Discrete Mathematics 25 (1979) 135-139.

@I North-Holland Publishing Company

A SHORT PROOF OF THE LAGRANGE-GOOD FORMULA

Josef HOFBAUER
Institut fiir Mathematik, Uniuersittit Wien, Strudlhofgasse 4, A-1090 Wien, Austria

Received 24 April 1978

A new proof of Good’s generalization to several varia’Bles of the Lagrange inversion formula
is given, which is mainly based on methods used in Roia’s theory of polynomial sequences of
binomial type.

1. Introduction

Since Good’s paper [5j appeared in 1960, :i lot of other works have been
published on this theme. Good’s proof of his theorem being analytical, <:helater
authors considered Lagrange inversion within the theory of formal power series.
The first attempt for a purely combinatorial proof was made by Chottin [ 1j, who
treated a special case of the two dimensional formula. Tutte gave an extensive
development of this subject in [ll j for an arbitrary number of variabks.
While studying Rota’s theory of polynomial sequences of binomial type (see
[8,9, lo]), Cigler [2], Garsia and Joni [3j realized its intimate connection with
Lagrange inversion (in the one dimensional case); &neralizing Rota’v theory to
higher dimensions they obtained an analogot:s setting for the Lagrange-Good
formula (see [2,4,6,73).
This paper is mainly based on these work, it follows the same ideas and gives
only a simplified version of the more technical part of their proofs. Readers
interested in the connection to Rota’s theory which will not be treated here are
referred to the above cited papers.

Relmark. We aid use the standard multi-irdex notation, i.e. for n = (n,, . . . , n,) E
N” (N is the set of nonnegative integers) we mean by n ! = n1! n, !, (r& = l l l

hh, l 8 (r&S (the lower factorials) and e, I=(aij)lsj<_s shall denote the ith basic
l

vector. For x =(x1, . . . , x,) we mean xn =5x);l x2, etc.

l l l

avange=Good

Let qi(Zl,. . . , Z,), i = 1,2,. . . , s be formal power series (over a field K of

characteristic zero) with Qi (0) f 01.Consi.der the formal power series Wi defined by
Wi(Z) =L i=l,...,s.
4pitz)’

135
136 .I. Hofbauer

If one expands any formal power series f(z,, . . . , z,) into a series in the variables
w=(w,***rW,),

f(z(w)) =’ c $ wn,l

then the coefficie,nts c,, are given by the formula

an,+n,+-**+nn
=- Q?(Z) det ( &j +yz_,
. . . 3% f (dQ;'(d
l l l

a nl
Z1 ZS

The intrinsic idea of the proof is the following:

Every formal power series

defines a linear operator

f(D) = c a,,D” = c a,,,. .,,,Dll l l l D>

ton the space P = K[x,, . . . , x,3 of all polynomials in s variables over the field K,
where Di denotes the partial derivation with respect to 4.
The mapping f(z) I+ f(D) gives an imbedding of the algebra K[[z,, . . . , zJ] of
formal power series into the algebra L(P) of K-linear operators on P. This
mapping is one-to-one, for Q = 1 uJn! D” implies Qx” IXCO = a,.
Let Qi be those operators, which correspond to the formal power series Wi(Z).
We will in the following construct a sequence of polynomials (p,JnENS with the
properties

(i) Pn(O) = son9

(ii) Q#n =Q&-ei (lSiSs, nENS).
Observing that (ii) implies Qkp, = (n)k&-_k for n, k E N” one obtains the coeffi-
. . . y w,) by applying the corresponding
cient of a formal power series in w = ( MF~,
L). v+~~~ 5 CJk !Qk to the polynomials p,, :

On the other hand we have

[fQn
det (&j - Wi$)Jn’(0) =f(D)Q"(D) det (6,j -V(D) 2 0)X” 1x=0

Computing these two terms we see that the only thing we have to show is that the
 A short prooj of the L.ugTange-Good j’omulz 137

polynomials defined by

(D)) X”

satisfy the above conditions (i) :and (ii).

Proof of (ii):

Oi
QiPn =- q”(D) det ( )xn = cpnmei(D) &t ( . jDig
l

Cpi(D)
= Q”-“‘(D) det ( )tZiXnvei = qpn_ei.
l

Proof of (i): p,(O) = Son means that the coefficient of D” in the expansion of
Q”(D) det ( ), that is l

This is clear for n = 0.

