Skip to main content

Full text of "On integrals of eigenfunctions over geodesics"

See other formats


ON INTEGRALS OF EIGENFUNCTIONS OVER GEODESICS 



XUEHUA CHEN AND CHRISTOPHER D. SOGGE 



Abstract. If (M,g) is a compact Riemannian surface then the integrals of L'^(M)- 
normalized eigenfunctions Cj over geodesic segments of fixed length are uniformly 
bounded. Also, if (M, g) has negative curvature and 7(t) is a geodesic parameterized 
by arc length, the measures ej{'y{t)) dt on R tend to zero in the sense of distributions 
as the eigenvalue \j — >■ oo, and so integrals of eigenfunctions over periodic geodesies 
tend to zero 
latter result. 



1. General results. 

If M is a compact hyperbolic surface Good [5] and Hejhal [7], using Kuznecov formulae, 
showed that if 7 is a periodic geodesic and ds is the associated arc length measure then 



(1.1) / , 

J J 



e\ ds 



with ex denoting the L^-normalized eigenfunctions on M, i.e., —Age\ = X^e\, and 

l|eA||L2(M) = 1- 

This result was generahzed by Zelditch |14) who showed, among many other things, 
that given any n-dimensional compact Riemannian manifold, one has uniform bounds for 
integrals of eigenfunctions over closed hypersurfacesQ Moreover, if Xj are the eigenvalues 
of y^—Ag and 0^(7) denotes the integral in (jl.ip with X — Xj, then [HI Lemma 3.1] says 
that J2xj<xWjil)\^ = A + 0(1), which implies (jl.ip . Note that since, by the Weyl 
law ; Xj < A} sa A^, Zelditch's formula says that most of the aj{'-f) are much smaller 
than 1. 

Reznikov [8 discussed this problem and, moreover, initiated work on the related prob- 
lem of obtaining restriction estimates for geodesies. The sharp estimates for general 
Riemannian surfaces were obtained by Burq, Gerard and Tzvetkov |^ . If 77 denotes the 
space of unit length geodesies in a two-dimensional compact Riemannian manifold, then 
one of the results in is that we have bounds of the form 

\ 1/2 

|2 - ^ ' 



(1-2) [J \exrdsj < A4||e^||^.(^,,), e 77. 

This estimate is sharp since it is saturated by the highest weight spherical harmonics on 
S^. Recently, improvements under the assumption of nonpositive curvature have been 
obtained by Sogge and Zelditch 13: and Chen and Sogge 4 . Work showing how these 
restriction estimates are related to L^(M) estimates for eigenfunctions is in Bourgain [2] 
and Sogge [TI] . 



The authors were supported in part by the NSF grant DMS-1069175 and the Simons Foundation. 
"'^We are grateful to Steve Zelditch for helpful comments and referring us to the work based on the 
Kuznecov formula. 



2 



XUEHUA CHEN AND CHRISTOPHER D. SOGGE 



Returning to note that the estimate cannot be improved when the compact 

hyperbohc surface M is replaced by the standard two-sphere 5^ or two-torus T^. For on 
zonal functions of even order saturate the bound, while for every periodic geodesic 
on one can find a sequence of eigenvalues Xj — >■ oo and corresponding L^-normalized 
eigenfunctions e\. having the property that e\. = 1 on 7. 

We shall start by giving a quick proof of a result in [Tl] saying that we have the analog 
of p.ip for all geodesic segments in any Riemannian surface. The proof will serve as a 
template for the improvements in the next section of the bounds in (jl.ip for Riemannian 
surfaces of negative curvature which appear to be new. 

Theorem 1.1. Let {M,g) be a two-dimensional compact Riemannian manifold. Then 
there is a constant C = C{M,g) so that 



(1.3) 



e\ ds 



< C'||eA|U2(M), J en. 



Proof. Fix an even function p e S{M.) satisfying p(0) = 1 and p{t) — 0, \t\ > 1/4, 
assuming, as we may, that the injectivity radius of {M,g) is ten or more. Then since 
p{X — ^y—Ag))ex = ex, in order to prove (11.21) . it suffices to show that 

(1.4) 



p(A-y^)/ds <C||/|U2(M), J en. 

