Generating Functions Instructor: Paddy

Lecture 3: Nonstandard Dice

Week 1 of 1 Mathcamp 2010

Definition. Define a k-die as a k-sided shape on which symbols s1 , . . . sk ∈ N+ are drawn.

Analogously, we can define a k-die to be a bucket with k balls in it, each stamped with a
symbol si ∈ N+ . In this sense, “rolling” our die corresponds to picking a ball out of our
bucket; for intuitive purposes, pick whichever model makes more sense and feel free to use
it throughout this lecture.

Question 1 Given a k-die D, let an denote the number of ways in which rolling D yields
a n. What is the generating function of {an }∞
n=0 ?

Answer: Just
∞
X
an xn ,
n=0

where an can be thought of as the number of balls in D’s bucket with a n stamped on them.

Definition. A standard k-die is a k-die for which si = i, for every i.

Question 2 If D is a standard k-die, what is D’s generating function?

Answer: D has exactly one ball labeled si for every 1 ≤ i ≤ k, and no other balls.
Consequently, we can write D’s generating function as
k k−1
X
n
X xk − 1
Sk (x) = x =x· xn = x · .
x−1
n=1 n=0

Question 3 Let D1 and D2 be a pair of dice, and let an denote the number of ways in
which rolling both dice can yield a pair of numbers that sum to n. What is the generating
function for {an }∞
n=0 ?

Answer: Let D1 ’s generating function be given by

∞
X
A1 (x) = bn xn ,
n=0

and D2 ’s generating function be given by

∞
X
A2 (x) = cn xn .
n=0

1
 If we roll both D1 and D2 and get a sum of n, we know that D1 has to have came up k
and D2 has to have came up n − k, for some k. For a fixed value of k, there are precisely
bk cn−k ways in which this can happen; thus, the generating function for rolling both dice is
given by
∞ n
!
X X
A1,2 (x) = bk cn−k xn .
n=0 k=0

But this is just the product of the two power series A1 (x) and A2 (x)! So, we have that

A1,2 (x) = A1 (x) · A2 (x);

in other words, to get the generating function for the sum of two dice, we can simply take
the product of their individual generating functions! (It bears noting that this will trivially
generalize to rolling n dice at once.)
So, we’ve finally cleared enough ground to state the driving question of this talk:

Question 4 For given k, n in N, are there n nonstandard k-dice D1 , . . . Dn with generating

functions A1 (x), . . . An (x) such that
n
xk − 1

D1 (x) · . . . · Dn (x) = x · ?
x−1

Equivalently: for given k and n, are there n nonstandard k-dice that have the same sum
probabilities as a collection of n standard k-dice?

Answer: First, notice that the equivalence of the above statements stems from the ob-
servation that if two generating functions are equal, then all of their coefficients are equal;
consequently, having the same generating functions is equivalent to having the same sum
probabilities.
k −1 n
 
What are we trying to do? Well, we want to factor the polynomial x · xx−1 into n
distinct polynomials A1 (x), . . . An (x) that are each equivalent to a nonstandard k-die. So,
what conditions do we then need on the Ai (x) to insure that they are nonstandard k-dice?

1. x is a factor of Ai (x) – i.e. Ai (0) = 0. This insures that it is a die, as (by definition)
we don’t allow symbols that correspond to 0 in our dice.

2. Ai (x) corresponds to a die with k faces: i.e. Ai (1), the sum of all of the coefficients
of Ai (x), is k.
xk −1
3. Ai (x) 6= x · x−1 . This insures that it’s nonstandard.
 n
xk −1
By condition 1 above, we know that each Ai (x) takes one of the x’s in x· x−1 .
xk −1
Thus, our question really boils down to finding ways to factor x−1 into distinct parts!
How can we do this?

2
Deus Ex Machina 5
xk − 1 Y
= Φd (x),
x−1
d|k,d>1

where the polynomials Φd (x) are the cyclotomic polynomials

Y
Φd (x) = (x − ω),
ω

where the product above is taken over all primitive d-th roots of unity1

Deus Ex Machina 6 The cyclotomic polynomials are all irreducible.

Deus Ex Machina 7 The following is a list of some of the cyclotomic polynomials:

Φ1 (x) = x − 1
Φ2 (x) = x + 1
Φ3 (x) = x2 + x + 1
Φ4 (x) = x2 + 1
Φ5 (x) = x4 + x3 + x2 x + x + 1
Φ6 (x) = x2 − x + 1
Φ7 (X) = x6 + x5 + x4 + x3 + x2 + x + 1
Φ8 (x) = x4 + 1
Φ9 (x) = x6 + x3 + 1
Φ10 (x) = x4 − x3 + x2 − x + 1

The proofs of the above observations are really beautiful and far beyond the reach of
this course; but their results are fantastically useful! For example, we have the immediate
observation:

Corollary 8 If k is prime, no collection of n nonstandard k-dice can have the same sum
probability as a collection of n standard k-dice.

Proof. Because k is prime, it has no nontrivial factors beyond itself; so, by the above,

xk − 1 Y
= Φd (x) = Φk (x)
x−1
d|k,d>1
 n
xk −1
is irreducible. Thus, by our earlier observations, there’s no way to factor x · x−1 into
anything that’s not a set of n standard k-dice!
1
A primitive d-th root of unity is a complex number ω such that ω d = 1 and ω i 6= 1, for any 1 ≤ i < d.

3
Proposition 9 If k = 6 n = 2, then there is precisely one pair of nonstandard 6-dice with
the same probability as a pair of standard 6-dice.

Proof. So: by the above, we seek to factor

 6 2
x −1
x = (xΦ2 (x)Φ3 (x)Φ6 (x))2
x−1
= x2 (x + 1)2 (x2 + x + 1)2 (x2 − x + 1)2

into two polynomials A1 (x), A2 (x) such that

1. Ai (0) = 0,

2. Ai (1) = 6, and

3. Ai (x) 6= x(x + 1)(x2 + x + 1)(x2 − x + 1).

So: because x + 1 is 2 at x = 1, x2 + x + 1 is 3 at x = 1, and x2 − x + 1 is 1 at x = 1, we

know that each Ai (x) has to have exactly one copy of both x + 1 and x2 + x + 1 in it in
order for Ai (1) to be 6.
Consequently, the only way we can have both of these dice not be standard is if

A1 (x) = x(x + 1)(x2 + x + 1)(x2 − x + 1)2 = x8 + x6 + x5 + x4 + x3 + x,

A2 (x) = x(x + 1)(x2 + x + 1) = x4 + 2x3 + 2x2 + x;

i.e. we have one die with faces {8, 6, 5, 4, 3, 1} and one die with faces {4, 3, 3, 2, 2, 1}.