Catalan Numbers, Their Generalization,

and Their Uses

Peter Hilton and Jean Pedersen

Probably the most prominent among the special in- stepping one unit east or one unit north of P,. We say
tegers that arise in combinatorial contexts are the bino- that this is a path from P to Q if Po = P, Pm= Q. It is
mial coefficients (~). These have many uses and, often, now easy to count the number of paths.
fascinating interpretations [9]. We would like to stress
one particular interpretation in terms of paths on the THEOREM 0.1. The number of paths from (0,0) to (a,b) is
integral lattice in the coordinate plane, and discuss the the binomial coefficient if+b).
celebrated ballot problem using this interpretation.
A path is a sequence of points Po P1 999 Pro, m >I O, Proof: We may denote a path from (0,0) to (a,b) as a
where each P, is a lattice point (that is, a point with sequence consisting of a Es (E for East) and b Ns (N for
integer coordinates) and Pz+l, i 1> 0, is obtained by North) in some order. Thus the number of such paths

64 THE MATHEMATICALINTELUGENCERVOL 13, NO 2 @ 1991 Spnnger-Verlag New York

 is the number of such sequences. A sequence consists
of (a + b) symbols and is determined w h e n we have
decided which a of the (a + b) symbols should be Es.
p=2 p=3
Thus we have (o+b) choices of symbol. k=3 k=2

3 4 5

2 3

Figure 1. Left: a binary (or 2-ary) tree with 3 source-nodes

(Q) and 4 end-nodes (o) and, at right, a ternary (or 3-ary) tree
with 2 source-nodes (o) and 5 end-nodes (o).

p=2 p=3
k=3 k=2

(Sl((S2 S3)S4)) (sl s2(ss

Figure 2. Left: an expression involving 3 applications of a
binary operation applied to 4 symbols and, at right, an ex-
pression involving 2 applications of a ternary operation ap-
plied to 5 symbols.
We will present another family of special integers,
called the Catalan n u m b e r s (after the nineteenth-cen-
tury Belgian 1 mathematician Eugene Charles Catalan;
see the historical note at the end of this ar~cle), which p=2 p=3
are closely related, both algebraically and concep- k=3 k=2
tually, to the binomial coefficients. These Catalan
numbers have many combinatorial interpretations, of
which we will emphasize three, in addition to their
interpretation in terms of 2-good paths on the integral
lattice. Because all these interpretations remain valid if
one makes a conceptually obvious generalization of
the original definition of Catalan numbers, we will
present these interpretations in this generalized form. Figure 3. Left: a S-gun subdivided into three 3-gons by 2
Thus the definitions we are about to give depend on a diagonals and, at right, a 6-gon subdivided into two 4-gons
parameter p, which is an integer greater than one. The by I diagonal.
original Catalan n u m b e r s - - o r , rather, characteriza-
tions of these n u m b e r s - - a r e obtained by taking p = 2. ak: rao = 1, ~a k = the number of p-ary trees with k
Let us then present three very natural combinatorial s o u r c e - n o d e s , k ~> 1 (see Fi-
concepts, which arise in rather different contexts, but gure 1).
which turn out to be equivalent. bk: pbo = 1, ~bk = the number of ways of associating
Let p be a fixed integer greater than 1. We define k applications of a given p-ary
three sequences of positive integers, depending on p, operation, k I> 1 (see Figure 2).
as follows: ck: f o = 1, t,ck = the n u m b e r of w a y s of subdi-
viding a convex polygon into k
disjoint (p + 1)-gons by means of
t H e lived his life m Belgium; but, b e i n g born in 1814, h e w a s n o t non-intersecting diagonals, k I> 1
b o r n a Belgian! (see Figure 3).