Let 7(t), < t < 1, be a parameterization of 7 by arc length and 

p{X- y/-Ag){x,y) = ^p(A- Aj)ej(x)ej(y) 

denote the kernel of the operator in (|1.4|) . Here, {ej} is an orthonormal basis of eigen- 
functions with eigenvalues {Xj}. 

By Schwarz's inequality, we would have (|1.4p if we could show that 

'dVg{y)<C. 



JmUo 



By orthogonality, if x(r) = (p(r))^, this is equivalent to showing that 



(1.5) 



/ / ^X(A-A,)e,(7(t))ej(7(s))rfids 
Jo Jo 



< C. 



Next, we note that the proof of [TOl Lemma 5.1.3] shows that if dg denotes the Riemannian 
distance then we can write 

(1.6) J2 - A,)e,(x)i;M = A^ J2 «±(^; ^si^^ y))e±^^'^«(-'^) + 0(1), 

J ± 
where for every fixed j = 0,1,2,... we have the uniform bounds 

d^ 



dr^ 



a±{X;r) 



if r> X- 



(1.7) 

and 

(1.8) |a±(A,r)| < A5, if re[Q,X-^]. 

To obtain (|1.6p - ()1.8p . as in [101 §5-1] one uses Hormander's parametrix for the half-wave 



operators e* 



-A, 



, as well as the fact that x(t) =0 for \t\ > 1 and our assumption that 



the injectivity radius of (M, g) is ten or more. 



ON INTEGRALS OF EIGENFUNCTIONS OVER GEODESICS 



3 



Since dg{'j{t),j{s)) — \t ~ s\, we conclude that we would have (ll.Sp if 



(1.9) A5 



f f e±*^l*-"la±(A;|t-s|)dtds 
^0 Jo 



< C. 



Since this is a trivial consequence of (|1.7p and (jl.Sp . the proof is complete. □ 

Reznikov in ^ and [9] also discusses the problem of integrals of eigenfunctions over 
geodesies circles in compact hyperbolic surfaces and obtains the analog of (|l.ip for them. 
More general results were obtained earlier by Zelditch Corollary 3.3], and the proof 
of Theorem 1 1.1 1 can also be used to obtain these special cases of the latter: 



Theorem 1.2. Let {M,g) be a compact two-dimensional Riemannian manifold. Then if 
a is a unit-length curve in M there is a constant so that 



(1-10) / exds <Ca\\ex\\L^(M)- 

Similar bounds with a uniform constant hold for small smooth perturbations of a. 

2. Improved results for negative curvature. 

We conclude our note by showing that we can improve the bounds in (jl.ip if we assume 
that the curvature of {M,g) is strictly negative. As noted before, this assumption is 
necessary since the corresponding result is false for the two-sphere and the two-torus. 
We shall now assume that j{t), i G M, is a geodesic in E parameterized by arc length, 
and our main result is the following 

Theorem 2.1. Let {M,g) be a negatively curved compact 2-dimensional Riemannian 
manifold. Then the measures {sjilit)) dt} on M. go to zero in the sense of distributions, 
by which we mean that if b £ (M) then 



(2.1) j b{t) e.j{-i{t)) dt 0, as j ^ oo. 

Consequently, if ^ per is a periodic geodesic of minimal period £ > 0, we have 

'-j{lper{t)) dt — > 0, as j — > oo. 



The second part of the lemma follows from the first part via a partition of unity 
argument. The proof of ()2.1|) shares some similarities with the related LP{'-f) restriction 
estimates for eigenfunctions of Sogge and Zelditch [13] and Chen and Sogge |4j. In 
particular, the oscillatory integral arguments and simple geometric facts that we shall 
employ are very similar to those in |4]. 

To prove (|2.1|) we may assume that the injectivity radius of (M, g) is ten or more and 
that 

supp b C [-i, \]. 

