Skip to main content

Full text of "About the blowup of quasimodes on Riemannian manifolds"

See other formats



Abstract. On any compact Riemannian manifold {M,g) of dimension n, the L^- 

normalized eigenfunctions ipx satisfy HvaIIoo < CA 2 where — Ai^;^ = X^fx- The 
bound is sharp in the class of all (M, g) since it is obtained by zonal spherical harmon- 
ics on the standard n-sphere 5". But of course, it is not sharp for many Riemannian 
manifolds, e.g. flat tori ]R"/r. We say that S", but not R^/F, is a Riemannian man- 
ifold with maximal eigenfunction growth. The problem which motivates this paper is 
to determine the (M, g) with maximal eigenfunction growth. In an earlier work, two 
of us showed that such an (M, g) must have a point x where the set Cx of geodesic 
loops at X has positive measure in S*M. We strengthen this result here by showing 
that such a manifold must have a point where the set 1^^ of recurrent directions for 
the geodesic flow through x satisfies YR,x\ > 0. We also show that if there are no 
such points, L^-normalized quasimodes have sup-norms that are o(X"~^^^'^), and, in 
the other extreme, we show that if there is a point blow-down x at which the first 
return map for the flow is the identity, then there is a sequence of quasi-modes with 
L°°-norms that are n(A("-l)/2)^ 

1. Introduction 

In a recent series of articles |SZ1 ITZ[ ITZ21 ITZ3| . the authors have been studying 
the relations between dynamics of the geodesic flow and estimates of i^-normalized 
eigenfunctions of the Laplacian on a compact Riemannian manifold {M,g). The general 
aim is to understand how the behavior of geodesies modifies the universal estimates of 
L°° of Avakumovic-Levitan-Hormander, and the general norms obtained by Sogge 
[Solj (see also |KTZ] and [SS] for recent and more general results). In particular, we 
wish to characterize the global dynamical properties of the geodesic flow of (M, g) which 
exhibit extremal behavior of eigenfunction growth. This problem is an example of global 
analysis of eigenfunctions as surveyed in 

This article continues the series. Its purpose is to sharpen the previous results on 
maximal eigenfunction growth and to prove they are sharp by giving converse results. 
To introduce our subject, we need some notation. Let { — A^} denote the eigenvalues of 
A, where < A§ < < A| < . . . are counted with multiplicity and let {^p\^{x)} be an 
associated orthonormal basis of L^-normalized eigenfunctions (modes). If A^ is in the 
spectrum of —A, let V\ ^ {ip : Aip = —X^ip} denote the corresponding eigenspace. We 

The authors were supported by the National Science Foundation, Grants DMS-0555162, DMS- 
0099642, and DMS-0354668. 



measure the growth rate of L°°-norms of modes by 

(1.1) i°°(A,g)= sup 

The general result of [Lei E] is that 

(1.2) i-(A,5) =0(A^). 

If this bound is achieved for some subsequence of eigenfunctions, i.e., L°°{X,g) — 
r2(A^"^^'/^), we say that [M,g) has maximal eigenfunction growth. The corresponding 
sequence of L^-normalized eigenfunctions have L°° norms which are comparable 

to those of zonal spherical harmonics on 5". The main result of |SZj is a necessary 
condition on maximal eigenfunction growth: there must then exist a point z such that 
a positive measure of geodesies emanating from z return to it at a fixed time T . In the 
case where all directions loop back, we will call z a blow-down point since the natural 
projection tt : S* M M has a blow-down singularity on S*M. For lack of a standard 
term, in the general case of a positive measure of loops we call z a partial blow-down 
point. Examples of blow-down points are poles of surfaces of revolution and umbilic 
points of two-dimensional tri-axial ellipsoids. In the case of surfaces of revolution, all 
geodesies emanating from poles smoothly close up while in the case of ellipsoids, the 
geodesies emanating from umbilic points loop back but with two exceptions do not close 
up smoothly. One can construct partial blow-down points by perturbing these metrics in 
small polar caps to obstruct some of the geodesies. 

A comparison of surfaces of revolution and ellipsoids shows that the necessary condition 
on maximal eigenfunction growth in [SZ| is not sharp and focusses attention on the 
distinguishing dynamical invariant. It is easily seen that surfaces of revolution are of 
maximal eigenfunction growth (cf. e.g. [SZj ) and that zonal eigenfunctions achieve 
the sup norm bound (|1.2p at the poles (see e.g. |So2] ). However, ellipsoids are not of 
maximal eigenfunction growth. It is proved in |T1|, IT2j that obvious analogues of zonal 

eigenfunctions on an ellipsoid only have the growth rate J^^^^ ' ^"^^ ^ ^ consequence of 


Theorem[31 it follows that \\^\^ ||oo = o{\J ) on such a surface. Indeed, although we will 
not prove it here, it is likely that L°°(A,g) — O ^-^L^ on the ellipsoid. 

The obvious difference between the geodesies of surfaces of revolution and ellipsoids is 
in the nature of the first return map $2 on directions 9 G S*M. This map is simplest to 
define when 2 is a blow-down point, i.e. if all directions loop back. The first return map 
is then the fixed time map of the geodesic flow G* acting on the sphere bundle: 

(1.3) ^, = G'^ : S*M ^ S;M. 

Below we will define it in the case of a partial blow-down point. 

In the case of a surface of revolution = id is the identity map on S*M, while in 
the case of an ellipsoid it has just two fixed points, one attracting and one repelling, and 
all directions except for the repelling fixed point are in the basin of attraction of the 
attracting fixed point. This comparison motivates the first theme of the present article: 
to study of the relation of maximal eigenfunction growth and the dynamics of this first 
return map. The relevance of the first return map to problems in spectral theory was 



already observed by Y. Safarov at al in studying clustering in the spectrum [Sj IGSi ISVj . 
It seems reasonable to conjecture that maximal eigenfunction growth can only arise in 
the identity case, or at least when $2 has a positive measure of fixed points, i.e. if there 
exists a positive measure of smoothly closed geodesies through a point z of M. 

However, this necessary condition is not sufficient. A counterexample was constructed 
in |SZj of a surface with a positive measure of closed geodesies through a point z but 
which does not have maximal eigenfunction growth. Further, we conjecture that even 
existence of blow-down points fails to be sufficient: for instance, every point is a blow- 
down point on a ZoU manifold, but we conjectue that generic ZoU manifolds fail to have 
maximal eigenfunction growth. 

We can close the gap between necessary and sufficient conditions on eigenfunction 
growth by generalizing the problem of eigenfunction growth to include approximate eigen- 
functions, or quasi-modes. As we shall see, the widest collection that one could hope to 
have pointwise o(A'"~^)/^) upperbounds are defined as follows: 

Definition 1.1. A sequence {tpx}, A ~ Xj, j = 1,2,... is a sequence of admissible 
quasimodes if \\'4>\\\2 = 1 o,''^d 

(1-4) m + \^)i^xh + \\Si^^Px\\oo^o{\). 

