 distinct points. Then Theorem 6 shows that there
exists a curve C of degree d passing through the remaining points, showing that L is
d-constructible.
As well, for all d > 0, the d-construction is dense in degree 2. The defining poly-
nomial of any conic can be expressed in the form [x, y, z]A[x, y, z]T , where A is a
symmetric matrix. The conic is irreducible if and only if rank(A) = 3. So the set of ir-
reducible conics is Zariski-open; it is the complement of the hypersurface det(A) = 0
in P(R2 ) = P5 . Now it is easy to show that any irreducible conic Q is d-constructible.
Just inscribe a polygon with 2(d + 2) edges in Q, color the edges alternately red and
blue and, if necessary, move the vertices so that the extensions of the red and blue
edges meet in distinct points, 0. Theorem 6 shows that the points in 0 that lie off Q
form a degree-d curve, so Q is d-constructible.
Using the group law on elliptic curves allows us to show that for each d, the d-
construction is dense in degree 3. An elliptic curve is a smooth plane curve of degree
3. In particular, each elliptic curve is irreducible; it is not the union of other curves.
The points on a fixed elliptic curve E form an abelian group; the sum of three distinct
points is equal to the identity element in the elliptic curve group if and only if they are
collinear.3 Figure 3 illustrates the group law on the elliptic curve E given by y 2 z
x 3 + x z 2 = 0. The point at infinity [0 : 1 : 0] E serves as the identity element 0 E .
The three points A, B, and C are collinear, so A + B + C = 0 E and C = (A + B).
The vertical line through C and 0 E meets the curve in one more point D, and since
C + 0 E + D = 0 E , D equals C = A + B.
Proof. Given an elliptic curve E, we produce a set of d + 3 blue lines and d + 3 red
lines meeting in a set of (d + 3)2 distinct points 0 with 3(d + 3) of them on E. Then
by Theorem 6, there exists a curve S of degree d through the points on 0 \ E, and so
E is d-constructible.
To produce the red and blue lines, we start with d + 4 properly selected points
A0 , B0 , P1 , P2 , . . . , Pd+2 on E (well say more on how to pick the points later). Well
3 Two of the points are the same if and only if the line is tangent to E at this point. All three points are equal
(to Q, say) if and only if Q is a flex pointthe tangent line to E at Q intersects E with multiplicity 3.
C
A
B
x
D=A+B
A2 Ad
A1 Ad+1
P2 Pd+1
P1 Pd+2 Ad+2
A0
Q Q
Bd+2 Pd+2 P1 B0
Pd+1 P2
Bd+1 B1
Bd B2
Figure 4 suggests that the line through Ad+2 and B0 meets E at the same point, Q,
where the line through Bd+2 and A0 meets E. To see this, first note that
A0 + A1 + P1 = 0 E A1 = A0 P1 ,
A1 + A2 + P2 = 0 E A2 = A1 P2 = A0 + P1 P2 ,
..
.
Ad+1 + Ad+2 + Pd+2 = 0 E Ad+2 = A0 + P1 P2 + Pd+2 ,
Q = B0 Ad+2
= B0 (A0 + P1 P2 + Pd+2 )
= A0 (B0 + P1 P2 + Pd+2 )
= A0 Bd+2 .
It follows that Q E is collinear with B0 and Ad+2 , as well as with A0 and Bd+2 .
Note that each point As and each point Bs lie on the intersection of one blue and
one red line. Because d is even, the number of edges from A0 to B0 is odd, so that each
point Ps (and the point Q) also lie on both a red and a blue line. So the red and blue
lines intersect E in the subset 0 consisting of all the points As and Bs , together with
the points P1 , . . . , Pd+2 , and Q.
Figure 4 is misleading when d is odd, because there are an even number of edges
from A0 to B0 , so each point Ps occurs on two lines of the same color. In this case, we
adopt Mobiuss approach: We inscribe two polygons in the elliptic curve, each with
d + 3 vertices, as in Figure 5.
B2 A2
B1 A1
P2 P2
P3 P3
P1 A3
P1 B3
A0 B0
P4 P4
Q
Q
B4 A4
Ad+2 Pd+2
Bd+2
Pd+2 Pd+1
Pd+1
Ad
Bd Ad+1
Bd+1
from which we conclude the collinearity claims. It follows that when d is odd, the red
and blue lines intersect the elliptic curve E in the subset 0 consisting of all the points
As and Bs , together with the points P1 , . . . , Pd+2 , and Q.
Bd+1 Ad+2 A0
Pd+2 Q
Ad+1 B0
Bd+2
Thus, Pd+2 lies on a line joining Ad+1 or Bd+1 to a point of 1. This case includes the
situation where Q E lies on a previously constructed line, because then the point of
intersection is in E and so is in 0d+1 . If one of the new lines passes through a point of
intersection of the previous lines, then again Pd+2 must lie on a line joining Ad+1 or
Bd+1 to a point of 1. If Q = Ad+2 , then 2Ad+2 + B0 = 0 E , so
It follows that Pd+2 lies on a line joining either Bd+1 or Ad+1 to a point of 3 (sim-
ilarly, if Q = Bd+2 , then the same conclusion holds). Finally, if Q = Pd+2 , then
A0 + Q + Bd+2 = 0 E = Bd+1 + Pd+2 + Bd+2 and canceling terms gives A0 = Bd+1 ,
which is impossible by our construction, since all the points of 0d+1 are distinct.
So the points in 0 = 0d+2 {Q} are distinct as long as Pd+2 does not lie on a line
joining Ad+1 or Bd+1 to a point in 0d+1 1 3 or on a line through Ad+1 or Bd+1
that is tangent to E. By avoiding poor choices of the points Ps , we can ensure that all
the red lines intersect the blue lines in distinct points. This completes the proof that
each elliptic curve is d-constructible.
Proof. To show that the d-construction is dense in degree 3, it is enough to show that
the elliptic curves form a dense open set in the set P(R3 ) = P9 parameterizing all
degree-3 curves. This is well known, but we sketch the proof. Consider the set
One way to check whether a point P is singular on the level curve F = 0 is to check
whether F(P) is zeroin this case, there is no well-defined tangent line (see Ex-
ercise 4.5). Since F(P) is a bihomogeneous polynomial, the set C is Zariski-closed
in the product of projective spaces P(R3 ) P2 (see Exercise 4.6 and Shafarevich [19,
Section 5.1] for details). Now, the image of a projective variety under the projection
1 : P(R3 ) P2 P(R3 ) is also Zariski-closed (see Shafarevich [19, Section 5.2] for
details), so the set 1 (C ) of singular degree-3 curves is Zariski-closed. It follows that
the set of smooth (nonsingular) curves is Zariski-open, and hence dense, in P(R3 ).
Theorem 9 suggests that for all d, the d-construction is also dense in degree 4.
Proving this result seems to require that we inscribe polygons in degree-4 curves, but
this appears to be difficult to do in general.
ACKNOWLEDGMENTS. I am grateful for conversations with my colleagues Mark Kidwell, Mark Meyer-
son, Thomas Paul, and Max Wakefield, and with my friends Keith Pardue and Tony Geramita. Amy Ksir and
Jessica Sidman provided very useful comments on an early draft, and several referees gave excellent feedback
that considerably improved the exposition. Many computations and insights were made possible using the
excellent software packages Macaulay2, GeoGebra, Sage, and Maple.
REFERENCES
1. A. Bashelor, A. Ksir, W. Traves, Enumerative algebraic geometry of conics, Amer. Math. Monthly 115
(2008) 701728.
WILL TRAVES grew up in Toronto, Canada and moved to the United States during graduate school. His
paper [1] in this M ONTHLY, co-authored with Andy Bashelor and Amy Ksir, won both the Lester R. Ford
and Merten M. Hasse prizes. He joined the faculty of the United States Naval Academy in 1999 and is a
brown-dot Project NExT fellow.
U.S. Naval Academy, Math Department, Mail Stop 9E, Annapolis, MD, 21402
traves@usna.edu
Abstract. Following A. Cuoco, a proof of the Law of Cosines is given using squares con-
structed on sides of the triangle. Then the configuration consisting of the original triangle and
the triangle defined by the centers of these squares is studied, and several remarkable relation-
ships are found.
