# A Geometric Series from Tennis

> An annotated edition of this paper on Fermat's Library, where scientists and readers explain its ideas, equations and history in the margins. Annotated edition: https://fermatslibrary.com/s/a-geometric-series-from-tennis

- **Authors:** James Sandefur
- **Published:** 2006
- **Topic:** Mathematics
- **Original:** https://s3-us-west-1.amazonaws.com/fermatslibrary/uploads%2F1656882227224-79zptxa60vc-aee8e79d7397edb3b73ba1eff7ad474f%2Fsandefur224.pdf
- **Pages:** 3
- **Annotations:** 5 comments in 5 threads
- **Featured:** Fermat's Library Journal Club
- **Annotated edition:** https://fermatslibrary.com/s/a-geometric-series-from-tennis

## Annotations

### Annotation 1 (page 1)

Beside the passage: “James Sandefur (sandefur@georgetown.edu), Georgetown Uni versity, Washington, DC 20057”

**Annotation** by Luís Batalha (2022-07-04):

> James Sandefur is currently a Professor in the Department of Mathematics and Statistics at Georgetown University. He got his PhD in Mathematics from Tulane University and in August 2006, he was awarded the MAA George Polya Award for my expository paper for this paper.
>
> ![](https://faculty.georgetown.edu/sandefur/Sandefur%20picture%202005.jpg)

### Annotation 2 (page 1)

Beside the passage: “3 0”

**Annotation** by Luís Batalha (2022-07-04):

> The way to think about this for a specific scenario, say 3-to-2 is to calculate the total number of possibilities of distributing the 2 points from player B in 5 different slots. You can do this using the binomial coefficient $\binom{5}{2}= 10$.
>
> ![](https://i.imgur.com/n5WbY0U_d.webp?maxwidth=760&fidelity=grand)

### Annotation 3 (page 1)

Beside the passage: “20 p 3 (1 − p) 3 % 2p(1 − p)”

**Annotation** by Luís Batalha (2022-07-04):

> There are 2 ways of splitting the next 2 points after a 3-to-3 tie: you either win the first point and lose the second or you lose the first point and win the second: $p(1-p)+(1-p)p=2p(1-p)$

### Annotation 4 (page 2)

Beside the passage: “& n−1”

**Annotation** by Luís Batalha (2022-07-04):

> For each pair of splitting points there are always 2 ways of  arranging them (depending on who wins the first point) so the probability is $(2p(1-p))^{n-1}$ and $p^2$ for the last 2 winning points.
>
> ![](https://i.imgur.com/J5MHfbm.jpg)

### Annotation 5 (page 2)

Beside the passage: “n=0”

**Annotation** by Luís Batalha (2022-07-04):

> For $|x|<1$, we know that the sum of the geometric series $\sum_{n=0}^{\infty}x^n = \frac{1}{1-x}$. Since $|2p(1-p)|<1$ we get
>
> $$
> \sum_{n=0}^{\infty}(2p(1-p))^n = \frac{1}{1-2p(1-p)}
> $$

## Full text

Extracted from the PDF. [Annotation N] marks where each annotation above sits.

#### Page 1

CLASSROOM CAPSULES

EDITOR

Michael K. Kinyon Indiana University South Bend South Bend, IN 46634

Classroom Capsules consists primarily of short notes (1–3 pages) that convey new mathematical insights and effective teaching strategies for college mathematics instruction. Please submit manuscripts prepared according to the guidelines on the inside front cover to the Editor, Michael K. Kinyon, Indiana University South Bend, South Bend, IN 46634.

AGeometricSeriesfromTennis

James Sandefur (sandefur@georgetown.edu), Georgetown Uni versity, Washington, [Annotation 1] DC 20057

During the Wimbledon tennis ﬁnals, the commentator mentioned that one of the players was winning 60% of his points on serve. I began wondering what fraction of the games a person should win if the probability of winning any particular point was p. In answering this question, I used some basic probability and summed a geometric series. Others might want to share this with their students.

As a reminder, the winner of a game is the ﬁrst person to score 4 points, unless the game reaches a 3-to-3 tie. Then it continues until someone goes ahead by 2. (The four points are called 15, 30, 40 and game, by the way.)

Assume that the probability that player A wins a point is p.Theprobabilitythat player A wins the game within the ﬁrst 6 points is the probability that A leads by a score of 3-to-0, 3-to-1, or 3-to-2, and then wins the next point, which is

!"

3 [Annotation 2] 0

\#

p 3 +

"

4 1

\#

p 3 (1 − p) +

"

5 2

\#

p 3 (1 − p) 2

$

p = p 4 (15 − 24 p + 10 p 2 ),

after simpliﬁcation. If neither player has won by the 6th point, then the score must be tied at 3-to-3. The probability of a 3-to-3 tie is

"

6 3

\#

p 3 (1 − p) 3 = 20 p 3 (1 − p) 3 .

After that, the game must be won after an even number of points. The probability of winning on the 8th point is the probability of a 3-to-3 tie, followed by winning the next two points, which is

20 p 3 (1 − p) 3 p 2 .

The probability of winning on the 10th point is the probability of a 3-to-3 tie, splitting the next two points, and then winning the 9th and 10th points, which is

20 p [Annotation 3] 3 (1 − p) 3 % 2p(1 − p)

&

p 2 .

