VOL. 92, NO.

2, APRIL 2019 123

A Probability Perspective to a Combinatorics

Problem
HIDEO HIROSE
Data Science Research Center
Hiroshima Institute of Technology
[email protected]

This paper gives a simple proof that the number of solutions to

x1 + x2 + · · · + xn ≡ k (mod p) (1)

is p n−1 for each k ∈ {0, 1, . . . , p − 1} = [p − 1], where xi ∈ [p − 1] for i = 1, . . . , n.

For example, when p = 2, and n = 2, then the solutions to x1 + x2 ≡ 0 (mod 2)
are (x1 , x2 ) = (0, 0) and (1, 1), while the solutions to x1 + x2 ≡ 1 (mod 2) are
(x1 , x2 ) = (0, 1) and (1, 0). The number of solutions is 22−1 = 2 in each case. As
another example, when p = 2, and n = 3, then the solutions to x1 + x2 + x3 ≡ 0
(mod 2) are (x1 , x2 , x3 ) = (0, 0, 0), (0, 1, 1), (1, 0, 1), and (1, 1, 0); and, the solu-
tions to x1 + x2 + x3 ≡ 1 (mod 2) are (x1 , x2 , x3 ) = (0, 0, 1), (0, 1, 0), (1, 0, 0), and
(1, 1, 1). In each case, the number of solutions is 23−1 = 4. It seems to be cumbersome
to write down the solutions for general p, k, and n.
Equation (1) is the same as

x1 + x2 + · · · + xn = k + jp (0 ≤ j < n), (2)

which can be regarded as an extension of the classic combinatorics problem of showing

that the number of nonnegative integer solutions to

x1 + x2 + · · · + xn = k (3)
n+k−1
is k
, as in [2]. For example, the solution x1 = 2, x2 = 1, x3 = 3 to x1 + x2 +
x3 = 6 can be represented as 11 ∗ 1 ∗ 111, where the six 1s are allocated among the
8 (= 6 + 3 − 1) places. This is equivalent to the allocation of 2 (= 3 − 1) “stars”
(given by the *s) into the eight places. (This method is referred to as the “stars and
bars” method
 in [1].) The number of ways to allocate the k 1s among the n + k − 1
places is n+k−1k
. However, the restriction of the xi ’s to [p − 1] seems to preclude this
approach to determine the number of solutions to equation (1).
A more general problem, sometimes called the “donut shop problem,” could be use-
ful to find the number of nonnegative integer solutions to equation (3) under inequality
constraints, where 0 ≤ ai ≤ xi ≤ bi for i = 1, . . . , n, as considered in [1, 8]. This is
equivalent to the problem of finding the number of nonnegative integer  solutions to
y1 + y2 + · · · + yn = k − s, where 0 ≤ yi ≤ bi − ai for each i and s = ni=1 ai . To
obtain the solution, we can use the inclusion–exclusion principle, as in [3], to get the
following formula

|A1 ∪ · · · ∪ An | = (−1)|I |+1 |AI |, (4)
∅=I ⊆{1,...,n}

where | · | denotes the number of elements in a set and AI = ∩i∈I Ai .

Math. Mag. 92 (2019) 123–125. doi:10.1080/0025570X.2019.1549890 c Mathematical Association of America
MSC: Primary 60C05, Secondary 05A19
124 MATHEMATICS MAGAZINE
For example, when p = 6, x1 + x2 ≡ 1 (mod 6), and xi ∈ [5] for each i, there
are two cases: x1 + x2 = 1 and x1 + x2 = 7. In the former case, the number of
cases is |U | = 2+1−1 1
= 2, where |U | represents the number of solutions without
any restrictions;
  here U can be thought
2+(7−6)−1  of as the universal set. In the latter case,
|U | = 2+7−1
7
= 8, |A 1 | = 7−6
= 2 = |A2 |, and |A1 ∩ A2 | = ∅, thus the num-
ber of solutions is 8 − 2 − 2 = 4. By combining these two cases, the total number of
solutions is 2 + 4 = 6.
Similarly, when x1 + x2 ≡ 2 (mod 6), we deal with two cases: x1 + x2= 2 and
x1 + x2 = 8. In the former case, the number of solutions is |U | = 2+2−1 = 3. In
2+8−1 2+(8−6)−1 2
the latter case, |U | = 8
= 9, |A1 | = 8−6
= 3 = |A2 |, and |A1 ∩ A2 | = ∅,
thus the number of solutions is 9 − 3 − 3 = 3. By combining these two cases, there
are 3 + 3 = 6 solutions.
This method seems to be cumbersome, too. Determining the number of solutions to
equation (1) seems to be tough to solve using a combinatorial approach.
Consider the following probabilistic interpretation of x1 + x2 ≡ k (mod 6). We
define a rule to move counterclockwise around the unit circle starting at (1, 0). Roll
a six-sided die and when it shows j dots, then move (j − 1)(2π/6) (j = 1, 2, . . . , 6)
counterclockwise along the circumference from your current position. What is the
probability that you will be at position k(2π/6) (k = 0, 1, . . . , 5) along the unit circle
after throwing the die two times?
In the next section, I show how counting the number of solutions to equation (1) is
easier in the probabilistic context.

