Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Proc. Nati. Acad. Sci.

USA
Vol. 74, No. 4, pp. 1328-1331, April 1977
Mathematics

Fourier analysis on the Heisenberg group


(Schwartz space/group contraction/asymptotics/spherical harmonics)
DARYL GELLER
Courant Institute of Mathematical Sciences, New York University, New York, New York 10012
Communicated by Elias M. Stein, January 24,1977

ABSTRACT We obtain a usable characterization of the S Hf(uv ')g(v)dv. If L is left invariant, fg E eV then L(f*g)
(group) Fourier transform of ef(H n) (Schwartz space on the -f*Lg.
Heisenberg group). The characterization involves writing ele- For X > 0, let 1x be Bargmann's space: 1X4 = If holomorphic
ments of C-(H ') as asymptotic series in Planck's constant. In the onCnI(2X/Ir)nfCIF(r)I2exp(-2X II2)dV = ODF 12<. 1x
process, we derive a new "discrete" version of spherical har- is a Hilbert space, and monomials Fa,>A = (V2/;4)a/vf/a[aCE
monics, and elucidate the theory of group contractions. We give (Z + )n, Z+ = 0,1, .. . j] form an orthonormal basis for 14.. Let
an application to Hardy space theory. eJ = (0.1, 0) E Rn with the 1 in the ith position. Define
PA to be the antirepresentation of the algebra (TZ1,Z1) on 14.
Objectives with 'x(T)O = -iX4, 2x(Zj)O =-21 XI A;+, 2x(ZZ)h = (a/
Our purpose is to derive machinery to make the group Fourier 8d')k, for k EE 1.. Then, P x(Z)Fa,\ =
transform on the Heisenberg group Hn a useful tool. In par- -V21A|(a; + 1)Fa+ejf, 9A(Zj)Fa,A = aVFa-ei,x,
ticular we derive a characterization of the Fourier transform P,\(Lo)Fa, IXI(21aI + n)Fa,
= (where lal = Eaj). ForX
(F.T.) of e(Hn) (Schwartz space), which we then apply to de- < 0, we obtain another antirepresentation Px on WI xI by re-
velop the real-variable theory of Hardy spaces on Hn. The main versing the roles of Zj and Zj in the above formulas.
For u E Hn, let u be the translation operator on functions
Downloaded from https://www.pnas.org by UNIVERSITEITSBIBLIOTHEEK GENT on May 14, 2024 from IP address 157.193.164.114.

ingredients are these:


(a) Exact Formulas. We define a certain space hn and con- f on Hn: [uf](v) = f(vu). If f is real analytic, [(t,z)f](u) =
sider the F.T. as mapping L2(Hn) onto L2(hn). We derive an exp(tT + 2xjXj + 2yjYj)f(u) = exp(tT + ZzjZj +
analogue, in this setting, for the Bessel function formulas which E2xjZj)f(u). u - (u-) is an antirepresentation of Hn, so u
give the Euclidean F.T. of a function in terms of its spherical x[ (U'-) ] is a representation of Hn. To compute it, say, for
harmonic decomposition. On the hW side, we use a new, "dis- ease, n = 1; one computes (by Eq. 1) that exp (tT + zZ + zZ)
crete" version of spherical harmonics. - e(t-1IzI2)TeizezZ, whence, formally, if X > 0, UX(t,z)F(r) =
(b) Contraction. Contraction is a "flattening out" process a9D(_t, _z)]F(v) = F(l7 -Z)elxt+2X(9IzI 2/2) for F EE Wx.
which we regard as a generalization of the approximate identity Ux(tz) is unitary because P'(tT + zZ + zZ) is skew-adjoint.
process. The F.T. on Cn is shown to be a limit of the F.T. on Hn, As X ranges over R, X $ 0, the Ux(t,z) give all nontrivial irre-
as Planck's constant suitably tends to zero, and the formulas of ducible unitary representations of Hn.
a have their Euclidean analogues as imits. Forf E LI(Hn), X 0, set (X) = f Hnf(u)Uxwdu; the op-
(c) Asymptotics. Set S(hn) = ;V(Hn). Contraction gives a erator family Vf(X)j is the Fourier transform of f. Let Cf be the
map from eV(hn) to ef(Cn). We study the extent to which this operator of convolution on the right by f, so that Cf(g) = g*f;
process can be reversed. Given R E ef(Cn), we give a function then formally, Cf = S Hnf(u)(u-)-dV, so that (X) = P\(Cf
R#CE(hn) which contracts to R; R# has a very simple form. If L is any left invariant differential operator, Lf(X) = PX,(CLf)
We show how to expand functions h C A(hn) asymptotically fx(LoCq) = Px(Cf)2x(L) = 1(X)Px(L) if f E Cc . So
in terms of functions of the form R #, and thereby reduce cer- g*f(Q=Cj (X) = R(X)Px(Cf) = g(X)J(X) if fg L'. If L E
tain questions about eV(hn) to questions about eV(Cn). IT,Zj,Z}j and R is the right analogue, g*Rf = Lg*f if fg E
Cc a; so R7(X)= I' (L ).
This definition is in analogy to the following definition of A,
Background
the standard Euclidean F.T. on C: for r E C, 1g(r) = cfI(Scg
Hn is the Lie group with underlying manifold R X Cn and exp[-(zO/Oz + z-l/0)]dV) where d1(O/cz) = -(1/2)it,
multiplication 1i.(d/ ) = -(1/2)i . It will be convenient to define T': L2(Cn)
(t,z) - (t',z') = (t + t' + 2Im z z', z + z')
- [1 L2(Cn) by 'g(71) = 3g(-2in) for E= Cn. Then
where z-z' = Xz1j'1. If 'y:R -- Hn is smooth, 7(0) = 0, then the
TO(8/1049)(n) = WkS7g(n), 5(a/"O kg)(n) = -%kTg(n).
Sayf C C2(n+2)(Hn), Lokf CL'forO<k <n +2.Then
equation [X,(f)](u) = (d/ds)f[u-"y(s)]Is=o(f C Co) defines a 1(X) is trace class for each A, and one has:
left invariant vector field X y. Use coordinates (t,x,y), where z Plancherel Theorem
= x + iy. Let y be any of the coordinate axes, to obtain the
vector fields T = (0/ct), Xj = (a/8x;) + 2yj(O/Ot), Y1 = (a/
III(X)12 XIndX = (2r)n+ 1f12
co

4y) - 2xj(O/Ot). Set Zj = 1/2(X - iY4) = (o//&0) + i1(O/Ot),


Zjfor=left(O/O&) - izj(O/at). Then TZJZZI(j = 1, . . . ,n) is a basis
invariant vector fields. We have [Zj,ZkI = -26jkT, [11(AX) 112 denotes the Hilbert-Schmidt norm of J(X)].
[ZJ,ZkL= [Z1,T] = 0. The right invariant analogues are Inversion Theorem
T,ZJR,ZR where ZJR = (a/lz1) - iz-(ad/t). For a e C, set La
=-1/2 X(Z1Z1 + Z1Z1) + iaT. 1 tr[Ui-xl(X)jIXIndx=(- ) PO,
If fg C L'(Hn), define their convolution by f*g(u) =
where the integral converges absolutely and uniformly for u
Abbreviations: F.T., Fourier transform; F.S., Fourier series. E Hn.
1328
Mathematics: Geller Proc. Natl. Acad. Sci. USA 74 (1977) 1329