1. THE LAW OF COSINES. The loveliest proof that I know for the Law of Cosines
was shown to me by Al Cuoco [1, pp. 489490]. It is a direct generalization of Euclids
proof of the Pythagorean Theorem, but it seems not to be well known. (For example, it
was not among the many proofs offered by Wikipedia as of May, 2012.) It proceeds by
drawing three squares on the sides of a triangle 1ABC (see Figure 1). It then draws the
three altitudes of 1ABC, and continues them through the squares, dividing each square
into two rectangles. The first notable fact about this configuration is the following.
J
K
b I
L C a
H
E D
b
G
a
A F B
c c
M N O
Figure 1. The Law of Cosines
Theorem 1.1A (A. Cuoco). The pair of rectangles that share a given vertex with
1ABC have equal area.
Theorem 1.1B. The area of these two rectangles is the product of the lengths of the
two sides adjoining the vertex, times the cosine of the angle at the vertex.
http://dx.doi.org/10.4169/amer.math.monthly.120.10.916
MSC: Primary 51M04, Secondary 51M15; 51-01
Proof of the corollary. Let the angles of 1ABC at the vertices A, B, and C be , ,
and , respectively. Let the sides opposite vertices A, B, and C have lengths a, b,
and c, respectively. The Law of Cosines expresses the square of the length of one side
in terms of the squares of the lengths of the other two sides and the angle between
them. For example, if the chosen side is c, it says c2 = a 2 + b2 2ab cos . We will
establish the equivalent equation a 2 + b2 c2 = 2ab cos .
Now look at Figure 1. The quantities a 2 , b2 , and c2 are represented by the areas of
the squares on the sides of 1ABC. In particular, c2 is the area of the square on AB, and
a 2 + b2 is the sum of the areas of the squares on the other two sides. Since the square
on AB is the union of rectangles AFNM and BFNO, which respectively have the same
areas as rectangles AEKL and BDHG, we see that the difference a 2 + b2 c2 is the
sum of the areas of rectangles CDHI and CEKJ, both of which have area ab cos
according to Theorem 1.1B. Thus, a 2 + b2 c2 = 2ab cos , as desired.
We will give a proof of Theorem 1.1B by direct computation, and then we will give
a proof of Theorem 1.1A that does not involve computation, and is very much in the
spirit of Euclids proof of the Pythagorean Theorem.
Proof of Theorem 1.1B. We will indicate the length of a given segment by vertical bars
enclosing the endpoints. Thus, for example, we would indicate the length of the side
AC by |AC| (so that |AC| = b).
Let D, E, and F be the feet of the altitudes of 1ABC, as shown in Figure 1. Then
we see, for example, that AF is a leg of the right triangle 1ACF, whose hypotenuse is
AC. The leg AF is adjacent to CAF = CAB = . By definition of cosine, we see
that |AF| = |AC| cos = b cos . Similar considerations give the formulas
From these formulas, we easily calculate the areas of the pairs of rectangles AFNM and
AEKL, BFNO and BDHG, and CDHI and CEKJ by length times width, with results as
stated in Theorem 1.1B. For example, for rectangle AFNM, the area is the product of
the lengths |AM| = |AB| = c and |AF| = b cos .
Proof of Theorem 1.1A. See Figure 2. Through corner M of square AMOB, draw a
line parallel to side AC of 1ABC. Let this line intersect the extended altitude CN in
point P. Likewise, through corner L of square ALJC, draw a line parallel to side AB of
triangle 1ABC, and let this line intersect the extended altitude BK in a point Q. Then:
(i) Rectangle AFNM has the same area as parallelogram ACPM, since they have
the same base (AM), and the same altitude (e.g., AF);
(ii) parallelogram ACPM is congruent to parallelogram ALQB (aka ABQL) by a
rotation through a right angle around point A, since this rotation takes AM to
AB, and AC to AL, and these segments determine the parallelograms; and
(iii) rectangle AEKL has the same area as parallelogram ABQL (aka ALQB), since
they have the same base (AL) and the same altitude (e.g., AE).
L Q
C
A F B
M N O
Figure 2. Euclidean style proof
Combining (i), (ii), and (iii), we conclude that rectangles AFNM and AEKL have the
same area, as claimed. Equality for the other pairs of rectangles is proved similarly.
Remark. Readers familiar with Euclids proof of the Pythagorean Theorem will see
the parallels to this one. Differences are: (i) This argument uses the full parallelograms
rather than the triangles formed by dividing each parallelogram in two by a diagonal;
and (ii) it is explicit in its use of the 90 rotation around A, while Euclid suppressed
transformations.
Theorem 2.1. If squares are built on the (out)sides of a quadrilateral, then the line
segments joining the centers of the squares on opposite sides of the quadrilateral are
equal and mutually perpendicular.
Proof. This curious fact has an elegant proof using transformations. Let the quadrilat-
eral be ABCD, and let the centers of the squares be R, T , U , and S (see Figure 3).
Given a point p and an (counterclockwise oriented) angle , let p, be the rotation
around p through the angle . Consider the product
8 = S, 2 U, 2 T, 2 R, 2 .
R
Z
D
U A
B
Applying these facts in succession, we find that 8(C) = C; that is, C is fixed by 8.
But a translation that fixes a point is the identity. Therefore,
S, 2 U, 2 T, 2 R, 2 = , (2.2)
T, 2 R, 2 = U, 2 S, 2 . (2.3)
The transformations on either side of equation (2.3) have total rotation angle ; thus,
they are half turns. Since they are equal, they are both a half turn around the same
point.
In [2, III.1, Ex. 1.6], the following method is explained for finding the center of
rotation of a product of rotations. Take rotations p, and q, . Draw the line segment
connecting the centers p and q. At p, draw a line ` that makes angle 2 with pq
(oriented from pq to `). At q, draw a line m, making an angle 2 with pq (oriented
from m to pq). Then the product p, q, is a rotation through angle + around the
point of intersection r = ` m.
For our case, in which = 2 = , the triangle pqr will be an isosceles right tri-
angle with pq as hypotenuse. Thus, the equality (2.3) says that the (appropriately
oriented) isosceles right triangles with RT and US as hypotenuses have the same third
vertex. Call it Z . See Figure 3 again.
Draw the isosceles right triangles 1ZRT and 1ZUS. Evidently, rotating around Z
through a right angle will take U to S, and will take R to T , so it will take UR to ST.
We can apply Theorem 2.1 to triangle geometry by imagining that the quadrilateral
DCBA degenerates into a triangle by letting D approach A, so that DA shrinks to a
point. The center U of the square on D A will also coalesce to A, and the centers of
the other three squares will form a triangle TSR. See Figure 4. Theorem 2.1 applied to
this situation gives the following result.
S C
A B
Corollary 2.4. If squares are built on the (out)sides of a triangle 1ABC, then the
line segment connecting the center of the square on one side to the opposite vertex of
1ABC is equal to and perpendicular to the segment joining the centers of the other
two squares.
Thus, in Figure 4, the line segment RA is (an extension of) the altitude from R to
ST in triangle 1RST, and is equal to ST in length, and similarly for the other vertices.
3. THE CUOCO CONFIGURATION. For any triangle 1ABC, we will call the tri-
angle 1RST whose vertices are the centers of the squares on the outsides of the sides
of 1ABC, the (outer) Cuoco triangle of 1ABC, and we will call the hexagon ATBRCS
the Cuoco hexagon of 1ABC. We call the triple, 1ABC, 1RST, and hexagon ATBRCS,
the Cuoco configuration of 1ABC (Figure 4).
Proof. Indeed, when D coalesces to A, the isosceles right triangle 1SDC becomes
1SAC, an isosceles right triangle with hypotenuse AC. If we drop the perpendicular
from S to AC, it will hit AC in its midpoint B 0 , and 1SAB0 and 1SCB0 will be two
congruent isosceles right triangles, with hypotenuses SA and SC, respectively.
On the other hand, since the point U also coalesces to A, the isosceles right triangle
1SUZ, which has SU as hypotenuse, becomes the isosceles right triangle 1SAZ, with
SA as hypotenuse. Thus, Z = B 0 .
