A Collection of Dice Problems

with solutions and useful appendices

(a work continually in progress)
current version May 5, 2011
q
q
q
q
q
q
q q
q q
q
q
q q
q
q
q
q
q
q
q
Matthew M. Conroy
doctormatt at madandmoonly dot com
www.madandmoonly.com/doctormatt
Thanks
A number of people have sent corrections, comments and suggestions, including Ryan Allen, Paul Elvidge,
Nick Hobson, Marc Holtz, Manuel Klein, Albert Natian and Elie Wolfe. Thanks, everyone.
2
Chapter 1
Introduction and Notes
This is a (slowly) growing collection of dice-related mathematical problems, with accompanying solu-
tions. Some are simple exercises suitable for beginners, while others require more sophisticated techniques.
Many dice problems have an advantage over some other problems of probability in that they can be
investigated experimentally. This gives these types of problems a certain helpful down-to-earth feel.
Please feel free to comment, criticize, or contribute additional problems.
1.0.1 What are dice?
In the real world, dice (the plural of die) are polyhedra made of plastic, wood, ivory, or other hard
material. Each face of the die is numbered, or marked in some way, so that when the die is cast onto a
smooth, at surface and allowed to come to rest, a particular number is specied.
Mathematically, we can consider a die to be a random variable that takes on only nitely many distinct
values. Usually, these values will constitute a set of positive integers 1, 2, ..., n; in such cases, we will refer
to the die as n-sided.
1.0.2 Terminology
A fair die is one for which each face appears with equal likelihood. A non-fair die is called xed. The
phrase standard die will refer to a fair, six-sided die, whose faces are numbered one through six. If not
otherwise specied, the term die will refer to a standard die.
3
Chapter 2
Problems
2.1 Standard Dice
1. On average, how many times must a 6-sided die be rolled until a 6 turns up?
2. On average, how many times must a 6-sided die be rolled until all sides appear at least once? What
about for an n-sided die?
3. Suppose we roll n dice and keep the highest one. What is the distribution of values?
4. How many dice must be rolled to have at least a 95% chance of rolling a six?
5. How many dice must be rolled to have at least a 95% chance of rolling a one and a two? What about
a one, a two, and a three? What about a one, a two, a three, a four, a ve and a six?
6. How many dice should be rolled to maximize the probability of rolling exactly one six? two sixes? n
sixes?
7. A single die is rolled until a run of six different faces appears. For example, one might roll the se-
quence 535463261536435344151612534 with only the last six rolls all distinct. What is the expected
number of rolls?
8. What is the most probable: rolling at least one six with six dice, at least two sixes with twelve dice,
or at least three sixes with eighteen dice? (This is an old problem, frequently connected with Isaac
Newton.)
9. Suppose we roll n dice, remove all the dice that come up 1, and roll the rest again. If we repeat this
process, eventually all the dice will be eliminated. How many rolls, on average, will we make? Show,
for instance, that on average fewer than O(log n) throws occur.
2.2 Dice Sums
10. Show that the probability of rolling 14 is the same whether we throw 3 dice or 5 dice. Are there other
examples of this phenomenon?
4
11. Suppose we roll n dice and sum the highest 3. What is the probability that the sum is 18?
12. A die is rolled once; call the result N. Then N dice are rolled once and summed. What is the
distribution of the sum? What is the expected value of the sum? What is the most likely value?
13. A die is rolled and summed repeatedly. What is the probability that the sum will ever be a given value
x?
14. A die is rolled once. Call the result N. Then, the die is rolled N times, and those rolls which are
equal to or greater than N are summed (other rolls are not summed). What is the distribution of the
resulting sum? What is the expected value of the sum?
2.3 Non-Standard Dice
15. Show that the probability of rolling doubles with a non-fair (xed) die is greater than with a fair die.
16. Find a pair of 6-sided dice, labelled with positive integers differently from the standaed dice, so that
the sum probabilites are the same as for a pair of standard dice.
17. Is it possible to have two non-fair n-sided dice, with sides numbered 1 through n, with the property
that their sum probabilities are the same as for two fair n-sided dice?
18. Is it possible to have two non-fair 6-sided dice, with sides numbered 1 through 6, with a uniform sum
probability? What about n-sided dice?
19. Suppose that we renumber four fair 6-sided dice (A, B, C, D) as follows: A = 0, 0, 4, 4, 4, 4,B =
1, 1, 1, 5, 5, 5,C = 2, 2, 2, 2, 6, 6,D = 3, 3, 3, 3, 3, 3.
(a) Find the probability that die A beats die D; die D beats die C; die C beats die B; and die B
beats die A.
(b) Discuss.
20. Find every six-sided die with sides numbered from the set 1,2,3,4,5,6 such that rolling the die
twice and summing the values yields all values between 2 and 12 (inclusive). For instance, the die
numbered 1,2,4,5,6,6 is one such die. Consider the sum probabilities of these dice. Do any of them
give sum probabilities that are more uniform than the sum probabilities for a standard die? What
if we renumber two dice differently - can we get a uniform (or more uniform than standard) sum
probability?
2.4 Games with Dice
21. Craps The game of craps is perhaps the most famous of all dice games. The player begin by
throwing two standard dice. If the sum of these dice is 7 or 11, the player wins. If the sum is 2,3 or
12, the player loses. Otherwise, the sum becomes the players point. The player continues to roll until
either the point comes up again, in which case the player wins, or the player throws 7, in which case
they lose. The natural question is: what is a players probability of winning?
5
22. Non-Standard Craps We can generalize the games of craps to allow dice with other than six
sides. Suppose we use two (fair) n-sided dice. Then we can dene a game analagous to craps in the
following way. The player rolls two n-sided dice. If the sum of these dice is n + 1 or 2n 1, the
player wins. If the sum of these dice is 2, 3 or 2n, then the player loses. Otherwise the sum becomes
the players point, and they win if they roll that sum again before rolling n + 1. We may again ask:
what is the players probability of winning?
23. Yahtzee There are many probability questions we may ask with regard to the game of Yahtzee. For
starters, what is the probability of rolling, in a single roll,
(a) Yahtzee
(b) Four of a kind (but not Yahtzee)
(c) Three of a kind (but not four of a kind or Yahtzee)
(d) A full house
(e) A long straight
(f) A small straight
24. More Yahtzee What is the probability of getting Yahtzee, assuming that we are trying just to get
Yahtzee, we make reasonable choices about which dice to re-roll, and we have three rolls? That is,
assume were in the situation where all we have left to get in a game of Yahtzee is Yahtzee, so that all
other outcomes are irrelevant.
25. Drop Dead In the game of Drop Dead, the player starts by rolling ve standard dice. If there are
no 2s or 5s among the ve dice, then the dice are summed and this is the players score. If there are
2s or 5s, these dice become dead and the player gets no score. In either case, the player continues
by rolling all non-dead dice, adding points onto the score, until all dice are dead.
For example, the player might roll 1, 3, 3, 4, 6 and score 17. Then they roll all the dice again and
get 1, 1, 2, 3, 5 which results in no points and two of the dice dying. Rolling the three remaining
dice, they might get 2, 3, 6 for again no score, and one more dead die. Rolling the remaing two they
might get 4, 6 which gives them 10 points, bringing the score to 27. They roll the two dice again,
and get 2, 3 which gives no points and another dead die. Rolling the remaining die, they might get
3 which brings the score to 30. Rolling again, they get 5 which brings this players round to an
end with 30 points.
Some natural questions to ask are:
(a) What is the expected value of a players score?
(b) What is the probability of getting a score of 0? 1? 20? etc.
26. Suppose we play a game with a die where we roll and sum our rolls as long as we keep rolling larger
values. For instance, we might roll a sequence like 1-3-4 and then roll a 2, so our sum would be 8. If
we roll a 6 rst, then were through and our sum is 6. Three questions about this game:
(a) What is the expected value of the sum?
(b) What is the expected value of the number of rolls?
(c) If the game is played with an n-sided die, what happens to the expected number of rolls as n
approaches innity?
6
27. Suppose we play a game with a die in which we use two rolls of the die to create a two digit number.
The player rolls the die once and decides which of the two digits they want that roll to represent. Then,
the player rolls a second time and this determines the other digit. For instance, the player might roll a
5, and decide this should be the tens digit, and then roll a 6, so their resulting number is 56.
What strategy should be used to create the largest number on average? What about the three digit
version of the game?
7
Chapter 3
Discussion, Hints, and Solutions
3.1 Single Die Problems
1. On average, how many times must a 6-sided die be rolled until a 6 turns up?
This problem is asking for the expected number of rolls until a 6 appears. Let X be the random
variable representing the number of rolls until a 6 appears. Then the probability that X = 1 is 1/6;
the probability that X = 2 is (5/6)(1/6) = 5/36. In general, the probability that X = k is
_
5
6
_
k1
1
6
(3.1)
since, in order for X to be k, there must be k 1 rolls which can be any of the numbers 1 through 5,
and then a 6, which appears with probabilty 1/6.
We seek the expectation of X. This is dened to be
E =

n=1
nP(X = n) (3.2)
where P(X = n) is the probability that X takes on the value n. Thus,
E =

n=1
n
_
5
6
_
n1
1
6
=
6
5

1
6

n=1
n
_
5
6
_
n
(3.3)
Using Equation B.3 from Appendix B, we can conclude that
E =
6
5

