A NOTE ON PASCAL-T TRIANGLES, MULTINOMIAL COEFFICIENTS,

AND PASCAL PYRAMIDS

Behrend

College,

RICHARD C. BOLLINGER
Pennsylvania
State
University,
(Submitted
August
1984)
1.

Erie,

PA 16563

INTRODUCTION

In what follows we give a formula for the entries in the Pascal-27 triangle
Tm in terms of the multinomial coefficients; this is the counterpart for these
arrays of the result of Philippou [1] on the elements of the Fibonacci &-sequences in terms of the multinomial coefficients. The proof is direct, and the
method also leads to a recurrence relation which gives the elements of a given
triangle Tm as a combination of certain elements of the "preceeding" triangle
Tm-i> t n e coefficients in the combination being binomial coefficients. Finally, because the multinomial coefficients provide the connection here, we offer
some remarks on those arrays of multinomial coefficients referred to in the
literature as "Pascal pyramids."
It will be convenient to recall the definition of the triangle Tm.
Definition 1.1: For any m ^ 0, Tm is the array whose rows are indexed by n = 0 ,
1, 2, ..., and columns by k = 09 1, 2, ..., and whose entries are obtained as
follows:
a)

T0 is the all-zero array;

b)

T1 is the array all of whose rows consist of a one followed by zeros;

c)

Tm9 777^2, is the array whose n = 0 row is a one followed by zeros, whose
n= 1 row is m ones followed by zeros, and any of whose entries in subsequent rows is the sum of the m entries just above and to the left in
the preceeding row.

The entry in row n and column k is denoted by Cm(n9

k), although we note that

C2(n9 k) = (),
since T2 is the Pascal Triangle. There will be n(m - 1)+ 1 nonzero entries in
row n9 and the principal property we need is that these are the coefficients
(see, e.g., [2], p. 66) in the expansion
n(m-

(1 + t + t2

+ + t m " V

1)

Cm(n9

k)tk.

(1.1)

k=o

Although it is easy to use property

relation

(c) to build the array Tm by means of the

m- 1

Cm(n9

k) = Cm(n - 1, k - j ) ,

(1.2)

j=o

the main result presented here evaluates Cm(n9

multinomial coefficients.
140

k) directly as a sum of certain

[May

A NOTE ON PASCAL-?7 TRIANGLES, MULTINOMIAL COEFFICIENTS, AND PASCAL PYRAMIDS

2*
Theorem 2,1: If Cm(n$
and 0 < k < n(m - 1),
C(. *> =

FORMULAS FOR Cm(n,

k)

k) is the (n, &)-entry in Tm9 m > 3 9 then for any n > 0

E

(n

).

(2.1)

where the summation is over all m-part compositions nls n2s ..., n m of n such
that (1) n x + n2 + .-. + nm = n 9 and (2) 0n1 + ln2 + + (m - l)nm = fc.
Proof:

The proof follows directly from the multinomial theorem9 for if in

(Xi

+Xz

...

+Xm)n

n
\ ri-^ $

\xixmxn.t

(2.2)

'frfi?/

where the summation is over all m-part compositions of ns we put x^ =

1 < i < m s we have
(1 + t +

... + t -l) = W

"
\ttlJ

...5

X t " 2+ 2n3+... + (m-l)n5

tz~1s

(2>3)

nmf

and when the coefficients of tk on the right-hand sides of (1.1) and (2.3) are
equated, (2.1) follows from conditions (1) and (2).
Example: ^ ( 4 , 4) - (^

Q>

Q)

+ (2>

2> 0

)+(

2 >

= 1 + 12 + 6 + 12 = 31
Another application of the multinomial expansions used partly as a binomial
expansions gives the following theorem.
Theorem 2.2:

Cm(n9 k) = t OK-i<J. k - j).

(2.4)

j=o w /
Proof: If the left side of (2.2) is grouped as [x1.+ (x2 + + xm)]n9
as a binomials and again tv~x is substituted for x^s the result is
nirn1)
n- I)

COT(n5 k)tk = E n U ' d + * +

k
=o

t/ = o ^

eee

+ tm-2)J.

expanded

(2.5)

But then the factors .(1 + t ++ tm~2)3

may be expressed in terms of C m -i f s s
using (1.1). When the coefficients of a given power of t on the right of (2.5)
are collected and equated to the corresponding coefficient on the left9 then
(2.4) follows.
Example: C(4, 4) = ( J J ^ C O . *) + (JJ^sC 1 * 3>

(2)^(2,

+ (3)^3(3, 1) + ( J H < 4 . 0)
= l-0 + 4 0 + 6-3 + 4-3 + l-l = 31

1986]

HI

A NOTE ON PASCAL-27 TRIANGLES, MULTINOMIAL COEFFICIENTS, AND PASCAL PYRAMIDS

3.