THE MATHEMATICALINTELLIGENCERVOL. 13, NO. 2, 1991 65

 However, relating the trees of Figure 1 (or the paren-
p=2 p=3 thetical expressions of Figure 2) to the corresponding
k=3 k=2 dissected polygons of Figure 3 is more subtle (and the
(1 ((23) 4)) (~ 2 s 4 5))
hint given in [8], in the case p = 2, is s o m e w h a t
cryptic), so we will indicate the proof that bk = Ck.
Thus, suppose that we are given a rule for associating
k applications of a given p-ary operation to a string of
(p - 1)k + 1 symbols sl, s2. . . . . s0,_l~,+ v Label the
successive sides (in the anticlockwise direction) of the
convex ((p - 1)k + 2)-gon with the integers from 1 to
3 (p - 1)k + 1, leaving the top, horizontal side unla-
Figure 4. The dissected polygons of Figure 3, with diag- beled. Pick the first place along the e x p r e s s i o n
onals and last side labeled. (counting from the left) w h e r e a succession of p
symbols is enclosed in parentheses. If the symbols en-
closed run from st+ 1 to sj+p, draw a diagonal from the
( 1 ( 2 3 ( 4 ( 5 6 7 ) 8))9) initial vertex of the (j + 1)st side to the final vertex
p=3 of the (j + p)th side, and label it (j + 1. . . . . j + p)
k=4 Also imagine the part (sj+a . . . . sj+p) replaced by a
3 ( 4 ( 5 6 7 ) 8)) / ~ single symbol. We now have effectively reduced our
word to a set of (k - 1) applications and our polygon
to a set of 2 polygons, one a (p + 1)-gon, the other a
[(p - 1)(k - 1) + 2]-gon, so we proceed inductively to
complete the rule for introducing diagonals. More-
over, the eventual label for the last side will corre-
'Y spond precisely to the string of (p - 1)k + 1 symbols
with the given rule for associating them, that is, to the
original expression. Figure 4 illustrates the labeling of
the dissections of the pentagon and hexagon that are
d e t e r m i n e d by the c o r r e s p o n d i n g expressions of
Figure 2. The converse procedure, involving the same
initial labeling of the original sides of a dissected
polygon, and the successive labeling of the diagonals
5 introduced, leads to an expression that acts as label for
the last (horizontal) side.
Figure 5. The polygonal dissection associated with We now illustrate a slightly more complicated situa-
(s~(s2s3(s,(sss~ )ss) )~ . tion:
Example O. 1. Let p = 3, k = 4 and consider the expres-
(As indicated, we suppress the 'p' from the symbol if sion
no ambiguity need be feared.) sl(s2s3(s,(sss6sT)sg))sg).
Note that, if k t 1,
(i) a p - a r y t r e e w i t h k s o u r c e - n o d e s h a s The dissection of the convex 10-gon associated with
(p - 1)k + 1 end-nodes and pk + 1 nodes in this expression, with the appropriate labeling, is
all (Figure 1); shown in Figure 5. The corresponding ternary tree is
(ii) the k applications of a given p-ary operation are shown in Figure 6, which also demonstrates how the
applied to a sequence of (p - 1)k + 1 symbols tree m a y be associated directly with the dissected
(Figure 2); polygon. The rule is "Having entered a (p + 1)-gon
(iii) the polygon has (p - 1)k + 2 sides and is sub- through one of its sides, we may exit by any of the
divided into k disjoint (p + 1)-gons by (k - 1) other p sides."
diagonals (Figure 3). In the case p = 2, any of at, bk, Ckmay be taken as the
A well-known and easily proved result (for the case definition of the kth Catalan number. Thus we may re-
p = 2 see Sloane [8]) is the following: gard any of pat, pbk, pck as defining the generalized kth
Catalan number. Moreover, the following result is
THEOREM 0.2. pak = pbk = pCk" known (see [6]).
The reader will probably have no trouble seeing THEOREM 0.3.
how the trees of Figure I are converted into the corre- Pak= k 1 (p-1)k + l '
sponding expressions of Figure 2 - - a n d vice versa. k~>l.

66 THE M A T H E M A T I C A L INTELLIGENCER VOL 13,N O 2, 1991

 The "classical" proof of this result is rather sophisti- particular stress on the flexibility of this fourth inter-
cated. One of our principal aims is to give an elemen- pretation, which we n o w describe.
tary proof of Theorem 0.3, based on yet a fourth inter- We say that a path from P to Q is p-good if it lies
pretation of the generalized Catalan n u m b e r s which, entirely below the line y = (p - 1)x. Let dk = pdk be
like that of the binomial coefficients given earlier, is the n u m b e r of p-good paths from (0, - 1) to (k, (p - 1)k
also in terms of paths on the integral lattice. We lay - 1). (Thus, by our convention, do = 1.) We extend
Theorem 0.2 to assert the following.

T H E O R E M 0.4. pak = pbk = pCk = p d k.

Proof: First we restate (with some precision) the defini-

1 O•j• 9
p=3
k=4
tions of bk a n d dk, which n u m b e r s we will prove equal.
bk = the n u m b e r of expressions involving k appli-
cations of a p-ary o p e r a t i o n (on a w o r d of
k(p - 1) + 1 symbols). Let E be the set of all
such expressions.
dk = t h e n u m b e r of g o o d p a t h s f r o m ( 0 , - 1 ) to
(k, (p - 1)k - 1); by following such a path by a
2 single vertical segment, we m a y identify such
paths with certain paths (which we will also
call 'good') from ( 0 , - 1) to (k, (p - 1)k). Let P
be the set of all such paths.
We set up functions ~: E---* P, ~F: P - - , E, which are
obviously mutual inverses. Thus <I>is obtained by re-
m o v i n g all final parentheses 2 a n d (reading the re-
sulting expression from left to right) interpreting a pa-
renthesis as a horizontal move a n d a symbol as a ver-
5 6 7 tical move. The inverse rule is ~ . It remains to show (i)
that ImCI>consists of elements of P (and not merely of
arbitrary paths from ( 0 , - 1) to (k, (p - 1)k); and (ii)
Figure 6 (a). The tree associated with (sl(s2ss(s,(ssssS~)s.))sg). that Im~F consists of elements of E (that is, of well-
formed, meaningful expressions).
(i) Let E be an expression. We argue by induction on
Ente[ here
k that ~ E is a good path. Plainly ~E is a path from
I (0, - 1) to (k, (p - 1)k), so we have only to prove it
good. This obviously holds if k = 1. Let k I> 2 a n d
let E involve k applications. There m u s t occur
somewhere in E the section

(st+is1+2.. 9 st+ p)
Call this the key section. We wish to show that if
(u,v) is a point on the path <I>Ethat is not the end-
point, t h e n v < (p - 1)u. Let E' be the expression
obtained from E by replacing the key section by s.
Then E' involves (k - 1) applications. Now if A is
the point on ~E corresponding to the parenthesis
of the key section, then the inductive hypothesis
tells us that v < (p - 1)u if (u,v) occurs prior to A.
A s s u m e n o w that (u,v) is not prior to A. There are
t h e n 2 possibilities:

2 This converts our expression into a meaningful reverse Polish ex-

pression (MRPE); see [2]. Note that the closing parentheses are,
strictly speaking, superfluous, prowded that we know that we are
Figure 6 (b). Associating a tree with a dissected p o l y g o n (in dealing with a p-ary operation. It is not even necessary to know this
this case, the tree s h o w n in (a) and the p o l y g o n of Figure 5). if we are sure that the expression is well-formed, since the 'arity' is
Note that 9 = interior node = source-node; o = exterior ~-1
K
+ 1, where ~ is the number of symbols and Kis the number of
node = end node. opening parentheses.