Here, S"^ denotes the projection onto the [/i, 00) part of the spectrum of V— A, and 
in what follows ^ I — Sj[, i.e., S^f = X^a <^ ^jif)^ where ej{f) is the projection of 
/ onto the eigenspace with eigenvalue Xj . 

There are many notions of quasimodes in the literature, but the above one seems to be 
new. We shall describe why it seems to give the natural class of "approximate eigenfunc- 
tions" in theorems that say maximal pointwise blowup implies the existence of certain 
types of dynamics (see below). We should also point out that the technical condition 
in the definition that ||>S'2aV'a||oo = o(A) is typically not included in the definitions of 
quasimodes. We need it for some of our results, and we note that, by Sobolev, when 
the dimension is smaller than 4, it is a consequence of the main part of the definition, 
i.e., ||(A -|- A^)V'A|i2 = o(A). A model case of functions satisfying ()1.4p would be a se- 
quence of L^-normalized functions {tp\j } whose A spectrum lies in intervals of the 
form [Xj — 0(1), Xj + 0(1)] as Xj — > 00. 

It is natural to consider quasi-modes because the methods of producing blowup apply 
in fact to quasi-modes. Results on modes are obtained only by specializing results on 
quasi-modes. In examples where one knows the eigenfunctions in detail, such as surfaces 
of revolution or ellipsoids, the reason is usually that the modes and quasi-modes are 
the same. We should also point out that while (|1.4p is a natural condition for classes 
of "approximate eigenfunctions" satisfying o(A^^"^^^/^) sup-norm upperbounds, it is not 
necessarily so for lowerbounds. Indeed, the quasimodes satisfying ri(A("~^^/^) lower- 
bounds that we shall construct will satisfy ||(A -|- A^)i/'a||2 = o(l) (quasimodes of order 
zero), which is weaker than (|1.4p . 



An important example of such a quasi-mode is a sequence of "shrinking spectral pro- 
jections" , i.e. the L^-normalized projection kernels 

with second point frozen at a point z G A/ and with width — > 0. Here, X[\j.\j+<Lj]{^i z) 
is the orthogonal projection onto the sum of the eigenspaces V\ with A £ [Xj, \j + ej] The 
zonal eigenfunctions of a surface of revolution are examples of such shrinking spectral 
projections for a sufhciently small e^, and when z is a partial focus such <i>|(x) are 
generalizations of zonal eigenfunctions. On a general ZoU manifold, shrinking spectral 
projections of widths ej = 0{\~^) are the direct analogues of zonal spherical harmonics, 
and they would satisfy the analog of (|1.4|) where o(A) is replaced by the much stronger 

1.1. General results. Our first result is quite general. It shows that the main result of 
[SZj extends to admissible quasi-modes and also gives a reasonable converse. 

Theorem 1. Let he a compact Riemannian manifold with Laplacian A. Then: 

(1) If there exists an admissible sequence of quasi-modes with \ \iP\^\\l°=' = ^(-^^^ ); 
then there exists a partial blow-down point z ^ M for the geodesic flow. If [M, g) 
is real-analytic, then there exists a blow-down point. 

(2) Conversely, if there exists a blow-down point and if the first return map is the 
identity, — id, then there exists a quasi-mode sequence {ipx^} order with 

As we mentioned before, Part (1) of the Theorem was proved in [SZj for modes. 
The improvement here is that there must be a partial blowdown point if a sequence of 
quasimodes has maximal sup-norm growth. We can make a further improvement and 
show that there must be a special type of partial blowdown point, a recurrent point for 
the geodesic flow. 

Let us be more specific. Given x € M, we let Cx the set of loop directions at x: 
(1.5) Cx^i^e S*M : 3T : exp, = x}. 

Thus, x is a partial blow-down point if > where \ ■ \z denotes the surface measure 
on S*M determined by the metric gx- We also let : S*M M+ U {oo} denote the 
return time function to x, 

TxiO = 

inf{i > : exp^, t^ = x}, if ^ G Cx] 
-l-oo, if no such t exists. 

The first return map is thus 

G'l- : Cx SIM. 



In the general case, Cx is not necessarily invariant under G^^. To obtain forward/backward 
invariant sets we put 

(1.6) £±-= fl {Gl-fLx, 


and also put = C^°° n C^°° ■ Then G!^"=) defines a dynamical system. We equip 
it with the restriction of the surface measure \ ■ \x, but of course this measure is not 
generally invariant under G^'' ■ We further define the set of recurrent loop directions to 
be the subset 

where w(^) denotes the w-limit set, i.e. the limit points of the orbit {(Gi^'")"'C • " G ^+}- 
Equivalently, ^ G belongs to TZx if infinitely many iterates, (0!^=")"^, n g Z, belong 
to r, whenever F is a neighborhood of ^ in S*M. Finally, we say that x is a recurrent 
point for the geodesic flow if \TZx\ > 0. 

Our improvement of the first half of the preceding theorem will be based on the 
following result that will give upperbounds for admissible quasimodes under a natural 
dynamical assumption. 

Theorem 2. Suppose that \TZx\ — for every x G M . Then, given e > 0, one can find 
A(e) < oo and S{e) > so that 

(1-7) IIX[A,A+5(e)]/llL-(M) <^A("-l)/2||/||i.(M), A>A(£). 

Under the stronger hypothesis that \Cx \ =0 for every x G M one has that for every 5 > Q 
there is a X{6) so that 

(1-8) IIX[A,A+5]/IU==(M) <C<5i/2A(»-i)/2||/||^,(^^,, X>X{S), 

for some constant C = C{M,g) which is independent of 5 and A. 

To show that Theorem [5] is indeed stronger than the result in [SZj , we note that it is 
well-known [Tl| that Liouville metrics on spheres (such as the triaxial ellipsoid) satisfy 
the condition \R.x\ = for each x G M. However, \Cz\ = 1 when z G M is an umbilic 
point for the metric and so, Theorem [2] applies to these examples as well and gives the 
L°°{X,g) = o(A("-i)/2)_bound. 

Also, as noted in |SZ] , if one uses an interpolation argument involving the estimates in 
[Solj . then (|1.7p implies that L^'-estimates are not saturated for p > 2(n-|-l)/(n— 1) under 
the assumption that \TZx\ = for all x G M. Recently, one of [So4| has formulated a 
sufficient condition for the non-saturation of L^-estimates in dimension two for 2 < p < 6 
that involves the concentration along geodesies. A condition for the endpoint case of 
p = 6 for dimension 2 or 2 < p < 2(n + l)/{n — 1) or p = 2{n + l)/(fT. — 1) in higher 
dimensions remains open. 