THE PASCAL PYRAMID

The device used in the previous theorem of bracketing off one term of a
multinomial in order to expand the result as a binomial cans of course, be repeated with the remaining multinomial parts, eventually running the unexpanded
part down to a binomial itself; this offers the possibility of obtaining the
multinomials entirely as products of binomial coefficients. In fact, this has
been done in [3] and [4] for a trinomial expansion, with the multinomial coefficients appearing in the successive powers of (x + x2 + x3) being associated
with points in triangular arrays, which form successive levels of a pyramidal
structurethe so-called Pascal pyramid. For example, Figure 1 shows the first
four levels, with each point labelled with both a multinomial coefficient and
the composition which gives rise to it (the compositions can be obtained by
designating the sides as first, second, third in some fashion, and letting n19
n29 n3 in the composition measure units of perpendicular distances from the
first, second, third sides, respectively). The law of formation for this trinomial case is clear (and also correct, as is easily verified by doing the reduction described earlier): just generate the ordinary Pascal triangle down to
level n9 and then multiply the rows successively upward by the numbers found
in the last line. For n = 3, e.g.,
1
1
1
1 3

_1 1
1

1
3

becomes

n=0

J 1
^
3

J 1
^

i (0,0, 0)

n= l

n=2

n=3

(3. 0. 0}

Figure 1.
142

(0, 3, 0)

Levels of the Pascal Pyramid for the Case (x1 + x2 +

x3)
[May

A NOTE ON PASCAL-27 TRIANGLES, MULTINOMIAL COEFFICIENTS, AND PASCAL PYRAMIDS

This works nicely, the idea is not restricted to trinomials (the generalization
is not just a bigger triangle9 but it is similarly simple), and there have been
several comments to the effect that it is surprising that the Pascal pyramids
(or hyperpyramids) are not more widely known or used.
Why should this be? An answer would seem to be that as soon as one gets
past the trinomials, the method, while still elegant, becomes computationally
unwieldy. That is, in the usual expansion of a multinomial (x1 + + xm)n
there are (

J terms corresponding to the m-part compositions of n.

To

see what is required to deal with these in terms of products of binomials, we

look at what might be called the Pascal square, in which we tabulate for m = 0,
1, 2, ... and n = 0, 1, 2, ... the number of m-part compositions of n (taking
the entry for m = n = 0 to be one). The first several lines are shown below:
Pascal Square
n
No. P a r t s m

0
1
2
3
4
5

1
1
1
1
1
1

0
1
2
3
4
5

3
0
1
3
6
10
15

0
1

0
1
6
21
56
126

1
4
10
20
35

5
15
35
70

Here, the law of formation is that each entry is the sum of alt
those entries
above and to the left of it in the preceeding row, and we recognize the m = 3
row as the triangular numbers, the m = 4 as the pyramidal numbers, and so on.
The point here is that the square shows that the trinomials (m = 3) are simple
sequences of products of binomials; as in the example, the ten trinomials in
(x + x2 + x3)3 reduce to 4 + 3 + 2 + 1 products of binomials. But for m > 3,
we find not sequences, but sequences of sequences. The thirty-five terms in
(x + x2 + x3 + xh)h
[the (4, 4) entry], e.g., have to be obtained using the
sequence
15 = 5 + 4 + 3 + 2 + 1 ,
10=
4 + 3 + 2 + 1 ,
6 =
3+2+1,
3 =
2 + 1,
1 =
1,
of sequences of products of binomials. It would seem that in spite of the appeal of an array for multinomial coefficients similar to the triangle for
binomials, one is better off for most purposes using a convenient algorithm
(e.g., [5], pp. 46-51) to generate the m-part compositions of n, from which the
exponents on the x^ and the multinomial associated with a given term are immediately available.

1986]

1*3

A NOTE ON PASCAL-T TRIANGLES, MULTINOMIAL COEFFICIENTS, AND PASCAL PYRAMIDS

REFERENCES
1.

A. N. Philippou. "A Note on the Fibonacci Sequence of Order k and the Mul-

2.

G. Chrystal. Textbook of Algebra3

Tart One. 5th edition. New York: Dover,
1961.
H. Staib & L. Staib. "The Pascal Pyramid." The Mathematics
Teacher 71
(1978):5Q5-510.
H. F. Lucas. "Paths and Pascal Numbers." College Mathematics
Journal
14
(1983):329-341.
A. Nijenhuis & H. S. Wilf. Combinatorial
Algorithms.
2nd edition. New
Yorks Academic Press, 1978.

3.
4.
5.

tinomial Coefficients/1 The Fibonacci

Quarterly

21 (1983):82-86.

^<>*o*

ANNOUNCEMENT

SECOND INTEENATIONAL CONFERENCE ON

FIBONACCI NUMBERS AND THEIR APPLICATIONS
August 13-16,1986
San Jose State University

San Jose, California 95192

LOCAL COMMITTEE
Bergum, Gerald, Chairman
Edgar, Hugh, Cochairman
Thoro, Dmitri
Johnson, Marjorie
Lange, Lester

INTERNATIONAL COMMITTEE
Philippou, A. (Greece) Chairman
Horadam, A.F. (Australia), Cochairman
Bergum, G. (U.S.A.)
Kiss, P. (Hungary)
Long, C. (U.S.A.)
Ando, S. (Japan)

CALL FOR PAPERS

The SECOND INTERNATIONAL CONFERENCE ON FIBONACCI NUMBERS AND THEIR
APPLICATIONS will take place at San Jose State University, San Jose, CA, Aug. 13-16, 1986. This conference is sponsored jointly by The Fibonacci Association and San Jose State University.
Papers on all branches of mathematics and science related to the Fibonacci numbers and their
generalizations are welcome. Manuscripts are requested by June 15,1986. Abstracts and manuscripts should
be send to the chairman of the local committee. Invited and contributed papers will appear in the Conference Proceedings, which are expected to be published.
The program for the Conference will be mailed to all participants, and to those individuals who
have indicated an interest in attending the conference, by June 15, 1986. All talks should be limited to one
hour.
For further information concerning the conference, please contact either of the following:

Professor G.E. Bergum, Editor

The Fibonacci Quarterly

Department of Mathematics
South Dakota State University
P.O. Box 2220
Brookings, S.D. 57007-1297

H4

Professor Calvin Long

Department of Mathematics
Washington State University
Pullman, Wash. 99163

[May