Plancherel enables one to extend the F.T. to L2(H ).'The Let h" = I(X,x)IX E R,x E cWI. On hn we use the element
proof of absolute and uniform convergence in the inversion of measure IXI dAxdX. If g C L2(Hn), g has F.S. type m and
theorem is particularly important. One notes that [(21al + g(X) = 2aRam(X)W, m (X), we now considerg to be the func-
n) ||]k III(X)FaXII = II -()Fa, | < |ILfokf 1 l for 0 < k < tion on hn given by g(X,2a XI ,w) = Rm(X)eim ¢ [here w =
n + 2, so (e.1,'... ,ein)]. This identification is extended linearly to a
constant multiple of an isometry from L2(Hn) onto L2(hn),
EIII| (X)Fa, x 1 I A I nd X with functions of F.S. type m going to functions of F.S. type m
a -co [we say h C L2(hn) has F.S. type m if h(X,2aXI,w) =
< 2C E (T+n'-) XAndX +
00
XndA S(X,2a I XI)eim 0 for some S].
0=0 nI SO 1,6 on+ 2Xn + 2 This is more than a matter of notational convenience. We can
give formulas for the F.T.s of a number of operations on
< 2C F, n+ I < co. functions on Hn; these formulas bear a very striking similarity
03if~ to their Euclidean analogues. The simplest is this: say n = 1;
For m zzn, say m > 0 if each mk > 0. For m > 0, A > 0, define the operators WxW*x on those G C L2(al#,,\) with F.S.
a E (Z+)n, define the partial isometry Wam(X) on lb. as type m > 0 as follows:
follows: Wasm (X)F3, = (-1)1 m a+ m,/Fa,A. WxG(2aIXI|,) = r2AX(a + m ) e-i0G(2aIXI,w), [4]
Assume now n = 1, g(t,z) = f(t, Iz )eimO (where z = Iz Ie'0)
(we say that g has Fourier series type m). Suppose g C L2(H1),
m > 0, A > 0; then, formally,
W*xG(2a I X ,w) = 2 AI(a + m + 1) eibG(2aIXI,w).[5]
Then if g C eS(Hn) has F.S. type m > 0, and X > 0, we
W = Px (SH ge(t+iIz 2)Te _Ze zZdV) have:
(J ( 12' 1zIz I2kzk+ m-k
X ) zg= V Zg = CW*k4 [6]
k=o (k + in)!k!dV This is immediate from the definitions. These relations are in
analogy to the fact that ' on C maps a/az to multiplication by
Downloaded from https://www.pnas.org by UNIVERSITEITSBIBLIOTHEEK GENT on May 14, 2024 from IP address 157.193.164.114.

