J OURNÉES É QUATIONS AUX DÉRIVÉES PARTIELLES

S HMUEL AGMON
On the spectral theory of the laplacian on noncompact
hyperbolic manifolds
Journées Équations aux dérivées partielles (1987), p. 1-16
<http://www.numdam.org/item?id=JEDP_1987____A17_0>

© Journées Équations aux dérivées partielles, 1987, tous droits réservés.

L’accès aux archives de la revue « Journées Équations aux dérivées partielles » (http://www.
math.sciences.univ-nantes.fr/edpa/) implique l’accord avec les conditions générales d’utili-
sation (http://www.numdam.org/conditions). Toute utilisation commerciale ou impression sys-
tématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fi-
chier doit contenir la présente mention de copyright.

Article numérisé dans le cadre du programme

Numérisation de documents anciens mathématiques
http://www.numdam.org/
 XVII-1

-1-

On the Spectral Theory of the Laplacian

on non-compact Hyperbolic Manifolds

Shmuel Agmon

The Hebrew University of Jerusalem

and The University of Virginia

In this lecture we shall discuss some problems which arise in the spectral theory of

the Laplacian on non-compact hyperbolic manifolds obtained as the quotient of hyper-

bolic n-space by discrete groups of hyperbolic isometrics. In particular we shall be

interested in objects such as the resolvent kernel, the generalized eigenfunctions and

the scattering matrix. The relations that exist among these objects are of a considerable

interest and are basic in the study of the continuous spectrum of the Laplacian.

The problems we discuss here have their counterparts in Euclidean space, in the

theory of SchrSdinger operators on :K'1. In some respects the spectral theory of the

Laplacian on hyperbolic manifolds is simpler and more complete than the corresponding

theory for general Schr55dinger operators. In fact, the theory we discuss here should

be compared with the theory of Schrodinger operators - A + V on tf?'1 for potentials

V having a compact support.

 XVI I-2

1. The Laplacian on the hyperbolic space sB"

We take as a model of the hyperbolic n -space the unit ball:

5 " = { ^ e J ? " : l x l < l } with the Poincare metric:

^ 2 =4(l-l;cl 2 )- 2 1a[x:| 2 .
The volume element is

^ = 2 M ( l - l ; c l ^-"dxi • • • ( & „ .
,5" is a complete Riemannian manifold with all sectional curvatures equal -1. It has

an ideal boundary 3jB" identified with the sphere 5"1"1 = [x :xe ^n, \x I = 1). One

refers to points coe3;S" as points at infinity.

We denote by p0c,x') the geodesic distance between two points in ,8" and

set: G = ( cosh p + 1)/2. <3(x,x') is a "two point invariant" with respect to the non-

Euclidean motions. It is given by:

G(x,x')= \x-x'\\\ - Ijcl 2 )-^!- Ix'l 2 )- 1 + 1. (I.I)

The basic invariant differential operator is the Laplace-Beltrami operator A. In

cartesian coordinates;

A—
4 d-ixi^ia^^
2 d-ixi^^a,
<=i • (-1
3; = 3/3^. It will be convenient to consider the modified Laplacian:

P = -A - ("jl)2.

P is an elliptic operator, formally self-adjoint with respect to d\i. It has a unique

self-adjoint realization in L^S"; d\i) also denoted by P . It is a non-negative operator

with a purely continuous spectrum consisting of the positive real axis.

 XVII-3

We describe briefly the main mathematical objects associated with P, all of which

can be given explicitly.

/. The resolvent kernel. This is the kernel of the operator (P-^)"1 defined for

Ke (C \tf?+. We set X = k2, Im k > 0, and denote the kernel by G(x, x'\ k). One has

the following explicit expression:

G(JC,JcU)=^(^G- J F(^^-' l j 2 ;2^-(n-2);a- l ) (1.2)

_i
where s = — — - ik, c = <7(.x,jc') is given by (1.1); F(a,b\c\t) is the hyper-
2^

geometric function, and

Cn (S ) ^(2S +l)^l)/2^(, WS --^j3-).