1
6
5/6
(1 (5/6))
2
= 6. (3.4)
Thus, on average, it takes 6 throws of a die before a 6 appears.
Heres another, quite different way to solve this problem. When rolling a die, there is a 1/6 chance that
a 6 will appear. If a 6 doesnt appear, then were in essence starting over. That is to say, the number
of times we expect to throw the dice before a 6 shows up is the same as the number of additional
times we expect to throw the die after throwing a non-6. So we have a 1/6 chance of rolling a 6 (and
8
stopping), and a 5/6 chance of not rolling a six, after which the number of rolls we expect to throw is
the same as when we started. We can formulate this as
E =
1
6
+
5
6
(E + 1) . (3.5)
Solving for E, we nd E = 6. Note that Equation 3.5 implicity assumes that E is a nite number,
which is something that, a priori, we do not necessarily know.
2. On average, how many times must a 6-sided die be rolled until all sides appear at least once? What
about for an n-sided die?
To roll until every side of the die appears, we begin by rolling once. We then roll until a different
side appears. Since there are 5 sides, this takes, on average,
1
5/6
= 6/5 rolls. Then we roll until a
side different from the two already rolled appears. This requires, on average,
1
4/6
= 6/4 = 3/2 rolls.
Continuing this process, and using the additive nature of expectation, we see that, on average,
1 +
6
5
+
6
4
+
6
3
+
6
2
+
6
1
=
147
10
= 14.7
rolls are needed until all 6 sides appear at least once. For an n-sided die, the number of rolls needed,
on average, is
1 +
n
n 1
+
n
n 2
+ +
n
1
=
n

i=1
n
i
= n
n

i=1
1
i
.
For large n, this is approximately nlog n.
3. Suppose we roll n dice and keep the highest one. What is the distribution of values?
Lets nd the probability that the highest number rolled is k. Among the n dice rolled, they must all
show k or less. The probability of this occuring is
k
n
6
n
.
However, some of these rolls dont actually have any ks. That is, they are made up of only the
numbers 1 through k 1. The probability of this occuring is
(k 1)
n
6
n
so the probability that the highest number rolled is k is
k
n
(k 1)
n
6
n
.
So, for instance, the probability that, if 7 dice are rolled, the highest number to turn up will be 3 is
3
7
2
7
6
7
=
2059
6
7
0.007355.
4. How many dice must be rolled to have at least a 95% chance of rolling a six? 99%? 99.9%?
Suppose we roll n dice. The probability that none of them turn up six is
_
5
6
_
n
9
and so the probability that at least one is a six is
1
_
5
6
_
n
.
To have a 95% chance of rolling a six, we need
1
_
5
6
_
n
0.95
which yields
n
log 0.05
log(5/6)
= 16.43 . . . > 16.
Hence, n 17 will give at least a 95% chance of rolling at least one six. Since log(0.01)/ log(5/6) =
25.2585 . . ., 26 dice are needed to have a 99% chance of rolling at least one six. Similarly, since
log(0.001)/ log(5/6) = 37.8877 . . ., 38 dice are needed for a 99.9% chance.
5. How many dice must be rolled to have at least a 95% chance of rolling a one and a two? What about
a one, a two, and a three? What about a one, a two, a three, a four, a ve and a six?
Solving this problem requires the use of the inclusion-exclusion principal. Of the 6
n
possible rolls of
n dice, 5
n
have no ones, and 5
n
have no twos. The number that have neither ones nor twos is not
5
n
+5
n
since this would count some rolls more than once: of those 5
n
rolls with no ones, some have
no twos either. The number that have neither ones nor twos is 4
n
, so the number of rolls that dont
have at least one one, and at least one two is
5
n
+ 5
n
4
n
= 2 5
n
4
n
and so the probability of rolling a one and a two with n dice is
1
2 5
n
4
n
6
n
.
This is an increasing function of n, and by direct calculation we can show that its greater than 0.95
for n 21. That is, if we roll at least 21 dice, there is at least a 95% chance that there will be a one
and a two among the faces that turn up.
To include threes, we need to extend the method. Of the 6
n
possible rolls, there are 5
n
rolls that have
no ones, 5
n
that have no twos, and 5
n
that have no threes. There are 4
n
that have neither ones nor
twos, 4
n
that have neither ones nor threes, and 4
n
that have neither twos nor threes. In addition,
there are 3
n
that have no ones, twos, or threes. So, the number of rolls that dont have a one, a two,
and a three is
5
n
+ 5
n
+ 5
n
4
n
4
n
4
n
+ 3
n
= 3 5
n
3 4
n
+ 3
n
.
Hence, the probability of rolling at least one one, one two, and one three is
1
3 5
n
3 4
n
+ 3
n
6
n
.
This is again an increasing function of n, and it is greater than 0.95 when n 23.
Finally, to determine the probability of rolling at least one one, two, three, four, ve and six, we extend
the method even further. The result is that the probability p(n) of rolling at least one of every possible
face is
p(n) = 1
5

j=1
(1)
(j+1)
_
6
j
__
6 j
6
_
n
= 16
_
1
6
_
n
+15
_
1
3
_
n
20
_
1
2
_
n
+15
_
2
3
_
n
6
_
5
6
_
n
.
This exceeds 0.95 when n 27. Below is a table showing some of the probabilities for various n.
10
n p(n)
6 0.0154...
7 0.0540...
8 0.1140...
9 0.1890...
10 0.2718...
11 0.3562...
12 0.4378...
13 0.5138...
14 0.5828...
15 0.6442...
16 0.6980...
17 0.7446...
18 0.7847...
19 0.8189...
20 0.8479...
21 0.8725...
22 0.8933...
23 0.9107...
24 0.9254...
25 0.9376...
26 0.9479...
27 0.9565...
30 0.9748...
35 0.9898...
40 0.9959...
6. How many dice should be rolled to maximize the probability of rolling exactly one six? two sixes? n
sixes?
Suppose we roll n dice. The probability that exactly one is a six is
_
n
1
_
5
n1
6
n
=
n5
n1
6
n
.
The question is: for what value of n is this maximal? If n > 6 then
(n+1)5
n
6
n+1
<
n5
n1
6
n
, so the maximum
must occur for some n 6. Heres a table that gives the probabilities:
n
n5
n1
6
n
1 1/6 = 0.1666...
2 5/18 = 0.2777...
3 25/72 = 0.3472...
4 125/324 = 0.3858...
5 3125/7776 = 0.4018...
6 3125/7776 = 0.4018...
This shows that the maximum probability is
3125
7776
, and it occurs for both n = 5 and n = 6.
For two sixes, the calculation is similar. The probability of exactly two sixes when rolling n dice is
_
n
2
_
5
n2
6
n
=
n(n 1)5
n2
2 6
n
A quick calculation shows that this is maximal for n = 12 or n = 11.
It seems that for n sixes, the maximal probability occurs with 6n and 6n 1 dice. Ill let you prove
that.
7. A single die is rolled until a run of six different faces appears. For example, one might roll the sequence
535463261536435344151612534 with only the last six rolls all distinct. What is the expected number
of rolls?
11
We may solve this by creating a set of linear recurrence equations. Let E
i
be the expected number of
rolls from a point where the last i rolls were distinct. We seek E
0
. We have then
E
0
= 1 + E
1
(3.6)
E
1
= 1 +
1
6
E
1
+
5
6
E
2
(3.7)
E
2
= 1 +
1
6
E
1
+
1
6
E
2
+
4
6
E
3
(3.8)
E
3
= 1 +
1
6
E
1
+
1
6
E
2
+
1
6
E
3
+
3
6
E
4
(3.9)
E
4
= 1 +
1
6
E
1
+
1
6
E
2
+
1
6
E
3
+
1
6
E
4
+
2
6
E
5
(3.10)
E
5
= 1 +
1
6
E
1
+
1
6
E
2
+
1
6
E
3
+
1
6
E
4
+
1
6
E
5
+
1
6
E
6
(3.11)
E
6
= 0 (3.12)
The last zero rolls are distinct only before the rolls have started, so E
0
= 1 + E
1
since there must
be a roll, and that takes us to the state where the last 1 roll is distinct. Then another roll occurs; at
this point, with probability 1/6 the roll is the same as the last roll, and so we remain in the same state,
or, with probability 5/6, a different face appears, and then the last two rolls are distinct. The pattern
continues this way.
Thus we have a system of seven linear equations in seven unknowns, which is solvable via many
methods. The result is
E
0
=
416
5
= 83.2
E
1
=
411
5
= 82.2
E
2
= 81
E
3
=
396
5
= 79.2
E
4
=
378
5
= 75.6
E
5
=
324
5
= 64.8
Thus, on average, it will take 83.2 rolls before getting a run of six distinct faces.
8. What is the most probable: rolling at least one six with six dice, at least two sixes with twelve dice,
or at least three sixes with eighteen dice? (This is an old problem, frequently connected with Isaac
Newton.)
One way to solve this is to simply calculate the probability of each. The probability of rolling exactly
m sixes when rolling r six-sided dice is
_
r
m
_
5
rm
6
r
so the probability of rolling at least m sixes when rolling r six-sided dice is
p(m, r) =
r

i=m
_
r
i
_
5
ri
6
r
.
12
Grinding through the calculations yields
p(1, 6) =
31031
46656
0.66510202331961591221
p(2, 12) =
1346704211
2176782336
0.61866737373230871348
p(3, 18) =
15166600495229
25389989167104
0.59734568594772319497
so that we see that the six dice case is the clear winner.
9. Suppose we roll n dice, remove all the dice that come up 1, and roll the rest again. If we repeat this
process, eventually all the dice will be eliminated. How many rolls, on average, will we make? Show,
for instance, that on average fewer than O(log n) throws occur.
We expect that, on average, 5/6 of the dice will be left after each throw. So, after k throws, we expect
to have n
_
5
6
_
k
dice left. When this is less than 2, we have, on average less than 6 throws left, so the
number of throws should be, on average, something less than a constant time log n.
Let M
n
be the expected number of throws until all dice are eliminated. Then, thinking in terms of a
Markov chain, we have the recurrence formula
M
n
=
1
6
n
+
5
n
6
n
+
_
5
6
_
n
M
n
+
n1

j=1
(1 + M
j
)
_
n
n j
_
5
j
6
n
which allows us to solve for M
n
:
M
n
=
1 + 5
n
+
n1