by expanding the exponentials and using Fourier series argu-


ments. Thus, one sees that g(X) = 2aRam(X)Wam(X) where v and ad/d to multiplication by -i.
We can go much farther. For example, say n = 1; define the
Ram() = f(t, I zI )eitlam(21X 1 I z1 2)dV [2] operator 6+/(62aI X) on functions G on cxTn as follows:
where 62a IG(2aXIW) -21X1
lam(X) = x/' a xm/2Lam(x)ex/2 X fG[2(a + 1)IXI,w] - G(2ajXj,w)).
(a + in)! This is in analogy to a/aInI12. If mi> 0, g EE e(Hn) has F.S. type
a (-X)k m, then
Lam(X) = ( +k )k
k=O
W,,m (A) is defined in general in such a way that Eq. 2 is true Z (2' a,w) =-M[ei
even if m < 0 or A < 0. This is the analogue of the Bessel func- 6+ 1 1 \ 1 \
tion formula for the Euclidean F.T. 5r acting on functions of
Fourier series (F.S.) type m [that is, functions of the form g(z) ba6 r-aT -)- \2 /
= f( z I )eImO]; that formula also admits a similar formal proof. where a(k) = [a!/(a - k)!] and M(*) denotes multiplication by
The lam are the Laguerre functions, the Lam are the Laguerre the function in brackets. (The formula for general A > 0 is ob-
polynomials. The $lam(x)-,a=, are an orthonormal basis of tained by scaling.) The formula is easily proven with the aid of
L2(R+) for any m, and the inversion formula becomes: the identity zT = -(1/2)i(ZR - Z). Eq. 7 is in analogy to the
4 co fact that 7' on C maps multiplication by z to-a/ai, because
fptz) =(21r)2 E f e-ixtRam(A)1a ml (21XAz I I 2)IXI dA. if h C C'(C) has F.S. type m, then (a/acl)h = ei'1 | m+
[3]
dajn2)lnI-mh for n # 0. [This is seen by writing h(n) =
IMPHt l 2) for some f, for tq # 9ne then may produce an
operator Ox on cxn such that I;12g = -Qg. Qx is a differ-
For general n, one obtains similar formulas with products of
Laguerre functions. Eqs. 2 and 3 also appear in the earlier pa- ence-operator analogue to A, the Laplacian on C".
pers (1-3). Now say
The spaces CA n and h n P(z) = E alzlpl(') ... Zn Pn(z)z li(-)... Zn qn(Y) [8]
We would like to think of the F.T. as mapping L2(Hn) onto
-y
L2(hn), where hn is a space we now define. is a (solid) spherical harmonic on c" of degree K. We say that
For A # 0, let c\n be the space (2IXIZ+ )n X (S')n where we P has type (p,qm) (where p,q C Z +, m C Zn) if P has holo-
use coordinates (2ai I XI, . . . 2a. I I| ,e 1, . . ,e 0n)(aj E Z+, morphic homogeneity p, antiholomorphic homogeneity q and
0 < Oj < 2r for each j). cxn is thought of in analogy to con = Cn F.S. type m [i.e., for each Y,2kPk(Y) = P, 2kqk(y) = q, Pk(Y)
- (R+)n X (Sl)n where we identify 77 C Cn with
(lql 12,. - qk(y) = mIk]. We let WKp denote the space of all such P.
Itn I2,ei,0 0e1n)
*
the "sphere of radius 2
y1
where = Jnj e'ki. For f'C- Z+,
in cxn is f(2a AI,CV) C Cxn I at
o2n-1l, Then it is easy to see that, if 14 denotes the space pf all spherical
harmonics of degree K, then 1-K = CDpq,,, [orthogonal direct
= ,B}; we use the measure v#,\ on this sphere, where vP,x is the sum over all (p,qm) with p + q = K, m EC Zn]. It is particularly
product of counting measure and Lebesgue measure, times the interesting to ask for the F.T. of functions g C L2(Hn) of the
factor I A n-1l The measure uAx on cxn is ^VO,X. form g(t,z) = f(t, |z 2)P(z).
1330 Mathematics: Geller Proc. Natl. Acad. Sci. USA 74 (1977)

To this end, if P is as in Eq. 8, P E Ifpqm, and m 2 0, define convolution on Hn,Hfn,Cn, respectively. Then Mzg*DJ(t,z)
r(P) on Cl/2n by = g*jE(t,z,) if we make the change of variables (t,z) =
n (t,Vz,) that enables one to pass from Hn to Hen. We lete
r(P)(a,w) = -Ya,, k=i
VaH(9k(Y))(az + mk)(Pk(E)) eim 0
II 0 and vary z with e in such a way that ZE Zo C Cn. Then
limE_-o g*JE(t'ZE) = g9()*4o(zo) where g(t)(Z'o) = g(tz'o),
and define rx(P) on cXn by: rx(P)(2ajXjw) = fo(z'o) = f if(t,z'o)dt.
r(P)(a,w)(2jX| ) I/2(p +q). (There is a similar definition for the We study the dual process on the F.T. side. We define hEn
case m I 0.) The rx(P) have the following properties (we as- {(X,X)IX C= R, x C CA n. It turns out that the (naturally de-
sume m > 0): fined) F.T. -E on L2(HEn) maps L2(HEn) to CVy, a certain sub-
space. of L2(hEn), and has the property that
(i) QArx(P) = 0. "-nMZeg(X,2a I X 1,W) = (,2alxlE|w). Then
(ii) If P C fpqm, Q C Wp'q'm, then M-nzg*D(X,2a I X |,,w) = goof (X,2al A| ,w). For fixed
X, we takethe limit ase --O0 and (2aE,w) In,(l12i-.InnI2,W)
(r(P),r(Q))#= (fi + p + n - )(P+q+n-l)(pQ) = C Cn [we have againused the identification (R+)n X (S')n
[the inner product on the left is in L2(4,2#nl71), on the right in - C"]. One finds that g*J E(XA,2a X |W,) (9'g)(x)(v'[XTn)
L2(S 2n -1)] 7'fo(V[ Tn) where we have written (51'g)(X)(v) =
(Mi) CD <#r(iq~,-T = n 2(--n1
type f5ei1tX7'g(t)(v)dt for v C Cn.
(iii) #q<\ A( m|f3fallfinL2(ar')with F.S. type The proof of the last statement is easy on a formal level; we
simply use the fact that VWA, and W*eaf strongly approach
(iv) If g(t,z) = f(t, Iz I 2)p(Z) E L 2(Hn), then multiplication by and n, respectively. Specifically, we assume
n = 1, and define Ex = so that EX(2aI XA,w) = 1. Now if (X,X)
g(X,2a I A |,w) = R(X,21alIX )rx(P)(2a I X ,w) C h',J(X,x) = Cf(6)(Xx), so
for some R; there is a Laguerre function formula for R.
Thus, we see that the TX(P) are a discrete version of ordinary DJf(X,2a1 XAI )= ( SH' exp (iXEt' - Z'WXE + F'*X )
Nwp,,, with iv being the analogue of the standard Bessel
Downloaded from https://www.pnas.org by UNIVERSITEITSBIBLIOTHEEK GENT on May 14, 2024 from IP address 157.193.164.114.