THE MATHEMATICAL INTELL1GENCER VOL 13, NO 2, 1991 67

 y=2x
/
/(.
p=3
k=4 ~2
,C

(0, -1) t
~=~bE
E =~b~

Figure 7 (a). The path associated w i t h the expression E = Figure 7 (b). The inductive step, in either direction.
(s~(s2s3(s,(sss6sT)ss))Sg).

If E ends with S/+p, so that E' e n d s with s, then b y AB, followed by ~2'" It is trivial that ~ ' is a
3(u',v'), the last point of ~E', such that u' = good path to (k - 1, (p - 1)(k - 1)). Thus, by the
u - 1, v' > v - p + 1, v' = (p - 1)u'. inductive hypothesis, ~ ' is a well-formed ex-
H e n c e v - p + 1 < (p - 1)(u - 1), s o y < pression involving (k - 1) applications. Let s be
( p - 1)u. the symbol in ~@' corresponding to AB. Replace s
If E does not end with st+ ~, so that E' does not end by (sj+ls1+ 2 . . . sj+p, w h e r e these symbols do not
with s, then 3(u',v'), not the last point of ~E', occur in ~@'. Then the n e w expression is well-
such that u' = u - 1, v' I> v - p + 1, v' < formed and is obviously ~@. This completes the
(p - 1)u' (by the inductive hypothesis). Hence induction in the ~-direction so that Theorem 0.4 is
v-p + l<(p- 1)(u- 1),sov<(p- 1)u. established.
This c o m p l e t e s the i n d u c t i o n in the ~ - d i r e c t i o n . With Theorem 0.4 established, we may proceed to
Figure 7, which gives a special, but not particular, calculate d k. This is achieved in Section 2; a more gen-
case, may be helpful in following the argument. eral algorithm for counting p-good paths is described
(ii) Let ~ be a good path to (k, (p - 1)k). We argue by in Section 3.
induction on k that ~@ is a well-formed expres- By considering the last application of the given p-ary
sion. This holds if k = 1, because then there is operation, it is clear that pbk satisfies the recurrence re-
only one good path @ and ~@ is (sis2. 9 9 Sp). Let k lation (generalizing the familiar formula in the case p
~> 2. Then b e c a u s e the path climbs altogether = 2)
k(p - 1) + 1 places in k jumps, there must be a Pbk = E pbtl pbi2 . . . pbip, k I> 1. (0.1)
j u m p s o m e w h e r e (not at the initial point) of at tl+t2+ . +tp=k-1
least p places. Let A be a point on ~ where the In Section 2 w e also enunciate a similar recurrence re-
path takes one horizontal step followed by p ver- lation for pdk, thus providing an alternative proof that
tical steps, bringing it to C. Let B be the point one ~bk = pdk. We will also s h o w in Section 2, in outline,
step above A (so that B is not on ~). Then B is on y o w (0.1) leads to a proof of Theorem 0.3 via gener-
= (p - 1)x if and only if C is on y = (p - 1)x, that ating functions and the theory of B/~rmann-Lagrange
is, if and only if C is the endpoint of ~; otherwise B series (see [5]).
is below y = (p - 1)x. Let ~1, ~2 be the parts of The case p = 2, dealing with the Catalan numbers
ending in A and beginning in C, respectively. Let as originally defined, is discussed in Section 1. In that
~2' be the translate of ~2, given by special case w e have available an elegant proof that 2dk
u' = u - 1, v' = v - (p- 1); = ~(k~l), k I 1, using Andre's Reflection Method [1];
however, w e have not discovered in the literature a
and let @' be the path consisting of ~a, followed means of generalizing this m e t h o d effectively.

68 T i m MATHEMATICAL INTELLIGENCER VOL 13, N O 2, 1991

 m

The authors wish to thank Richard Guy, Richard composition, @ = ~1~2; and let @1 be the path ob-
Johnsonbaugh, H u d s o n Kronk, and Tom Zaslavsky tained from @1 by reflecting_in the line y = x. Then, if
for very helpful conversations and communications = ~ 2 , @ is a path from P(d,c) to Q(a,b); and the rule
during the preparation of this article. ---* @ sets up a one-one correspondence between the
set of bad paths from P to Q and the set of paths from
1. Catalan Numbers and the Ballot Problem to Q. It follows that there are ((a+b)2(~+a)) bad paths
from P to Q, and hence
The ballot problem (see box) was first solved by Ber-
trand, but a very elegant solution was given in 1887 by ((a + b) - c(c + d)) ((a + b) - d(c + (1.1)
the French mathematician D4sir6 Andr6 (see [1], [4]).
In fact, his method allows one easily to count the
number of 2-good paths from (c,d) to (a,b), where (c,d), good paths from P to Q. The argument, illustrated in
(a,b) are any two lattice points below the line y = x. Figure 8, is called Andre's Reflection Method.
Because p = 2 throughout this discussion, we will Notice, in particular, that if c = 1, d = 0, this gives
speak simply of good paths; any path from (c,d) to (a,b) the number of good paths from (1,0) to (a,b) as
in the complementary set will be called a bad path. We
will assume from the outset that both good paths and (o+b
a-1
l) (a+b l)
bad paths exist; it is easy to see that this is equivalent a = a