But in Figure 3, Z is also the right-angle vertex of the isosceles right triangle built
on RT as hypotenuse, and as Figure 3 degenerates to Figure 5, this relation is preserved.
Thus, in the degenerated figure, B 0 = Z is the right angle vertex of an isosceles right
triangle with RT as hypotenuse. In particular, the line connecting B 0 to the midpoint V
of RT is the altitude of this isosceles right triangle, and therefore is perpendicular to
RT. Thus, it is the perpendicular bisector of RT. Since this line goes through B 0 , it also
bisects AC. See Figure 5 for an illustration of this. Of course, the parallel facts hold for
the midpoints of AB and BC.
S C
R
B0 =Z
A B
These isosceles right triangles have another relation to the configuration as a whole,
expressible in terms of area.
Theorem 4.2. The area of 1RTB0 plus the area of 1SAB0 equals half the area of the
Cuoco hexagon ATBRCS.
Proof. The area of 1SAB0 is clearly half the area of 1SAC. Also, the hexagon ATBRCS
decomposes into 1SAC and the pentagon ATBRC. Hence, to prove the theorem, it will
suffice to show that 1RTB0 has area equal to half the area of pentagon ATBRC.
a2 c2 a c a2 c2
+ 2 cos + = + + ac sin
2 2 2 2 2 2 2
2
a c2 ac sin
=2 + +
4 4 2
= 2 (Area (1RBC) + Area (1TBA) + Area (1ABC))
= 2 Area (ATBRC). (4.3)
Corollary 4.4.
(a) The sum of the areas of the squares on RT and AS is twice the area of the Cuoco
hexagon ATBRCS.
(b) The squares with RT and AS as diagonals have the midpoint B 0 of AC as a
common vertex, and have area equal to the Cuoco hexagon.
Parallel facts of course hold for sides RS and ST of triangle 1RST. Hence
(c) |RT |2 + |AS|2 = |TS|2 + |CR|2 = |SR|2 + |BT |2 ; or
2|RT |2 + |AC|2 = 2|TS|2 + |CB|2 = 2|SR|2 + |BA|2 .
CONCLUDING REMARKS.
(a) The outer Cuoco triangle 1RST of Figure 5 may be regarded as resulting from
1ABC by a certain construction: Build squares on the outsides of 1ABC and
form the triangle defined by the centers. Evidently, we could also define an
inner Cuoco triangle, formed from the centers of squares built on the insides of
1ABC. This language lets us reformulate the first statement of Theorem 4.1 as:
The inner Cuoco triangle of the outer Cuoco triangle of 1ABC is the triangle
formed by the midpoints of the sides (the medial triangle) of 1ABC.
(b) Construction of the outer Cuoco triangle is an operation of Euclidean similarity
geometry, in the sense that, if triangle 1A0 B 0 C 0 is similar to triangle 1ABC,
then the outer Cuoco triangle of 1A0 B 0 C 0 is similar to the outer Cuoco triangle
of 1ABC. Indeed, if 8 is a similarity taking 1ABC to 1A0 B 0 C 0 , then 8 will
also take the outer Cuoco triangle of 1ABC to the outer Cuoco triangle of
1A0 B 0 C 0 . Similar remarks of course apply to the inner Cuoco triangle. Thus,
either of these constructions can be thought of as a mapping from similarity
classes of triangles to similarity classes of triangles. Since, as is well known
[2, VI.3], the medial triangle is similar to the original triangle, Theorem 4.1
asserts, that as an operation on similarity classes, the inner Cuoco construction
is inverse to the outer Cuoco construction. It should be noted that the outer
construction is not onto: Not all similarity classes of triangles can be outer
Cuoco triangles. In particular, it follows from formula (4.3) that an outer Cuoco
triangle is always acute.
(c) The reader may already have noted that in both triangle 1ABC and its outer
Cuoco triangle 1RST, the points on the sides that are implicated in the relation-
ships described above are the midpoints and the feet of the altitudes, the same
points that define the nine-point circle.
ACKNOWLEDGMENTS. The author is grateful to Bill Barker and a referee for comments that improved
the exposition. Thanks also to Melody Hachey and Kevin Waterman for help with the figures.
REFERENCES
ROGER E. HOWE has been teaching and doing research at Yale University for nearly 40 years. His research
has been mainly concerned with symmetry, especially representation theory. He is a longtime fan of Kleins
Erlanger Programm. In recent years, he has also devoted a lot of attention to issues of mathematics education.
He should be thinking about retiring, but the pleasures of teaching cloud his mind.
Department of Mathematics, Yale University, PO Box 208283, New Haven, CT 06520
howe@math.yale.edu
Complex Knowledge
I took a statistics course at MIT. I would study and do problems, and have high confidence that
I understood the material. Then Id go to the lecture, and be more confused than I was when
I entered the classroom. Thus, I discovered that some teachers were capable of conveying
negative knowledge, so that after listening to them, I knew less than I did before.
It was also clear that knowledge varies considerably in quantity among people, and this
convinced me that real knowledge varies over a very wide range.
Then I encountered people who either did not know what they were talking about, or
were clearly convinced of things that were wrong, and so I learned that there was imaginary
knowledge.
Once I understood that there was both real and imaginary knowledge, I concluded that
knowledge is truly complex.
Submited by Hillel J. Chiel, Ph.D.
Professor of Biology, Neurosciences and Biomedical Engineering,
Department of Biology, Case Western Reserve University
hjc@case.edu
Abstract. We present a generalization of the Leibniz Rule. One of its consequences is the
celebrated Abel identity.
for sufficiently often differentiable functions, is a topic in most calculus courses. Also,
its generalization to several functions f i (for i = 1, . . . , r ),
r
!(n) r
Y X n Y (k )
fi = f i i (n = 0, 1, 2, . . .), (1)
i=1 |k|=n
k i=1
(for n = 0, 1, 2, . . .). It turns out that the latter equation remains valid if the monomials
x k are replaced with powers of arbitrary linear functions `(x) = ax + b. Furthermore,
it can be extended to several functions. Moreover, we study sums of the type
n
n
(h k f )(k) (h nk g)(nk) ,
X
k=0
k
for arbitrary analytic functions h, and present closed expressions. The purpose of this
short note is to prove those intriguing formulas. They are useful in the derivation of
asymptotic expansions for sequences of approximation operators. As applications, we
obtain remarkable identities. The most prominent example is the Abel identity
n
n
a(a kc)k1 (b + kc)nk = (a + b)n ,
X
k=0
k
published in 1826 [1]. This deep generalization of the binomial formula follows as a
direct consequence of Theorem 3.
http://dx.doi.org/10.4169/amer.math.monthly.120.10.924
MSC: Primary 26A24, Secondary 05A19
Remark 4. The identity is of an algebraic nature among the derivatives (Taylor coeffi-
cients) and hence automatically extends to non-analytic functions of sufficient smooth-
ness by a general principle. Therefore, Theorem 3 is valid also for real functions h and
f i possessing a continuous derivative of order n in a (real) neighborhood of z 0 R.