Next, we notice that if, as before, p G 5(K) is even and satisfies p(0) = 1 and p{t) = 0, 
1^1 > 1/4, then given T ^ 1 we have p{T{\ ~ Ag)))eA — e\. As a result, in order to 
prove (|2.ip it suffices to verify that if T S> 1 then 

(2.2) /fe(s)(p(T(A-y^))/)(7(s))ds < (CT-UCtA-i)||/|U2(m), 



4 



XUEHUA CHEN AND CHRISTOPHER D. SOGGE 



where C is independent of T 3> 1, but not Ct- Repeating the argument which showed 
how p.5p imphes (11.41) . we conclude that if x(r) — (/9(t))^ and 

then we would have (|2.2I) if we could show that 



(2.3) I j I h{t,s)Y,x{T{\~\j))eMt))^3{l{s))dtds 



< CT-^ + Ct^ 



Note that 



5^x(T(A-A,))e,(x)e,(j/) = ^ j x{r/T) 



e ^^^e'^V-'^s dr. 



If we pick a bump function (3 G (R) satisfying 

(3(t) = 1, |r| < 3, and /3(r) = 0, |r| > 4, 



the proof of (|1.5|) shows that if denotes the inverse Fourier transform of t 
/?(t)x(t/T), then 



1 



27rT 



hit, s) /3(T)x(T/r)e-*^^(e'V-^.) ^(s)) drcitds 



Here, (6*"^^ » ) (x, y) = ^ ■ e*"^"^^ ej (x)ej (y) denotes the kernel of the half- wave operator 



Based on the preceding inequality, in order to prove (j2.3p . it suffices to show that 



(2.4) 



1 

2^ 



6(i,s)(l-/?(r))x(r/r)e 



-ir\ ( At -y./ ~ IS.C 



)ilit),j{s))dTdtds 



< 1 + CtX~ 



We shall need to use the fact that since x = (27r) * p, we have 



(2.5) 



x(t)=0, |t|>-. 



which means that the r integrand in the left side of (|2.4p vanishes when |t| > T/A. 
We can make one more easy reduction. If $t denotes the inverse Fourier transform of 
T ^ (1 - /3(t))x(t/T) then $t € S{R) and consequently 



5] $t(A + A,) e,(7(i))e,-(7(5)) = Ot,w((1 + A)" 



for any N = 1, 2, 3, ... . Thus, by Eulcr's formula and (|2.5p . in order to prove (|2.4p . it 
suffices to show that 



(2.6) 



T/2 



T/2 



5(f, s) (1 - /3(T))x(T/T)e-'^^ (cos rV^) (7(i) ,7(5)) drdids 



< 1 + CtA"2. 



ON INTEGRALS OF EIGENFUNCTIONS OVER GEODESICS 



5 



Here (cosT^-Ag)(a;, y) is the kernel for the map C°°{M) 9 / -> w e C°°(Rx Af), where 
u(t, x) is the solution of the Cauchy problem with initial data (/, 0), i.e. 

(2.7) (df - Ag)u = 0, u(0, • ) = /, dMO, • ) = 0. 

To be able to compute the integral in (|2.6I) we need to relate this wave kernel to the 
corresponding one in the universal cover for (M, g). Recall that by a theorem of Hadamard 
(see Chapter 7]) for every point P € M, the exponential map at P, expp : TpAI — )■ M 
is a covering map. We might as well take P = 7(0) to be the midpoint of the geodesic 
segment{7(f) : \t\ < i}. If we identify TpM with R^, and let k denote this exponential 
map then k : — M is a covering map. We also will denote by g the metric on 
which is the puUback via k of the the metric g on M. Also, let F denote the group of 
deck transformations, which are the diffeomorphisms a from M.'^ to itself preserving k, 
i.e., K = Ko a. Next, let 