to assuming
Thus the probability that a path from (0,0) to (a,b)
d<c<~b<a. proceeds first to (1,0) and then continues as a good
(Note that this condition is certainly satisfied in our path to (a,b) is a~bb;this then is the solution to the ballot
ballot problem, provided the losing candidate receives problem.
at least I vote!) We return now to Catalan numbers. The kta Catalan
Now a simple translation argument shows that the number ck is expressible as 2dk, the number of good
total number of paths from P(c,d) to Q(a,b) is the bino- paths from ( 0 , - 1 ) to (k,k - 1). According to (1.1)
mial coefficient this is
1
((a + b) - c(c +

thus it suffices to count the number of bad paths from Thus the kth Catalan number ck is given by the formula
P to Q; see Figure 8. If ~ is a bad path, let it first make
contact with the line y = x at F; let ~1, ~2 denote the 1 ( 2k ) k~>l. (1.2)
subpa~hs PF, FQ, so that, using juxtaposition for path- co = 1, ck =-~ k - 1 "

Notice that an alternative expression is

Ck=k+ 1 (1.3)

Notice, too, that, by translation, ck may also be inter-

preted as the n u m b e r of good paths from (1,0) to
(k + 1,k).
Andr4's Reflection Method does not seem to be
readily applicable to obtaining a formula corre-
sponding to (1.2) in the case of a general p I> 2. In-
deed, in the next section, where we calculate pck, p/>
2, we do not obtain a general closed formula for the
number of p-good paths from (c,d) to (a,b). We do in-
troduce there and calculate explicitly the number dqk of
p-good paths from (1,q - 1) to Pk(k,(p - 1)k - 1) for q
~< p - 1. This enables us, by translation, to calculate
the number of p-good paths from any lattice point C
below the line y = (p - 1)x to Pk. For if C = (c,d), then
the number of p-good paths from C to Pk is the number
of p-good paths from (1,q - 1) to Pk-c+l, where q =
Figure 8. Andr4's reflection method. d + 1 - (p - 1)(c - 1). On the other hand, although

THE MATHEMATICAL INTELLIGENCER VOL 13, N O 2, 1991 69

a convenient explicit formula w h e n the terminus of for a n e a s y t r a n s l a t i o n a r g u m e n t s h o w s t h a t the
the path is an arbitrary lattice point below the line y = n u m b e r of p-good paths from (1, p - 2) to Pk is the
(p - 1)x does not seem available, we will nevertheless same as the n u m b e r of p-good paths from ( 0 , - 1) to
derive in Section 3 an algorithmic formula in the form of Pk-l" Thus dp-l,1 = 1 = d00 if k = 1, and, if k I-- 2, the
a finite sum, d e p e n d i n g on the quantities dqk. This for- n u m b e r of p-good paths from ( 0 , - 1) to Pk-1 is, as ar-
mula, while readily applicable, has a mysterious con- g u e d above, the number of p-good paths from (1, - 1)
ceptual feature to which we draw attention at the e n d to Pk- 1, that is, do,k- 1.
of Section 3. We m a y write ~dk for pd0k, so that, if k I> 0,
The standard procedure for calculating Ck (we here
2dk = Ck, the kth Catalan number, and (2.3)
revert to the original case p = 2) is based on the evi-
dent recurrence relation =/k =/0,k =/p-l,k§
the generalized kth Catalan number. (2.4)
Ck = ~ c,cl, k >~ l; co = 1. (1.4)
z+/=k-1 We will also suppress the 'p' from these symbols a n d
This relation is most easily seen by considering bk and from the term 'p-good' if no ambiguity need be feared.
concentrating on the last application of a given binary We n o w enunciate two fundamental properties of
operation within a certain expression. If we form the the n u m b e r s dqk, for a fixed p 1 2. A special case of the
power series first property was pointed out to us by David Jonah
oo during a Mathematical Association of America Short
S(x) = (1.5) C o u r s e , c o n d u c t e d by us in the s u m m e r of 1987,
k=O u n d e r the auspices of the Michigan Section. His orig-
inal f o r m u l a concerned the case p = 2 and was re-
then (1.4) shows that S satisfes stricted to q = 0 (and the parameter n appearing in the
xS 2 - S -t- 1 -~ 0, S(0) = 1, formula w a s restricted to being an integer not less
t h a n 2k instead of an arbitrary real number).
so that Recall that, for any real n u m b e r n, the binomial coef-
1 - X/1 - 4 x
S = (1.6) ficient (8) m a y be interpreted as 1, and the binomial
2x coefficient (~) m a y be interpreted as the expression
n(n-1) ....
r!
(n-r+1), provided r is a positive integer (see,
It is n o w straightforward to expand (1.6) a n d thus to
e.g., [3]). Using this interpretation, we prove the fol-
obtain confirmation of the value of ck already obtained
lowing.
in (1.2). As w e will d e m o n s t r a t e , this a n a l y t i c a l
m e t h o d for calculating ck does generalize, but the gen-
T H E O R E M 2.2 (Generalized Jonah Formula). Let n be
eralization is highly sophisticated a n d we prefer to
any real number. If k >I 1, and q ~ p - 1, then
emphasize a m u c h more elementary, combinatorial ar-
g u m e n t to calculate pck. -
k