From the definition of Wi (Z) = Zi/Qi( 2) follows

[Q"(Z)det
(S,-W.(Z)z)l(n)(0)

= [ det (QF6ij - ZiS?>-‘Qi,j)]‘“‘(O)= C sgn UP(O) (2)

u
with

P(O) = ( n (Qysia(i)- ziQ?-lQi,~,~)~n)(o)

lsiss

and Qi.j = aQi/aZj for short.

If some ni = 0, then thla corresponding factor in P(a) has the fOMl (Sim(i)-
ZiQ~lQi.~(i,). Not taking l;he derivative with respect to Zi, for z = 0 we get
P(O) = 0, if o(i) # i. That means that the sum (2) reduces to permutations of the
set W = {i : n, # 0). Losing a generalized version of the Leibniz product rtile
(n)

( rIf
lSi=Sp
i) = ‘C
~IGi:cP~=n
--
k,! l
n!
’ l
nr
kp! lSi<p ’
.ki)
’

P(a) takes the form (non indexed c and n will run over all i E W)

n!
P(a)=
&
1 -
=n Ilki !
n (QFaio(i) - >)‘k”(0)
ziQ1'-'Qi,a(l

= c --
l-B
xk,=n
n!

I!
Zi(Qp-lQi 1
9a(i))(ki’
z=i!

The term with Zi vanishes for z = 0.

138 J. Hoj’bauer

= kii = ith coordinate of the multi index ki.

Bbservivg

and using the abbreviation cpp= 6, we get

k-
P(u)= c 2!- n(r
iki)Siu(i.--L! f~ki-e8+ecrC6J

I
.

~~=nllki! 4

Substituting the indices ki by li = ki - ei + e,(i) gives in view of C Ii = yt

for (l/ki!)kii = (1/1,!)1,,,,.

Notice that the delicate cases iii = -1 and lb(i) = 0 do not disturb the calculation.
In this last representation of P(a) the only term still depending on the special
permutation 0 is n(&[i]- (ib(iJni)). Summing (2) over all permutations of the
set W, we get just the determinant of the matrix (Sii - (lij/ni))i,jEw, which is zero,
for its rows are linearly dependent:

Zi) =O.
i&ni(&j-~)znje~ lij =(lZ.-~

Thus (1) is established for II> 0 and the proof is completed.

Ill 1 Chottin, Une demonstration

.L_ . combinatoire de la formule de Lagrange a deux variables,
Llrscrete Math. 13 (1975) 215-224.
PI J. Cigler, Sequences of polynomials of binomial type and the Lagrange-Good formula, preprint,
Universi*Qt Wien (1977).
[31 A.M. Garsia and S.A. Joni, A new expression for umbra1 operators and power series inversion,
Proc. AMS 64 (1977) 179-185.
WI A.M. Garsia and S.A. Joni, Higher dimensional polynomials of binomial type and formal power
series inversion, Comm. Algebra, to appear.
Fl. I.J. Good, Generalization to several variables of Lagmnge’s expansion, with applications to
stochastic processes, Proc. Cambridge Philos. Sot. 56 (1960) 367-380.
PI S.A. Joni, Polynomials of binomial type and the Lagrange inversion formula, Ph.D. Thesis,
University of CaXornia, San Diego, CA (1977).
r71 S.F.. Joni, Lagrange inversion in higher dimensions and umbra1 operators, Linear and Multilinear
Algebra, 6 (1970) 111-122.
 A short probf of the Lagrange-Good fornaula 139

[8] R. Mullir. and G.-C. Rota, On the foundations of combinatorial theory, III: Theory of binomial
enumeraf.ion, in: Graph Theory and its Applications (Academic Prees, New York, 1970)
167-213.
[9] St. Roman and G.-C. Rota, The umbral calculus, Advances in Math. 27 (1978) 95-188.
[lo] G.-C. Rota, D. Kahaner and A. Odlyzko, Finite operator caIcuhrs, J. Math. Anal. Appl. 42
(1973) 685-760.
[ll] W.T. “I’utte, On elementary calculus and the Good formula, J. Combin. Theory 18(B) (1975)
97-13’7.