Probability problem
Suppose that the probability question from the previous section is generalized to
a p-sided die, where rolling a j ∈ [p − 1] results in moving counterclockwise
(j − 1)(2π/p) along the circumference. Each xi from equation (1) corresponds to
a roll of the p-sided die. Take a second to think about how determining the probability
of landing on k(2π/p) after n rolls is equivalent to the number of solutions to equation
(1).
Theorem 1 (Probability). Let Xi be the random variable for which P (Xi = j ) = p1

for j = 0, 1, . . . , p − 1. Define Sn = ni=1 Xi . Then, P (Sn ≡ k (mod p)) = p1 for
k = 0, 1, . . . , p − 1.
Proof. When you throw a p-sided die one time, the probability that you are at the
position k(2π/p) is p1 for all k (k = 0, 1, . . . , p − 1). This means that a second throw
provides equal probability to each position you proceed regardless of the last position;
this is the Markov property. Therefore, throwing a die n times also provides equal
probability of p1 to each position of k(2π/p) for all k (k = 0, 1, . . . , p − 1). 
Since each Xi in Theorem 1 has p possibilities, the total number of cases of
X1 , X2 , . . . , Xn is p n . Because the probability P (Sn ≡ k (mod p)) = p1 , then the
total number of cases in which Sn ≡ k (mod p) is p n−1 for all k ∈ [p − 1].

Combinatorial problem
To interpret the probability problem to the combinatorial problem, we only regard
xi ∈ {0, 1, . . . , p − 1}, (i = 1, . . . , n) in equation (1) as the samples from a discrete
uniform distribution. This delivers the answer to the combinatorial problem.
VOL. 92, NO. 2, APRIL 2019 125
Theorem 2 (Combinatorics). For k ∈ [p − 1] and xi ∈ [p − 1] for each i, there are
p n−1 solutions x = (x1 , x2 , . . . , xn ) to
x1 + x2 + · · · + xn ≡ k (mod p).

Conclusion

In this note, we provided a relatively easy solution to a problem in combinatorics via

a probabilistic approach. There are other instances in which probabilistic proofs of
nonprobabilistic results are easier, too. See, for example, Rosalsky [7], Peterson [6],
and Kataria [4, 5].

Acknowledgment The author would like to thank the Editor and referees for valuable comments.

REFERENCES
[1] Brualdi, R. A. (2017). Introductory Combinatorics, 5th ed. Englewood Cliffs, NJ: Pearson.
[2] Fomin, D., Genkin, S., Itenberg, I. V. (1996). Mathematical Circles: Russian Experience. Mathematical
World, Vol. 7. Providence, RI: American Mathematical Society.
[3] Jukna, S. (2011). Extremal Combinatorics, 2nd ed. Heidelberg: Springer.
[4] Kataria, K. K. (2016). A probabilistic proof of the multinomial theorem. Amer. Math. Monthly. 123: 94–96.
[5] Kataria, K.K. (2017). Some probabilistic interpretations of the multinomial theorem. Math. Mag. 90: 221–
224.
[6] Peterson, J. (2013). A probabilistic proof of a binomial identity. Amer. Math. Monthly. 120: 558–562.
[7] Rosalsky, A. (2007). A simple and probabilistic proof of the binomial theorem. Amer. Statist. 61: 161–162.
[8] University of North Dakota Mathematics Department (2010). Math408: Combinatorics, p. 9. http://arts-
sciences.und.edu/math/ files/docs/courses/supp/math408notesold.pdf

Summary. A problem that seems to be tough in a field sometimes becomes easy to solve by looking at it from
a different field. In this note, a problem in combinatorics is framed as a problem in probability, where it becomes
easier to solve.

HIDEO HIROSE (MR Author ID: 620160) received his Ph.D. in Engineering from Nagoya University. He held
visiting positions at Stanford University before joining the faculty at Kyushu Institute of Technology, where he
was eventually appointed vice dean of the university. Then, he moved to Hiroshima Institute of Technology, and
was appointed director of Data Science Research Center.