A corollary of Theorem [2] will be given in Theorem 12.41 below which says that if there 
is a sequence of admissible quasimodes with maximal sup-norm blowup, then there must 
be a recurrent point for the geodesic fiow. Equivalently, if there is no such point, then a 
sequence of admissible quasimodes must have sup-norms that are o 

We can write the conclusion of Theorem [2] in the shorthand notation 

(1-9) IIX[A,A+o(l)]llL^(Af)^L~(M) = 0(A("-l)/2). 



This result is optimal in one sense because the well known sup-norm estimate 

||X[a,a+i]||l^(m)^l^(m) =0(A("-i)/2) 

cannot be improved on any compact Riemannian manifold (see e.g., [So3]), and this 
together with (|1.9p provides a motivation for Definition On the other hand, it might 
be the case that (|1.9p holds under the weaker hypothesis that \Cx\ =0 for every x € M, 
if Cx C Cx denotes the set of periodic directions, i.e. initial directions for smoothly closed 
geodesies through x. Also, because of the sharp Weyl formula, the bounds in (jl.Sp are 
clearly sharp in the sense that one cannot take a larger power of 5, and one also needs 
the hypothesis that A is large depending on 5. 

We should point out that Theorem [2] is related to the error estimates for the Weyl law 
of Duistermaat and Guillemin fDG] and Ivrii |Ivlj and the error estimates of Safarov [S] 
for a local Weyl law. Like Ivrii's argument, ours are just based on exploiting the nature 
of the singularity of the wave kernel e**^/^^ at t = 0. Unlike these other works, though, 
we can prove our main estimate, (jl.7|) . without using Taubcrian lemmas. Traditionally, 
sup-norm estimates like ()1.7p were obtained by deducing them from stronger asymptotic 
formulas, e.g. appropriate Weyl laws with remainder bounds. Two of us in [SZj used this 
approach to prove the weaker variant of Theorem [5] where one deduces (II. 7p under the 
stronger assumption that there are no partial blowdown points. In the present work, we 
are able to prove these stronger results using a simpler argument that yields the main 
estimate (|1.7|) directly but does not seem to yield a correspondingly strong local Weyl 
law under the assumption that \TZx\ = for all x. 

1.2. Invariant tori and surfaces with mELximal eigenfunction growth. An easy 
consequence of our results is the following: 

Theorem 3. // a real analytic Riemannian n-manifold {M,g) has maximal growth of 
eigenf unctions or admissible quasi-modes, then its geodesic flow has an invariant La- 
grangian submanifold A ~ S'^ x S"~^ C S*M. Hence, a surface with ergodic geodesic 
never has maximal growth of eigenf unctions or admissible quasi-modes. 

By jSZj . a real analytic surface with maximal eigenfunction growth must be a topo- 
logical sphere, so the last result only adds new information when M ~ S*^. Real analytic 
ergodic metrics on 5*^ have been constructed by K. Burns - V. Donnay |BD| and by 
Donnay-Pugh [DP) IDP2| . There even exist such surfaces embedded in K.^ (see |BDj for 
computer graphics of such surfaces). Note that such metrics must have conjugate points, 
so the logarithmic estimates of |Bej do not apply. 

2. Recurrent points and upperbounds for quasimodes 

We shall first prove Theorem O To do so, we first note that 
(2-1) IIX[A,A+5]lli2(M)^L-(M) = sup ^ \ejix)f, 

if {cj} is an orthonormal basis of eigenfunctions with eigenvalues {Aj}. By compactness, 
we conclude that the first inequality in Theorem [5] follows from the following local version 



Proposition 2.1. Suppose that xq e M satisfies \R-xo\ ~ 0- Then, given e > we can 
find a neighborhood Me of xq, a < oo and a (5(e) > so that 

(2.2) J2 |ej(x)|2 <e2A(«-i)/2^ A4, A> Ae. 


In [SZ| we exploited the lower semicontinuity of L{x,^) where L{x,£,) equaled the 
shortest loop in the direction ^ S S*M if there was one and L{x,£,) = +oo if not. 

To prove our improvement of the main result of [SZj . instead of watching all loops, 
we shall just watch all loops of a given length length i and initial direction f which 
have the property that dist (0^(0:0 ^ with, dist (• , •) being the distance induced 
by the metric, and, as before, Gf,(^) being the terminal direction. So we let Ls{x,^) 
be the length of the shortest such loop fulfilling this requirement if it exists and +oo 
otherwise. Then Ls{x,^) : S*M — > (0, +cxd] is lower semicontinuous and l/Ls(x,S,) is 
upper semicontinuous. We then let TZ^ then is all ^ for which 1/Ls{x,^) ^ 0. 

To exploit this, if xq is as in the proposition, we shall choose S large enough so that 
\TZ^\ < £^/2 and then take /(x, <^) to be 1/Ls{x, ^) in the following variant of Lemma 3.1 
in [SZj . We shall take the parameter p in the lemma to be 1/lOT where T is much larger 
than l/(5("-i\ 

Lemma 2.2. Let f be a nonnegative upper semicontinuous function on O x S"^^ , where 
O C K" is open. Fix xq O and suppose that G 5*"^^ : /(xo,$) 7^ 0} has measure 
< e/2, with e > being fixed. Let p > be given. Then there is a neighborhood J\f of xq 
an open set Qb C 5'"""'^ satisfying 

\f{x,0\<P, (x,0eAAx5"-i\r!f, 

Furthermore, there is a b{^) G C°° supported in fit, satisfying < b < 1, and having the 
property that if B{^) = 1 — &(^) then f{x, ^) < p on J\f x suppB . 

Proof: The proof is almost identical to Lemma 3.1 in [SZj . 

By assumption the set i?p = G S*"^^ : /(a;o,^) > p} satisfies \Ep\ < e/2. Let 

EpU) = e : f{x,0 > P. some x G B{xo. 1/j)}, 

where B{xQ,r) is the closed ball of radius r about xq. Then clearly Ep{j + 1) C Ep{j). 
Also, if ^ G r\j>iEp{j) then for all j one can find Xj G B{xo, such that f{xj,S,) > p, 
which means that 

p < limsup/(xj,0 < f{xa,£,), 
by the upper semicontinuity of /. Thus, 

nj>iEp{j) C Ep. 