Dotr {y e : dg{0,y) < d~g{0,a{y)), Va G T, Identity} 

be the Dirichlet domain for (M^,g), where dg{- , •) denotes the Riemannian distance 
function for corresponding to the metric g. We can then add to Duir a subset of 
dDoir = -DDir\Int {D Dir) to obtain a natural fundamental domain D, which has the 
property that is the disjoint union of the a{D) as a ranges over F and S : 
dg{0,y) < 10} C D since we are assuming that the injectivity radius of {M,g) is more 
than ten. It then follows that we can identify every point x € M with the unique point 
X € D having the property that k{x) = x. Let also j{t), \t\ < i similarly denote 
those points in D corresponding to our geodesic segment j{t), \t\ < ^ in M. Then 
{l{t) '■ \t\ ^ 5} is a line segment of unit length whose midpoint is the origin, and we shall 
denote just by 7 the line through the origin containing this segment. Note that 7 then 
is a geodesic in for the metric g, and the Riemannian distance between two points 
on 7 agrees with their Euclidean distance. Finally, if Ag denotes the Laplace-Beltrami 
operator associated to g then since solutions of the Cauchy problem (|2.7p correspond 
exactly to periodic (i.e. F-invariant) solutions of the corresponding Cauchy problem 
associated to df — Ag, we have the following important formula relating the wave kernel 
on {M, g) to the one for the universal cover (R^ , g) : 

(2.8) (cos T^/^A^ (x, y)^Yl (^^^ ^V^) "(^))- 

qGE 

The simple geometric facts that we require is in the following variation of [U Lemma 3.2] 

Lemma 2.2. Let 71 (t) and 72(3) he two distinct geodesies in (R^,^) each parameterized 
by arc length. Put 

(l){t,s) = dg(7i(t),72(s)). 
Then if there is a point {to,so) G R x R such that i9j0(to,so) = ds4>{to, sq) — and 
71 (^o) 7^ 72 (so), then 

(2.9) \dscj){t,s)\ + \dt(bit,s)\^0 if ^i{t)^^2{s) and {t,s)^ito,so), 
and 

(2.10) dtdM,so)^0. 

Proof. To TO'Ove (12.91) we first note that if n-i (tn) ^ 72(50) then |9t0(io, So)| + |3s0(io, so)| — 
if and only if the geodesic connecting the points 71(^0) and 72(50) is perpendicular to 
both 71 and 72 at the unique intersection points. Since (R^,g) has negative curvature 



6 



XUEHUA CHEN AND CHRISTOPHER D. SOGGE 



there cannot be another point (ti,si) G M x M with this property. For if ^ ti and 
So 7^ si the geodesic quadrilateral with vertices 71(^0): 7i(ii), 72(so) and 72(51) would 
have total angle 2tt, which is impossible due to the fact that is negatively curved, 

and similarly if to — ti but sq 7^ si then the geodesic triangle with vertices 71 (to), 72(^0) 
and 72(51) would have total angle of more than tt, which is also impossible. 

To prove (j2.10p we may assume that = Q and work in geodesic normal coordi- 
nates vanishing at 71 (0) so that 71 is the xi-axis, i.e., 71 (t) = (t, 0). Then if 72(3) = 
{xi{s),X2{s)) ^ (0,0), 

5^ ^ -xi{s) 

Our assumption that |j^(0,so) = and 71 (0) 7^ 72(^0) means that xi(so) = and 
2^2(30) 7^ 0. In our coordinates the geodesic connecting (0,0) = 71 (0) and 72(^0) is the 
a;2-axis, and for it to be orthogonal to 72 at 72(50)1 we must have 2:2(50) = and so 
x[{so) ^ 0. But then 

^ ^i(go) / n 
— — - 0,so = -] — 1—^, + 0, 
osot F2(so)| 

which is ((2I0I) . □ 

We also need the following simple stationary phase lemma. 
Lemma 2.3. Let a G C^(R^) and assume that (f> e C°^{M?) is real. Put 

/(A) = // e'^'f'^^''^a{t,s)dtds, A > 1. 



Then 

(2.11) |/(A)| < CA-i if Vt,,0(i, s) ^ 0, {t, s) G supp a. 

Also, if there is a unique point (toi^o) G supp a at which V t,s4'{toT ^a) = and if 
'Qt§s{^Oi So) 7^ then given any e > there is a constant so that 

(2.12) \I{X)\<e\-^ +Ce\'^. 

Proof. The first assertion, (|2.11l) . just follows via integration by parts. To prove (|2.12p 
we assume that there is a (^oi^o) G supp a at which V(/) vanishes but V0(t, s) 7^ 0, 
(t, s) G supp a\{(io5'So)} and dtds(l>{to, sq) ^ 0. We can split matters into two further 
cases: (i) df(l){to,so) = and (ii) df(f>{to,so) 7^ 0. 