function formula. X f(t,',z')dt-'dz,'dz-] Ex,) (XE,2aJXAl ,@).


i is easy, and iif follows from ii by a dimension count. ii and
iv are proven by use of the operators Mjk = Zjg/Ozk - zkO/0Zj As E 0, (2aE,w) - n, the limit is then just 7'fo(V[$$).
(1 < j, k < n) which enable us to handle the unitary group U(n)
in infinitesimal form. The F.T. of Mjk is a difference-operator Similarly, (_nMzzg contracts to (97g)(x).
analogue of -Mjk and is easy to work with. There are similar limiting formulas for A instead of -. The
The Mjk operators are important. In another paper, we will justification is easy if the Fourier series of f and g converge
compute the spherical Laplacian on forms on S2n-I and Ob on absolutely in L'(Hn). The classical formula which results is lim
S2n-' in terms of the Mik and Mjk. lam(21 X Je) = Jm(2V/]Ti), as e -*0 and 2ac- p.
In preparation for the next section, we note that Eq. 6 does The contraction process explains why Qx contracts to A,
not hold if m < 0 or X < 0. However, the definition of the F.T.: etc.
L2(HI) - L2(hl) can be modified slightly to make Eq. 6 hold Schwartz space
in general. We write f --- for this modified F.T.; then maps
-

L2(H') onto V, a certain subspace of L2(hl), and Zg = Jag, If R = 2;tnRpqm (jn7)Ppp,


I ( is a function on Cn, with PPqw
Zg = -s*ta for any g C *(Hl) (we have extended the defi- C Ip,,M,, let R # be the function on hn defined by:
nitions 4 and 5 linearly and by passing to closures).
R#(X,2aJ XI,w)
Contraction of H n
Group contractions were studied by Inonu and Wigner in ref. Z Rpq,[(2a
= Pqm + n + ImJ)|XA|]rx(Ppqm)(2alXl,w)
4, mainly in connection with the rotation and Lorentz groups.
In our case, the F.T. and the dilation structure considerably (provided the series converges).
clarify the theory. We will see that, not only is there a close Recall e?(hn) = f|f CE eV(Hn)l. Then we have:
analogy between the F.T. on Hn and on Cn, but that the F.T. Theorem 1. R C J'(C n) R# CE R(h n). The proof of The-
=