Consequently, if j is large \Ep{j)\ < e. Fix such a j = jo and choose an open set Vlb 
satisfying Ep{j) C ^Ib and \^lb\ < e. Then clearly f{x,S) < p if (x,^) G B{xQ,l/jQ) x 

For the last part, note that the argument we have just given will show that the sets 
Ep{j) are closed because of the upper semicontinuity property of /. Thus, if Ep{jo) and 



rif, are chosen as above, we need only apply the C°° Urysohn lemma to find a smooth 
function b{^) supported in fif, with range [0, 1] and satisfying h{S) — 1, ^ G Ep{jo), which 
then will clearly have the required properties. □ 

To apply this lemma we first choose a coordinate patch /C with coordinates k{x) around 
Xq, which we identify with an open subset of R". Also, fix a number 5 > small enough 
so that < £^/2. We then let f{x,£,) denote the image of 1/Ls{x,£^) in the induced 

coordinates for {{x,(^) € S*M : x G /C}. Then, given a large number T (to be specified 
later), we can find a function b G C°°(S'"~^) with range [0, 1] so that 

(2.3) / b{Odi<s', 

(2.4) Ls{x,^)>2T onTVxsuppB, 
where A/" C k(/C) is a neighborhood of xq and 

BiO - 1 - 6(0- 

Choose a function ip £ C°°(R") with range [0,1] which vanishes outside of Af and 
equals one in a small ball centered at k{xq). Using these functions we get zero-order 
pseudo-differential operators on R" by setting 

b{x,D)f{x)^i:{x){2TT)-" 11 e'^--y>^b{^/\£,\)i,{y)f{y)dydi, 


B{x,D)f{x)=^{x){2^r' [[ e^^--y>^B{im^{y)f{y)dyd^. 

Note that both variables of the kernels of these operators have support in /C. If we let 
b{x, D) and B{x, D) in ^'''(A/) be the puUbacks of b and B, respectively, then 

b{x,D) + B{x,D) = ^/{x). 

Since ip^X[\.x+s] = b{x, D)x[\,\+s] + B{x, D)x[\,\+s] it is clear that (|2.2p would follow if 
we could show that there is a T = T(e) > S, A(e) < oo and 6{e) > so that 

(2.5) II foX[A,A+5(e)] II L^->L=^ <^^£A("-i)/2, A > A(e), 

(2.6) ||i?X[A,A+5(e)]llL^^L=° <CeA("-i)/^ A > A(e), 
for some uniform constant C which is independent of e. 

Note that 

(2-7) ll&X[A,A+5(e)]lli2^i» = sup \bej{x)\^ 

^ \je[\,\+S{e)] 

(2.8) \\Bxix,x+sie)]\\h^L^ ^ snp ^ \Be,{x)\^ 

^ \je[\,\+S{e)] 

To exploit this we shall use a standard trick of dominating these truncated sums by 
smoothed-out versions in order to use the Fourier transform and the wave operator. To 
this end, we choose p g C^(R) which vanishes for \t\ > 1/2 and satisfies p > and 
p{0) = 1. If we then take T to be a fixed multiple of 1/S{e), we conclude from (|2.7p and 


([^ that and ([^ would follow from showing that if T = T{e) and A(£) are large, 


(2.9) 5^(p(T(A-A,)))'|6e,(a:)p<C£2A"-\ A > A(£) 


(2.10) 5](p(r(A-A,)))'|i3e,(a:)p<C£2A"-\ A > A(£). 

To prove these, we shall require the following standard result which is based on the 
singularity of the wave kernel restricted to the diagonal at t = 0. To state the notation, 
we let U = e**^ denote the wave group and U(t,x,y) its kernel. Then we need the 
following result which follows from Proposition 2.2 in [SZj . 

Lemma 2.3. Let {M,g) have injectivity radius > 10 and let A{x,D) E '^''^{M) be a 
pseudo- differential operator of order 0. Let a € C^(M) vanishes for \t\ > 2 and satisfies 
a{0) = 1. Then, if Ao{x,S,) denotes the principal symbol of A, 

a{t)e''^' {AUA*){t,x,x) dt 

- (27r)-«A"-i / \Ao{x, 0\d<T{0 = 0(A""2). 

In what follows, we may assume without loss of generality that the hypothesis on the 
injectivity radius of M is satisfied. 

Note that we can rewrite the left side of (|2.11[) as 

a{t)e-'^' {AUA*) {t, x, x)dt = ^ a(A - Xj)\Aej{x)\^ . 

If we choose a as above so that a > 0, q;(0) = 1, we conclude from (|2.1ip and (j2.12p that 

(2.13) ^ C\-'\\A,{x, OIli-^tsjA/) +^^aA"-2, 


where C is independent of A = A{x, D) E 'I''^(Af ). This will prove to be a useful estimate 
in what follows. 

Using (|2.13[) we can get (j2.9p if we assume, as we may, that T > 1. For then (/5(T(A — 
Aj)))^ < C7v(l + |A - Xj\)-^ for any N, which yields (EH) as \\b{x, ■ )\\l2^s'M) ^ 
by (1231). 

To finish the proof of (jl.7p by proving (|2.10p . we first exploit (|2.4p to see that we can 
construct a smooth partition of unity 1 = J2k V'fc(?) of the unit sphere which consists of 
terms each of which has range in [0, 2] and is supported in a small spherical cap 
of diameter smaller than S/10. We then let Bk{x, D) be the zero-order pseudo-differential 
operator whose symbol equals B{x,£,)'ipk{£,/\^\) in the coordinates used before. Since i5 is 
fixed, we would have (j2.10p if we could show that 


(2.14) $:(p(T(A - X,))Y\Bke,{x)\' < CT-'X^-' + Cb,.tA"-^ 



Indeed, if T is chosen large enough so that C(5-("^i)r^i < then, since B = ^ Bj, by 
applying the Cauchy-Schwarz inequality, we get (|2.10p for large enough A. As we shall 
see, the constant C in (|2.14p can be taken to be 0(1) as (5 ^ 0; however, the reduction to 
estimates for each single Bk contributes an additional factor 0((5"(""^)) to the constant 

To prove (|2.14p . we note that we can rewrite the left side as 


T-\p* p) {t/T) {BkUBl){t, X, x)e'''^ dt 

To estimate this, we need to exploit the fact that our hypothesis (I2.4p implies that {t, x) — > 
{BkUBl){t,x,x) is smooth when < |t| < T. Also, by construction, [p * p){t/T) = 
for \t\ > T. To use these facts, we choose (3 G C5^(M) satisfying f3{t) 1, |i| < 1 and 
l3{t) = 0, |<| > 2 and then split the left side of ([^1^ as 



mT-\p*p){t/T){BkUB*k){t,x,x)e-'^' dt 

+ ^Ji^- P^))T-\p * p){t/T){BkUBl){t, X, x)e-'^' dt ^ I + II. 