In case (i), we note that our assumptions mean that at (ioi^o) the mixed Hessian of 
(j) satisfies 



and, therefore, by two-dimensional stationary phase we have |/(A)| < CX~^. 
To finish, it suffices to show that we have (|2.12l) under the assumption that 
\7<j){t, s) 7^ 0, {t, s) G supp a\{{to, So)}, and d^cpito, sq) 7^ 0. 
If we let [3 be as above, it follows from (|2.1ip that given any fixed e > 0, we have 

(2.13) /i(A)= [[ e'^''''-*-'\l-l3{e''\t-to))l3{e~\s-so)))a{t,s)dtds = Oe(A-i). 



ON INTEGRALS OF EIGENFUNCTIONS OVER GEODESICS 



7 



Furthermore, if e > is chosen so that ^ on supp /3{e - — tQ))fi{e - — so)): it 
follows from one-dimensional stationary phase that for each fixed < s < 1 we have 

/•OO 

P{e~\s - So)) / e'^*^''-'^ (3{e-\t - to)) a{t, s) dt 



where C is independent of e > (cf. the proof of TU', Theorem 1.1.1]). This clearly 
implies that we have the uniform bounds |/(A) — /i(A)| < AeA^a for all small e > 0. By 
combining this inequality with (j2.13p we deduce that (|2.12p holds in this case as well, 
which finishes the proof. □ 

To use Lemmas 12.2112.31 we also require another result which is essentially Lemma 3.1 
in a. 

Lemma 2.4. Given a G T set 

(2.14) Kl^Jt,s) 
_ 1 

TT 



(l-/3(r))x(r/T)e-'^-(cosry^)(7(t),a(7(s)))dr, \t\,\s\ < 1/2. 



Then if a ^ Identity and we set 

(2.15) </)^,„(t, s) = dg{j{t), a(7(s))), \t\, \s\ < 1/2, 
then we can write for \t\, \s\ < 1/2 

(2.16) Kl^^Jt,s) ^ x'^wm),a{%s)))Y,a±iT,X; 0^,o(i, s))e±^^'^-"(*-^) 

± 

where w{x, y) is a smooth bounded function on M2 X M2 and where for each j = 0, 1, 2, . . . 
there is a constant Cj independent of T, X > 1 so that 

(2.17) \dia±{T,X;r)\<Cjr-i-\ r > 1, 
and for a constant Ct which is independent of 7, a and X 

(2.18) \Rl^Jt,s)\<CTX-\ 

We shall postpone the proof of this result until the end and use it now, along with 
Lemmas 12.21 and 12. 3[ to prove (12. 6p . Recall that 



(2.19) (cosTv/^)(£,y) =0 if dg(i,y) > |r|. 



and (cos r-^/— Ag) (x, y) is smooth if dg{x,y) 7^ |r|. 

Therefore since I3(t) = 1 for |t| < 3 we have (1 - /3(r))(cos ryA^)(7(t), 7(3)) € C°° if 
< s,t < 1, and so 

(1 - P{T))x{r/T)e-^^^{cosry^,)m,%s))dT = Ot(A-^), 

-T/2 

for any = 1,2,3, .. .. 



8 



XUEHUA CHEN AND CHRISTOPHER D. SOGGE 



As a result, if we use 
that 



we conclude that we would have 



if we could show 



(2.20) 



E 

iGT\Identity 



T/2 



-T/2 



b{t,s)il-l3iT))xir/T)e-'^\cosr^g){^{t),a{%s)))dTdtds 



< 1 + CtA"2. 



By (|2.19p there are are only finitely many nonzero summands here (actually 0(exp(cT)) 
ones). Consequently, we would have (|2.20p if we could show that given e > and 
Identity =/= a T we have 



(2.21) 