//. The generalized eigenfunctlons. These functions, defined in B", play in hyperbolic

space the role that exponential functions play in Rn. Denoted by E (x ,0); k) they

depend on an arbitrary point coe 3 Bn and on the eigenvalue parameter k. They are

distinguished solutions of the equation: Pu = k^u in B^1. With a chosen normaliza-

tion they are given explicitly by

_ , ,. —(n—\V1/ l~ljC I vc yi""l ., ... ^.

E(x, co; k) = n (n l)/2
(———-y) 5
, s = ———ik, (1.3)
21o)-jcl 2 2

///. The scattering matrix. This is an operator valued function, denoted by S (k), defined

for ke <T. S (k) acts on functions, distributions and hyperfunctions on QBB^1. It has a

distributional kernel 5(0), co'; k), co, co' e3iB'1, given by

 XVII-4

n~l n-^-Hfc)
2
5(CO,0)U)=7l ——— 2 —————ICO-CO'I-^ 1 ^ 2 ^
ro70
S (k) is meromorphic function of k in a weak sense with poles on the positive imaginary

axis. It verifies the functional equation: S (k)S (-/:) = id.

S ( k ) is related to the generalized eigenfunctions through the functional equa-

tion:

£(x,co;/:)=5(^)£(^,-; -k).

2. The Laplacian on J^/r

We consider a Riemannian hyperbolic manifold M" of the form:

M" = ^/r (2.1)

n
where F is a discrete group of non-Euclidean isometrics acting on B . We assume that

M'1 is not compact. We also assume that r has no elements of finite order (elliptic ele-

ments). This ensures that M'1 ia a smooth Riemannian manifold. (The last assumption

may be relaxed).

We denote by Ar the Laplace-Beltrami operator onM' 1 . We shall be interested

in studying spectral properties of Ap related to the continuous spectrum. To obtain

interesting results we have to impose restrictions on the nature of the "boundary at

infinity" of M". The main restriction that we impose is that V has a geometrically

finite fundamental domain in 5?^ (i.e. a fundamental domain bounded by a finite

number of totally geodesic hypersurfaces). We shall compactify M"' by adding to it

its boundary at infinity. Roughly and informally this is done as follows. One represents

M" by a canonical fundamental domain j^in 5'1 with sides identified according to
 XVII-5

the action of r. The boundary at infinity ^Mn is then represented by the set y^y^B1^

( y denotes closure in IB n ) with the proper identifications.

The compactified manifold: Mn =Mf^^J^Mn is a real analytic Riemannian

manifold with a boundary having a finite number of singularities known as cusps. The

cusps correspond to parabolic elements of F. We shall impose on F a last restriction

that it does not contain parabolic elements. This last assumption implies that ()Mn

is a compact real analytic manifold without singularities of dimension n-1 (possibly

disconnected).

Remark: The theory we describe here can be extended to the case where M11 is

allowed to have cusps which are all of maximal rank. In particular for n = 2 the

theory can be extended to all finitely generated discrete groups F.

Some historical remarks. The interest in the spectral theory of the Laplacian

on hyperbolic manifolds of the form (2.1) arose originally in number theory, for the case

n = 2, for special groups F having a finite area fundamental domain. This followed

an observation made by Maass [11] that Eisenstein series are automorphic gen-

eralized eigenfunctions of the Laplace-Beltrami operator in the hyperbolic plane. The

work of Selberg [19] on the trace theorem gave a special impetus to the study of the

spectral properties of the Laplacian on automorphic functions in the hyperbolic plane.

Among the many important contributions to this study we mention (a partial list):

Roelcke [18], Faddeev [4], Faddeev and Pavlov [5], Elstrodt [3], Patterson [16], Fay

[6], Lax and Phillips [9], Venkov [20].

 XVII-6

Spectral properties of the Laplacian on non-compact n -dimensional hyperbolic

manifolds were studied in recent years by various authors. We mention in particular

the work of Miiller [15] dealing with spectral properties of the Laplacian on general

Riemannian manifolds with cusps, and the work of Lax and Phillips [10] dealing with

spectral properties of Ap on B"^ IV for general geometrically finite discrete groups r.