If we integrate by parts we see that // must be 0{X~^) for any N, which means that we 
are left with showing that / enjoys the bounds in (j2.14p . However, since we are assuming 
that T > 1, one can check that the inverse Fourier transform of t l3(t)T''^{p * p){t/T) 
must be < CArr^^(l + |t|)~^ for any if r is the variable dual to t. Thus, for every A^, 


/ < ^(1 + |A - A,|)-^|i3,e,(x)p, 

which means that our remaining estimate (|2.14p also follows from l|2.13p . 

One proves (|1.8p by the above argument if one takes 5 in the last step to be equal to 
1. □ 

2.1. Blowup rates for quasimodes: Proof of Theorem [T] (i). Next, we shall show 
that we can extend the blowup results of [?SZ] for eigenfunctions to include the admissible 
quasimodes (defined in Definition II. ip and also allow one to conclude that there must 
points through which there is a positive measure of recurrent directions for the geodesic 

Theorem 2.4. Suppose that ip\ is a sequence of admissible quasimodes satisfying 

IIVaIIoo - r!(A("-i)/2). 
Then there must be a point x ^ M with \TZx\ > 0. 

Since TZx C Cx this result is stronger than the first part of Theorem [T] 
The proof of Theorem 12.41 is based on Theorem [3 and the following lemma. 

Lemma 2.5. Fix B > and suppose that for A = Aj — > oo we have 

(2.15) IIVaIIoo > BA("-i)/2. 



Then if < S < 1 there exists e > so that if X — Xj and 

(2.16) ||(A + X^)ij^h + < eX, 
then if X[x-5,\+s]f = Y.x^e[x-5,\+s] 

(2.17) \\xix-5,x+s]Moo > |a("-i)/2, 
for all sufficiently large X — Xj. 

Before proving Lemma 12.51 let us see why it and Theorem [2] imphes Theorem 12.41 
To do this, let us suppose that we have a sequence of admissible quasimodes satisfying 
IIV^aIIoo = ri(A*^"^^^/^) If we apply Lemma [^751 we conclude that there is a positive constant 
c > so that for any < (5 < 1 we have 

\\xix-s,x+s]Mo. > cA("-i)/2, 
for some sequence A = Aj, if A is large enough (depending on 6). 

Let p > 0. If there were no recurrent points, we could apply Theorem [2] to conclude 
that there is a, S = d{p) so that for large enough A (depending on p) 

IIX[a-5,a+5]^a||co <CpA("-i)/2, 

which leads to a contradiction if p is chosen small enough so that Cp < c. Thus, we 
conclude that there must be a recurrent point under the hypotheses of Theorem l2.4l □ 

Proof of Lemma [275} To simplify the notation, let us set Xx = X[x-S.x+S]- We need to 
see under the hypotheses of Lemma l2.5l we have 

(2.18) ||(/-xi)^A||oo<f A("-l)/^ 

if A = Xj is large enough. 

This would follow from a couple of estimates. The first one says that there is a constant 
A which is independent of < (5 < 1 and A > 1 so that 

(2.19) \\x\{I - xi)/l|oo < AA("-i)/2(A<5)-i||(A + X')fh, 
while the second one says that 

(2.20) ||(/-x1)^2a/||oo < CA("-i)/2^-i||(A + A2)/|l2. 

To see how these imply (|2.17p . we take f = ipx- Then since (5 < 1 we have 

(2.21) (/ - xi)^x = x\{I - xi)^x + xi)S2X^x + S^^iJx- 

If n > 4 we estimate the last piece by the second part of our admissible quasimode 
hypothesis H'S'^VaHcc- = o{X) = o(A("^-'^)/^). If n < 3 we use Sobolev to get that for a 
given < cr < 1/2 

< C||(V^)t-2+-5J-^(A + X')M2 

<CA2-^+-||(A + A2)V;a||2 

< CAt-2+'^A = o(A("-^)/2)^ 


as desired since a < 1/2. 

Using (|2.16p . (|2.19p and (|2.20[) we can estimate the remaining pieces in (|2.2ip 

Wxiil - x'Mlo. + xi)^2A^A||oo <{A + C)A("-i)/2(A<5)-ie(A/log A) 

< 2(^ + C)(£/(5)A("-i)/2, 

if A is large. Since this estimate and our earlier bounds for S2xtp\ yield (|2.18p . we are 
left with proving ((2l9)) and ((2:20)) . 

The estimate (|2.19l) is easy. Using the fact that \\x\\\l2^l°° < AA^""^)/^, we get 


< ^A("-i)/2||(A + X')-'{I - xi)(A + A2)/I|2 

< AA("-i)/2(A5)-1||(A + A2)/||2. 

To prove (|2.20p . let H^jj^i-f denote the projection onto the [j, j+1) part of the spectrum 
of \/~A. Then we can write 


[I - X\)S2\f = ^^(^^l\+k,\+k+l)S2\f + n[A-fc-l,A-fc)'^2A/) , 



ll(/-xi)^2A/|loo<E(lln[A+^^-A+fc+l)^2A/||oo + ||%_fe_i,;,_fe)52A/||oo) 
= 1 + 11. 

We shall only estimate / since the same argument will yield the same bounds for //. 
To estimate /, we first note that for 1 < fc < A 

||n[A+fc,A+fe+l)5l|oo < C'A^""^^/^||n[A+/c,A+fc+l)5ll2 

= CA("-i)/2||(A + X^)-%^+k,x+k+i) (A + X')g\\2 
< CA("-i)/2(Afc)-i||n[A+fc,A+fc+i)(A + A2).g||2 
Therefore, by applying the Schwarz inequality, we get 


I = ^k^^ {k\m\+k,\+k+i)S2\f\\oo) 

<CA("-i)/^A-i(^||n[,+,,,+,+i)(A + A^)/||^)'^' 

< CA(""1)/2^-1||(A + a2)/||2, 

as desired. Since, as we noted, the same argument works for //, we have completed the 
proof of Lemma [2T5l □ 

Let us conclude this section by pointing out that the conclusion of the lemma is not 
valid for dimensions n > 4 if one just assumes ||(A + ADi/'Ai-lb = o{Xk) or even 

(2.22) ||(A + A2)V;aJ|2 = 0(1) 



for the quasimode definition. 

Let us first handle the case where n > 5 since that is slightly simpler than the rt = 4 
one. To handle this case, we fix a nonnegative function rj € C^(R) satisfying 77(10) = 1 
and r]{s) = 0, s ^ [5, 20]. We then set 

where s > 1 is large and will be chosen later. Put 

where xq & M is fixed and e > is small. 