We close this section with some direct consequences. If h(x) = x and f i (x) =
x ai (i = 1, . . . , r ) we obtain the following.
k=0
k
3. PROOFS.
Proof of Theorem 1. We prove Theorem 1 only for monomials, i.e., `(x) = x; the full
result follows by translation and scaling. The formula is trivial for r = 1. We proceed
by mathematical induction with respect to r . First we show the formula for r = 2. We
start with the left-hand side. Application of the Leibniz Rule yields
n
n (k) (nk)
x k f (x) x nk g(x)
X
LHS =
k=0
k
n X
k nk
n k k! n k (n k)!
i
x j g ( j) (x)
X X
= x f (x) (i)
k=0
k i=0
i i! j=0
j j!
n X
k nk
k 1 i (i) nk 1 j ( j)
X X
= n! x f (x) x g (x)
k=0 i=0
i i! j=0
j j!
n
XX 1 1 ( j) k nk
i+ j (i)
X
= n! x f (x) g (x) .
i0 j0
i! j! k=0
i j
we obtain
1 XX n
= f (i) (x)g ( j) (x)(x n+1 )(ni j)
x i0 j0 i, j
1 n+1 (n+1)
= x f (x)g(x) ,
x
n r
!(nk1 )
X n! k1 nk1 +r 2
Y
=` (` f 1 ) (k1 )
` fi
k ! (n k1 )!
k =0 1 i=2
1
r
!(n)
n+1 r 2
Y
= ` f1 ` fi ,
i=2
Proof of Theorem 3. The formula is trivial for r = 1. First we show the formula for
r = 2. We start with the left-hand side. Application of the Cauchy integral formula
yields that
n
n
(h k f )(k) (z 0 )(h nk g)(nk) (z 0 )
X
LHS =
k=0
k
n
n k! h k (w) f (w) (n k)! h nk (z)g(z)
X Z Z
= dw dz
k=0
k 2i D1 (z0 ) (w z 0 )k+1 2i D2 (z 0 ) (z z 0 )
nk+1
2 Z n
1 f (w)g(z) h k (w)h nk (z)
Z X
= n! dz dw,
2i D2 (z 0 ) (w z 0 )(z z 0 ) k=0 (w z 0 ) (z z 0 )
k nk
D1 (z 0 )
= 0,
because the inner integral vanishes by the Cauchy Integral Theorem. Two applications
of the Cauchy Integral Formula to the second integral yield, in view of H (w, w) =
h 0 (w), that
which implies the desired formula. As in the proof of Theorem 1, the general case
follows by mathematical induction with respect to r .
ACKNOWLEDGMENTS. The author is grateful to the anonymous referees, whose valuable suggestions led
to an improved exposition of the paper.
REFERENCE
1. N. H. Abel, Beweis eines Ausdruckes, von welchem die Binomial-Formel ein einzelner Fall ist, J. Reine
Angew. Math. 1 (1826) 159160.
Abstract. The arbelos is a classical geometric shape bounded by three mutually tangent semi-
circles with collinear diameters. We introduce a parabolic analog, the parbelos. After a review
of the parabola, we use theorems of Archimedes and Lambert to demonstrate seven properties
of the parbelos, drawing analogies to similar properties of the arbelos, some of which may
be new.
Just as all circles are similar, so too all parabolas are similar. (See, e.g., [4, p. 118].
The same is not true for the other conic sections, because the similarity class of an
ellipse or a hyperbola depends on its eccentricity.) For that reason, one might expect
to find a parabolic analog of the arbelos in the literature of the past two millennia.
However, extensive searches have failed to uncover any mention of one. This note
provides and studies such an analog.
To define it, recall first that the latus rectum of a parabola is the focal chord par-
allel to the directrix (see Section 2). Now, replace the semicircles of the arbelos with
the latus rectum arcs of parabolas, all opening in the same direction, whose foci are
the centers of the semicircles. The region bounded by the three arcs is the parbelos
associated to the arbelos. See Figure 2.
http://dx.doi.org/10.4169/amer.math.monthly.120.10.929
MSC: Primary 51M04, Secondary 51M15; 51M25, 51N20
2. THE PARABOLA. Recall that a parabola P is the locus of points equidistant from
a point F, called the focus, and a line L, the directrix. The distance p > 0 from F to L
is the focal parameter. The point V at a distance a := 2p from both F and L is the
vertex of P. The chord C1 C2 of P passing through F and parallel to L is the latus
rectum. Since C1 and C2 lie on P, the length of C1 C2 equals 2 p = 4a; one half of that
is the semi-latus rectum p = 2a. The arc of P with endpoints C1 and C2 is the latus
rectum arc.
These notations are illustrated in Figure 3, which shows the unique downward-
opening parabola P whose latus rectum is the interval [2a, 2a] on the x-axis. The
x2
equation of P is y = a 4a .
y = 2a L
(0, a) V
x2
y=a P
4a
C1 C2
(2a, 0) (0, 0) F (2a, 0)
Just as the ratio of the length of any semicircle to its radius is always , the ratio of
the length s of the latus rectum arc of any parabola to its semi-latus rectum (and focal
parameter) p is also a constant, namely, the Universal Parabolic Constant P = sp see
Reese and Sondow [7]. For a geometric method of computing the length of a parabolic
arc, due to Isaac Barrow (16301677), see Dorrie [5, Problem 58]; for the sweeping
tangents method, see Apostol and Mnatsakanian [1].
Property 1. The upper and lower boundaries of the parbelos have the same length.
Proof. (We paraphrase Boas [2, p. 237] on the arbelos.) This is immediate from the
knowledge that the length of the latus rectum arc of a parabola is proportional to its
Our second property of the parbelos is the direct analog of part of a deeper property
of the arbelos discovered by Schoch in 1998see [8, Figure 6].
Property 2. Under each lower arc of the parbelos, construct a new parbelos similar to
the original (see Figure 4). Of the four new lower arcs, the middle two are congruent,
and their common length equals one half the harmonic mean of the lengths of the
original lower arcs.
Proof. Denote by ` L and ` R the lengths of the original left and right lower arcs, and
by `1 , `2 , `3 , `4 the lengths of the four new lower arcs. By similarity, we have the
equalities
`L 1 2 `R
`2 = `R = 1 1
= ` L = `3 .
`L + ` R 2 `L
+ `R
`L + ` R
The next property of the parbelos is analogous to the fact that the area of the arbelos
equals /2 times the area of its cusp-midpoints rectangle, determined by the middle
cusp and the midpoints of the three semicircular arcs (see Figure 5). The proof of this
fact is similar to that of Property 3.
Property 3. The middle cusp of the parbelos and the vertices of its three parabolas
determine a parallelogram, the cusp-vertices parallelogram. The area of the parbelos
equals 4/3 times the area of its cusp-vertices parallelogram.
Proof. A glance at Figure 3 reveals that, in Figure 6, the cusp C1 and the vertices V1
and V2 all lie on a line with slope 1/2, as do the cusp C2 and the vertex V3 . Likewise,
C3 , V3 , and V2 all lie on a line with slope 1/2, as do C2 and V1 . The first statement
follows.
V1
V3
C1 C2 C3
area C1 V2 C3 V3 C2 V1 = C1 V2 C3 ( C1 V1 C2 + C2 V3 C3 )
Property 4. The four tangents to the parbelos at its three cusps enclose a rectangle,
the tangent rectangle. The parbelos has two thirds the area of its tangent rectangle.
Proof. By Figure 3 and calculus (or by the parabolas reflection property), the latus
rectum of a parabola makes an angle of /4 with the tangent line at each endpoint.
The first statement follows.
The area of the tangent rectangle in Figure 7 is
From Figure 3, each 4 equals twice the corresponding 4 in the previous proof, and
we are done.
T2
T1
T3
C1 C2 C3
Figure 7. The tangent rectangle of the parbelos, a diagonal, and an angle bisector
Proof. In Figure 7, choose coordinates C1 = (0, 0), C2 = (2b, 0), and C3 = (4a, 0).
The endpoints of the diagonal opposite C2 are then T1 = (b, b) and T3 = (2a + b,
2a b). Now, the equation of the line through T1 and T3 , and the equation of the
upper parabola, are
ab b2 (x 2a)2
y = f (x) := x+ and y = g(x) := a .
a a 4a
ab
Setting f (x) = g(x), the only solution is x = 2b. As g 0 (2b) = a
, we infer Prop-
erty 5.
Property 6. The circumcircle of the tangent rectangle of the parbelos passes through
the focus of the upper parabola (see Figure 8).
Figure 8. The circumcircle of the tangent rectangle and the focus of the upper parabola
We give two proofs. The first uses the statement of Property 5; the second uses its
proof.
Proof 1. Any three lines tangent to a parabola bound a tangent triangle. By Property 5,
a diagonal and two adjacent sides of the tangent rectangle form a tangent triangle of
the upper parabola. Property 6 now follows from Lamberts Theorem, which asserts
that the circumcircle of any tangent triangle of a parabola passes through the focus.
Figure 9. The common tangent to the lower Figure 10. The circumcircle of the cusp-
parabolas of the parbelos, their foci, and the cir- midpoints rectangle of the arbelos and the center
cumcircles of similar tangent triangles of the upper semicircle
Our final property describes how to construct a parbelos directly from an arbelos
via a locus.
