Multinomial coefficients

Posted on March 22, 2017

This is the fourth post in a series of posts on combinatorial analysis. The post is opened with the
following problem. This post builds on the previous post on binomial coefficients.

Figure 1 – three choices for lunch

Multinomial Lunch Problem

Each of the twelve people have 3 choices for lunch – McDonald, Burger King and IN-N-OUT, all
fast food restaurants. The choice for each person is independent of the choices of the other diners.
1. In how many ways can the lunch choices for this 12-person group be made?
2. In how many ways can the lunch choices for this 12-person group be made in such a way that
3 people choose McDonald, 4 people choose Burger King and 5 people choose IN-N-OUT?
3. In how many ways can the lunch choices for this 12-person group be made in such a way that
3 people choose one restaurant, 4 people choose another restaurant and 5 people choose the
remaining restaurant?
The previous posts on combinatorial analysis are: binomial coefficients, combination and
permutation and multiplication principle.
___________________________________________________________________________

Multinomial Coefficients
The problem for lunch choices is a multinomial coefficient problem. The second question in the
problem is equivalent to any one of the following question.

1. How many ways can a set of 12 distinct objects be divided into 3 subgroups, one consisting of
3 objects, one consisting of 4 objects and one consisting of 5 objects?
2. How many different 12-letter strings are there consisting of 3 M’s, 4 B’s and 5 I’s?
3. Suppose each of the questions in a 12-question multiple choice test has three choices (M, D
and I). A student chooses the answers by pure guessing. How many of the possible test
answers consist of 3 M’s, 4 B’s and 5 I’s?
 4. How many ways can 12 people be assigned to 3 committees such that one committee consists
of 3 people, one committee consists of 4 people and one committee consists of 5 people?
Assume that each person is assigned to only one committee.
5. An urn contains three letters M, B and I. Randomly sample 12 letters from the urn with
replacement. What is the number of sample results that consist of 3 M’s, 4 B’s and 5 I’s?
The first question is basically the definition of multinomial coefficient. It is the total number of ways
to divide a set of objects into several subsets such that each subset is of a pre-determined size. In the
second question, the objects are the 12 positions in the letter strings. In the third question, the objects
to be divided are the 12 multiple choice questions. In the fourth question, the objects to be divided
are the 12 people to be assigned into committees.

The fifth question is from a random sampling perspective. In any of the question, the dividing can be
thought of as a choice (or decision) for each object. For each of the 12 objects, should the object be
put into group 1, group 2 or group 3 (or should choice 1 be taken, choice 2 be taken, choice 3 be
taken)? In the lunch problem, each diner has 3 choices. The choices in their totality would be viewed
as the results from a random sampling of the population of 3.

The answer to any of the above questions is a multinomial coefficient.

Multinomial Coefficient

If  , the notation for the multinomial coefficient is   and is

defined by  . The multinomial coefficient is the number of ways

a set of   distinct objects can be divided into   subsets, one of which consists of   objects, one of
which consists of   objects, …., one of which consists of   objects.

It is clear that multinomial coefficients are a generalization of binomial coefficients. Instead of

having to choose from one of two choices in each random selection, there are multiple choices to

choose from. Even the notation   is a generalization of the notation  , the

notation for binomial coefficients. Note that  .

___________________________________________________________________________

Multinomial Lunch Problem

Viewing the lunch choices as random sampling (each diner randomly chooses from M, B and I),
there are   531,441 many ways can the lunch choices be made for the whole 12-person group.
This count contains all the possible choices including possibilities that some restaurant is not chosen.
The following is one possible outcome that fall under question 2.

B-M-B-I-I-M-B-B-M-I-I-I
The above string shows that the first person chooses Burger King, the second person chooses
McDonald, the third person chooses Burger King and so on. How many other strings are like this one
in the sense that there are 3 M’s, 4 B’s and 5 I’s? This would be a multinomial coefficient.

The answer to the third question is more than 27,720. For example, there could be three diners
choosing IN-N-OUT, 4 diners choosing Burger King and 5 diners choosing McDonald. In other
words, we need to apply multinomial coefficient on the three positions (three restaurants). There are
6 ways to divide 3 positions resulting in three subsets with one each since  . Then there are
6 x 27,720 = 166,320 ways for 12 diners making choices so that 3 of them go to one place, 4 of them
go to another place and five of them go to the remaining place. This is an example of the double use
of the multinomial coefficients – the first application is on dividing the objects into subgroups and
the second application on the dividing the subgroups. The following example gives another
illustration.