One of the interesting problems in the theory (related to Selberg trace formula) is

the problem of the meromorphic continuation of the Eisenstein series associated with

r ( = generalized eigenfunctions of Ap ). In the two dimensional case the problem

was solved by Patterson [16] and Fay [6] for finitely generated groups F having no par-

abolic elements (see also [1]). For n = 3 Mandouvalos [12] has derived the mero-

morphic continuation for a certain class of groups F. Recently Mazzeo and Melrose

[13] have shown that in any dimension n, for geometrically finite groups F with no

parabolic elements, Eisenstein series admit meromorphic continuation to the entire com-

plex plane. They derive the result as a corollary of a theorem on the meromorphic

continuation of the resolvent kernel of Ap. In the following we shall describe an

alternative solution to the problem of meromorphic continuation of Eisenstein series

which follows more closely Selberg's original ideas. This approach uses technics

which proved successful in the study of Schrodinger operators. We mention that Peter

Perry has informed us in a letter that he had obtained a proof of the Mazzeo - Mel-

rose result using similar technics.

Returning to our main discussion we consider the operator Ap on M'1 = [ B n / ^

where r verifies the restrictions imposed before. We set:

 XVII-7

n
?n r = --Ar-(-y-).
A / ~~^ \2

Prhas a unique self-adjoint realization in L^M^; d^) also denoted by Pp I1 can be

shown that its spectrum o"(Pr) = ^(^r^U^900! where o^(Pr) consists of a finite

set of negative eigenvalues. There are no eigenvalues imbedded in the positive spec-

trum ^J.[ 10]).

/^ -i
Next, we consider the resolvent operator Gy(k) = (P ? - < : ) . It follows from

standard results in elliptic theory that G^k) is an integral operator with a smooth ker-

nel G p(jc ^c'; k) for x , x / ^ M n , x ^x', for any k in the half-plane

Im k > 0, A:2^ <^(Pp). We refer to GrC^'; ^) as the resolvent kernel. It is a mero-

morphic function of k in the upper half-plane with simple poles at the points k. verify-

ing: ^/eo^Pr)-

There is an explicit series representation for G^iflm k > (^-1)/2. Identifying

M'1 with ^c^B'1 it can be shown that

Gr(x^U)= ^ G ( g x , x ^ k ) ^
^er
where G is the resolvent kernel of the modified Laplacian on fi?" given by (1.2).

Next we look for the functions which should play the role of the generalized

eigenfunctions of the Laplacian onMn. Natural candidates for such functions are the

functions defined by the series:

ey(x,w,k)^ ^E(gx,w,k) ^^
g^r
where E(x,w,k) denotes the generalized eigenfunctions of the Laplacian on /5'1,

defined by (1.3). (Here, M1^ is represented by a fundamental domain (

]pcLBn and
 XVII-8

coe 3M'1, identified with ^j3 S n ).

The series (2.3) are the Eisenstein series associated with F. They can be shown to

converge in ^B^, uniformly on compact subsets, for any k in the half-plane

Im k > (/t-l)/2 and any coe^iB n , co not in the limit set of F, to a F - automorphic solu-

tion of the differential equation: Pu = k^u in Bn.

The functions ey are essentially the desired generalized eigenfunction for k in the

half-plane Im k > (^-1)/2. However, to be useful in spectral problems one needs to

renormalize the ey for the following reason. When one considers ey(x, CO; k) as a func-

tion of (x, co) on Mnxc)Mn, one finds that e p is smooth in x on M^ but is not smooth in

CO on 3M'1. For this and other reasons we shall use another method to define the general-

ized eigenfunctions.