Property 7. The locus of the centers of circles inscribed in a semicircle of the arbelos
is the boundary of a parbelos with its cusps deleted. The arbelos and parbelos share
the same cusps.
Proof. We claim that the locus of the centers of circles inscribed in any semicircle
is the open latus rectum arc of a parabola whose latus rectum is the diameter of the
semicircle. A more general fact was discovered by Byer, Lazebnik, and Smeltzer [4,
p. 118]. (Added in proof : It has been discovered earlier by Tahir [9, p. 30].) We adapted
their elegant proof as follows.
Let the semicircle have center O, radius R, and diameter AB, and let an inscribed
circle have center K and radius r. Drawing a line L parallel to AB at a distance R as in
r
K R
A O B
Hence, K lies on the parabola with focus O and directrix L. Since O lies on AB, it
is the latus rectum, implying our claim. Finally, applying it to each semicircle of the
arbelos leads to Property 7.
ACKNOWLEDGMENTS. I thank both referees for suggesting changes that improved the exposition, and
Harold Boas for correcting two numerical errors.
REFERENCES
1. T. Apostol, M. Mnatsakanian, The method of sweeping tangents, Math. Gazette 92 (2008) 396417.
2. H. P. Boas, Reflections on the arbelos, Amer. Math. Monthly 113 (2006) 236249, available at http://
dx.doi.org/10.2307/27641891; also available at http://www.math.tamu.edu/~harold.boas/
preprints/arbelos.pdf.
3. A. Bogomolny, Arbelosthe Shoemakers Knife, from Interactive Mathematics Miscellany and Puzzles
(2012), available at http://www.cut-the-knot.org/proofs/arbelos.shtml.
4. O. Byer, F. Lazebnik, and D. L. Smeltzer, Methods for Euclidean Geometry. Mathematical Association of
America, Washington, DC, 2010.
5. H. Dorrie, 100 Great Problems of Elementary Mathematics: Their History and Solution, second edition.
Translated by D. Anton. Dover, New York, 1965.
6. T. C. Hull, Solving cubics with creases: the work of Beloch and Lill, Amer. Math. Monthly 118 (2011)
307315.
7. S. Reese, J. Sondow, Universal Parabolic ConstantFrom MathWorld, A Wolfram Web Resource, edited
by E. W. Weisstein, http://mathworld.wolfram.com/UniversalParabolicConstant.html.
8. T. Schoch, A Dozen More Arbelos Twins (1998), available at http://www.retas.de/thomas/
arbelos/biola/.
9. H. Tahir, Conic Sections and Tangent Circles. Hussein Tahir, Melbourne, 1999.
10. E. Tsukerman, Solution of Sondows problem: a synthetic proof of the tangency property of the parbelos.
Amer. Math. Monthly (forthcoming), available at http://arxiv.org/abs/1210.5580.
including results obtained independently by Artin, Wendel, and Gautschi. Applications in-
clude inequalities for binomial coefficients and the BohrMollerup theorem.
0(x + y)
R(x, y) = .
0(x)
have interesting applications, and have been considered by a number of writers over a
long period. In a M ONTHLY article [7], Wendel showed that
The lower bound in (2) is weaker than the one in (1), and is not exact when y = 0.
Gautschi only stated the result for integer values of x, but his method does not require
this. He also established the more elaborate lower bound x exp[(y 1)(x + 1)],
where (x) = 0 0 (x)/ 0(x), which implies (2) because (x) < log x. Refinements
have appeared in many later articles, e.g., [3, 5, 6]. Most of them take the form of
expressions in terms of (x), but one of Kershaws bounds is x(x + 2y ) y1 . Here, we
confine our attention to bounds of the simple type seen in (1) and (2).
Artins classic book [2] was published in 1931, long before either Wendel or
Gautschi. In it (p. 14) we find the statement
(actually only stated for integers x). Again, the lower bound is weaker than the one in
(1), though this is not quite so transparent, and is not exact at 1. Artin did not state (3) as
a result in its own right, but only as a step in the proof of another theorem. Perhaps for
this reason, his result appears to have been overlooked by most later writers, including
Wendel and Gautschi. Indeed, inequalities of this type have generally been referred to
as Gautschi-type inequalities, with scant respect to either Artin or Wendel.
http://dx.doi.org/10.4169/amer.math.monthly.120.10.936
MSC: Primary 33B15, Secondary 26D07
Theorem 1. Let f be any function such that log f (x) is convex and f (x + 1) =
x f (x) for all x > 0 (in particular, the gamma function). Write R(x, y) = f (x +
y)/ f (x). Then
Proof. For fixed x, let F(y) = log f (x + y) y log x for all y > x. Then F is con-
vex and
It follows that F(y) log f (x) for 0 y 1 and F(y) log f (x) for y 1 and for
y 0. This equates to (4) and (5).
Clearly, equality holds when y is 0 or 1, and if log f (x) is strictly convex, then strict
inequality holds for other y.
Both (6) and (7) are exact at the end points of the stated intervals for y.
We reassemble these inequalities for the three intervals for y, in some cases choos-
ing the better of two available options.
Note that (9) reproduces (1), and (10) amounts to the reverse of (2).
This system of inequalities amounts to a comprehensive set of bounds for R(x, y)
of the type we are considering. If it seems rather complicated, reflect that it describes
the true situation, and that we have arrived at it by a very simple process. It is, of
course, entirely typical for inequalities involving powers to reverse at certain values of
the index. At the cost of a loss of accuracy, we can, in fact, extract one simple pair of
bounds valid for all y 0:
It is worth restating (9) specifically for the gamma function in the case where x is a
positive integer n. Since 0(n) = (n 1)!, it equates to
Application 1. 0(x + y)/ 0(x) x y as x with y fixed. (This was the objective
of Wendels article.) To show this, first suppose that 0 y 1. Then (9) gives
y y1 R(x, y)
1+ 1,
x xy
hence, R(x, y) x y . For y > 1, the statement now follows from the fact that
y
R(x, y + 1) = 1 + R(x, y).
x
y n+y1 y(y + 1) . . . (y + n 1)
n
K n (y) = (1) = = .
n n n!
0(n + y) R(n, y)
K n (y) = = ,
0(y)n! n0(y)
(n + y) y1 n y1
K n (y) for 0 y 1, (14)
0(y) 0(y)
n y1 (n + 1) y1
K n (y) for 1 y 2, (15)
0(y) 0(y)
1 1.3. . . . (2n 1) 1
, (16)
[(n + 21 ) ]1/2 2.4. . . . (2n) (n )1/2
Application 3 (Bounds for 0( y n)). If 0 < y < 1, then 0(y n) alternates sign
in such a way that (1)n 0(y n) is positive. Using Eulers reflection formula
0(x)0(1 x) = / sin x, we can give bounds for this quantity that form a natu-
ral companion to (13). Indeed, by Eulers formula,
0(y n)0(n + 1 y) = = (1)n
sin (y n) sin y
and by (13), applied to 1 y, we have (n + 1)y n! < 0(n + 1 y) < n y n!. Hence,
for 0 < y < 1 and integers n 1,
ny (n + 1) y
< (1)n 0(y n) < . (17)
sin y n! sin y n!
We outline Artins method. It uses Eulers limit expression for the gamma function,
which can be stated as follows: 0(x) = limn G n (x), where
n x1 n!
G n (x) = .
x(x + 1) . . . (x + n 1)
(n + 1)x1 n!
Hn (x) = .
x(x + 1) . . . (x + n 1)
It is clearly enough to prove the theorem for 1 < x < 2 (for consistency with our
earlier notation, we now write y for x). By (10), for integers n we then have n y
R(n, y) n(n + 1) y1 . Clearly, f (n) = (n 1)!, so
n y1 n! f (n + y) (n + 1) y1 n!.
ACKNOWLEDGMENTS. The author is indebted to Prof. Andrea Laforgia for some very helpful comments.
REFERENCES
1. G. E. Andrews, R. Askey, R. Roy, Special Functions, Cambridge Univ. Press, Cambridge, 1999.
2. E. Artin, Einfuhrung in die Theorie der Gammafunktion, Teubner, Leipzig, 1931; English translation: The
Gamma Function, Holt, Rinehart and Winston, New York, 1964.