2. Generalized Catalan Numbers

Proof: We first assume (see Figure 9) that n is an in-
We will base our calculation of ~ck on that of a more teger not less than pk, so that (k, n - k) is a lattice
general quantity. Let p be a fixed integer greater than point not below the line y = (p - 1)x. Partition the
1, and let Pk be the point (k,(p - 1)k - 1), k ~ 0. We paths from (1, q - 1) to (k, n - k) according to where
define the n u m b e r s ~/qk = dqk, as follows. they first meet the line y = (p - 1)x. The n u m b e r of
these paths that first meet this line where x = i is ~i ~ i ,
Definition 2.1 Let q ~< p - 1. Then dq0 = I and dqk is the where
n u m b e r of p-good paths from (1, q - 1) to Pk, if k >1 1. ~/z = the number of paths from (1, q - 1) to
(i, (p - 1)i)
The quantities dqk will play a crucial role in counting which stay below y = (p - 1)x except at the
p-good paths a n d turn out to be no more difficult to endpoint
calculate than the quantities dk. Of course, we have = the number of good paths from (1, q - 1) to

~Ck = ~/k = pd0k, k ~> 0, (2.1)

P,
t/q ;
where pck is the kth (generalized) Catalan number. For,
if k ~> 1, every p-good path from (0, - 1) to Pk m u s t first and
proceed to ( 1 , - 1). We further note that we have the the n u m b e r of paths from (i, (p - 1)i) to
relation (k, n - k)

6_1, k = d0,k_l, k ~ 1; (2.2)

70 THE MATHEMATICAL INTELLIGENCER VOL 13, NO 2, 1991

 (Recall that k t> 2.)
( k , n - k) N o w we h a v e the universal identity (r"+l) = 7~-i(r)"-r
n
Hence
(k, (p - :) k)
- pi- 1) pk- pi- k + i _ 9
(p k k - 1 = k -~i (P~ PZ---11)

kk i-1]'
so that

(p-- 1)z)
' Pz

= dqk (o 1) = dqk.
Finally,

1 (P 1)( pk k - 2

( 1 , q - 1)
pk- q -
Figure 9. Partitioning paths from (1, q - 1) to (k, n - k)
according to where they first meet the line y = (p - 1)x. = P- q(pk- q~
pk- q\k- 1]"
Because i ranges from I to k and there are (~--0 paths
in all from (1, q - 1) to (k, n - k), formula (2.5) is and (2.6) is proved.
proved in this case. Note that nothing prevents us from taking q nega-
We n o w observe that each side of (2.5) is a polyno- tive in Theorems 2.2 and 2.3. Taking q = 0, we obtain
mial in n over the rationals Q of degree (k - 1), and the values of the generalized Catalan numbers:
we have proved that these two polynomials agree for
infinitely m a n y values of n. They therefore agree for Corollary 2.4 ~lo = 1 and ~dk = ~_~),
1
k >I 1.
all real values of n.
We m a y use Theorem 2.2 to calculate the value of We come n o w to the second fundamental property
dqk; recall that q ~< p - 1. of the n u m b e r s dqk, for a fixed p ~> 2.

THEOREM 2.3 If k >>-1, then THEOREM 2.5 (Recurrence relation for dqk). Choose a
fixed r >>-1. If k >I 1, and q < p - r, then

Proof: Because, from its definition, dqa = 1, formula dqk = Z dp-r,'dq+r,J' where i >~ 1, j >I 1, i + j = k + 1.
z,!
(2.6) holds if k = 1. We therefore assume k / > 2. We
(2.7)
first substitute 3 n = pk - 1 into (2.5), obtaining
The proof proceeds by partitioning the good paths
1 2 d~ (Pk k _ 1 from (1, q - 1) to Pk according to where they first meet
the line y = (p - 1)x - r (see Figure 10). We suppress
We next substitute n = pk - 1 in (2.5), but n o w re- the details of the argument, which resemble those for
place k by (k - 1), obtaining Theorem 2.2. However, notice that formula (2.7) gen-
eralizes the familiar recurrence relation (1.4),
2 ~ d , ( P ~ - /P'_--11).
Ck= ~ c,q, k > ' l ,
i+y=k-1
3 It is a m u s i n g to note that t h e combinatorial a r g u m e n t in t h e proof of
T h e o r e m 2.2 required n >t pk, a n d we are here s u b s t i t u t i n g a value of for the Catalan numbers, in the light of (2.3) and (2.4)
n less t h a n pk. - - w e have merely to substitute p = 2, q = 0, r = 1. In

THE MATHEMATICALINTELLIGENCERVOL 13, NO 2, 1991 71

 Corollary 2.7 If k >I 1, then

= E A1 A2 . . .
,l+Z2+...+Zp=k-1

+ Because the numbers pbk are easily seen to satisfy the

same relation, Corollary 2.7 provides another proof of
t h e e q u a l i t y of pd k w i t h t h e k th g e n e r a l i z e d
Catalan number. It also s h o w s us that, if S(x) is the
p o w e r series oo
S(x) =
(p - - 1 ) 1 - - r ) k=O
then
-"-X xSp = S - 1. (2.9)
Klamer [6] attributes to P61ya-Szeg6 [7] the observa-
tion that we may invert (2.9) to obtain