j=1
(1 + M
j
)
_
n
n j
_
5
j
6
n
5
n
Here are a few values of M
n
.
n Mn
1 6
2 8.72727272727273
3 10.5554445554446
4 11.9266962545651
5 13.0236615075553
6 13.9377966973204
7 14.7213415962620
8 15.4069434778816
9 16.0163673664838
10 16.5648488612594
15 18.6998719821123
20 20.2329362496041
30 22.4117651317294
40 23.9670168145374
50 25.1773086926527
13
We see that M
n
increases quite slowly, another suggestion that M
n
= O(log n). To show this, suppose
M
j
< C log j for all 2 j < n. Then we have
M
n
<
1 + 5
n
+ max1 + 6, 1 + C log(n 1)
n1

j=1
_
n
n j
_
5
j
6
n
5
n
=
1 + 5
n
+ C log(n 1)(6
n
5
n
1)
6
n
5
n
= C
_
1
1
6
n
5
n
_
log(n 1) +
1 + 5
n
6
n
5
n
< C log n
if and only if
_
1
1
6
n
5
n
_
log(n 1)
log n
+
1 + 5
n
C log n(6
n
5
n
)
< 1
Since M
2
/ log 2 < 13, we may suppose C = 13. It is not hard to show the above inequality holds for
all n, and hence M
n
< 13 log n for all n 2.
3.2 Dice Sums
10. Show that the probability of rolling 14 is the same whether we throw 3 dice or 5 dice.
This seems like a tedious calculation, and it is. To save some trouble, we can use a computer algebra
system to determine the coefcient of x
14
in the polynomials (x + x
2
+ x
3
+ x
4
+ x
5
+ x
6
)
3
and
(x +x
2
+x
3
+x
4
+x
5
+x
6
)
5
(see Appendix C for an explanation of this method). They are 15 and
540, respectively, and so the probability in question is
15
6
3
=
540
6
5
=
5
72
.
Are there other examples of this phenomenon?
Yes. Let p
d
(t, n) be the probability of rolling a sum of t with n dsided dice. Then:
p
3
(5, 2) = p
3
(5, 3) =
2
9
p
3
(10, 4) = p
3
(10, 6) =
10
81
p
4
(9, 3) = p
4
(9, 4) =
5
32
p
6
(14, 3) = p
6
(14, 5) =
5
72
p
9
(15, 2) = p
9
(15, 4) =
4
81
p
20
(27, 2) = p
20
(27, 3) =
7
200
Questions: Are there others? Can we nd all of them?
11. Suppose we roll n dice and sum the highest 3. What is the probability that the sum is 18?
In order for the sum to be 18, there must be at least three 6s among the n dice. So, we could calculate
probability that there are 3,4,5,. . . ,n 6s among the n dice. The sum of these probabilities would be the
probability of rolling 18. Since n could be much greater than 3, an easier way to solve this problem is
14
to calculate the probability that the sum is not 18, and then subtract this probability from 1. To get a
sum that is not 18, there must be 0, 1 or 2 6s among the n dice. We calculate the probability of each
occurence:
zero 6s: the probability is
5
n
6
n
one 6: the probability is
n5
n1
6
n
two 6s: the probability is
_
n
2
_
5
n2
6
n
Hence, the probability of rolling a sum of 18 is
1
_
5
n
6
n
+
n5
n1
6
n
+
_
n
2
_
5
n2
6
n
_
= 1
_
5
6
_
n
_
1 +
9
50
n +
1
50
n
2
_
= p(n)
say. Then, for example, p(1) = p(2) = 0, p(3) = 1/216, p(4) = 7/432, and p(5) = 23/648.
12. A die is rolled once; call the result N. Then N dice are rolled once and summed. What is the
distribution of the sum? What is the expected value of the sum? What is the most likely value?
Since each of the possible values 1, 2, 3, 4, 5, 6 of N are equally likely, we can calculate the distri-
bution by summing the individual distributions of the sum of 1, 2, 3, 4, 5, and 6 dice, each weighted
by
1
6
. We can do this using polynomials. Let
p =
1
6
(x + x
2
+ x
3
+ x
4
+ x
5
+ x
6
).
Then the distribution of the sum is given by the coefcients of the polynomial
D =
6

i=1
1
6
p
i
=
1
279936
x
36
+
1
46656
x
35
+
7
93312
x
34
+
7
34992
x
33
+
7
15552
x
32
+
7
7776
x
31
+
77
46656
x
30
+
131
46656
x
29
+
139
31104
x
28
+
469
69984
x
27
+
889
93312
x
26
+
301
23328
x
25
+
4697
279936
x
24
+
245
11664
x
23
+
263
10368
x
22
+
691
23328
x
21
+
1043
31104
x
20
+
287
7776
x
19
+
11207
279936
x
18
+
497
11664
x
17
+
4151
93312
x
16
+
3193
69984
x
15
+
1433
31104
x
14
+
119
2592
x
13
+
749
15552
x
12
+
2275
46656
x
11
+
749
15552
x
10
+
3269
69984
x
9
+
4169
93312
x
8
+
493
11664
x
7
+
16807
279936
x
6
+
2401
46656
x
5
+
343
7776
x
4
+
49
1296
x
3
+
7
216
x
2
+
1
36
x.
To get the expected value E, we must calculate
E =
36

i=1
id
i
where D =
36

i=1
d
i
x
i
. This works out to E =
49
4
= 12.25.
15
More simply, one can calculate the expected value of the sum as follows, using the fact that the
expected value of a single roll is 3.5:
E =
1
6
(3.5 + 2 3.5 + 3 3.5 + + 6 3.5) = 12.25.
0
0.01
0.02
0.03
0.04
0.05
0.06
5 10 15 20 25 30 35
Coefficients of D
i
c
o
e
f
f
i
c
i
e
n
t
You can see from the plot of the coefcients of D that 6 is the most likely value. It is perhaps a bit
surprising that there are three local maxima in the plot, at i = 6, 11, and 14.
13. A die is rolled and summed repeatedly. What is the probability that the sum will ever be a given value
x?
Lets start by considering 2-sided dice, with sides numbered 1 and 2. Let p(x) be the probability that
the sum will ever be x. Then p(1) = 1/2 since the only way to ever have a sum of 1 is to roll 1 on the
rst roll. We then have p(2) = 1/2 + 1/2p(1) = 3/4, since there are two mutually exclusive ways to
get a sum of 2: roll 2 on the rst roll, or roll a 1 followed by a 1 on the second roll. Now, extending
this idea, we have, for x > 2,
p(x) =
1
2
p(x 1) +
1
2
p(x 2). (3.13)
This equation could be used to calculate p(x) for any given value of x. However, this would require
calculating p for all lower values. Can we get an explicit expression for p(x)?
Equation 3.13 is an example of a linear recurrence relation. One way to get a solution, or explicit
formula, for such a relation is by examining the auxiliary equation for equation 3.13:
x
2
=
1
2
x +
1
2
or
x
2

1
2
x
1
2
= 0
The roots of this equation are
= 1 and =
1
2
16
A powerful theorem (see Appendix E) says that
p(n) = A
n
+ B
n
= A+ B
_

1
2
_
n
for constants A and B. Since p(1) = 1/2 and p(2) = 3/4 we can solve for A and B to nd that
p(n) =
2
3
+
1
3
_

1
2
_
n
.
For 3-sided dice, we have
p(1) =
1
3
, p(2) =
4
9
, and p(3) =
16
27
with, for n > 3,
p(n) =
1
3
(p(n 1) + p(n 2) +p(n 3)) =
1
3
3

i=1
p(n i).
The characteristic equation for this recurrence equation can be written
3x
3
x
2
x 1 = 0
which has roots
= 1, =
1
3

2
3
i, and =
1
3
+

2
3
i.
Using these, and the fact that
p(1) =
1
3
, p(2) =
4
9
, and p(3) =
16
27
,
we nd
p(n) =
1
2
+
1
4

n
+
1
4

n
.
Since and are complex conjugates, and, in any case, p(n) is always real, we might prefer to write
p(n) like this:
p(n) =
1
2
+
1
2
_
1

3
_
n
cos
_
n
_

2
+ tan
1
1

2
__
Using this formula to generate a table, we see that while p(n) is asymptotic to the value 1/2, it wobbles
quite a bit:
17
x p(x) p(x) p(x 1)
1 0.3333333333333333333333333333
2 0.4444444444444444444444444444 0.1111111111111111111111111111
3 0.5925925925925925925925925925 0.1481481481481481481481481481
4 0.4567901234567901234567901234 -0.1358024691358024691358024691
5 0.4979423868312757201646090534 0.04115226337448559670781893002
6 0.5157750342935528120713305898 0.01783264746227709190672153636
7 0.4901691815272062185642432556 -0.02560585276634659350708733424
8 0.5012955342173449169333942996 0.01112635269013869836915104404
9 0.5024132500127013158563227150 0.001117715795356398922928415384
10 0.4979593219190841504513200901 -0.004453928093617165405002624938
11 0.5005560353830434610803457015 0.002596713463959310629025611496
12 0.5003095357716096424626628355 -0.0002464996114338186176828660183
13 0.4996082976912457513314428757 -0.0007012380803638911312199598200
14 0.5001579562819662849581504709 0.0005496585907205336267075952194
15 0.5000252632482738929174187274 -0.0001326930336923920407317435396
16 0.4999305057404953097356706913 -0.00009475750777858318174803604667
17 0.5000379084235784958704132966 0.0001074026830831861347426052109
18 0.4999978924707825661745009051 -0.00004001595279592969591239145842
19 0.4999887688782854572601949643 -0.000009123592497108914305940764722
20 0.5000081899242155064350363887 0.00001942104593004917484142432929
Lets skip over 4- and 5-sided dice to deal with 6-sided dice. Let p(x) be the probability that the sum
will ever be x. We know that:
p(1) =
1
6
p(2) =
1
6
+
1
6
p(1) =
7
36
p(3) =
1
6
+
1
6
p(2) +
1
6
p(1) =
49
216
p(4) =
1
6
+
1
6
p(3) +
1
6
p(2) +
1
6
p(1) =
343
1296
p(5) =
1
6
+
1
6
p(4) +
1
6
p(3) +
1
6
p(2) +
1
6
p(1) =
2401
7776
p(6) =
1
6
+
1
6
p(5) +
1
6
p(4) +
1
6
p(3) +
1
6
p(2) +
1
6
p(1) =
16807
46656
and for x > 6,
p(x) =
1
6
6