3. N. Elezovic, C. Giordano, J. Pecaric, The best bounds in Gautschis inequality, Math. Ineq. Appl. 3 (2000)
239252.
4. W. Gautschi, Some elementary inequalities relating to the gamma and incomplete gamma function, J.
Math. and Phys. 38 (1959) 7781.
5. C. Giordano, A. Laforgia, J. Pecaric, Unified Treatment of Gautschi-Kershaw type inequalities for the
gamma function, J. Comp. Appl. Math. 99 (1998) 167175, available at http://dx.doi.org/10.1016/
S0377-0427(98)00154-X.
6. D. Kershaw, Some extensions of W. Gautschis inequalities for the gamma function, Math. Comp. 41
(1983) 607611.
7. J. G. Wendel, Note on the gamma function, Amer. Math. Monthly 55 (1948) 563564, available at http:
//dx.doi.org/10.2307/2304460.
Department of Mathematics and Statistics, Lancaster University, Lancaster LA1 4YF, United Kingdom
g.jameson@lancaster.ac.uk
PROBLEMS
11740. Proposed by Cosmin Pohoata, Princeton University, Princeton, NJ. Let p, q be
prime ideals in a commutative Noetherian ring R with unity. Suppose that p q. Let
I be the set of all prime ideals j in R such that p j q. Prove that I is either empty
or infinite.
11741. Proposed by Chindea Filip-Andrei, Unversity of Bucharest, Bucharest, Roma-
nia. Given a ring A, let Z (A) denote the center of A, which is the set of all z in A that
commute with every element of A. Prove or disprove: For every ring A, there is a map
f : A Z (A) such that f (1) = 1 and f (a + b) = f (a) + f (b) for all a, b A.
11742. Proposed by Alexandr Gromeko, Odessa, Ukraine. For 0 p < q < 1, find all
zeros in C of the function f given by
(aq n , p/(aq n ); p)(z)n q n(n1)/2 ,
X
f (z) =
n=
11743. Proposed by Francois Capaces, Nancy, France. Let n be a positive integer, let
x be a real number, and let B be the n-by-n matrix with 2x in all diagonal entries, 1
in all sub- and super-diagonal entries, and 0 in all other entries. Compute the inverse,
when it exists, of B as a function of x.
11744. Proposed by C. P. Cholkar and M. N. Deshpande, Nagpur, India. Flip a fair
coin until the start of the r th run. (For instance, if r = 3 then TTTHHT is one possible
outcome.) Let Y be the number of runs consisting of one head. Find the expected value
and variance of Y .
http://dx.doi.org/10.4169/amer.math.monthly.120.10.941
Q4
P2 E0
P3
Q1 E
M P1
Q3
Q2
P4
SOLUTIONS
Solution by Richard Stong, Center for Communications Research, San Diego, CA. Let
ak,n count the n-words with k occurrences of 00 and 01 that end with 0, and let bk,n
count those that end with 1. With k occurrences of 01, words counted by ak,n have
k + 1 runs of 0s. They arise from the alternating word 010 10 of length 2k + 1 by
k insertions of a 0 before the beginning of a 0-run and n 3k 1 insertions of a 1
before the beginning of a 0-run. Since the number of available locations is k + 1 and
the resulting words aredetermined by how many insertions go into each location, we
have ak,n = 2kk n2k1 , where uk is 0 when u < k. Thus
k
X
X
n
ak,m+3k+1 t m+3k+1
X X
ak,n t =
n=1 k=0 k=0 m=0
X
2k m + k m+3k+1 X 2k t 3k+1
X
= t = . (1)
k=0 m=0
k k k=0
k (1 t)k+1
Similarly, a word counted by bk,n arises from the alternating word 01 01 of length
2k by k insertions of 0 and n 3k insertions
n2k of 1, again before the starts of 0-runs.
Hence for k > 0 we have bn,k = 2k1 k k
, so
X X X
2k 1 m + k m+3k
bk,n t n = bk,m+3k t m+3k =
X X X
t
n=1 k=1 k=1 m=0 k=1 m=0
k1 k
2k 1 t 3k 1 2k t 3k
X X
= = . (2)
k=1
k1 (1 t)k+1 k=1
2 k (1 t)k+1
Since b0,n = 1 for n > 0 and f (0) = 1, the desired generating function
P is obtained
2k
k
by summing (1), (2), and 1/(1 t). Using the well-known expression k=0 k x =
(1 4x)1/2 with x = t 3 /(1 t), we have
" !#
X
n 1 t 1 1
f (n)t = 1+ p + 1 ,
1t 1 4t 3 /(1 t) 2 1 4t 3 /(1 t)
p
n=0
Minimize a Functional
11618 [2012, 68]. Proposed by Pal Peter Dalyay, Szeged, Hungary. Let a, b, c, and d
be real numbers such that a < c < d < b and b a = 2(d c). Let S be the set of
twice-differentiable functionsR from [a, b] to R with continuous second derivative such
b
that f (c) = f (d) = 0 and x=a f (x) d x 6 = 0. Let p be a real number with p > 1.
Show that the map from S to R given by
R b 00
| f (x)| p d x
( f ) = aR p
b
a f (x) d x
The function h is continuous on [a, b]. Integration by parts shows that if f S, then
Z b
c d b
h(x) f 00 (x) d x = (x a)2 f 0 (x) a + (x )2 f 0 (x) c + (x b)2 f 0 (x) d
a
Z c Z d Z b
2 (x a) f (x) d x 2
0
(x ) f (x) d x 2
0
(x b) f 0 (x) d x
a c d
and
cx x d
(b d)2q +
L(x) = (c a)2q .
d c d c
We may therefore conclude that
Z b p/q
q
( f ) 2 p h(x) d x ,
a
Also solved by M. Bataille (France), R. Chapman (U. K.), P. P. Dalyay (Hungary), P. De (India), D. Fleischman,
V. V. Garca (Spain), O. Geupel (Germany), M. Goldenberg & M. Kaplan, A. Habil (Syria), J. G. Heuver
(Canada), O. Kouba (Syria), P. Nuesch (Switzerland), C. R. Pranesachar (India), R. Stong, Z. Voros (Hungary),
GCHQ Problem Solving Group (U. K.), and the proposer.
A Clopen Family
11627 [2012, 162]. Proposed by Samuel Alexander, The Ohio State University, Colum-
bus, Ohio. Let N be the set of nonnegative integers. Let M be the set of all functions
from N to N. For a function f 0 from an interval [0, m] in N to N, say that f extends f 0
if f (n) = f 0 (n) for 0 k m. Let F( f 0 ) be the set of all extensions in M of f 0 , and
equip M with the topology in which the open sets of M are unions of sets of the form
F( f 0 ). Thus, { f M : f (0) = 7 and f (1) = 11} is an open set.
Let S be a proper subset of M that can be expressed both as iN jN X i, j and as
S T
jN Yi, j , where each set X i, j or Yi, j is a subset of M that is both closed and open
T S
iN
(clopen). Show that there is a family Z i, j of clopen sets such that S = iN jN Z i, j
S T
and S = iN jN Z i, j .
T S
Solution by Richard Stong, San Diego, CA. We apply the lemma below once for each
i N. Given i, take A j = X i, j and B j = Yi, j . The lemma then yields {Ck }kN and
Thus
[\ [\ \[ \[
Z i,k = X i, j = S, Z i,k = Yi, j = S.
i k i j i k i j
Proof. Let {Ck }kN be the union of the families {A j } jN and {A j B j } jN , and let
{Dk }kN be the union of the families {B j } jN and {A j B j } jN . Now
\ \
Ck = Aj,
k j
since every Ck contains some A j and the A j are among the Ck . Also
\ \ \ \
Dk = Aj Bj Aj.
k j j j
Therefore,
\ \ \
Ck Dk = Aj.
k k j
Next,
[ [
Dk = Bj,
k j
since every Dk is contained in some B j and the B j are among the Dk . Also
[ [ [ [
Ck = Aj Bj Bj.
k j j j
Therefore,
[ [ [
Ck Dk = Bj.
k k j
Real Analysis Through Modern Infinitesimals. By Nader Vakil. Cambridge University Press,
2011, xix + 565 pp., ISBN 978-1-107-00202-9, $127.00.