T/2 



T/2 



b{t, s){l - /?(r))x(T/r)e-'^^ (cos r ^Aj) {j{t), a{j{s))) drdtds 



Note that since 7 is a geodesic in (R^,7) and a is an isometry, it follows that a{'^) 
is also a geodesic. It is a geodesic which is different from 7 if a is not in the stabilizer 
subgroup of r of all deck transformations preserving 7. If our geodesic 7 in (Af , g) is 
not a periodic geodesic then the stabilizer subgroup is just the identity element. If 7 is 
a periodic geodesic in M of minimal period £ > 0, then we must have i > 10, because of 
our assumption regarding the injectivity radius. In this case, every nontrivial element of 
the stabilizer group satisfies a{'j{s)) — 7(5 + k£) for some k E Z\0. By Lemma [2.41 for 
such a a with fc 7^ 0, modulo a term which is Oq,(A^^) we have that the left side of (|2.21|) 
is equal to the sum over ± of 



b{t, s) w{^{t), 7(s + k£)) a±{T, A; |t - s - k(i\) e 



±i\\t-s-U\ 



dtds 



with a± as in p.l7|) . Since b e C(j"(R^) vanishes when \t\ or \s\ is larger than i and 
w € C°°(]R^), by a simple integration by parts argument this term is Oa(A~2)^ which 
means that we have (|2.2ip for all nontrivial elements of the stabilizer group of 7. 

To prove that we also have (j2.21l) for the remaining case where a G F is not in the 
stabilizer group, by the above, it is enough to show that, if 4>^_a{t, s) is as in (j2.15p . then 
given e > there is a constant Ca,e so that 



(2.22) 



b{t, s) w{j{t), a{j{s))) a±{T, A; 0^,„(t, s)) e±^^'^^.°(*'^) dtds 



< £A^5 + Ca eA" 



By Lemma 12.21 the phase function here either satisfies Vt.s4'j.a 7^ on the support of 
the integrand, or there is a unique point (to, sq) in the support where Vt.s4'i.a vanishes 
and at that point (to, 5o) 7^ 0. In the former case by (|2.1ip the left side of (I2.22p 
Oa(A~^), which is more than is required. In the latter case, we obtain the estimate from 
Lemma [^751 Thus, we have established (|2.22l) . which except for the proof of Lemma 
completes the proof of Theorem 12.11 

Proof 0/ Lemma \2.4\ Since the injectivity radius of {M,g) is 10 or more, it follows that 
dg{'l{t),ct{'j{s))) > 10 if \t\,\s\ < 1/2. Also, as noted before, for each T, by Huygens 
principle, there are only finitely many terms in (j2.14p that we must consider. 



ON INTEGRALS OF EIGENFUNCTIONS OVER GEODESICS 

Next, let for x £ M^, |a;| > 1, 

Koi\^\) = - / r Xir/T)e'^^ cos(T|e|)e"-« drd^ 

TxinX - |^|))e"-« f Tx{T{\ + |C|))e"-« d^. 



Also let $t(C) G 5(M) be defined by the Fourier transform $t(t) = (i(T)x{T/T) and put 
^i(kl) = - / r mx{r/T)e^^^ cos(rlel) drd^ 



= / $T(A-|C|)e"-«df+ / $T(A+|e|)e*"-«de 
Recall that the Fourier transform of Lebesgue measure on the circle is of the form 
Te{y)^ [ e^«-(™^^'^-^)d0=|yr^ Va±(|y|)e±'l^l, |y| > 1, 

where for every j = 0, 1, 2, . . . we have 

\dla±{r)\<Cjr-\ r > 1. 

Also M e C°°(]R). Therefore, if A,r > 1, modulo a term which is 0((A|^|)-^) for any 
N independent of T, we have 

(2.23) Ka{\x\) = \x\'^- 



nOO 

5 / rx(r(A-r)) ^a±(|x|r)e±*''l^lr^dr 
= A^|xr^^6±(T,A;|x|)e±*^l-l, 



± 

where one easily sees that 

(2.24) \dib±{T,\-r)\<C,r-\ r. A, T > 1, j = 0, 1, 2 . . . . 
Similar arguments show that, modulo an 0{{\\x\)^^) error we also have 

(2.25) K^{\x\) = 1:^1-5 ^6±(T, A; |x|)e±^^l-l, 

± 

where h± satisfy the bounds in p.24p . 

To use this we shall use the Hadamard parametrix (see [H]). For x,?/ e we can 
write 

(2.26) {cosTy^g){x,y) = {2Tr)-^w{x,y) [ e'^^^^^'^^^i cos(t|$|) 



V / e^'^^(^'«)«^e±-l«la±(r,a;,2/,|^|)d^ + i?(r,x,y). 