(1,p - 1 - r ) S(x) = 1 + ~ , - s k- 1
(o, --r) indeed, this is the solution given to Problem 211 on p.
(1, q - - l )
125 of [71, based on the theory of Bfirmann-Lagrange
series. Thus, Corollary 2.7 leads to a different, but far
more sophisticated, evaluation of the generalized Ca-
Figure 10. Proof of Theorem 2.5.
talan n u m b e r s pdk, though it does not, in any obvious
way, yield Theorem 2.3.
fact, we may use Theorem 2.5 to express dqk in terms of Note that Corollary 2.4 could, of course, have b e e n
the generalized Catalan numbers ~di, at the same time o b t a i n e d w i t h o u t introducing the quantities pdqk for
generalizing (2.4): q # 0. H o w e v e r , these quantities have an obvious
combinatorial significance and satisfy a recurrence re-
T H E O R E M 2.6. If q ~ p - 1, and if k >! 1, then lation (Theorem 2.5) m u c h simpler than that of Corol-
lary 2.7. In the light of (2.1) the quantities pdqk them-
selves deserve to be regarded as generalizations of the
dqk = E rdq ~d,2 . . . ~dlp_q. (2.8) Catalan n u m b e r s - - t h o u g h certainly not the ultimate
q+i2 + . +ip_q=k-1
generalization! We will see in the next section that
Proof: The case q = p - 1 is just (2.4), so w e assume they are useful in certain important counting algo-
q < p - 1. Then, by Theorem 2.5, with r = 1, rithms.

3. C o u n t i n g p-good Paths
dqk = ~ dp-l,, dq+l a,
',] We will give in this section an algorithm for counting
wherei/1, j1>l,i+j = k + 1 the n u m b e r of p-good paths from (c,d) to (a,b), each of
those lattice points being a s s u m e d below the line y =
(p - 1)x. As explained in Section 1, it suffices (via a
= ~ ~d,ldq+ 1,12, where j2 ~ 1, il + j2 = k, translation) to replace (c,d) b y the point (1, q - 1),
'1,12
with q ~ p - 1. Further, in order to blend our notation
using (1.4). with that of Section 2, w e write (k, n - k) for (a,b), so
that n < pk. We assume there are paths from (1, q - 1)
Iterating this formula, we obtain to(k,n - k),i.e.,thatk1>l,n- k / > q - 1.
dqk Of course, there are (~-~) paths from (1, q - 1) to
(k, n - k). It is, moreover, easy to see that there exist
= E pdH p d z 2 . . , pdzp_q_l pdp_l,lp_q , p-bad paths from (1, q - 1) to (k, n - k) if and only if n
Zl+Z2+ . +,p-q- l +lp-q=k - k I> p - 1, so we assume this--otherwise, there are
where jp_q >i 1. only p-good paths and we have our formula, namely,
(~--0.
A second application of (2.4) yields the formula (2.8). To s u m up, we are counting the p-good paths from
Setting q = 0 in (2.8), and again using (2.4), yields (1, q - 1) to (k, n - k) u n d e r the (nontrivializing) as-
our next result. sumption that 1 ~ k ~ n - p + 1 ~ p(k - 1); notice

72 THE MATHEMATICALINTELLIGENCERVOL 13, NO 2, 1991

 that this composite inequality implies that, in fact, k/>
2, so that we assume
y = (p-l)/
2~k~n- p + l <-p(k- 1). (3.1)

g =
J 9 (k,. - k ) There are h~-~) paths in all, so we count the p-bad
paths. Let (~ = [~-1~], where [x] is the integral part of x.
Then (3.1) ensures that (~/> 1, a n d a p-bad path from
(1, q - 1) to (k, n - k) will meet the line y = (p - 1)x
first at some point Q(j,(p - 1)j), j = 1, 2 . . . . .
Figure 11).
e (see

Because there are dqj paths from (1, q - 1) to Qi that

stay below the line y = (p - 1)x except at QI' and be-
cause there are (~-ro paths from Qj to (k, n - k), it
follows that there are dq1(~-rO bad paths from (1, q - 1)
to (k, n - k) that first meet the line y = (p - 1)x at
Qj. Thus the number of p-bad paths from (1, q - 1) to
(k, n - k) i s
(1, q - 1

Figure 11. A bad path from (1, q - 1) to (k, n - k), n < pk.
Notice that Qe is the last possible initial crossing point of
the line y = (p - 1)x for such a path. We have proved the following result.

THEOREM 3.1. The number of p-good paths from (1, q - 1)

to (k, n - k), under the conditions (3.1), is

Yl y = 2x
, [n,]
We m a y express this differently, however, and in a
w a y that will be more c o n v e n i e n t if ~ is relatively
xxxx [ (5,
large. If we invoke Theorem 2.2, we obtain the fol-
b x lowing corollary.

Corollary 3.2 The number of p-good paths from (1, q - 1) to

(k, n - k), under the conditions (3.1), is
(3, 6 ) ~ \
NN
y=4 ' -,,J r ,-k 1
\ " ~ (5, 4) l=e+l J ' LP-lJ

It is interesting to observe that the binomial coeffi-

1, 2) d e n t s entering into the formula of Corollary 3.2 are
certainly 'generalized,' because k - j > n - pj if j > 2.
Thus if e + 1 ~ j ~< k, then (~--ro = o, so long as n - pj
/ >/0, i.e., j ~ In~p]. Because n/p > (n - k)/(p - 1), it
/ 9 (1, -1) follows that In~p] <~ [(n - k)/(p - 1)], so that we m a y
improve Corollary 3.2, at least formally, by replacing (~
by m = In~p]. Thus we obtain our final corollary.
Figure 12. Corollary 3.3 with q = 0, k = 5, p = 3, n = 9; Corollary 3.3. The number of p-good paths from (1, q - 1)
h e n c e l = 2 , m = 3.
to (k, n - k), under the conditions (3.1), is
all paths bad paths good paths
-- + r -3 6
1=m+1 J '
= 1(15) + 3(1) + 12(0) + 55(-3) + 273(1)
i.e., 126 = 18 + 108 We give an example (see Figure 12).