We notice that the conclusion of the lemma is false for these functions since X[\k-5,\k+s]''l'\k — 
if s > 1 is fixed and if Afe is large, due to the fact that the spectrum of ipx^ is in [5A^, 20 A^] 
and Afc does not lie in this interval for large k. Also, it is not hard to verify that 

V'A.(xo)«i^^ = Ar, 

and so by choosing s = -^^j^, we have one of the assumptions of the lemma that \\ipXk Hoc = 
^^^^n-i)/2-j^ We also have (P?^ if n > 5. For then 

||(A + ADV'aJU « ||(A + 1)7/^aJ|2 « A^II^.JI^ « K^K^K-K-/^ = o(l), 
if, as we may, we choose e < 1/2. 

Minor modifications of this argument show that things break down for n = 4 as well 
if one just assumes (|2.22p . Here one would take jo — 2^'= so that logjo — A^^"'^, where 
log is the base-2 log. Then, with the above notation, one sets 

^A,(a;) = (logjo)-'/' E 2-^",7(x/A/2^)(xo,x). 

is [log jo, 2 log jo] 

Then, one can see that 

^A.(xo)«(log,o)^/^ = Ai"^^)/^ 
Xl\k-s,Xk+s]'4^\k = if Afc is large. Finally, if n = 4, (|2.22|) is valid since 

||(A + A2)^,J|2«||(A + l)Kll2 
«(logjo)-^ E 2^^2-^^\fj{VA/2^){xor)\\l^{logjor' ^ ^ ^ 

j£ [log jo ,2 log jo] je [log jo ,2 log jo] 

These constructions will also show that when n > 4 one cannot use 

||(A + AD^Aj|2 = 0(Ar) 
for any large s as the condition for quasimodes {^Xk} a-nd have the conclusions of the 
lemma be valid. 

2.2. Quasi-modes associated to blow-down points: Proof of Theorem [1] (2): 

We now prove the converse result in Theorem [T] under the assumption that = Id . 
The method is to construct quasi-modes associated to the "blow-down" Lagrangian 
in Definition |4] (see below). The analysis generalizes the one in [Z] in the ZoU case. The 
key point is the existence of an invariant 1 /2-density on A^ for the geodesic flow. In this 
case, the invariant 1/2-density is |c?/idt|^/^ where, d/i is Liouville measure on S^M. 



2.2.1. The Blow-down Lagrangian. Since z is a blow-down point, the geodesic flow in- 
duces a smooth first return map (|1.3p . Let Ct denote the mapping cy finder of Gj, 

(2.23) Ct = SIM x [0,T]/ where {^,T) ^ (GJ(C),T). 

The Ct is a smooth manifold. It naturally fibers over by the map 

-k-.Ct^S^, 7r(^,t)=t mod 2ttZ. 

Proposition 4. Let {M, g) be an n- dimensional Riemannian manifold, and assume that 
it possesses a blow down point z. Let Lz : Ct ^ T* M be the map 


Then iz is a Lagrange embedding whose image is a geodesic-flow invariant Lagrangian 
manifold, Kz, diffeomorphic to x S"^^ ~ Ct- 

Proof. We let uj denote the canonical symplcctic form on T* M . Then, under the map 
(2.24) l: X SIM ^T*M, L{t,x,^) ^ G*{x,^), 

we have 

L*uj = oj - dH hdt, i7(a;,C) = l^lg. 

The map iz is the restriction of t to K x S*M. Since dH = on S* M and w = on 
S'*M, the right side equals zero. 

Thus, Lx is a Lagrange immersion. To see that it is an embedding, it suffices to prove 
that it is injective, but this is clear from the fact that G* has no fixed points. 


Let aA denote the action form a = ■ dx restricted to A. Also, let ttia denote the 
Maslov class of A. A Lagrangian A satisfies the Bohr-Sommerfeld quantization condition 

(2.25) = niod H\K,Z), 

ZTT 4 


with (3 equal to the common Morse index of the geodesies G*(z,^),^ g S^M. 
Proposition 5. A^ satisfies the Bohr-Sommerfeld quantization condition. 

Proof. We need to identify the action form and Maslov class. 
Lemma 6. We have: 

(1) t*Q:A = dt. 

T L 

(2) blmjy^ — S;[dt]. 



Proof. (1) Let S^h denote the Hamiltonian vector field of H. Since {G*)*a — a for all 
t, we may restrict to i = T and to S^AI to obtain {G'^)*a\szM — a|s;Af- But clearly, 
C ■ dx\s'M = 0. 

(2) We recall that ttia^ S H^{Az,'Z) gives the oriented intersection class with the singular 
cycle S C A2 of the projection w : ^ M. Given a closed curve a on A^, we deform it 
to intersect S transversally and then rriA^ is the oriented intersection number of the 
curve with E. Our claim is that toa^ = /? where (3 is the common Morse index of the 
(not necessarily smoothly) closed geodesic loops 7^(i) = G*(z,^),^ S SfM. 

The inverse image of the singular cycle of under consists of the following com- 

i^^E = SlM\JConj{z), 


Conj{z) = {{t, ^) -.0 <t <T, ^ e S*, I det 4 expt^l = 0} 

is the tangential conjugate locus of z. All of S^M consists of self-conjugate vectors at 
the time T. 

If dimAf > 3, then H^{Ct,'Z) = Z is generated by the homology class of a closed 
geodesic loop at z and in this case mj^^ — (3 hy definition of the Morse index. If 
dimAf = 2, then H^{Ct,'^) has two generators, that of a closed geodesic loop and that 
of S'*M. The value of toa, on the former is the same as for dim A/ > 3, so it suffices to 
determine /g.jv^WA^- To calculate the intersection number, we deform S*M so that it 

intersects lJ^'S transversally. We can use C^S^M as the small deformation, and observe 
that it has empty intersection with ilT^E for small e since the set of conjugate times and 
return times have non-zero lower bounds. 


The Lemma immediately implies (j2.25p . completing the proof. 


2.2.2. Construction of quasi-modes. We now 'quantize' A^ as a space of oscillatory inte- 

Lemma 7. There exists £ (A/, A^, {rfc}) with i*a{<^k) = e"'^'=*|dt|i/2 (g) jd/il^/^, 

where dfj, is Liouville measure on S*M. 

We will refer to $fc as quasi-modes associated to the blow down point z G M. 


(1) In the case of S"" and z the north pole, ^k{z) is the zonal spherical harmonic 
of degree k. Equivalently, it equals, up to L^-normalization, the orthogonal 
projection kernel nfe(-, z) onto /cth order with second variable fixed at z. In this 
case, it is an eigenfunction. 

(2) On a general ZoU manifold, with z any point, the projection kernel onto the 
fcth eigenvalue cluster is a quasi-mode of this type, see |Z]. In general, it is a 



zeroth order quasi-mode, reflecting the width k^^ of the fcth cluster, and not an 

(3) On a surface of revolution diffeomorphic to 5^, the zonal eigenfunctions are os- 
cillatory integrals of this type. 