Example 1
A fair die is rolled 15 times. How many of the outcomes consist of four 2’s, five 3’s and six 4’s?
How many of the outcomes consist of four of one face, five of another face and six of another face
(different from the other two faces)?
The first question is a straight application of one multinomial coefficient.

To understand the second question, consider the notation (0, 4, 5, 6, 0, 0), which means that there are
zero 1’s, four 2’s, five 3’s, six 4’s, zero 5’s and zero 6’s from rolling the die 15 times. Basically it
shows how many rolls fall into each position (face). Now we are interested in counting other
outcomes that have 4, 5 and 6 of another 3 positions. Basically we are trying to divide 6 positions
into four groups, one groups with 3 positions that do not appear in the 15 rolls, one groups with one
position that appear 4 times, one group with one position that appear 5 times, and one group with one
position that appears 6 times. The following shows how many ways to divide the 6 positions.

For each of the 120 times, there are 630,630 outcomes. Thus the answer to the second question is 120
x 630,630 = 75,675,600. 

___________________________________________________________________________

Multinomial Sampling
We now explore the random sampling angle of multinomial coefficients. To make it easier to follow,
we continue to use the Multinomial Lunch Problem. The problem, as discussed above, is like
sampling 12 times with replacement from an urn with 3 letters M, B and I. Each of the 12 random
selections has 3 outcomes. Thus the experiment in total has   = 531,441 outcomes, each of which
can be written as a string of 12 letters. Three examples:

B-M-B-I-I-M-B-B-M-I-I-I

I-M-M-I-M-M-M-I-M-I-I-M

M-M-M-M-M-M-M-M-M-M-M-M

These three strings can be called ordered samples for the experiment of randomly selecting from the
urn 12 times with replacement. These display the results of each draw sequentially. We can also use
unordered samples, which do not contain the orders but contain the number of times each letter is
drawn. For the first string, the unordered sample would be (3, 4, 5) where the first number is the
number of people choosing McDonald, the second number is the number of people choosing Burger
King and the third number is the number of people choosing IN-N-OUT. For the second string, the
unordered sample would be (7, 0, 5). For the third, everyone goes to McDonald and so it is (12, 0, 0).

Note that (3, 4, 5) is called an unordered sample because it does not tell us the order of the chosen
letters. It only tells us how many times each letter is chosen. The number of ordered samples
resulting in an unordered sample would be precisely the multinomial coefficient. For the unordered
sample (3, 4, 5), there would be

many ordered samples. In randomly selecting letters 12 times with replacement from an urn
containing the letters M, B and I, how many unordered samples are there? To answer this question, it
is helpful to look at unordered samples using a star-and-bar diagram. The star-and-bar diagrams for
above three unordered samples are:

***|****|*****

* * * * * * * |   | * * * * *

* * * * * * * * * * * * |   | 

In the random experiment in question, a star-and-bar diagram has 12 stars, representing the 12 letters
selected, and 2 bars, creating three spaces for the three letters M, B and I. In each star-and-bar
diagram, there are 12 + 2 = 14 positions, 12 of which are for the stars and the remaining 2 positions
are for the bars. In other words, each of the 14 positions can be either a star or a bar. Thus star-and-
bar diagrams are the results of a binomial experiment. How many possible star-and-bar diagrams?
There are
many different diagrams. Thus there are 364 many unordered samples from drawing letters 12 times
from an urn containing the letters M, B and I.

A related question is: how many unordered samples are there in this experiments such that each letter
is selected at least once? This means that each space in the star-and-bar diagram has at least one star.
Then the remaining 12 – 3 = 9 stars will have to be distributed in the three spaces. There are

unordered samples in which each letter is chosen at least once.

There is another way to interpret the unordered samples. What is the total number of non-negative
integer solutions to the equation  ? The solution x = 3, y = 4, z = 5 would be the
unordered sample (3, 4, 5). Thus there are 364 different solutions. What is the total number of
positive integer solutions to the equation  ? There are 55 such solution. Each such
solution would be an unordered sample where each letter is selected at least once.

___________________________________________________________________________

Multinomial Theorem
We now generalize the discussion on multinomial sampling.