i=1
p(x i).
Heres a table of the values of p(x) and p(x) p(x 1) for x 20:
18
x p(x) p(x) p(x 1)
1 0.1666666666666666666666666666
2 0.1944444444444444444444444444 0.02777777777777777777777777777
3 0.2268518518518518518518518518 0.03240740740740740740740740740
4 0.2646604938271604938271604938 0.03780864197530864197530864197
5 0.3087705761316872427983539094 0.04411008230452674897119341563
6 0.3602323388203017832647462277 0.05146176268861454046639231824
7 0.2536043952903520804755372656 -0.1066279435299497027892089620
8 0.2680940167276329827770156988 0.01448962143728090230147843316
9 0.2803689454414977391657775745 0.01227492871386475638876187573
10 0.2892884610397720537180985283 0.008919515598274314552320953784
11 0.2933931222418739803665882007 0.004104661202101926648489672418
12 0.2908302132602384366279605826 -0.002562908981635543738627618117
13 0.2792631923335612121884963084 -0.01156702092667722443946427417
14 0.2835396585074294008073228155 0.004276466173868188618826507133
15 0.2861139321373954704790406683 0.002574273629966069671717852795
16 0.2870714299200450923645845173 0.0009574977826496218855438489728
17 0.2867019247334239321389988488 -0.0003695051866211602255856684957
18 0.2855867251486822574344006235 -0.001115199584741674704598225314
19 0.2847128104634228942354739637 -0.0008739146852593631989266598476
20 0.2856210801517331745766369062 0.0009082696883102803411629425406
Notice that p(x) seems to be settling down on a value of about
2
7
. Lets prove the following (proof
idea from Marc Holtz):
lim
x
p(x) =
2
7
First, lets dene a sequence of vectors v(i):
v(i) = p(i), p(i 1), p(i 2), p(i 3), p(i 4), p(i 5)).
If we then dene the matrix M:
M =
_
_
_
_
_
_
_
_
1
6
1
6
1
6
1
6
1
6
1
6
1 0 0 0 0 0
0 1 0 0 0 0
0 0 1 0 0 0
0 0 0 1 0 0
0 0 0 0 1 0
_
_
_
_
_
_
_
_
Then its not hard to show that
Mv(i) = v(i + 1)
What we are interested in, then is M

v(j) = lim
i
v(i), where j is any nite value (but we may as
well take it to be six, since weve calculated p(1),...,p(6) already
Note that each entry of M is between 0 and 1, each row of M sums to one, and M
6
has no zero entries:
M
6
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
16807
46656
9031
46656
7735
46656
6223
46656
4459
46656
2401
46656
2401
7776
2401
7776
1105
7776
889
7776
637
7776
343
7776
343
1296
343
1296
343
1296
127
1296
91
1296
49
1296
49
216
49
216
49
216
49
216
13
216
7
216
7
36
7
36
7
36
7
36
7
36
1
36
1
6
1
6
1
6
1
6
1
6
1
6
_
_
_
_
_
_
_
_
_
_
_
_
_
_
19
So, we can consider M to be a transition matrix of a regular Markov system. Hence M

is a matrix
with all identical rows given by the vector w where the sum of the entries of w equals 1, and
wM = w.
A little simple algebra shows that
w =
_
2
7
,
5
21
,
4
21
,
1
7
,
2
21
,
1
21
_
Hence, v() is a vector of six identical probabilities equal to
w v(6) =
2
7
Thus, lim
i
=
2
7
.
More questions:
(a) Notice that while p(x) is settling down on
2
7
, it does so quite non-monotonically: p(x) increases
to its maximum at x = 6, and then wobbles around quite a bit. Is the sequence p(i) eventually
monotonic, or does it always wobble?
14. A die is rolled once. Call the result N. Then, the die is rolled N times, and those rolls which are
equal to or greater than N are summed (other rolls are not summed). What is the distribution of the
resulting sum? What is the expected value of the sum?
This is a perfect problem for the application of the polynomial representation of the distribution of
sums.
The probability of a sum of k is the coefcient on x
k
in the polynomial
1
6
_
1
6
_
x + x
2
+ x
3
+ x
4
+ x
5
+ x
6
_
_
+
1
6
_
1
6
_
1 + x
2
+ x
3
+ x
4
+ x
5
+ x
6
_
_
2
+
1
6
_
1
6
_
2 +x
3
+ x
4
+x
5
+ x
6
_
_
3
+
1
6
_
1
6
_
3 + x
4
+ x
5
+ x
6
_
_
4
+
1
6
_
1
6
_
4 +x
5
+ x
6
_
_
5
+
1
6
_
1
6
_
5 + x
6
_
_
6
=
1
279936
x
36
+
1
7776
x
30
+
5
46656
x
29
+
5
23328
x
28
+
5
23328
x
27
+
5
46656
x
26
+
1
46656
x
25
+
59
31104
x
24
+
13
5832
x
23
+
5
1296
x
22
+
11
2916
x
21
+
67
23328
x
20
+
1
486
x
19
+
1117
69984
x
18
+
23
1296
x
17
+
7
288
x
16
+
16
729
x
15
+
1
54
x
14
+
1
72
x
13
+
6781
93312
x
12
+
47
729
x
11
+
377
5832
x
10
+
67
1296
x
9
+
19
432
x
8
+
1
36
x
7
+
8077
46656
x
6
+
565
5832
x
5
+
7
108
x
4
+
5
108
x
3
+
1
27
x
2
+
1
36
x +
27709
279936
So, thats the distribution.
The expected value is simply the sum of i times the coefcient on x
i
in the distribution polynomial.
The result is
133
18
= 7.38888....
20
3.3 Non-Standard Dice
15. Show that the probability of rolling doubles with a non-fair (xed) die is greater than with a fair
die.
For a fair, n-sided die, the probability of rolling doubles with it is n
1
n
2
=
1
n
. Suppose we have
a xed n-sided die, with probabilities p
1
, ..., p
n
of rolling sides 1 through n respectively. The
probability of rolling doubles with this die is
p
2
1
+ + p
2
n
.
We want to show that this is greater than
1
n
. A nice trick is to let

i
= p
i

1
n
for i = 1, ..., n.
Then
p
2
1
+ + p
2
n
= (
1
+
1
n
)
2
+ + (
n
+
1
n
)
2
=
2
1
+ +
2
n
+
2
n
(
1
+ +
n
) +
1
n
.
Now, since p
1
+ + p
n
= 1, we can conclude that
1
+ +
n
= 0. Hence,
p
2
1
+ + p
2
n
=
2
1
+ +
2
n
+
1
n
>
1
n
precisely when not all the
i
s are zero, i.e. when the die is xed.
16. Find a pair of 6-sided dice, labelled with positive integers differently from the standaed dice, so that
the sum probabilites are the same as for a pair of standard dice.
Number one die with sides 1,2,2,3,3,4 and one with 1,3,4,5,6,8. Rolling these two dice gives the same
sum probabilities as two normal six-sided dice.
A natural question is: how can we nd such dice? One way is to consider the polynomial
(x + x
2
+ x
3
+ x
4
+ x
5
+ x
6
)
2
.
This factors as
x
2
(1 + x)
2
(1 + x + x
2
)
2
(1 x + x
2
)
2
.
We can group this factorization as
_
x(1 + x)(1 + x + x
2
)
_ _
x(1 + x)(1 + x +x
2
)(1 x + x
2
)
2
_
= (x + 2x
2
+ 2x
3
+ x
4
)(x + x
3
+ x
4
+ x
5
+ x
6
+ x
8
).
This yields the weird dice (1,2,2,3,3,4) and (1,3,4,5,6,8). See Appendix C for more about this
method.
See [1] for more on renumbering dice.
21
17. Is it possible to have two non-fair n-sided dice, with sides numbered 1 through n, with the property
that their sum probabilities are the same as for two fair n-sided dice?
Another way of asking the question is: suppose you are given two n-sided dice that exhibit the prop-
erty that when rolled, the resulting sum, as a random variable, has the same probability distribution as
for two fair n-sided dice; can you then conclude that the two given dice are fair? This question was
asked by Lewis Robertson, Rae Michael Shortt and Stephen Landry in [2]. Their answer is surprising:
you can sometimes, depending on the value of n. Specically, if n is 1,2,3,4,5,6,7,8,9,11 or 13, then
two n-sided dice whose sum acts fair are, in fact, fair. If n is any other value, then there exist pairs
of n-sided dice which are not fair, yet have fair sums.
The smallest example, with n = 10, gives dice with the approximate probabilities (see [Rob 2] for the
exact values)
(0.07236, 0.14472, 0.1, 0.055279, 0.127639, 0.127639, 0.055279, 0.1, 0.14472, 0.07236)
and
(0.13847, 0, 0.2241, 0, 0.13847, 0.13847, 0, 0.2241, 0, 0.13847).
Its clear that these dice are not fair, yet the sum probabilities for them are the same as for two fair
10-sided dice.
18. Is it possible to have two non-fair 6-sided dice, with sides numbered 1 through 6, with a uniform sum
probability? What about n-sided dice?
No. Let p
1
, p
2
, p
3
, p
4
, p
5
and p
6
be the probabilities for one 6-sided die, and q
1
, q
2
, q
3
, q
4
, q
5
and q
6
be
the probabilities for another. Suppose that these dice together yield sums with uniform probabilities.
That is, suppose P(sum = k) =
1
11
for k = 2, ..., 12. Then
p
1
q
1
=
1
11
and p
6
q
6
=
1
11
.
Also,
1
11
= P(sum = 7) p
1
q
6
+ p
6
q
1
so
p
1
1
11p
6
+ p
6
1
11p
1