2.3. Proof of Theorem [3j By the results of [SZj . (M, g) possesses a point m such that 
all geodesies issuing from the point m return to m at some time £ (which with no loss of 
generality may be taken to be 2tt). By Proposition [H the map l of (|2.24p is a Lagrange 
immersion with image A^. 

If dimM = 2, the image i([0,27r] x S^M) is a Lagrangian torus, the mapping torus 
of the first return map G'^^'ls:^!^ : S^M S^M. Obviously, G*(A) = A for aU i, so A is 
an invariant torus for the geodesic flow. Moreover, M is diffeomorphic to S'^ or to KP^. 
Since S*M = RP^ when M = S'^ (or in the case RP^ is a quotient by a Z2 action), 
we have H'^{S*M) = {0}. Hence, A = dfl where ft C S*M is a singular 3-chain. Since 
dimS'*Af = 3, il has a non-empty interior, so A is the boundary of an open set. But 
G*ri C ft. Hence, there exists an open invariant set, and G* cannot be ergodic. □ 

In higher dimensions, we do not see how ergodicity rules out existence of invariant 
Lagrangian x S*™^^ or blow down points. Hyperbolicity of the geodesic flow is in- 
consistent with existence of such Lagrangian submanifolds. But, as mentioned in the 
introduction, there are better estimates in the case of (M, g) with Anosov (hyperbolic) 
geodesic flows. These never have conjugate points, and the generic sup norm estimate 

can be improved to ||t/?j||L=° = ^{Xj ^ /logXj) (|Be|). But of course such flows do not 
exist for metrics on 5^, and the previous result provides new information for analytic 
metrics with ergodic geodesic flow on S*^. 

2.3.1. Point-wise asymptotics of the quasimode ^ ■ In the following we let £ {'rk}~^]k — 
1,2,... and let Bj C M\j — 1,...,N be small geodesically convex balls with tt{Az) C 
UjLiBj. Let Xj G dQ°{Bj) be a partition of unity subordinate to this covering and 
Xr{s) G C'o°(]^) be a cutoff equal to 1 when |s| < P with P > 1 and zero when \s\ > 
2P. One then constructs the quasimode ^k{x) as a sum J2jLi Xj^k'' where the e 
C°°{Bj) are local oscillatory integrals of the form 

<^^\x) = (27r?i)^/jj„ e"^"'(^''')/'"'a(^')(x,6»;n) XR{\S\)de. 

Without loss of generality, we assume that z g Pi and let x = (xi, ...,a;„) G Pi be 
geodesic normal coordinates with x{z) = G M". Consider first 

$W(.t) = (2^^)^ / e''P'"'^--'^"'a^^\x,0-K)xRm)de. 

The L^-normalized quasimode <I>fc is constructed to solve the equation || — Ag$fe — 
ft^kWh^ = C'(l) and for this, one needs to globally solve the eikonal equation and the 
first transport equation. 

For the eikonal equation, we choose the phase {x, 6) positive homogeneous 

of degree zero in the ^^-variables. Since S^M C A^ n 7r~^(Pi) is non-characteristic for 
the geodesic flow, it follows that there exists a locally unique solution ip'^'^\x,9) to the 



initial value problem 

(2.26) \W,^('Hx,0)\l^l 

(2.27) (^(i)(O,0)-O, 

(2.28) A, n7r-i(Bi) = {{x,d^ip<^^\x,9)) G Bi x R"; ^^^^(x,^) = 0}. 
Consider the function 

V.W(x,0) = (x,^), MO. 

By the Gauss lemma, 

(2.29) 9ij{x)xj = ^ g2jiO)xj = x^, 

and so, (x, 0)g x^^i. Consequently, for ^^ 7^ we have that 


Then, from (P?^ it follows that if'-^^Q, 0) = and 

" 1^12 

(2.31) |v.^(i)(x,0)|^ = ^ 5^^(x)a.^a,^(i) = ^ - 1. 

i,j=l I Iff 

Thus, (p'-^^x,9) = {x, ill) satisfies the initial value problem in (|2.26p and (|2.27p . More- 
over, a direct computation shows that 

{{x,d^ip^^Hx,e)) e Bi X M"; dgip^^^{x,0) 0} = {{tuj,uj) e R" x S"~^;\t\ < e^}. 

Here, Cq is the geodesic radius of the ball Bi. The latter set is just n tt^^{Bi) written 
in normal coordinates. 

The transport equation for al^\x,9) is 

(2.32) g^'d,,^ ■ d,^4''> = g^'d^^d^^^ ■ 4^'' = g'^d^^d^^{{x,e)) ■ 4^'' = 0, 

where, we impose the initial condition a'i^\o, 9) = 1. It follows that 

(2.33) a\^\x,9)^l. 

2.3.2. -normalization. Consider first the local quasimode and choose 
<5e (1- i,l). Clearly, 

(2.34) / |$i'^(a;)P dx = 0{h^~^^-^^''). 


In the annulus As{K) := {x G _Bi; fi,* < |a;| < eo}i we introduce polar coordinates and 

$^^^(a;) = (27rn)T^ / e'^'^^^\'W) Xr{\0\) d9 


(2.35) ={2TTh)^ (^J^ e^^^^^'^^dwj XflW?■""^rf^■ 
Since — > oo as ft — > 0"*", one makes a stationary phase expansion in the inner cj-integral 
in The result is that for x e As(h), 

(2.36) ix) = \x\^{ c+ + c_ e~'^ +0{\x\-^h)) 
Here, c± e C with |c±| ^ 0. It foUows from ((2?36)) that 

(2.37) / \<i>i'\x)\'dx^{\c+\' + \c^\')eo + 0{h'). 


From (|2.37p and (|2.34p it foUows that there is a constant C(eo) > such that for ft 
sufficiently small, 

(2.38) / \^']^\x)\'^dx = C{eo)+0{h^'), ,5' = min (1 - (1 - ^)n, (5). 

The computation for the other quasimodes is the same and so, there exist constants 
Cj > 0,j = 2, ...,iV such that also H^i^^lU^ ^ Q for aU j ^ 1. Since satisfies 
the Bohr-Sommerfeld quantization conditions in Proposition O the local quasimodes 
satisfy (x) = ^ (x) for all a; e Bm n i?m' and so they patch together to form 
a global quasimode After possibly multiplying $fc by a postiive constant, it follows 
that ||<I>a;||l2 ^ 1 with |l(-Ag - rl)^k\\L^ = 0{1) as k ^ oo. 

2.3.3. Symbol computations. In normal coordinates, the map is given by the formula 

L^{t,uj) = {tuj,uj); t e K/[0,T]. 