Multinomial Sampling

Suppose that a multinomial experiment consisting of   independent trials with each trial resulting in
exactly   outcomes is performed. For convenience, these   outcomes are labeled by the
symbols  . The result of the experiment can be recorded by an ordered sample, which
is a string of   symbols with each symbol being one of  . The result of the experiment
can also be recorded by an unordered sample, which is a sequence of   numbers with the first
number being the number occurrences of the outcome  , the second number being the number of
occurrences of the outcome   and so on. The number of possible ordered samples in the
multinomial experiment is  . The number of ordered samples resulting in a given unordered

sample   is the multinomial coefficient  . The

number of unordered samples in the experiment is   based on a combinatorial argument

using stars and bars. The number of unordered samples such that in each sample, each of the
outcomes   occurs at least once

is  .

Number of Integer Solutions to Equations

The number of non-negative integer solutions to the equation   

is  , which is the total number of unordered samples as indicated above. The number of

positive integer solutions to the equation   is  , which is the

number of unordered samples such that each symbol from   occurs at least once in the
multinomial sampling.

There is a lot to unpack in the above two paragraphs. It is helpful to follow the random sampling idea
on a specific example, either the lunch problem discussed above or another example. The ideas in the
above paragraphs contains all the information for stating the multinomial theorem.

Sum of all Multinomial Coefficients

In the multinomial experiment described above, the sum of all possible multinomial coefficients is 

, i.e. 

Multinomial Theorem

The formula shown below shows how to raise a multinomial to a power. 

The count   is identical to the sum of all the possible multinomial coefficients in the experiment.
This fact follows from the observations we make about the multinomial sampling. The count   is
the total number of all ordered samples. The ordered samples can be divided into groups where each
group consists of of all ordered samples associating with the same unordered sample. Note that the
size of each of these groups is a multinomial coefficient.

The multinomial theorem is a compact way of describing the multinomial experiment. When it is
expanded out, the left-hand-side lists out all the   ordered samples. A typical ordered sample
consists of   many  ,   many   and so on, which can be represented by  , which is
a term on the right-hand-side. Thus a term such as   is another way to notate an
unordered sample and it represents   many ordered samples. Thus   
is the result of aggregating all the ordered samples for a given unordered sample. Summing all these
aggregated results give all the possible ordered samples in the multinomial experiment.

___________________________________________________________________________

Examples
The best way to absorb the concepts discussed here is to work excises. Three more examples are
presented. Several exercises are given in the section below.

Example 2
Suppose that 11 new hires are assigned to four office locations – the headquarter, branch office
North, branch office South and branch office East.
 How many assignments can be made?
 How many assignments can be made such that 3 new hires are assigned to the headquarter, 3
new hires are assigned to branch office North, 4 new hires are assigned to branch office South
and 1 new hire is assigned to branch office East?
 How many assignments can be made such that 3 new hires are assigned to one location, 3 new
hires are assigned to another location, 4 new hires are assigned to another location and 1 new
hire is assigned to the remaining location?
The answer to the first question is   = 4,194,304. The answer to the second question is a
multinomial coefficient.

Each assignment can be viewed as a string of 11 letters, each of which is chosen from H, N, S and E
(these are the ordered samples discussed above). An unordered sample is a sequence of 4 numbers.
For example, the unordered sample for the second question is (3, 3, 4, 1). The third question would
include other unordered samples with the numbers summing to 11, e.g. (1, 3, 4, 3). Basically we are
trying to divide the 4 letters into 3 groups, one group of one letter appearing once, one group with
two letters appearing 3 times each, one group with one letter appearing 4 times. This is another
multinomial coefficient.

Thus the answer to the third question is 12 x 46,200 = 554,400. 

Example 3
Suppose that 16 new hires are assigned to four office locations – the headquarter, branch office
North, branch office South and branch office East. Six of the new hires are engineers and only work
in the headquarter or branch office East. The other ten new hires are technicians and can be assigned
to any one of the four locations.
  How many assignments can be made?
 How many assignments can be made such that four of the engineers will work in the
headquarter and two of the new technicians will work in the headquarter, 3 new technicians
are assigned to branch office North, 4 new technicians are assigned to branch office South and
1 new technician is assigned to branch office East?
 How many assignments can be made such that four of the engineers will work in either the
headquarter or branch office East and two of the new technicians will work in one of the 4
locations, 3 new technicians are assigned to another location, 4 new technicians are assigned to
a third location and 1 new technician is assigned to the remaining location?
There are two assignments, one for engineers and one for technicians. The answer would be obtained
by multiplying the two assignment counts. The answer to the first question is   = 640,000. The
following is the answer to the second question.