1
11
i.e.,
p
1
p
6
+
p
6
p
1
1.
Now, if we let x =
p
1
p
6
, then we have
x +
1
x
1
which is impossible, since for positive real x, x +
1
x
2. Thus, no such dice are possible.
An identical proof shows that this is an impossibility regardless of the number of sides of the dice.
19. Suppose that we renumber four fair 6-sided dice (A, B, C, D) as follows: A = 0, 0, 4, 4, 4, 4,B =
1, 1, 1, 5, 5, 5,C = 2, 2, 2, 2, 6, 6,D = 3, 3, 3, 3, 3, 3.
(a) Find the probability that die A beats die D; die D beats die C; die C beats die B; and die B
beats die A.
22
(b) Discuss.
20. Find every six-sided die with sides numbered from the set 1,2,3,4,5,6 such that rolling the die twice
and summing the values yields all values between 2 and 12 (inclusive). For instance, the die numbered
1,2,4,5,6,6 is one such die. Consider the sum probabilities of these dice. Do any of them give sum
probabilities that are more uniform than the sum probabilities for a standard die? What if we
renumber two dice differently - can we get a uniform (or more uniform than standard) sum probability?
The numbers 1, 2, 5 and 6 must always be among the numbers on the die, else sums of 2, 3, 11 and
12 would not be possible. In order to get a sum of 5, either 3 or 4 must be on the die also. The last
place on the die can be any value in 1,2,3,4,5,6. Hence there are 11 dice with the required property.
Listed with their corresponding error, they are:
1,2,4,5,6,6 0.0232884399551066
1,2,4,5,5,6 0.0325476992143659
1,2,4,4,5,6 0.0294612794612795
1,2,3,5,6,6 0.0232884399551066
1,2,3,5,5,6 0.026374859708193
1,2,3,4,5,6 0.0217452300785634
1,2,3,3,5,6 0.0294612794612795
1,2,2,4,5,6 0.026374859708193
1,2,2,3,5,6 0.0325476992143659
1,1,2,4,5,6 0.0232884399551066
1,1,2,3,5,6 0.0232884399551066
The error here is the sum of the square of the difference between the actual probability of rolling each
of the sums 2 through 12 and 1/11 (the probability we would have for each sum if we had a uniform
distribution). That is, if p
i
is the probability of rolling a sum of i with this die, then the error is
12

i=2
(p
i

1
11
)
2
.
Note that the standard die gives the smallest error (i.e., the closest to uniform sum).
If we renumber two dice differently, many more cases are possible. One pair of dice are 1,3,4,5,6,6
and 1,2,2,5,6,6. These two dice give all sum values between 2 and 12, with an error (as above) of
0.018658810325477, more uniform than the standard dice. The best dice for near-uniformity are
1,2,3,4,5,6 and 1,1,1,6,6,6 which yield all the sums from 2 to 12 with near equal probability: the
probability of rolling 7 is 1/6 and all other sums are 1/12. The error is 5/792, or about 0.00631.
3.4 Games with Dice
21. Craps What is the probability of winning a round of the game Craps?
The probability of winning a round of craps can be expressed as
P(rolling 7 or 11) +

b{4,5,6,8,9,10}
P(rolling b)P(rolling b again before rolling 7).
We now evaluate each probability. The probability of rolling 7 is
6
36
=
1
6
, and the probability of
rolling 11 is
2
36
=
1
18
. Hence,
P(rolling 7 or 11) =
6
36
+
2
36
=
2
9
.
23
The following table gives the probability of rolling b, for b 4, 5, 6, 8, 9, 10. (This is the probability
of b becoming the point.)
4 3/36 = 1/12
5 4/36 = 1/9
6 5/36
8 5/36
9 4/36 =1/9
10 3/36 =1/12
Finally, we need to determine the probability of rolling b before rolling 7. Let p be the probability of
rolling b on any single roll. Rolling b before roling 7 involves rolling some number of rolls, perhaps
zero, which are not b or 7, followed by a roll of b. The probability of rolling k rolls which are not b or
7, followed by a roll of b is
_
1 p
1
6
_
k
p =
_
5
6
p
_
k
p.
Since k may be any non-negative integer value, we have
P(rolling b before rolling 7) =

k=0
_
5
6
p
_
k
p =
p
1
6
+p
.
See Appendix B for some formulas for simplifying series such as the one above. Another way of
looking at this is that the probability of rolling b before rolling a 7 is the conditional probability of
rolling b, given that either b or 7 was rolled.
We can calculate the following table:
b P(rolling b) P(rolling b again before rolling 7) P(rolling b)P(rolling b again before rolling 7)
4 1/12 1/3 1/36
5 1/9 2/5 2/45
6 5/36 5/11 25/396
8 5/36 5/11 25/396
9 1/9 2/5 2/45
10 1/12 1/3 1/36
Thus, the probability of winning a round of craps is
2
9
+
1
36
+
2
45
+
25
396
+
25
396
+
2
45
+
1
36
=
244
495
= 0.4929.
Since
244
495
=
1
2

7
990
, the odds are just slightly against the player.
22. Non-Standard Craps We can generalize the games of craps to allow dice with other than six
sides. Suppose we use two (fair) n-sided dice. Then we can dene a game analagous to craps in the
following way. The player rolls two n-sided dice. If the sum of these dice is n + 1 or 2n 1, the
player wins. If the sum of these dice is 2, 3 or 2n the player loses. Otherwise the sum becomes the
players point, and they win if they roll that sum again before rolling n + 1. We may again ask: what
is the players probability of winning?
24
23. Yahtzee There are many probability questions we may ask with regard to the game of Yahtzee. For
starters, what is the probability of rolling, in a single roll,
a) Yahtzee
b) Four of a kind (but not Yahtzee)
c) A full house
d) Three of a kind (but not Yahtzee, four of a kind or full house)
e) A long straight
f) A small straight
These questions arent too tricky, so Ill just give the probabilities here:
(a) Yahtzee:
6
6
5
=
1
1296
0.07716%
(b) Four of a kind (but not Yahtzee):
_
5
4
_
6 5
6
5
=
25
1296
1.929%
(c) A full house:
_
5
3
_
6 5
6
5
=
25
648
3.858%
(d) Three of a kind (but not Yahtzee, four of a kind or full house) :
_
5
3
_
6 5 4
6
5
=
25
162
15.432%
(e) A long straight:
2 5!
6
5
=
5
162
3.086%
(f) A small straight (but not a long straight):
2(4! (6 1) 5) + 4! (6 2) 5
6
5
=
1680
6
5
=
35
162
21.60%.
24. More Yahtzee What is the probability of getting Yahtzee, assuming that we are trying just to get
Yahtzee, we make reasonable choices about which dice to re-roll, and we have three rolls? That is,
if were in the situation where all we have left to get in a game of Yahtzee is Yahtzee, so all other
outcomes are irrelevant.
This is quite a bit trickier than the previous questions on Yahtzee. The difculty here lies in the large
number of ways that one can reach Yahtzee: roll it on the rst roll; roll four of a kind on the rst roll
and then roll the correct face on the remaining die, etc. One way to calculate the probability is to treat
the game as a Markov chain (see Appendix D for general information on Markov chains).
We consider ourselves in one of ve states after each of the three rolls. We will say that we are in
state b if we have b common dice among the ve. For example, if a roll yields 12456, well be in state
1; if a roll yields 11125, well be in state 3. Now, the goal in Yahtzee is to try to get to state 5 in three
rolls (or fewer). Each roll gives us a chance to change from our initial state to a better, or equal, state.
We can determine the probabilities of changing from state i to state j. Denote this probability by P
i,j
.
Let the 0 state refer to the initial state before rolling. Then we have the following probability matrix:
P = (P
i,j
) =
_
_
_
_
_
_
_
_
_
_
0
120
1296
900
1296
250
1296
25
1296
1
1296
0
120
1296
900
1296
250
1296
25
1296
1
1296
0 0
120
216
80
216
15
216
1
216
0 0 0
25
36
10
36
1
36
0 0 0 0
5
6
1
6
0 0 0 0 0 1
_
_
_
_
_
_
_
_
_
_
(3.14)
25
The one representing P
5,5
indicates that if we reach yahtzee, state 5, before the third roll, we simply
stay in that state. Now, the probability of being in state 5 after 3 rolls is given by

P
0,i
1
P
i
1
,i
2
P
i
2
,5
= (M
3
)
1,5
where the sum is over all triples (i
1
, i
2
, i
3
) with 0 i
j
5. Calculating M
3
gives us the probability
2783176
6
10
=
347897
7558272
0.04603.
Since
347897
7558272
=
1
21