A little over fifty years ago, Abraham Robinson discovered and developed a new ap-
proach to analysis [11]. It was both a beautiful application of mathematical logic and
an epochal moment in the history of mathematics. For thousands of years, humans had
disputed the existence, relevance, and consistency of infinite and infinitesimal num-
bers. Robinsons construction clearly demonstrated their consistency. Some thought it
might settle the remaining questions and revolutionize analysis all the way from the
teaching of calculus to the frontiers of research.
The way has not been smooth, however. An issue with such a long history is not eas-
ily resolved. Standing at times against infinitesimals are Aristotle, the Catholic Church,
Berkeley, Gauss, DeMorgan, Kronecker, and Archimedes. Standing at times in favor
are Democritus, Cavalieri, Leibniz, Euler, Cantor, and Archimedes. The disputes then
and now are fundamental: variously mathematical, philosophical, and pedagogical.
Views on them are sometimes keenly felt.
Its not clear that any of these questions have been resolved. Mathematicians still
face a spectrum of versions of the real numbers, from the constructive real numbers of
Bishop and Bridges [3], based on well-known criticisms of classical mathematics, to
the nonstandard reals, including infinitesimals, whose existence rests on the Axiom of
Choice. In the middle, of course, are the classical real numbers of Cauchy, Dedekind,
and Cantor.
The book under review is a text on nonstandard analysis, the name most commonly
given to the field pioneered by Robinson, and which we abbreviate NSA. The publica-
tion of this book seemed to us an occasion to reflect on the promises and controversies
of NSA as well as its current state, both of which are of interest to more than specialists
in analysis.
We start by discussing the field at the research end, move to its place in the curricu-
lum, and then to philosophical or sectarian issues. There are interesting developments.
There are ironies. As we proceed, we are assisted by a longish list of mathematicians
whom we polled on the present and future state of NSA. Our correspondents repre-
sent a spectrum of views, but cant be considered responsible for our observations
and conclusions. So heated has the discourse on these issues been at times that some
correspondents opted for anonymity. Weve decided to treat all their responses as con-
fidential. We return to Vakils book at the end.
Those who would like a quick overview of nonstandard analysis might look at the
Wikipedia article [15]. The parallel article on constructivism [14] may also be of inter-
est. Full disclosure: We have both written on this subject ([6], [7]), but are not attached
http://dx.doi.org/10.4169/amer.math.monthly.120.10.949
in analysis is just a change in classification, removing differential equations from analysis. But in the year of
the change, 1994, analysis fell only slightly, from 140 to 127, while the number of applied dissertations fell
dramatically, from 188 to 106. Most probably, differential equations were previously classified with applied
mathematics.
4. A NEW TEXT IN ANALYSIS. We return, finally, to the book under review. Real
Analysis Through Modern Infinitesimals is not written to grow nonstandard analysts.
Rather, it implements the principle that a subject should be pursued by all effective
means, implying there should be textbooks that introduce and develop analysis with
nonstandard as well as standard methods.
While at one time it was thought that all the results of NSA could be duplicated
by standard methods, an argument can be made that this is no longer the case. The
framework for Vakils book is Nelsons internal set theory (IST). IST is an extension
of ZermeloFraenkel set theory with choice (ZFC)the system in which all standard
mathematics is done. IST has, in Vakils words, features that ZFC lacks. He points
out that IST can express properties of the reals, function spaces, and so on, that cant be
described in ZFC. Work with these properties has produced new results in functional
analysis, stochastic analysis, and many areas of applied analysis [9].
Vakils text is appropriate for advanced undergraduates and beginning graduate stu-
dents. The first two chapters introduce the reader to IST and the resulting number line.
Chapters 39 contain the elements of calculus and advanced calculus; this is Part I.
Part II deals with real analysis, topology, metric spaces, linear operators, and function
REFERENCES
Thomas H. Foregger offers the following comment about the paper A new proof of a
classical formula, by Habib Bin Muzaffar (Vol. 120, No. 4, April 2013).
Another simple proof of this result can be found in
http://www.jstor.org/stable/pdfplus/2320072.pdf
which uses just a mean value theorem for integrals.
Gerald Folland offers the following observation about the filler piece Differentiat-
ing Iteratively, by Philip Mummert and Ken Constantine (Vol. 120, No. 6, JuneJuly
2013).
As a point of information: This theorem is an exercise in Section 2.3 (p. 70) of my
book Advanced Calculus (Prentice Hall, 2002), and the proof is the one suggested in
my hint. I agree, this observation deserves to be more widely known.
With respect to the same filler piece, Gerry Bilodeau offers the following.
Two papers closely related to this filler piece are An exponential rule, College
Mathematics Journal (September 1993), pp. 350351 (with a similar narrative al-
though a different proof) and in Logarithmic differentiation: two wrongs make a
right, College Mathematics Journal (November 2004), pp. 388390. Both appeared
in the Calculus Collection, Mathematical Association of America, 2010.
We received the following comments from Robin Chapman concerning the paper A
Probabilistic Proof of a Binomial Identity, by Jonathon Peterson (Vol. 120, No. 6,
JuneJuly 2013).
I very much enjoyed this article about applying conditional probability to prove this
identity. The author observes that the ChuVandermonde identity and the Rice integral
http://dx.doi.org/10.4169/amer.math.monthly.120.10.954
The following was submitted by Randy Schwartz regarding the paper Origin and
Evolution of the Secant Method in One Dimension, by Richard Tapia and Joanna
Papakonstantinou, which appeared in the JuneJuly 2013 issue.
The article states that when ancient Egyptians dealt with relationships of the form
ax + b = c, They gave instructions on how to obtain the solution using the relation
x0 e1 x1 e0
x= .
e1 e0
This method for solving for x is now most commonly referred to as the Rule of Double
False Position (p. 508). The authors conclude, . . . the Rule of Double False Position
dates back to the 18th century BC. . . (pp. 512513). In fact, however, no instance
of any double false position technique being used in ancient Egypt, with or without
instructions, is known to exist. The problems from Smeur cited on 508509 are not
from Egypt but from J. van der Schueres arithmetic from Haarlem in 1611 AD.
The article also leaves readers with the impression that isab al-khaaayn (the double
false position method that was transmitted from the Arab world to Europe in medieval
times) might have first reached the Middle East from China and/or India (pp. 509
511). While a double false position method was certainly used in Chinese antiquity,
it was so different from the Arab method that this is not likely an instance of borrow-
ing, as I showed in detail in a 2004 paper (Issues in the origin and development of
isab al-Khaaayn (calculation by double false position), Actes du Huitieme Colloque
Maghrebin sur lHistoire des Mathematiques Arabes, Tunis, les 18-19-20 Decembre
2004 (Tunis: Tunisian Association of Mathematical Sciences, 2006), pp. 275296).
Professors Papakonstantinou and Tapia offer the following response.
We thank Professor Schwartz for his interest in our recent paper. Moreover, we
view him as an expert in the history of the rule of double false position and in the
general area of Arab and Islamic contributions to science. Our two statements brought
into question by Professor Schwartz were made as considered opinions based on our
readings. However, the situation could well be exactly as Professor Schwartz has de-
scribed. The establishment of correct history is often far more challenging than is the
establishment of correct mathematics.
Jose Hernandez Santia offers the following observations about the paper Variations on
a theme: Rings satisfying x 3 = x are commutative by Stephen Buckley and Desmond
MacHale, which appeared in the May 2013 issue.
