- EISENSTEIN SERIES 37
we may as well assumes to be real (i.e. in X(A)). If x varies in a compact set D
then, by Section 5.7, there exists constants c, c' > 0 such that
(99) ca(r)s+p :S a(rx)s+p :S c'a(x)^5 +Pa(r)5+P (x ED,'}' Er)
therefore the uniform convergence on D and the convergence at one point are equiv-
alent (and the moderate growth will follow); these conditions are also equivalent
to
(100) in L a(rx)^5 +Pdx < oo
,,Er p 0 \r
Assume D small enough so that nn-^1 n r = {l}. Then:
L r a(rx)^5 +Pdx = L 1 a(x)^5 +Pdx:::; 1 a(y)5+Pdy = J
,,Erp 0 \ r Jn " ,,n <rPo \r)O.
By Proposition 11.5, we can write a(r) E a 0 - Ap, for some a 0 E Ap. Therefore,
since the 1D are disjoint:
J:::; r a(x)s+p dx
lrp\NpMp(aoA?)K
The Haar measure can be written as a-^2 Pdmdndkda. Since rP \MpNpK has
finite volume, we are reduced to showing that
r a^5 -pda = a~-p J a^5 -Pda < 00
Jao -Ap -A
Take as coordinates on A the ta = a/3"', notation of Section 11. l. Write
P = l:a r af3a, s = l:a saf3a. Then the last integral is
II {l t~"'-r"' dt a
a Jo ta
which converges if Sa - r a > 0 for all a E b...
12.6. Bernstein's reduction of the general case to Section 12 .5 relies on the following
crucial observation:
(*) Let f be an automorphic form on r\G. There exists t E X(A), in the
domain of convergence of Ec,p 0 (l, s), such that
(101) lf(x)I-< Ec,P 0 (1,t)(x) (x E r\G)
Proof. First note that for s real, the series defining Ec,p 0 (l, s) consists of positive
terms, the first of which is a(x)^5 +P. Since f has moderate growth on r \^0 G, we can
find ti, in the domain of convergence of Ee ,Po ( 1, s), such that
(102) lf(x)I-< Ea,p 0 (l,t1)(x) (x Er\^0 G).
Combined with Proposition 5.8, this shows the existence of finitely many μi E
X(AG) (i EI) such that
(103) lf(x)I-< Ea,P 0 (l, t1)(x)(L a(x)μ'), (x E r \ G).
But ifμ E X(AG), it is left-invariant under r (or even °G), so that it follows from
the definition that