347897
7558272
, a player will get Yahtzee about once out of every twenty-one at-
tempts.
25. Drop Dead
(a) What is the expected value of a players score?
(b) What is the probability of getting a score of 0? 1? 10? 20? etc.
(a) The player begins with ve dice, and throws them repeatedly, until no dice are left. The key factor
in calculating the expected score is the fact that the number of dice being thrown changes. When
throwing n dice, a certain number may die (i.e. come up 2 or 5), and leave j non-dead dice. The
probability of this occuring is
P
n,j
=
_
n
n j
_
2
nj
4
j
6
n
.
The following table gives P
n,j
for n and j between 0 and 5.
nj 0 1 2 3 4 5
1 1/3 2/3 0 0 0 0
2 1/9 4/9 4/9 0 0 0
3 1/27 6/27 12/27 8/27 0 0
4 1/81 8/81 24/81 32/81 16/81 0
5 1/243 10/243 40/243 80/243 80/243 32/243
When throwing n dice, the expected sum is 3.5n, if none of the dice come up 2 or 5. Let E(n)
represent the expected score starting with n dice (so were ultimately concerned with E(5)). Consider
E(1). Rolling a single die, the expected score is
E(1) = 3.5P
1,1
+ E(1)P
1,1
.
That is, in one roll, we pick up 3.5 points, on average, if we dont drop dead (so we get 3.5P
1,1
expected points), and then were in the same position as when we started (so we pick up E(1)P
1,1
expected points). We can solve this equation to get
E(1) = 3
_
2
3
__
7
2
_
= 7.
Now, suppose we start with 2 dice. The expected score is
E(2) = (2 3.5 + E(2)) P
2,2
+ E(1)P
2,1
.
26
That is, on a single roll, we pick up 2 3.5 points on average if none of the dice die, in which case
were back where we started from (and then expect to pick up E(2) points), or exactly one of the dice
die, and so we expect to pick up E(1) points with the remaining die. This equation yields
E(2) =
1
1 P
2,2
(7P
2,2
+ E(1)P
2,1
) =
56
5
.
Continuing in this way, we have the general equation
E(n) = 3.5 n P
n,n
+
n

j=1
E(j)P
n,j
which we can rewrite as
E(n) =
1
1 P
n,n
_
_
3.5 n P
n,n
+
n1

j=1
E(j)P
n,j
_
_
With this formula, we can calculate E(n) for whatever value of n we want. Here is a table of E(n):
n E(n)
1 7
2
56
5
= 11.2
3
1302
95
13.70526
4
3752
247
15.19028
5
837242
52117
16.06466
6
4319412
260585
16.57583
10
9932935945754444
577645434482545
17.19556
20 17.26399
30 17.26412371400800701809841213
100 17.26412423601867057324993502
250 17.26412422187783220247082379
6
8
10
12
14
16
18
0 5 10 15 20 25 30

27
So we see that a game of Drop Dead, using 5 dice, will have, on average, a score of about 16.06.
Further questions: Notice that if we play the game with more than 5 dice, the expected score does
not increase very much. In fact, it appears as if there is an upper bound on the expected score; that is,
it seems that there is some B so that E(n) < B for all n. What is the smallest possible value for B?
Also, we expect E(n) to always increase as n increases. Can we prove this is so?
(b) Calculating the exact probabilities of scores seems to be a bit of a pain. The easiest score to work
out is zero. To get zero, the player must roll at least one 2 or 5 on every roll. If we dene a Markov
process, with states 0, F, 5, 4, 3, 2, 1 (in that order), where 0 means a score of zero has been achieved,
F means a score greater than 0 has been achieved, and 5 through 1 are the current number of dice
being rolled, we have the following transition matrix:
P
0
=
_
_
_
_
_
_
_
_
_
_
1 0 0 0 0 0 0
0 1 0 0 0 0 0
1
243
32
243
0
80
243
80
243
40
243
10
243
1
81
16
81
0 0
32
81
8
27
8
81
1
27
8
27
0 0 0
4
9
2
9
1
9
4
9
0 0 0 0
4
9
1
3
2
3
0 0 0 0 0
_
_
_
_
_
_
_
_
_
_
Since the games takes at most ve rolls, the fth power of this matrix tells us what we want to know:
P
5
0
=
_
_
_
_
_
_
_
_
_
_
1 0 0 0 0 0 0
0 1 0 0 0 0 0
978853
4782969
3804116
4782969
0 0 0 0 0
4163
19683
15520
19683
0 0 0 0 0
55
243
188
243
0 0 0 0 0
7
27
20
27
0 0 0 0 0
1
3
2
3
0 0 0 0 0
_
_
_
_
_
_
_
_
_
_
Thus we see that the probability of achieving a score of zero is
978853
4782969
=
978853
3
14
, which is about
0.2046538457598....
The probability of achieving a score of 1 is calculatable in a similar way. Our transition matrix is
P
1
=
_
_
_
_
_
_
_
_
_
_
1 0 0 0 0 0 0
0 1 0 0 0 0 0
0
11
81
0
80
243
80
243
40
243
10
243
0
17
81
0 0
32
81
8
27
8
81
0
1
3
0 0 0
4
9
2
9
0
5
9
0 0 0 0
4
9
1
18
17
18
0 0 0 0 0
_
_
_
_
_
_
_
_
_
_
A note on this lower-left most entry: once the player has only one die left, they have a 1/6 chance of
rolling a one; but then, the die must die, which occurs with probability 1/3. Hence the 1/18 probability
of getting a score of 1 after the state of one die is attained.
28
Raising this matrix to the fth power yields
P
5
1
=
_
_
_
_
_
_
_
_
_
_
1 0 0 0 0 0 0
0 1 0 0 0 0 0
305285
14348907
4477684
4782969
0 0 0 0 0
1300
59049
18383
19683
0 0 0 0 0
17
729
226
243
0 0 0 0 0
2
81
25
27
0 0 0 0 0
1
18
17
18
0 0 0 0 0
_
_
_
_
_
_
_
_
_
_
and thus the probability of a score of 1 in this game is 305285/14348907 = 0.0212758365497804118....
Scores higher than 1 are more difcult, since it will not be necessary to reach a single die. On the
other hand, to achieve a score of n, there can be at most n+4 rolls, so the problem is nite. Appendix
F has experimental data for comparison to any exact calculations you might attempt.
26. Suppose we play a game with a die where we roll and sum our rolls as long as we keep rolling larger
values. For instance, we might roll a sequence like 1-3-4 and then roll a 2, so our sum would be 8. If
we roll a 6 rst, then were through and our sum is 6. Three questions about this game:
(a) What is the expected value of the sum?
(b) What is the expected value of the number of rolls?
(c) If the game is played with an n-sided die, what happens to the expected number of rolls as n
approaches innity?
We can consider this game as a Markov chain with an absorbing state. If we consider the state to be
the value of the latest roll, or 0 if the latest roll is not larger than the previous one, then we have the
following transition matrix:
P =
_
_
_
_
_
_
_
_
_
_
1 0 0 0 0 0 0
1/6 0 1/6 1/6 1/6 1/6 1/6
2/6 0 0 1/6 1/6 1/6 1/6
3/6 0 0 0 1/6 1/6 1/6
4/6 0 0 0 0 1/6 1/6
5/6 0 0 0 0 0 1/6
1 0 0 0 0 0 0
_
_
_
_
_
_
_
_
_
_
(3.15)
Using the notation of Appendix D, we have
Q =
_
_
_
_
_
_
_
_
0 1/6 1/6 1/6 1/6 1/6
0 0 1/6 1/6 1/6 1/6
0 0 0 1/6 1/6 1/6
0 0 0 0 1/6 1/6
0 0 0 0 0 1/6
0 0 0 0 0 0
_
_
_
_
_
_
_
_
29
so that N = (I Q)
1
is
N =
_
_
_
_
_
_
_
_
1 1/6 7/36 49/216 343/1296 2401/7776
0 1 1/6 7/36 49/216 343/1296
0 0 1 1/6 7/36 49/216
0 0 0 1 1/6 7/36
0 0 0 0 1 1/6
0 0 0 0 0 1
_
_
_
_
_
_
_
_
The row sum of N is
_
_
_
_
_
_
_
_
16807/7776
2401/1296
343/216
49/36
7/6
1
_
_
_
_
_
_
_
_
and so the expected number of rolls before absorbtion (i.e., the number of rolls that count in the sum)
is
(1/6) (16807/7776 + 2401/1296 + 343/216 + 49/36 + 7/6 + 1) = 70993/7776 1.521626.
We use N to calculate the expected sum as well. If the rst roll is a 1, the expected sum will be
1 + 2
1
6
+ 3
7
36
+ 4
49
216
+ 5
343
1296
+ 6
2401
7776
= 6.
In fact, for any rst roll, the expected sum is 6. Hence, the expected sum is 6.
Now, suppose the game is played with an n-sided die. Let E be the expected number of rolls. Let
E(j) be the expected number of rolls if the rst roll is j. Then,
E(j) = 1 +
1
n
E(j + 1) +
1
n
E(j + 2) + +
1
n
E(n)
and so
E(j + 1) = 1 +
1
n
E(j + 2) +
1
n
E(j + 3) + +
1
n
E(n)
from which we can conclude
E(j) =
_
1 +
1
n
_
E(j + 1).
Since E(n) = 1, we have
E(j) =
_
1 +
1
n
_
nj
.
Thus,
E =
1
n
n