REFERENCES
T ITLE I NDEX
Another Irreducibility Criterion, Anthony J. Foreword: The High Priest of Game Theory,
Bevelacqua, 648 Ehud Kalai, 384
Another Proof for Non-Supercyclicity in Finite Frobenius Result on Simple Groups of Order
Dimensional Complex Banach Spaces, F. p3 p
2
, Paul Monsky, 725
Galaz-Fontes, 466 From Pascals Theorem to d-Constructible
Another Proof of (2) = 6 Using Double
2
Curves, Will Traves, 901
Integrals, Daniele Ritelli, 642 Generalization of Rouths Triangle Theorem,
Binomial Identity via Differential Equations, A, Arpad Benyi and Branko Curgus, 841
A, D. Aharonov and U. Elias, 462 Generalization of the Leibniz Rule, A, Ulrich
BorsukUlam Equivalent that Directly Abel, 924
Implies Sperners Lemma, A, Kathryn Geometric Multiplicities and Gersgorin Discs,
L. Nyman and Francis Edward Su, Rachid Marsli and Frank J. Hall, 452
346 Greedy Galois Games, Joshua Cooper and
Buoyancy-Driven Perpetual Motion Machine, Aaron Dutle, 441
A, Tadashi Tokieda, 564 Heronian Tetrahedra Are Lattice Tetrahedra,
Certain Inequalities Associated with the Susan H. Marshall and Alexander R. Perlis,
Divisor Function, Jorge Luis Cimadevilla 140
Villacorta, 832 Higher-Dimensional Analogues of the Map
Christianes Hair, Jacques Levy Vehel and Coloring Problem, Bhaskar Bagchi and
Franklin Mendivil, 771 Basudeb Datta, 733
Class of Continued Radicals, A, Costas J. History of the Undergraduate Program in
Efthimiou, 459 Mathematics in the United States, The, Alan
College Admissions and the Stability of Tucker, 689
Marriage, D. Gale and L. S. Shapley, 386 How would Riemann Evaluate (2n)? Marco
Commuting and Noncommuting Infinitesimals, Dalai, 169
Mikhail G. Katz and Eric Leichtnam, 631 Identities for sin x that Came from Medical
Coxeter Friezes and Triangulations of Imaging, Peter Kuchment and Sergey Lvin,
Polygons, Claire-Soizic Henry, 553 609
Cuoco Configuration, The, Roger E. Howe, Illuminating a Network from Its Nodes, Steve
916 Alpern and Robbert Fokkink, 358
Direct Proof of the Existence of Eigenvalues Inequalities for Gamma Function Ratios,
and Eigenvectors by Weierstrasss Theorem, G. J. O. Jameson, 936
A, Jean Van Schaftingen, 741 Intersecting Curves (Variation on an
Equation x 2 + m y 2 = k in Z/ pZ, The, Kurt Observation of Maxim Kontsevich), Etienne
Girstmair, 546 Ghys, 232
Euclidean Algorithm and the Linear Irreducible Factorization Lengths and the
Diophantine Equation ax + by = gcd(a, b), Elasticity Problem within N, Matthew
The, S. A. Rankin, 562 Jenssen, Daniel Montealegre, and Vadim
Evolutes and Involutes of Spatial Curves, Ponomarenko, 322
Dmitry Fuchs, 217 Lehmers Interesting Series, Freeman J. Dyson,
Example of a Monotonic, Everywhere Norman E. Frankel, and M. Lawrence
Differentiable Function on R Whose Glasser, 116
Derivative Is not Continuous, Manas R. Linear Algebra via Complex Analysis,
Sahoo, 566 Alexander P. Campbell and Daniel Daners,
Factorization Theory and Decompositions of 877
Modules, Nicholas R. Baeth and Roger Maximum Principle for High-Order
Wiegand, 3 Derivatives, A, David Pan, 846
http://dx.doi.org/10.4169/amer.math.monthly.120.10.957
AUTHOR I NDEX
Abel, Ulrich, A Generalization of the Leibniz Bernardi, Olivier, A Short Proof of Rayleighs
Rule, 924 Theorem with Extensions, 362
Aharonov, D., and U. Elias, A Binomial Bevelacqua, Anthony J., Another Irreducibility
Identity via Differential Equations, 462 Criterion, 648
Aldous, David J., Using Prediction Market Bliss, Nathan, Ben Fulan, Stephen Lovett, and
Data to Illustrate Undergraduate Probability, Jeff Sommars, Strong Divisibility,
583 Cyclotomic Polynomials, and Iterated
Alm, Jeremy, Michael Gramelspacher, and Polynomials, 519
Theodore Rice, Rubiks on the Torus, 150 Brams, Steven J., Michael A. Jones, and
Alexanderson, Gerald, see Klosinski Christian Klamler, N -Person Cake-Cutting:
Alpern, Steve, and Robbert Fokkink, There May Be No Perfect Division,
Illuminating a Network from Its Nodes, 358 35
Anghel, Vinicius Nicolae Petre, A Use of Buckley, Stephen M., and Desmond MacHale,
Symmetry: Generalization of an Integral Variations on a Theme: Rings Satisfying
Identity Found by M. L. Glasser, 62 x 3 = x Are Commutative, 430
Apostol, Tom M., and Mamikon A. Butler, Steve, and Ron Graham, Subdivision
Mnatsakanian, New Balancing Principles Using Angle Bisectors Is Dense in the Space
Applied to Circumsolids of Revolution, and of Triangles, 622
to n-Dimensional Spheres, Cylindroids, and Campbell, Alexander P., and Daniel Daners,
Cylindrical Wedges, 298 Linear Algebra via Complex Analysis, 877
Arnold, Matthias, and Werner Gla, Simple Casey, Stephen D., and Brian M. Sadler, Pi, the
Approximation Formulas for the Birthday Primes, Periodicities, and Probability, 594
Problem, 645 Chen, Yi Jun, A Proof that Zeilberger Missed:
Baeth, Nicholas R., and Roger Wiegand, A New Proof of an Identity by Chaundy and
Factorization Theory and Decompositions of Bullard Based on the WilfZeilberger
Modules, 3 Method, 69
Bagchi, Bhaskar, and Basudeb Datta, Conway, John H., On Unsettleable
Higher-Dimensional Analogues of the Map Arithmetical Problems, 192
Coloring Problem, 733 Cooper, Joshua, and Aaron Dutle, Greedy
Bak, Wodzimierz, A Simple Characterization Galois Games, 441
of Differentiation, 74 Curgus, Branko, see Benyi
Bayless, Jonathan, and Dominic Klyve, Dalai, Marco, How Would Riemann Evaluate
Reciprocal Sums as a Knowledge Metric: (2n)?, 169
Theory, Computation, and Perfect Numbers, Daners, Daniel, see Campbell
822 Datta, Basudeb, see Bagchi
Benyi, Arpad, and Branko Curgus, A Dragovic, Vladimir, Polynomial Dynamics
Generalization of Rouths Triangle and a Proof of the Fermat Little Theorem,
Theorem, 841 171
P ROBLEMS S OLVED
S OLUTIONS
January, 2013 May, 2013
11540, 77 11574, 470
11558, 78 11576, 471
11559, 78 11577, 472
11560, 79 11596, 472
11561, 80 11597, 473
11564, 81 11600, 474
11565, 82 11603, 475
11571, 83 11604, 476
February, 2013 JuneJuly, 2013
11569, 175 11595, 570
11572, 176 11605, 571
11575, 177 11607, 572
11578, 177 11608, 573
11579, 178 11612, 574
11580, 178 11614, 576
11581, 179 AugustSeptember, 2013
11586, 180 11584, 661
11589, 181 11588, 662
April, 2013 11592, 662
11562, 366 11611, 664
11563, 367 11615, 665
11566, 368 11616, 666
11567, 369 11621, 667
11568, 371 October, 2013
11570, 371 11557, 755
11573, 372 11582, 756
The M ONTHLY expresses its appreciation to the following people for their help in
refereeing during the past year. We could not function successfully without such people
and their hard work.
B
Beginning withh one off the
h most remarkable ecological collapses of recent time, that
of the passenger pigeon, Hadlock goes on to survey collapse processes across the
entire spectrum of the natural and man-made world. He takes us through extreme
weather events, technological disasters, evolutionary processes, crashing markets
and companies, the chaotic nature of Earth's orbit, revolutionary political change, the
spread and elimination of disease, and many other fascinating cases. His key thesis
is that one or more of six fundamental dynamics consistently show up across this
wide range. These "six sources of collapse" can all be best described and investigated
using fundamental mathematical concepts. They include low probability events, group
dynamics, evolutionary games, instability, nonlinearity, and network effects, all of which
are explained in readily understandable terms.