Note that there are   = 15 ways to 4 of the engineers to the headquarter (and thus assigning the
other two engineers to the other office). The number of ways to assign 10 technicians to the 4
locations in the indicated numbers is the multinomial coefficient indicated above (the one resulting in
12,600). The following is the answer to the third question.

Note that for the third question, a second multinomial coefficient on the locations is required for each
type of workers. 

Example 4
Fifteen identical brand new mail delivery trucks are assigned to 5 post offices.
 How many assignments can be made?
 How many assignments can be made such that each post office is assigned at least one truck?
The problem can be done using the combinatorial argument with stars and bars discussed above. The
15 trucks are stars and there 4 bars creating 5 spaces representing the post offices. Any assignment of
trucks would be like a string of 15 stars and 4 bars. The following is an example.

****|******||*****

The above star-and-bar diagram shows that 4 trucks are assigned to the first post office, 6 trucks are
assigned to the second post office, zero trucks are assigned to the third post office and 5 trucks are
assigned to the fourth post office. What is more important to notice is that there are 19 positions in
the diagram. The problem is to choose 15 of them to place the stars. The following is the answer to
the first question.
The second question requires that each space has at least one star. We place one star in each space.
There are 11 stars left. We now consider 11 stars and 3 bars. The following is the number of ways of
arranging 11 stars in 14 positions.

Thus there are 364 ways to assign 15 trucks to four post offices such that each office gets at least one
truck. 

___________________________________________________________________________

Exercises
Exercise 1
A father is to distribute 9 gifts to his three children.
1. In how many ways can the gifts be distributed if the eldest child is to receive 2 gifts, the
middle child is to receive 2 gifts and the youngest child is to receive 5 gifts?
2. In how many ways can the gifts be distributed if one child is to receive 2 gifts, another child is
to receive 2 gifts and the remaining child is to receive 5 gifts?
Exercise 2
Eleven job assignments are randomly distributed to four workers Marcus, Issac, Samantha and Paul.
In how many ways can these jobs be assigned to the four workers such that Marcus will receive one
job, Issac will receive 4 jobs, Samantha will receive 4 jobs and Paul will receive 2 jobs?
Exercise 3
1. Ten students are to be assigned to two math classes. How many ways can the students be
divided into the two math classes with 5 students in each class?
2. Fifteen students are to be assigned to three math classes. How many ways can the students be
divided into the three math classes with 5 students in each class?
Exercise 4
An investor has 25 thousand dollars to invest among 5 possible investments. The amount to invest in
each investment is in the unit of one thousand dollars. Suppose that the entire amount of 25 thousand
dollars is to be invested.
1. How many different investment strategies are possible?
2. How many different investment strategies are possible if at least one unit is put into each
investment choice?
Exercise 5
Recall the Multinomial Lunch Problem. Each of the twelve people have 3 choices for lunch –
McDonald, Burger King and IN-N-OUT, all fast food restaurants. The choice for each person is
independent of the choices of the other diners. In how many ways can the lunch choices for this 12-
person group be made in such a way that each restaurant is visited by at least 3 diners?
Answers
Exercise 1
1. 756
2. 2,268
Exercise 2
34,650
Exercise 3
1. 252
2. 756,756
Exercise 4
1. 23,751
2. 10,626
Exercise 5
256,410
Hint: There are three cases, each of which requires the use of two multinomial coefficients, one for
the diners and one for the restaurants.
___________________________________________________________________________
 2017 – Dan Ma

SHARE THIS:

 Twitter
 Facebook


RELATED
Binomial coefficientsMarch 9, 2017In "Combinatorics"
Strength in numbersAugust 23, 2017In "Combinatorics"
A banquet menu can be a lesson in countingFebruary 16, 2017In "Combinatorics"

This entry was posted in Combinatorics and tagged Binomial

Coefficients, Combination, Combinatorics, Factorial, Multinomial Coefficients, Permutation by Dan Ma.
Bookmark the permalink.
Leave a Reply