j=1
E(j) =
1
n
n

j=1
_
1 +
1
n
_
nj
=
1
n
_
n + 1
n
_
n n

j=1
_
n
n + 1
_
j
=
_
n + 1
n
_
n
_
1
_
n
n + 1
_
n
_
=
_
n + 1
n
_
n
1.
And so we see that
lim
n
E = e 1 = 1.718281828459....
30
27. Suppose we play a game with a die in which we use two rolls of the die to create a two digit number.
The player rolls the die once and decides which of the two digits they want that roll to represent. Then,
the player rolls a second time and this determines the other digit. For instance, the player might roll
a 5, and decide this should be the tens digit, and then roll a 6, so their resulting number is 56.
What strategy should be used to create the largest number on average? What about the three digit
version of the game?
A strategy in this game is merely a rule for deciding whether the rst roll should be the tens digit or
the ones digit. If the rst roll is a 6, then it must go in the tens digit, and if its a 1, then it must
go in the ones digit. This leaves us with what to do with 2,3,4 and 5. If the rst roll is b, then using
it as the ones digit results in an expected number of
7
2
10 +b. Using it as the tens digit results in
an expected number of 10b +
7
2
. So, when is 10b +
7
2
>
7
2
10 + b? When b 4. Thus, if the rst
roll is 4, 5 or 6, the player should use it for the tens digit. With this strategy, the expected value of
the number is
1
6
(63.5 + 53.5 + 43.5 + 38 + 37 + 36) = 45.25.
In the three-digit version of the game, once we have decided what to do with the rst roll, well be
done, since we will then be in the two-digit case which we solved above. Note this is obviously true
if we place the rst roll in the hundreds digit. If we place the rst roll in the ones digit, then the
strategy to maximize the resulting number is the same as the two-digit case, simply multiplied by a
factor of ten. If we place the rst roll in the tens digit, then our strategy is to put the next roll b in
the hundreds digit if
100b + 3.5 > 350 + b
i.e., if b 4. Thus we have the same strategy in all three cases: put the second roll in the largest digit
if it is at least 4.
Now, if the rst roll, b, is placed in the hundreds digit, then the expected value will be 100b +45.25.
If the rst roll is placed in the ones digit, then the expected value will be 452.5 + b. If the rst roll
is places in the tens digit, then the expected value will be
10b + (351 + 352 + 353 + 403.5 + 503.5 + 603.5)/6 = 427.75 + 10b.
Our strategy thus comes down to maximizing the quantities 100b+45.25, 427.75+10b, and 452.5+b.
From the graph below, we see that 100b + 45.25 is the largest when b 5; 427.75 + 10b is largest
when 3 b 4, and 452.5 + b is largest when b < 3. Thus our strategy for the rst roll is this: if it
is at least 5, put it in the hundreds digit; if it is 3 or 4, put it in the tens digit; otherwise, put it in
the ones digit. If the second roll is 4, 5, or 6, place it in the largest available digit.
200
250
300
350
400
450
500
2 2.5 3 3.5 4 4.5
31
The expected value using this strategy is thus
(645.25 + 545.25 + (40 + 427.75) + (30 + 427.75) + (452.5 + 2) + (452.5 + 1))/6 = 504.
32
Chapter 4
Problems for the future
Here are some problems that I intend to add to this collection some time in the future, as soon as I get
around to writing decent solutions.
1. the game of Threes (see wikipedia). Best strategy for minimal score?
2. more drop dead: probability of getting zero? probability of any particular value?
3. If a die is rolled 100 times (say), what is the probability that all six sides have appeared at least one?
If a die is rolled 100 times (say), what is the probability that there is a run of 6 that has all six possible
faces? What about a run of 10 with all six faces at least once?
4. A die is rolled repeatedly and summed. What is the expected number of rolls until the sum is:
(a) a prime? Experimentally it appears to be around 2.432211 (one million trials).
(b) a power of 2? Hmmmm...
(c) a multiple of n? The answer appears to be n.
5. For every composite n, there appear to pairs of weird dice with n sides (i.e. a pair of dice not
numbered in the usual way with sum probabilities equal to the standard dice). Prove this. For many
n, there are many such pairs. Give useful bounds on the number of such pairs in terms of n.
For n = 4k+2, it appears that the dice 1, 2, 2, 3, 3, ..., 2k + 3, 2k + 3, 2k + 4, 1, 3, 5, ..., 2k + 1, 2k + 2, 2k + 3, ..., n
do the trick.
33
Bibliography
[1] Duane M. Broline. Renumbering of the faces of dice. Mathematics Magazine, 52(5):312314, 1979.
[2] Lewis C. Robertson, Rae Michael Shortt, and Stephen G. Landry. Dice with fair sums. Amer. Math.
Monthly, 95(4):316328, 1988.
34
References
[Ang 1] Angrisani, Massimo, A necessary and sufcient minimax condition for the loaded dice problem
(Italian), Riv. Mat. Sci. Econom. Social. 8 (1985), no. 1, 3-11
[Ble 1] Blest, David; Hallam Colin, The design of dice, Math. Today (Southend-on-Sea) 32 (1996), no. 1-2,
8-13
[Bro 1] Broline, Duane M., Renumbering of the faces of dice, Math. Mag. 52 (1979), no. 5, 312-314
[Dia 1] Diaconis, Persi; Keller, Joseph B., Fair dice, Amer. Math. Monthly, 96 (1989), no. 4, 337-339
[Faj 1] Fajtlowitz, S., n-dimensional dice, Rend. Math. (6) 4 (1971), 855-865
[Fel 1] Feldman, David; Impagliazzo, Russell; Naor, Moni; Nisan, Noam; Rudich, Steven; Shamir, Adi, On
dice and coins: models of computation for random generation, Inform. and Comput. 104 (1993), 159-174
[Fer 1] Ferguson, Thomas S., On a class of innite games related to liar dice, Ann. Math. Statist. 41 (1970)
353-362
[Fre 1] Freeman, G.H., The tactics of liar dice, J. Roy. Statist. Soc. Ser. C 38 (1989), no. 3, 507-516
[Gan 1] Gani, J., Newton on a question touching ye different odds upon certain given chances upon dice,
Math. Sci. 7 (1982), no. 1, 61-66
[Gri 1] Grinstead, Charles M., On medians of lattice distributions and a game with two dice, Combin.
Probab. Comput. 6 (1997), no. 3, 273-294
[Gup 1] Gupta, Shanti S.; Leu, Lii Yuh, Selecting the fairest of k 2, m-sided dice, Comm. Statist. Theory
Methods, 19 (1990), no. 6, 2159-2177
[Ito ] Itoh, Toshiya, Simulating fair dice with biased coins, Inform. and Compu. 126 (1996), no.1, 78-82
[Koo ] Koole, Ger, An optimal dice rolling policy for risk, Nieuw Arch. Wisk. (4) 12 (1994), no. 1-2, 49-52
[Mae 1] Maehara, Hiroshi; Yamada, Yasuyuki, How Many icosahedral dice?, Ryukyu Math. J. 7 (1994),
35-43
[McM 1] McMullen, P., A dice probability problem, Mathematika 21 (1974), 193-198
[Pom 1] Pomeranz, Janet Bellcourt, The dice problem - then and now, College Math. J. 15 (1984), no. 3,
229-237
[Rob 1] Roberts, J. D., A theory of biased dice, Eureka 1955 (1955), no. 18, 8-11
[Rob 2]* Robertson, Lewis C., Shortt, Rae Michael; Landry, Stephen G., Dice with fair sums, Amer. Math.
Monthly 95 (1988), no. 4, 316-328
[Sav 1] Savage, Richard P., Jr., The paradox of non-transitive dice, Amer. Math. Monthly 101 (1994), no. 5,
429-436
[She 1] Shen, Zhizhang; Marston, Christian M., A Study of a dice problem, Appl. Math. Comput. 73 (1993),
no. 2-3, 231-247
[Sol 1] Sol de Mora-Charles, Leibniz et le probleme des partis. Quelches papiers inedits [Leibniz and the
problem of dice. Some unpublished papers], Historia Math. 13 (1986), no. 4,352-369
[Wys 1] Wyss, Roland, Identitaten bei den Stirling-Zahlen 2. Art aus kombinatorischen Uberlegungen bein
Wurfelspiel [Identities for Stirling numbers of the second kind from combinatorial considerations in a game
of dice], Elem. Math. 51 (1996), no. 3,102-106
35
Appendix A
Dice sum probabilities
Sums of 2, 6-Sided Dice
Sum Probability
2 1/36
3 2/36 = 1/18
4 3/36 = 1/12
5 4/36 = 1/9
6 5/36
7 6/36 = 1/6
8 5/36
9 4/36 =1/9
10 3/36 =1/12
11 2/36 =1/18
12 1/36
Sums of 3, 6-Sided Dice
Sum Probability
3 1/216
4 3/216 = 1/72
5 6/216 = 1/36
6 10/216 = 5/108
7 15/216 = 5/72
8 21/216 = 7/72
9 25/216
10 27/216 = 1/8
11 27/216 = 1/8
12 25/216
13 21/216 = 7/72
14 15/216 = 5/72
15 10/216 = 5/108
16 6/216 = 1/36
17 3/216 = 1/72
18 1/216 = 1/216
36
Appendix B
Handy Series Formulas
For [r[ < 1,

n=0
ar
n
=
a
1 r
(B.1)
For [r[ < 1,
N

n=0
ar
n
=
a(1 r
N+1
)
1 r
(B.2)
For [r[ < 1,

n=1
nr
n
=
r
(1 r)
2
(B.3)
For [r[ < 1,
N

n=1
nr
n
=
r
_
1 r
N
(1 + N Nr)
_
(1 r)
2
(B.4)
For [r[ < 1,

n=1
(n + k)r
n
=
r(1 + k kr)
(1 r)
2
(B.5)
37
Appendix C
Dice Sums and Polynomials
Very often in mathematics a good choice of notation can take you a long way. An example of this is the following
method for representing sums of dice. Suppose we have an n-sided die, with sides 1, 2, ..., n that appear with proba-
bility p
1
, p
2
, ..., p
n
, respectively. Then, if we roll the die twice and add the two rolls, the probability that the sum is k
is given by
n

j=1
p
j
p
kj
=
nk

j=k1
p
j
p
kj
(C.1)
if we say p
i
= 0 if i < 1 or i > n.
Now consider the following polynomial:
P = p
1
x + p
2
x
2
+ + p
n
x
n
(C.2)
If we square P, we get
P
2
= a
2
x
2
+ a
3
x
3
+ + a
2n
x
2n
(C.3)
where a
k
, for k=2, 3,. . . , 2n, is given by
a
k
=
nk