We start by defining for x, x 'e jB n, the function:

_jl_
OrOc.jQ^Sa^.x')-' ) n 1
(2.4)
^r
where cjOc.jc') is the pair invariant function (1.1). It can be shown that Cy is a well

defined positive, smooth, F automorphic function in x and x'. Next, considering <7p

as a function on MnxMn we fix a point XQ^Mn and set:

T(jc)=(Jr(^,^o)- (^S)
The function t(x) is a positive smooth function ofx on M". It verifies the growth

relation:

Pr(^ V ~ c < logW ^ Pr(^ ^) + c

for some constant C, p^x.Xo) denotes the geodesic distance between x and XQ on
 XVII-9

M". For any point CD€ 3M" and ^ in the half-plane Im k > Oi-l)/2, we now define:

Er(x, co; k) = lim ^xVG^x,^ k), s = -^ - ^ (2.6)

x'-xa 2

It follows from (2.2) and (1.2), and properties of the function t, that the limit (2.6)

exists for k in the half-plane Im k > (/i-l)/2. It follows further that E^(x, w, k) is a

real analytic function in (x,co) on M"x3M", analytic in k in the half-plane

Im k > (n-l)/2, satisfying the differential equation:

PyU = k^u in M". (2.7)

It can be shown that the generalized eigenfunctions Ey are renormalized Eisenstein

senes: E^(x, w, k) = w (co. A: )e r(-^, co; k) where w is a non-vanishing entire analytic

function of k.

The next step is to continue analytically the generalized eigenfunction f p beyond

the half-plane Im k > (n-l)/2. Now, using relatively simple P.D.E. estimates one can

show that the limit on the r.h.s. of (2.6) exists (uniformaly on compact subsets) for all k

in the half-plane Im k > 0, ^c^(Pp). Hence formula (2.6) gives the meromorphic

continuation of Ey{x, w, k) to the half-plane Im k > 0.

We come to a crucial intermediate step in the analytic continuation proof. It

consists in showing that both Gy and Ey possess continuous boundary values on the

real axis in the k -plane. The proof of this result requires some delicate estimates "on

the continuous spectrum" of the type used to prove the limiting absorption principle for

Schrodinger operators (c.f. [2], [8]). Such estimates in the hyperbolic case are dis-
 XVI I-10

cussed in [17].

Summing up, one finds at this stage that both G r ( x , x ' ; k ) andE^x,w,k)

exist and are analytic in the half-plane Im k > 0, continuous in Im k > 0, except for

a finite number of poles at the points [kj ] such that k^e o^(Pr). The f p satisfy (2.7)

and are related to Gr- by (2.6) (also for k real). An important point is that £:r is a

smooth (real analytic) function in (x, co) on M" x3M".

We denote by JTOM") and ^OM") the classes the hyperfunctions and distri-

butions on 3M", respectively. We shall denote by ^(M"), ^e (T, the class of solu-

tions of the differential equation: Pyu = ^ M i n M ' 1 . The generalized eigenfunctions in

'E\ is a distinguished subclass. We have the following representation theorem which

generalizes for the class of manifolds M" a theorem proved by Helgason [7] and

Minemura [14] in the case Mn = SB'1.

Theorem 1 : The functions u(x) in ^(M"), for any k in the half-plane

/s

Imk >.0,k ^ <j^ (Pr)» ar^ precisely the functions given by

u(x)= <f,E^x,^k)> (2.8)

where f e JTOM") (facts on Ey as a function of^\ Moreover, the mapping: f -> u

is a bijection of JTOM") onto £^(M'1). Also, fe ^DW) if and only if u<ix)

grows at most exponentially, i.e. if I u (x) I < C T(;c )aonMn for some constants C, a.

We shall give now some indications on the scattering matrix S^lc) which is a more

subtle mathematical object appearing in the spectral theory of the Laplacian on M'1.

S^k) is an operator valued function defined initially for k in the half-plane Im k >. 0
 XVII-11

11-

except for a discrete set of points S on the positive imaginary axis, consisting of

9 n —1
points k with k e<j^(Pr) ^d t^ set K—— + jZ^). (5r(^) acts on functions, distri-

butions and hyperfunctions on c>Mn. One finds that for any k Sy(k) is an elliptic

pseudo-differential operator on SM"' of complex order - 2ik. S]-(k) is an analytic

function of k for Im k > 0 (in a weak sense), and continuous for Im k > 0, except for

poles in the set S. The scattering matrix makes its appearance in the following

asymptotic expansion theorem for functions in ^E^M"). The theorem can be used

to give the definition of the scattering matrix.