THE MATHEMATICALINTELLIGENCERVOL 13, NO 2, 1991 73

Example 3.1. Let q = 0, k -- 5, p = 3, n = 9. The in- him in connection with his solution of the problem of
equalities (3.1) arc certainly satisfied and m = 3. Thus dissecting a polygon by means of non-intersecting di-
the number of 3-good paths from (1, - 1) to (5,4) is agonals into triangles; thus his definition of the kth Ca-
talan number is our Ck. In fact, the problem of enumer-
ating such dissections had already been solved by
Segner in the eighteenth century, but not in conve-
Now dj is the jth (generalized) Catalan number for - nient or perspicuous form. Euler (see the bibliog-
3; thus, by Corollary 2.4, d4 = 88 = 55 and ds = ~(i~) raphy) had immediately obtained a simpler solution
= 273. Thus the number of 3-good paths is 108. Notice involving generating f u n c t i o n s - - w e have sketched
that the total number of paths is (9) = 126. Notice also this approach in Section 1. Catalan's own solution,
that, in this case, e = [(n - k)/(p - 1)] = 2, so that achieved almost simultaneously with that of Binet
there is an advantage in replacing e by m. a r o u n d 1838, was even simpler, not requiring the
Circumstances will determine whether it is more theory of generating functions, which was regarded at
convenient to use Theorem 3.1 or Corollary 3.3. In the the time as rather 'delicate' in view of its (apparent)
example above there is little to choose. However, dependence on the notion of convergence of series.
whereas in Theorem 3.1 each term has an obvious A h o s t of other interpretations of the Catalan
combinatorial meaning, it is difficult to assign a com- numbers have been given, largely inspired by work in
binatorial meaning to the terms in Corollary 3.3; the combinatorics and graph theory, and have been col-
binomial coefficient (~2r0, j I> m + 1, does not count lected by Gould (see the bibliography). The interpreta-
the paths from Q1 to (k, n - k), because there are none tions have led to generalizations, of which the most
- - m i g h t it be said to count 'phantom paths'? obvious and important, treated in this article, have
Example 3.1 is really "generic" and thus merits been studied by probabilists (e.g., Feller) and others.
closer study. Theorem 2.2 tells us, in this case, that The ballot problem itself leads to an evident general-
ization, considered by Feller, in which the line y = x is
5
(9- replaced by the line y = (p - 1)x.
An interesting generalization of a somewhat dif-
ferent kind is treated by Wenchang Chu (see the bibli-
The expression of the right of (3.2) breaks up into ography).

33/ References
1. D. Andr6, Solution directe du probl~me r6solu par M.
N o w the binomial coefficient (9) of the left of (3.2) Bertrand, Comptes Rendus Acad. Sc~. Paros 105 (1887),
436-7.
counts the number of paths from ( 1 , - 1) to (5,4); the 2. A. P. Hillman, G. L. Alexanderson and R. M. Grassl,
first block in (3.3) counts the bad paths from (1, - 1) to Discrete and Combinatorial Mathematzcs, San Francisco:
(5,4), partitioning them into those that first meet the Dellen (1987).
danger line y = 2x at (1,2) and those that first meet the 3. Peter Hilton and Jean Pedersen, Extending the binomial
coefficients to preserve symmetry and pattern, Com-
danger line at (2,4); the second block in (3.3) makes a puters Math. Apphc. 17 (1-3) (1989), 89-102.
zero contribution; and the third block in (3.3) thus 4. Peter Hilton and Jean Pedersen, The ballot problem and
counts the good paths from ( 1 , - 1) to (5,4). In some Catalan numbers, Nzeuw Archief voor Wiskunde, 1990 (to
mysterious sense the third block--which intrigues us appear).
e n o r m o u s l y - - i s also partitioned by the points (4,8) 5. A. Hurwitz and R. Courant, Allgemeine Funktzonentheorze
und elliptische Funktionen. Geometrische Funktionentheorie,
and (5,10) on the line y = 2x. But this is only a nota- Berlin: Springer (1922), 128.
tional partitioning, because we have 108 represented 6. David A. Klarner, Correspondences between plane trees
as (55)(-3) + (273)(1). Here the coefficients d,(i = 4,5) and binary sequences, J. of Comb. Theory 9 (1970),
have an obvious combinatorial significance, but the bi- 401-411.
nomial coefficients, (]3) and (56), do not. They result 7. G. P61ya and G. Szego, Aufgaben und Lehrs~tze aus der
Analyszs, Vol. I, Berlin, Gottingen, Heidelberg: Springer
from applying the formula (~_-p') for the number of (1954), 125.
paths from (i, (p - 1)i) to (k, n - k) outside its domain 8. N. J. A. Sloane, A Handbook of Integer Sequences, New
of validity 1 ~ i ~ 2. There is here much food for York and London: Academic Press (1973).
thought! 9. Marta Sved, Counting and recounting, Mathematzcal In-
telligencer 5, no. 4 (1983), 21-26.

Historical Note Bibliography

The versatile Belgian mathematician Eugene Charles [Items [2], [6] and [8] among the References may also be re-
Catalan (1814-1894) defined the numbers named after garded as general sources on Catalan numbers.]