j=k1
p
j
p
kj
. (C.4)
In other words, the probability of rolling the sum of k is the same as the coefcient of x
k
in the polynomial given by
squaring the polynomial P.
Heres an example. Suppose we consider a standard 6-sided die. Then
P =
1
6
x +
1
6
x
2
+
1
6
x
3
+
1
6
x
4
+
1
6
x
5
+
1
6
x
6
(C.5)
and so
P
2
=
x
2
36
+
2x
3
36
+
3x
4
36
+
4x
5
36
+
5 x
6
36
+
6x
7
36
+
5 x
8
36
+
4x
9
36
+
3x
10
36
+
2x
11
36
+
x
12
36
(C.6)
=
x
2
36
+
x
3
18
+
x
4
12
+
x
5
9
+
5 x
6
36
+
x
7
6
+
5 x
8
36
+
x
9
9
+
x
10
12
+
x
11
18
+
x
12
36
(C.7)
And so, we see that the probability of rolling a sum of 9, for instance, is 1/9.
For two different dice the method is the same. For instance, if we roll a 4-sided die, and a 6-sided die, and sum
them, the probability that the sum is equal to k is give by the coefcient on x
k
in the polynomial
_
x
4
+
x
2
4
+
x
3
4
+
x
4
4
__
x
6
+
x
2
6
+
x
3
6
+
x
4
6
+
x
5
6
+
x
6
6
_
(C.8)
which, when expanded, is
x
2
24
+
x
3
12
+
x
4
8
+
x
5
6
+
x
6
6
+
x
7
6
+
x
8
8
+
x
9
12
+
x
10
24
. (C.9)
38
Notice that this can be written as
1
24
_
x
2
+ 2 x
3
+ 3 x
4
+ 4 x
5
+ 4 x
6
+ 4 x
7
+ 3 x
8
+ 2 x
9
+ x
10
_
(C.10)
In general, a fair n-sided die can be represented by the polynomial
1
n
_
x + x
2
+ x
3
+ + x
n
_
(C.11)
With this notation, many questions about dice sums can be transformed into equivalent questions about polyno-
mials. For instance, asking whether or not there exist other pairs of dice that give the same sum probabilities as a pair
of standard dice is the same as asking: in what ways can the polynomial
(x + x
2
+ x
3
+ x
4
+ ... + x
n
)
2
be factored into two polynomials (with certain conditions on the degrees and coefcients of those polynomials)?
39
Appendix D
Markov Chain Facts
A Markov chain is a mathematical model for describing a process that moves in a sequence of steps through a set
of states. A nite Markov chain has a nite number of states, s
1
, s
2
, . . . , s
n
. When the process is in state s
i
, there
is a probability p
ij
that the process will next be in state s
j
. The matrix P = (p
ij
) is called the transition matrix for
the Markov chain. Note that the rows of the matrix sum to 1.
The ij-th entry of P
k
(i.e. the k-th power of the matrix P) gives the probability of the process moving from state
i to state j in exactly k steps.
An absorbing state is one which the process can never leave once it is entered. An absorbing chain is a chain
which has at least one absorbing state, and starting in any state of the chain, it is possible to move to an absorbing
state. In an absorbing chain, the process will eventually end up in an absorbing state.
Let P be the transition matrix of an absorbing chain. By renumbering the states, we can always rearrange P into
canonical form:
P =
_
_
J O
R Q
_
_
where J is an identity matrix (with 1s on the diagonal and 0s elsewhere) and O is a matrix of all zeros. Q and R are
non-negative matrices that arise from the transition probabilities between non-absorbing states.
The series N = I + Q + Q
2
+ Q
3
+ . . . converges, and N = (I Q)
1
. The matrix N gives us important
information about the chain, as the following theorem shows.
Theorem 1 Let P be the transition matrix for an absorbing chain in canonical form. Let N = (I Q)
1
. Then:
The ij-th entry of N is the expected number of times that the chain will be in state j after starting in state i.
The sum of the i-th row of N gives the mean number of steps until absorbtion when the chain is started in state
i.
The ij-th entry of the matrix B = NRis the probability that, after starting in non-absorbing state i, the process
will end up in absorbing state j.
An ergodic chain is one in which it is possible to move from any state to any other state (though not necessarily
in a single step).
A regular chain is one for which some power of its transition matrix has no zero entries. A regular chain is
therefore ergodic, though not all ergodic chains are regular.
Theorem 2 Suppose P is the transition matrix of an ergodic chain. Then there exists a matrix A such that
lim
k
P + P
2
+ P
3
+ + P
k
k
= A
For regular chains,
lim
k
P
k
= A.
40
The matrix A has each row the same vector a = (a
1
, a
2
, . . . , a
n
). One way to interpret this is to say that the
long-term probability of nding the process in state i does not depend on the initial state of the process.
The components a
1
, a
2
, . . . , a
n
are all positive. The vector a is the unique vector such that
a
1
+ a
2
+ + a
n
= 1
and
aP = a
For this reason, a is sometimes called the xed point probability vector.
The following theorem is sometimes called the Mean First Passage Theorem.
Theorem 3 Suppose we have a regular Markov chain, with transition matrix P. Let E = (e
ij
) be a matrix where,
for i ,= j, e
ij
is the expected number of steps before the process enters state j for the rst time after starting in state i,
and e
ii
is the expected number of steps before the chain re-enters state i. Then
E = (I Z + JZ

)D
where Z = (I P A)
1
,A = lim
k
P
k
, Z

is the diagonal matrix whose diagonal entries are the same as Z, J is

the matrix of all 1s, and D is a diagonal matrix with D
ii
= 1/A
ii
.
41
Appendix E
Linear Recurrence Relations
Heres a useful theorem:
Theorem 1 Consider the linear recurrence relation
x
n
= a
1
x
n1
+ a
2
x
n2
+ + a
k
= 0 (E.1)
If the polynomial equation
x
n
a
1
x
n1
a
2
x
n2
a
k
= 0 (E.2)
has k distinct roots r
1
, ...r
k
, then the recurrence relation E.1 has as a general solution
x
n
= c
1
r
n
1
+ c
2
r
n
2
+ + c
k
r
n
k
. (E.3)
Proof: We can prove this with a bit of linear algebra, but well do that some other time.
Example: May as well do the old classic. The Fibonacci numbers are dened by
f
0
= 1, f
1
= 1, and f
n
= f
n1
+ f
n2
for n > 1.
The characteristic equation is
x
2
x 1 = 0
which has roots
r
1
=
1 +

5
2
and r
2
=
1

5
2
.
So
f
n
= A
_
1 +

5
2
_
n
+ B
_
1

5
2
_
n
for constants A and B. Since
f
0
= 1 = A + B
and
f
1
= 1 = A + B +

5
2
(A B)
we conclude that
A =
1 +

5
2

5
and B =
1 +

5
2

5
so that
f
n
=
(1 +

5)
n+1
(1

5)
n+1
2
n+1

5
.
42
Appendix F
Drop Dead Simulation Data
In 10
7
simulated games of Drop Dead, the following frequencies of scores were achieved.
score (n) number of games with score n number of games with score n
0 2044719 2044719
1 212885 2257604
2 69266 2326870
3 226785 2553655
4 353402 2907057
5 180453 3087510
6 338023 3425533
7 335320 3760853
8 220094 3980947
9 258632 4239579
10 321143 4560722
11 233060 4793782
12 259303 5053085
13 257301 5310386
14 240123 5550509
15 225187 5775696
16 241364 6017060
17 223916 6240976
18 211570 6452546
19 208928 6661474
20 199351 6860825
21 184160 7044985
22 179823 7224808
23 170278 7395086
24 158398 7553484
25 150177 7703661
26 144061 7847722
27 133669 7981391
28 126951 8108342
29 120721 8229063
30 112686 8341749
31 107082 8448831
32 101481 8550312
33 95625 8645937
34 90543 8736480
35 85022 8821502
43
36 80163 8901665
37 74323 8975988
38 70833 9046821
39 66156 9112977
40 61407 9174384
41 57766 9232150
42 53754 9285904
43 50170 9336074
44 47367 9383441
45 44263 9427704
46 41087 9468791
47 38652 9507443
48 35709 9543152
49 32950 9576102
50 31230 9607332
51 28709 9636041
52 26898 9662939
53 25022 9687961
54 23160 9711121
55 21400 9732521
56 19890 9752411
57 18385 9770796
58 17360 9788156
59 16018 9804174
60 14872 9819046
61 13627 9832673
62 12830 9845503
63 11798 9857301
64 10968 9868269
65 10345 9878614
66 9332 9887946
67 8584 9896530
68 8028 9904558
69 7478 9912036
70 6846 9918882
71 6220 9925102
72 5928 9931030
73 5411 9936441
74 5091 9941532
75 4660 9946192
76 4263 9950455
77 3796 9954251
78 3750 9958001
79 3401 9961402
80 3077 9964479
81 2940 9967419
82 2674 9970093
83 2455 9972548
84 2174 9974722
85 2087 9976809
86 2019 9978828
87 1701 9980529
88 1650 9982179
89 1504 9983683
44
90 1334 9985017
91 1234 9986251
92 1149 9987400
93 1056 9988456
94 975 9989431
95 845 9990276
96 817 9991093
97 772 9991865
98 676 9992541
99 615 9993156
100 583 9993739
101 539 9994278
102 501 9994779
103 460 9995239
104 395 9995634
105 354 9995988
106 326 9996314
107 318 9996632
108 285 9996917
109 247 9997164
110 223 9997387
111 225 9997612
112 194 9997806
113 160 9997966
114 175 9998141
115 148 9998289
116 140 9998429
117 119 9998548
118 138 9998686
119 105 9998791
120 99 9998890
121 96 9998986
122 105 9999091
123 64 9999155
124 75 9999230
125 54 9999284
126 54 9999338
127 65 9999403
128 52 9999455
129 48 9999503
130 45 9999548
131 28 9999576
132 32 9999608
133 33 9999641
134 37 9999678
135 27 9999705
136 27 9999732
137 34 9999766
138 12 9999778
139 23 9999801
140 21 9999822
141 12 9999834
142 17 9999851
143 12 9999863
45
144 13 9999876
145 14 9999890
146 15 9999905
147 5 9999910
148 13 9999923
149 6 9999929
150 5 9999934
151 7 9999941
152 5 9999946
153 8 9999954
154 3 9999957
155 2 9999959
156 7 9999966
157 1 9999967
158 2 9999969
159 1 9999970
160 2 9999972
161 1 9999973
162 2 9999975
163 3 9999978
164 3 9999981
165 2 9999983
166 4 9999987
167 2 9999989
170 2 9999991
172 1 9999992
174 2 9999994
175 1 9999995
177 2 9999997
178 1 9999998
181 1 9999999
192 1 1000000
46