Theorem 2: Let u(x)e ^(A^) where Im k >. 0, ^c^(Pr)* Suppose also

that 2ik ^ .Z. Then the following asymptotic formula holds:

-(nz±^) 00
2 A
u (x )-c(;c) S ^) ^ (^)/ + (2.9)
w=0

.n~l ... ^
2 f
+T(^)" " s^r^m^^r^)/
w=0

as x —> c>Mn ,for some hyperfunction f on 3M", where A^(k) and B^(k) are differen-

tial operators of order <m on 3M" with smooth (real analytic) coefficients. The opera-

tors A^ and B^ depend on F, T, k and m. A o and B o ^r^ constants given by

W-^B^--^.

Remark: The asymptotic relation (2.9) holds in some generalized sense. If

Fe COOOMn) then (2.9) holds in the usual pointwise sense.

 XVII-12

Note that for k real, taking the two dominant terms in (2.9), we have:

,n-l 4 ., . /i-l ., N
i -(—o—
2 -^) rak)7 -(—r~-^)
2 (2.10)
u(x)^(x) f -h p.^ ^) 5rW
as;c ->3M'1.

The relation between Theorem 1 and Theorem 2 is given in

Theorem 3: There exists on 3M" a measure do)' of the form: do' = \(/(co)dco,

vv/^r^ v(co) is a positive real analytic function on 3M'1 and dco ^ the induced Lebesgue

measure on 'QMn {considered as a subset ofQB'1), such that if ue rE^(Mn) has the

asymptotic expansion (2.9) withfe. ^['(9M'1), then u has the representation:

u(x) = J /(co)£:r(^ ^ k)d^ ^^

BM'1
where (2.11) is interpreted in an obvious way (by continuity) iff is not a function.

As easy applications of the proceeding theorems we derive two basic formulas.

Letk bereal,/:^0.

Consider the function:

^00=Gr(x,^U)-Gr(^';-^),

jc' being fixed. Clearly ue fEh2(Mn)- From

(^ it follows that

,n-l .,. /n-1 ., .

-(^--ik)
2
- ( —25 — + ^ ) /9 i9\
(2 12)
u(x)^(x) £rOc',co;^)-T(;0 E^(x\w^k) •

as x —> coe3M'1. Hence, applying Theorem 3, it follows from (2.11) that

Gr(x,x^k)-Gr(x,x^ -k)=ki J Er(x\w, -k)E^(x,w, k)d(^\ ^ ^

W1

Also, by comparing (2.12) and (2.10), we find that

 XVI I-13

E^(x, co; k) = - -^^SrW^x, •; - ^). (2.14)

1 ( ^}

We note that an easy application of formula (2.13) is the spectral reprsenta-

tion theorem for Ap. Another immediate application of the formula is a proof

that G^(x,x'\k) admits a meromorpic continuation to the entire complex plane

once it is shown that Ey(x, co; k) has this property.

Formula (2.14) is the functional equation for the generalized eigenfunc-

tions. An easy application of the formula is that Ey(x, co; k), and thus also the Eisen-

stein series ey(x, co; k), admit meromorphic continuations to the entire complex

plane once it is shown that S^k) has this property.

We conclude by giving some very brief indications concerning the proof that S ( k )

admits a meromorphic continuation to the entire complex plane. As explained above

this is a main (and last) step in the proof of the theorem that both the Eisenstein series

and the resolvent kernel admit such meromorphic continuations.

To derive the meromorphic continuation of S ( k ) into the lower half- plane one

starts by showing that for all real k the pseudo-differential operator Sy(k) has an

inverse satisfying the functional relation:

S^kr^SrW. (2.15)
(This follows essentially from the functional equation (2.14)).

Next one uses the fact that Sy(k) in an elliptic pseudo-differential operator of

order -2ik for any k in the half- plane Im k ^ 0, kt^L, and that Sj-(k) is a ( weakly)

analytic operator valued function in k for Im k > 0, continuous for Im k >. 0, except
 XVI I-14

for poles in £. Using the calculus of pseudo-differential operators and the analytic

Fredholm theory, it then follows by standard arguments that S^(k)~1 exists as a mero-

morphic operator valued function for 7m k > 0, Sv(k)~'1 admitting also continuous boun-

dary values on the real axis. These properties together with the functional equation (

2.15) yield the meromorpnic continuation of 5 r to the entire complex plane.