± 

where we can take the remainder to satisfy 

\R{T,x,y)\ + \drR{T,x,y)\ < Ct, if |r| < T, 
and a± is a symbol of order —2 which, in particular, satisfies 
(2.27) |a^a±(r,x,y,|^|)|<CT,,(l + |C|)-^ if |r| < 2T, j = 0, 1, 2, . . . , 

and where the leading coefficient w is smooth, nonnegative and satisfies 

w{x,y) < C 



10 



XUEHUA CHEN AND CHRISTOPHER D. SOGGE 



independent of x, y € M^, by volume comparison theorems (see [T], |13|). 

Clearly if we replace (cosT-y/A^)(7(t), a(7(s))) by i?(r, 7(t), 0(7(5))) in (|2.14p we can 
integrate by parts in r to see that the resulting expression satisfies the bounds in (|2.18|) . 

If we take x — j{t) and y — 0(7(5)), \t\, \s\ < 1/2 for the first term in the right side of 
(|2.26p and replace the cosine-transform kernel in (|2.14l) by this expression then we will 
exactly obtain {2tt)^ times 

w{j{t), a{j{s)))Ko{cb^At, s)) - w{^{t), a{j{s)))Ki{cj>^^o.{t, s)), 

which can be taken as the first term in the right side of (|2.16p . with (|2.17p being valid. 
Finally since (|2.27p holds, the proof of this last assertion also shows that if in (|2.14[) 
we replace the cosine-transform kernel by the second term in the right side of (I2.26|) 
we obtain another term satisfying the bounds in (|2.18p (in fact it is Ot{^~^)), which 
completes the proof. □ 

References 

[1] p. H. Berard, On the wave equation on a compact manifold without conjugate points, Math. Z. 
155 (1977), 249-276. 

[2] J. Bourgain, Geodesic restrictions and -estimates for eigenf unctions of Riemannian surfaces, 
Linear and complex analysis, 27—35, Amer. Math. Soc. Tranl. Ser. 2, 226, Amer. Math. Soc, 
Providence, RI, 2009. 

[3] N. Burq, P. Gerard and N. Tzvetkov, Restriction of the Laplace- Beltrami eigenf unctions to sub- 
manifolds, Duke Math. J. 138 (2007), 445-486. 

[4] X. Chen and C. D. Sogge, A few endpoint geodesic restriction estimates for eigenf unctions, 
arXiv:1210.7520 

[5] M. Do Carmo, Riemannian geometry, Birkhauser, Basel, Boston, Berlin, 1992. 

[6] A. Good, Local analysis of Selberg's trace formula. Springer Lecture Notes 1040 (1983). 

[7] D. Hejhal, Sur certaines series de Dirichlet associees aux geodesiques fermees d'une surface de 
Riemann compacte, 0. R. Acad. Sci. Paris 294 (1982), 273-276. 

[8] A. Reznikov, Norms of geodesic restrictions for eigenfunctions on hyperbolic surfaces and repre- 
sentation theory, arXiv:math.AP/0403437 

[9] A. Reznikov, A uniform bound for geodesic periods of eigenfunctions on hyperbolic surfaces, Forum 
Math., to appear. 

[10] C. D. Sogge, Fourier integrals in classical analysis, Cambridge Tracts in Mathematics 105, Cam- 
bridge University Press, Cambridge, 1993. 

[11] C. D. Sogge, Kakeya-Nikodym averages and LP-norms of eigenfunctions, Tohoku Math. J. 63 
(2011), 519-538. 

[12] C. D. Sogge, Hangzhou lectures on eigenfunctions of the Laplacian, to appear in Annals of Math 
Studies, Princeton Univ. Press. 

[13] C. D. Sogge and S. Zelditch, On eigenfunction restriction estimates and L'^-bounds for compact 
surfaces with nonpositive curvature, arXiv;1108.2726 

[14] S. Zelditch, Kuznecov sum formulae and Szego limit formulae on manifolds. Comm. Partial Dif- 
ferential Equations 17 (1992), 221-260. 

Department of Mathematics, Johns Hopkins University, Baltimore, MD 21218