on Cn is a limit of the F.T. on Hn. orem 1 is lengthy and computational. We obtain preliminary
The simplest example of the contraction process is the fa- criteria for a function to be in V(hn), and verify that R# satisfies
miliar approximate identity process, which we briefly review the criteria; we will say no more here.
on Hn. Iff C L'(Hn), g & C,, Djf(t,z) = E-(n+ ')f(t/Ez/V/_) In the discussion that follows, we restrict to the case of radial
if e > 0, then (g*DJ)(u) -- (5fdV)g(u) as E 0; we take the functions [functions on Cn,Hn,hn which are functions of
view thatf on Hn has contracted to the function c = ffdV on nl2,(t, lz 12),(X,2 Ial Xl ), respectively], though almost every-
the zero-dimensional space, the origin. On the F.T. side, one thing extends to general functions. Let So = {radial functions
dually finds g*Dsj(X,x) cg(Xx) as e 0. [Note in SI?.
DJ(X,2aI XIw) = f(XE,2a AI Je,@). ] The canonical example of a function g C eV(Hn) with g =
Matters are more interesting if we also dilate g in the z R# for some R C SV(Cn), is the kernel of the heat equation.
variables: we set MZg(t,z') = g(t,z'/1V'), and study Mzg*DJ Thus, if Rh = e-hl 1712 for each h > 0, and gh = Rh#, then (d/
as e -O 0 suitably. We will be contracting to Cn, passing through cdh)gh = -.Logh and ghEC eV (Hn) for each h by Theorem 1.
Lie groups which are isomorphic to Hn. Thus, we let HEn de- That ghC eV (Hn) can also be verified by computing the F.T.
note the Lie group with manifold R X Cn and multiplication of gh in the t variable.
(tZE)-(t',z,') = (t + t' + 2cImz,-z'Ez, + ZE,). The change of The importance of Theorem 1 is that it gives us a simple way
variables (t,z) = (t,v'cz) gives an isomorphism between H" of picking out an element of eV(hn) (namely, R#), which con-
and HE". We let fE(t,zE) = (I/e)f(t/E,z,), and let *,*E,*o denote tracts to a given element R of S(Cn).
Mathematics: Geller Proc. Natl. Acad. Sci. USA 74 (1977) 1331
The main result is Theorem 2 below, which provides an as- this, one expands the general function asymptotically and es-
ymptotic expansion of functions in e?(hn) in terms of the easily timates the IIGk I .| by appropriate &(Cn) norms and error
handled functions R #. Let norms.]
Inversion. Given f C *o(Hn), fH. fdV = 1. It is required
ev'#o(hl)= hE '?(hn)lh = k=OEN xkRk # to construct a g C ejo(Hn) such that, for some E > 0,
1g(X,2a|IX 1) = 1 if (a + n/2)1AI <,E. The analogous problem
for some N E Z +,R1,... ,RN C 'O(Cn)J. on Cn is readily solved by setting 9Ig = 0/.7f where 0 is an
appropriate cutoff function. Theorem 2 allows one to analo-
Theorem 2. The following are equivalent: gously solve the problem for Hn, because the inverse of the
asymptotic series in ii has the form specified in iii.
(i) h C 4V(hn). These techniques enable one to imitate the arguments in ref
(ii) There exist functions Rk C eO(Cn) (for each k E Z+) 5 and do real-variable HP theory on Hn. As on Rn, one says that
and hj C No(hn) (for each j C Z+) such that, for each a tempered distribution f C HP(Hn) if its area integral (ap-
K C Z+: propriately defined) is in LP(Hn). We prove an analogue of
K Theorem 11 of ref. 5. The difficult part of the proof, for which
h = Z XkRk#+ AK+lhK- we use the techniques we have sketched, in imitation of the
k=O Euclidean case, is the following. Say p > 0, f C 8e'(Hn) and /i
(iii) For each N E Z +, h = GN + EN, where GN C eV#o(hn) C o(Hn) with .fHnk(v)dv = 1. Suppose that
and EN C 60N.
W0N is an error class, which we will define presently. For the u*(v) = sup Jf*Dq(w)I C LP.
Iw'lvl2<r
proof of i = ii one uses Theorem 1: say that h contracts to R C Then
NCn9), h = g, R# = I. Then g-J contacts to zero, so f ,(g -
f)(t,z)dt = 0 for each z. Then g - f = Tgj where gl(t,z) = f*(v) = sup sup If*D,4(w)I C LP
-Jt (g-f)(t,z)dt = S (g-f)(t,z)dt C V(Hn). This proves D-&A lw-lv12<r
Downloaded from https://www.pnas.org by UNIVERSITEITSBIBLIOTHEEK GENT on May 14, 2024 from IP address 157.193.164.114.