References

[1] Agmon, S., Spectral theory of Schr'odinger operators on Euclidean and

non-Eulidean spaces, Comm. Pure Appl. Math., 39, 1986, pp. S3-S16.

[2] Agmon, S. and Ho'rmander, L., Asymptotic properties of solutions of differential

equations with simple characteristics, J. Analyse Math. 30, 1976, pp. 1-38.

[3] Elstrodt, J., Die Resolvente mm Eigenwert problem der automorphen For-

men in der hyperbolischen Ebene, I, 11, III, Math. ann. 203, 1973, pp. 295-330;

Math. Z., 132, 1973 pp. 99- 134; Math. Ann. 208, 1974, pp. 99-132.

[4] Faddeev, L. D., Expansion in eigenfunctirons of the Laplace operator in the

fundamental domain of a discrete group on the Lobachevski plane, Trudy Moscow

Mat. Obsch. 17, 1967, pp. 323-350; English transl., Trans. Moscow Math. Soc.,

17, 1967,pp.357-386.

[5] Faddeev, L. D., and Pavlov, B. S., Scattering theory and automorphic functions,

Seminar of the Steklov Mathematical Institute of Leningrad 27, 1972, pp. 161-

193.
 XVII-15

[6] Fay, J. D., Fourier coefficeints of the resolvent/or a Fuchsian group, J. Fur die

Reine und Angewandte Math., 293-294, 1977, pp. 143-203.

[7] S. Helgason, Eignespaces of the Laplacian; Integral representations and

irreducibility, J. Func. Analysis, 17, 1974, pp. 328-353.

[8] Ho'rmander, L., The Anaysis of Linear Partial differential Equations II,

Springer-Verlag, New York, 1983.

[9] Lax, P., and Phillips, R. S., Scattering Theory for Automorpic Functions,

Annals of Mathematics Studies 87, Princeton University Press, Princeton, 1976.

[10] Lax, P., and Phillips, R. S., Translation representation for automorphic solu-

tions of the wave equation in non-Euclidean spaces, I, II, in, Comm. Pure Appl.

Math. 37, 1984, pp. 303-324, pp. 779-813 and 38, 1985, pp. 179-208.

[11] Maass, H., Uber eine neue Art van nichtanalytischen automorphen Funktionen

und die Bestimmung Dirichletscher Reihen durch Funktionalgleichungen, Math.

Ann.121,1949,pp.141-183.

[12] Mandouvalos, N. , The theory of Elisenstein series for Kleinian groups, Con-

temporary Mathematics, 53, 1986, pp. 357-370.

[13] Mazzeo, R. R., and Melrose, R. B., Meromorphic extension of the resolvent

on complete spaces with asymptotically constant negative curvature, M. /. T. pre-

print, 1986.

[14] Minemurea, K., Eigenfunctions of the Laplacian on a real hyerbolic space, J.

Math. Soc. Japan, ^ 1975^ ^^. 8^-1°^,

 XVII-16

[15] Muller, W., Spectral theory for Riemannian manifolds with cusps and

related trace formula, Math. Nachtr. Ill, 1983, pp. 197-288.

[16] Patterson, S. J., The Laplacian operator on a Riemann surface, I, II, III,

Comp. Math. 31, 1975, pp. 83-107; 32, 1976, pp. 71-112; 33, 1976, pp. 227-259.

[17] Perry, P. A., The Laplace operator on a hyperbolic manifold, /. Spectral and

scattering theory, Univ. of Kentucky preprint, 1986.

[18] Roelcke, W., Das Eigenwertproblem der automorphen Formen in der hyper -

bolischen Ebene, I, II, Math. ann. 167, 1966, pp. 292-337; 168, 1967, pp. 261-324.

[19] Selberg, A., Discontinuous groups and harmonic analysis, Proc. Intemat.

Congress of Mathematicians, Stockholm, 1962, pp. 177-189.

[20] Venkov, A. B., Spectral Theory of Automorphic Functions, Proc. Steklov Inst.

of Math. 153, 1981; English translation, Amer. Math. Soc., 1982.