224 c " THE MATHEMATICAL ASSOCIATION OF AMERICA

#### Page 2

More generally, the probability of winning on the (2n + 6)th point is the probability of a3-to-3tie,20p 3 (1 − p) 3 ,splittingeachpairofthenext2n − 2points[2p(1 − p)] n−1 , and then winning the last 2 points p 2 ,whichis

20 p 3 (1 − p) 3 % 2p(1 − p)

& [Annotation 4] n−1

p 2 .

Thus, the probability of A’s winning the game is

f ( p) = p 4 (15 − 24 p + 10 p 2 ) + 20 p 3 (1 − p) 3 p 2

∞ '

n=0 [Annotation 5]

% 2p(1 − p)

& n

.

Summing the geometric series gives the function

f ( p) = p 4 (15 − 24 p + 10 p 2 ) +

20 p 5 (1 − p) 3

1 − 2 p(1 − p)

,

whose graph is shown in Figure 1.

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

Figure 1. Graph of probability function f of winning a game as function of the probability p of winning a point.

The answer to my original question is that players who win 60% of their points on serve will win f (0.6) ≈ 0.74 or about 74% of their service games.

The graph suggests that f has rotational symmetry about the point (0.5, 0.5),although this is not apparent from the form of f .Wecanverifytherotationalsymmetry by noting that if q is the probability that the second player wins a point, then f (q) gives the probability that the second player wins the game. But p + q = 1, so f (q) = f (1 − p).Sinceoneofthetwoplayersmustwin, f ( p) + f (1 − p) = 1. If we take p = 0.5 + x and g(x) = f (0.5 + x) − 0.5, we ﬁnd that

g(x) + g(−x) = f (0.5 + x) + f (0.5 − x) − 1 = f ( p) + f (1 − p) − 1 = 0,

so g(−x) =−g(x).Thus,g is an odd function, and hence f has rotational symmetry.

From the graph, it also appears that there is a point of inﬂection at p = 0.5. It would be tedious to check that f %% (0.5) = 0. Ho wever, if we remember the derivation

f %% (x) = lim

h→0

f (x + h) + f ( x − h) − 2 f (x)

h 2

VOL. 36, NO. 3, MA Y 2005 THE COLLEGE MATHEMA TICS JOURNAL 225

#### Page 3

and apply that to g(x),weseethat

f %% (0.5) = g %% (0) = lim

h→0

g(h) + g(−h) − g(0)

h 2 = lim

h→0

0 h 2 = 0,

which shows that p = 0.5givesapointofinﬂection.Thismeansthatplayersgainthe most in the number of games they win by increasing p when p ≈ 0.5. In other words, if you are a weak or a strong player relative to your opponent ( p small or large), then a small improvement in your serve (increasing p)doesn’tresultinasmuchimprovement in the number of games you win as an improvement against a comparable opponent (p about 0.5) does. A simple computation shows that f % (0.5) = 2.5, so a 1% gain in p will result in about a 2.5% increase in your likelihood of winning a game.

◦

On Sums of Cubes

Hajrudin Fejzi ´ c(hfejzic@csusb.edu),DanRinne(drinne@csusb.edu),andBobStein (bstein@csusb.edu), Department of Mathematics, California State University, San Bernardino, CA 92407

Awell-knownidentityforthesumoftheﬁrstn cubes is

1 3 + 2 3 +···+n 3 = (1 + 2 +···+n) 2 . (1)

Some of our students noticed that, curiously, equality still holds if n − 1isreplaced by 2, that is

1 3 + 2 3 +···+(n − 2) 3 + 2 3 + n 3 = ( 1 + 2 +···+(n − 2) + 2 + n ) 2 .

This observation led us to ask whether other such switches are possible. In this note, we investig ate those triples (k, m, n) for which

k−1 '

j=1

j 3 + m 3 +

n '

j=k+1

j 3 =

( k−1 '

j=1

j + m +

n '

j=k+1

j

) 2

. (2)

Clearly (2) holds (because of (1)) if m = k, so in what follows we assume m (= k. Furthermore, as our students pointed out, (n − 1, 2, n) is a solution for every n ≥ 2. Do other solutions exist, and if so, do they ﬁt nice patterns? Our inquiry led us to some interesting and unexpected answers, and ultimately to a connection with an unsolved problem in number theory.

Our ﬁrst result gives a necessary and sufﬁcient condition for a triple (k, m, n) to be asolutionof(2).(Weassumethatk and m are positive integers not exceeding n.)

Theorem 1. Atriple(k, m, n) with m (= ksatisﬁes(2) if and only if either

(a) k = n − 1 and m = 2,or (b) there exists an integer p ≥ 2 and a positive divisor s of 3p( p − 1) for which

(k, m, n) =

"

2(p − 1) +

3(p − 1) p

s

, 2p + s, m + k − p

\#

. (3)

Furthermore, different pairs ( p, s) yield different solutions.

226 c " THE MATHEMATICAL ASSOCIATION OF AMERICA

---

Source: [Fermat's Library](https://fermatslibrary.com/) — https://fermatslibrary.com/s/a-geometric-series-from-tennis. More annotated papers: https://fermatslibrary.com/llms.txt