74 THE MATHEMATICAL INTELLIGENCER VOL 13, NO 2, 1991

 R. Alter, Some remarks and results on Catalan numbers, series inversion, Trans. Amer. Math. Soc. 94 (1960),
Proc. 2nd Louisiana Conf. on Comb., Graph Theory and Comp. 441-451.
(1971), 109-132. D. G. Rogers, Pascal triangles, Catalan numbers and re-
D. Andre, Solution directe du probl~me r~solu par M. Ber- newal arrays, Discrete Math. 22 (1978), 301-310.
t-rand, Comptes Rendus Acad. Sc~. Parrs 105 (1887), 436-7. A.D. Sands, On generalized Catalan numbers, Dzscrete
W. G. Brown, Historical note on a recurrent combinatorial Math. 21 (1978), 218-221.
problem, Amer. Math Monthly 72 (1965), 973-977.
Wenchang Chu, A new combinatorial interpretation for gen- Memoirs by Eugene-Charles Catalan relevant to the theme
eralized Catalan numbers, DiscreteMath. 65 (1987), 91-94. of Catalan numbers may be found in Journal de Math&na-
K. L. Chung and W. Feller, On fluctuations in coin-tossing, tlques pures et appliqu~es de Liouville, (1), III, 508-816; W,
Proc. Nat. Acad. Sc~., U.S.A. 35 (1949), 605-608. 91-94, 95-99; VI, 74, 1838-41.
John H. Conway and Richard K. Guy, The Book of Numbers, For details of Catalan's life and work see "Les travaux math-
Sci. Amer. Library, W. H. Freeman (to appear 1990). ~matiques de Eugene-Charles Catalan," Annuaire de L'Aca-
I. Z. Chomeyko and S. G. Mohanty, On the enumeration of ddmie Royale des Sczences, des Lettres et des Beaux-Arts de Bel-
certain sets of planted trees, J. Combm. Theory Ser. B18 gique, Brussels (1896), 115-172.
(1975), 209-21.
T. T. Cong and M. Sato, One-dimensional random walk Peter H~lton Jean Pedersen
with unequal step lengths restricted by an absorbing bar- Department of Mathematical Department of Mathematzcs
rier, DzscreteMath. 40 (1982), 153-162. Sciences Santa Clara University
R. Donoghey and L. W. Shapiro, Motzkin numbers, J. State University of New York Santa Clara, CA 95053 USA
Combin. Theory Ser. A23 (1977), 291-301. Binghamton, NY 13901 USA
N. Dershowitz and S. Zaks, Enumeration of ordered trees, and
Discrete Math. 31 (1980), 9-28. Abteilung fi~r Mathematik
L. Euler, Novz Commentarii Academiae Sczentarzum Imperialis E.T.H., 8092 Ziirich, Switzerland
Petropohtanque 7 (1758/9), 13-14.
Roger B. Eggleton and Richard K. Guy, Catalan strikes
again! How likely is a function to be convex?, Math. Mag.
61 (1988), 211-219.
A. Erd~lyi and I. M. H. Etherington, Some problems of non-
associative combinatorics, Edinburgh Math. Notes. 32
(1941), 7-12.
I. J. Good, Generalizattons to several variables of Lagrange's
expansion, Proc. Camb. Phil. Soc. 56 (1960), 367-380.
Henry W. Gould, Bell and Catalan Numbers, Research bibliog-
raphy of two special number sequences, available from the au-
thor (Department of Mathematics, West Virginia Univer-
sity, Morgantown, WV 26506). The 1979 edition sells for
$3.00 and contains over 500 references pertaining to Ca-
talan numbers.
Henry W. Gould, Combinatorial Identities, available from the
author (Department of Mathematics, West Virginia Uni-
versity, Morgantown, WV 26506).
Richard K. Guy, Dissecting a polygon into triangles, Bulletin
of the Malayan Mathematical Society 5 (45) (1958), 57-60.
Richard K. Guy, A medley of Malayan mathematical memo-
ries and mantissae, Math. Medley (Singapore), 12, no. 1
(1984), 9-17.
V. E. Hoggatt, Jr., and Marjorie Bicknell, Pascal, Catalan,
and general sequence convolution arrays in a matrix, Fi-
bonaccz Quarterly 14, no. 2, 135-143.
V. E. Hoggatt, Jr., and Marjorie Bicknell, Sequences of ma-
trix inverses from Pascal, Catalan, and related convolution
arrays, Fibonaccz Quarterly 14, no. 3, 224-232.
V. E. Hoggatt, Jr., and Marjorie Blcknell, Catalan and re-
lated sequences arising from inverses of Pascal's triangle
matrices, Fibonacci Quarterly (December 1976), 395-405.
M. S. Klamkin, Problem 4983, Amer. Math. Monthly 69
(1962), 930-931.
Mike Kuchinski, Catalan structures and Correspondences, M.Sc.
thesis, Department of Mathematics, West Virginia Univer-
sity, Morgantown, WV 26506.
Th. Motzkin, Relations between hypersurface cross ratios,
and a combinatorial formula for partitions of a polygon,
and for non-associative products, Bull. Amer. Math. Soc. 54
(1948), 352-360.
G. P61ya, On picture-writing, Amer. Math. Monthly 63 (1956),
689 697.
-

G. N. Raney, Functional composition patterns and power

THE MATHEMATICAL INTELLIGENCER VOL 13, N O 2, 1991 7 5