ii for K = 0; the general case is proven at once by induction. The where .A = (4 C R1 114'11 = 1,... ,Nol. [For (t,z) C Hn,
k < 1, k
expansion in ii is seen to be unique. we write I (t,z)I = (t2 + IZ 1 4)1/4. No depends only on p and
ii * iii follows from the fact that XK+ Igo C 60K+ 1. iii n. The conditions are equivalent to: f C HP(Hn).] We then
i is technical. derive an analogue of Theorem 12 of ref. 5, that PV distribu-
Via Theorem 2, we see how to understand the behavior of tions map HI(Hn) to HI(Hn).
h C No(hn) near X = 0, which is critical since it, in part, reflects As a separate application of the exact formulas, we are also
the decrease of g at co (where A= h). able to give a formula for the F.T.s of a wide class of regular
Write I 1 Ilkj(k C Z+) for the norms that define the topol- homogeneous distributions. We give a full classification by
ogy of eV(hn) _ e(Hn). For any k C Z+, there exists K > 0 such asymptotics. A number of earlier formulas, notably those of
that if L > K, then for any h C (h n),Ih IIk can be estimated Folland and Stein (6), are special cases. Certain kernels on the
by the e?(Cn) norms of the Rj(I < j < L) and the 60L+1 norm generalized upper half plane can also be dealt with; Whittaker
of XL+ 'hL. (The Rj and hL are as in ii.) This is the importance functions play a role here, as in Greiner and Stein (7).
of eoN-the error norm is very easy to handle.
To define eoN, one first sets I would like to thank my thesis advisor, E. M. Stein, for his advice
and encouragement. This work was supported by a Sloan Foundation
0 N = Ih radial on hn Ih(X,2a X) Grant to the Courant Institute.
< CkjXjN[(21a| + n)IXI] k
1. Miller, W., Jr. (1968) Lie Theory and Special Functions (Academic
for each k C Z+}. Set @o = @oA; clearly eoN c @o for all N. @0 Press, New York).
is important because the proof of the inversion theorem shows 2. Peetre, J. (1972) "The Weyl transform and Laguerre polynomials,"
that @o c L2(hn), and if I = h E @o, then f satisfies the con- Le Mathematiche, Universita di Catania Seminario 27, 301-
clusion of the inversion theorem. Set 323.
3. Vilenkin, N., Jr. (1968) "Laguerre polynomials, Whittaker func-
6'N = JE radial I (d/dX)kE C eoN-2k whenever 2k < N1. tions and representations of the group of bounded matrices," Mat.
Sbornik 75,432-444.
The rather trivial behavior of G0N functions near X = 0, and 4. Inonu, E. & Wigner, E. P. (1953) "On the contraction of groups
their differentiability, guarantee some order of decrease at o and their representations," Proc. Natl. Acad. Sci. USA 39,510-
of their inverse F.T.s. 524.
Here are some applications of Theorem 2. 5. Fefferman, C. & Stein, E. M. (1972) "HP spaces of several bari-
Partitions of Unity. Given radial functions Rk(77) & C,(Cn) ables." Acta Math. 129, 137-193.
for k C Z+, such that, for each iq C Cn, 2;k=ORk(q) = 1 and only 6. Folland, G. B. & Stein, E. M. (1974) "Estimates for the 5b complex
and analysis on the Heisenberg group," Commun. Pure Appl.
finitely many Rk(fl) # 0. Then the Rk# provide a convenient Math. 27,429-522.
partition of unity on hn. If the Rk are chosen properly, the Rk# 7. Greiner, P. & Stein, E. M. (1975) "A parametrix for the a-Neumann
enable one to cut up (by "multiplication", i.e., convolution on problem," Symp. Rencontre Sur L'Analyse Complexe d Plusieurs
the Hn side) a general function G in eV(hn) into a sequence of Variables et les Equations Indetermines (University of Montreal
functions Gk which decrease rapidly in 1 Ili for all j. [To prove Press, Montreal).

You might also like