Let : Az be the standard immersion {x,9) (x^d^^), {x,6) G C^f, where 

:= {{x,e) e M X K", deip{x,e) = 0}. Then, 

il{r/y^{t,uj) = {tuj,uj)=t, cjeS"-\ 

This is the phase function of the principal symbol L*a{^k) in Proposition [T] For the 
amplitude of the symbol t*(T($fc), one looks for a half-density solution a e C°^{Ct', \^\^) 
of the equation 

—G*a{t,u;)\s=o = 0, 

and in view of (|2.33p . the required solution is given by 

(2.39) d{t,Lu) ^ {2TTh)^\dtd^ji. 

Consequently, t*cr($fc)(i, w) = {2TTrk)^^ e'^*'^''\dij^dt\i as in Lemma[7]and moreover, 
we have proved 

Proposition 8. Let e {M,Az,{rk}) be the -normalized quasimode con- 

structed above. Then, t*(T($fe)(t, cj) = (27rrfe)^~e~**'"'= \dt\^ (g) |c?/i„|2 and 

|$fe(z)| = (27rrfc)'^ / a'^^\0,d;h)xRi\0\)de^k-..oCRr^. 

Here, Cr > is a constant depending only on R. 

This completes the proof of Theorem [T] (ii) . □ 




[A] G. V. Avakumovic, Uber die Eigenfunktionen auf geschlossenen Riemannschen Mannig- 

faltigkeiten, Math. Z. 65 (1956), 327-344. 
[Be] P. H. Berard, On the wave equation on a compact Ricmannian manifold without conjugate 

points. Math. Z. 155 (1977), no. 3, 249-276. 
[BD] K. Burns and V. J. Donnay, Embedded surfaces with ergodic geodesic flows. Internal. J. Bifur. 

Chaos Appl. Sci. Engrg. 7 (1997), no. 7, 1509-1527. 
[D] J. Duistermaat, Fourier integral operators, Courant Inst. Lecture Notes (1973). 

[DG] J. J. Duistermaat and V. W. Guillemin, The spectrum of positive elliptic operators and periodic 

bicharacteristics, Invent. Math. 29 (1975), 39-79. 
[DP] V.J. Donnay and C. C. Pugh, Anosov geodesic flows for embedded surfaces. Geometric methods 

in dynamics. II. Astrisque No. 287, (2003), xviii, 61-69. 
[DP2] V.J. Donnay and C. C. Pugh, Finite horizon Riemann structures and ergodicity. Ergodic Theory 

Dynam. Systems 24 (2004), no. 1, 89-106 
[GS] T. E. Gureev and Yu. G. Safarov, Exact spectral asymptotics for the Laplace operator on a 

manifold with periodic geodesies, (Russian) Translated in Proc. Steklov Inst. Mth. 1989, no. 

2, 35-53. Boundary value problems of mathematical physics, 13 (Russian). Trudy Mat. Inst. 

Steklov. 179 (1988), 36-53, 241. 
[Ho 1] L. Hormander, The spectral function of an elliptic operator. Acta Math. 121 (1968), 193—218. 
[Ho III] L. Hormander, The Analysis of Linear Partial Differential Operators, Volume III, Springer- 

Verlag Berlin Heidelberg, 1983. 
[Ho IV] L. Hormander, The Analysis of Linear Partial Differential Operators, Volume IV, Springer- 

Verlag Berlin Heidelberg, 1983. 
[Ivl] V. Ivril, The second term of the spectral asymptotics for a Laplace Beltrami operator on man- 
ifolds with boundary, (Russian) Funktsional. Anal, i Prilozhen. 14 (1980), no. 2, 25-34. 
[KTZ] H. Koch, D. Tataru and M. Zworski, Semiclassical LP estimates, Ann. Henri Poincar 8 (2007), 


[Le] B. M. Levitan, On the asymptoptic behavior of the spectral function of a self-adjoint differential 
eguaiton of second order, Isv. Akad. Nauk SSSR Ser. Mat. 16 (1952), 325-352. 

[S] Yu. G. Safarov, Asymptotics of a spectral function of a positive elliptic operator without a 

nontrapping condition, (Russian) Funktsional. Anal, i Prilozhen. 22 (1988), no. 3, 53—65, 96; 
translation in Funct. Anal. Appl. 22 (1988), no. 3, 213-223 (1989) 

[SV] Yu. Safarov and D. Vassiliev, The asymptotic distribution of eigenvalues of partial differential 
operators. Translated from the Russian manuscript by the authors, Translations of Mathemat- 
ical Monographs, 155. American Mathematical Society, Providence, RI, 1997 

[SS] H.F. Smith and C.D. Sogge, On the LP norm of spectral clusters for compact manifolds with 
boundary. Acta Math. 198 (2007), 107-153. 

[Sol] C.D. Sogge, Concerning the LP norm of spectral clusters for second-order elliptic operators on 
compact manifolds, J. Funct. Anal. 77 (1988), no. 1, 123—138. 

[So2] C.D. Sogge, Oscillatory integrals and spherical harmonics, Duke Math. J. 53 (1986), no. 1, 

[So3] C.D. Sogge, Fourier integrals in classical analysis, Cambridge Tracts in Math., 105. Cambridge 

Univ. Press, Cambridge, 1993. 
[So4] C. D. Sogge, Kakeya-Nikodym averages and LP-norms of eigenf unctions, |arXiv;0907.4827| 
[SZ] C. D. Sogge and S. Zelditch, Ricmannian manifolds with maximal eigenfunction growth, Duke 

Math. J. 114 (2002), no. 3, 387-437. 
[Tl] J.A.Toth, Eigenfunction localization in the quantized rigid body, J.Diff.Geom. 43(4)(1996), 844- 


[T2] J.A.Toth, On the quantum expected values of integrable metric forms, J.Diff.Geom. 52 (1999), 
no. 2, 327-374. 

[TZ] J. A. Toth and S. Zelditch, Ricmannian manifolds with uniformly bounded eigenfunctions, Duke 

Math. J. Ill (2002), no. 1, 97-132. 
[TZ2] J. A. Toth and S. Zelditch, Norms of modes and quasi-modes revisited. Harmonic analysis at 

Mount Holyoke (South Hadley, MA, 2001), 435-458, Contemp. Math., 320, Amer. Math. Soc, 

Providence, RI, 2003. 



[TZ3] J. A. Toth and S. Zelditch, LP -norms of eigenfunctions in the completely integrable case, An- 

nales Henri Poincare 4 (2003), 343-368. 
[Z] S. Zelditch, Fine structure of Zoll spectra, J. Punct. Anal. 143 (1997), 415-460. 

[Z2] S. Zelditch, Wave invariants for non-degenerate closed geodesies, GAFA 8 (1998), 179-217. 
[Z3] S. Zelditch, Local and global analysis of eigenfunctions on Riemannian manifolds. Handbook of 

geometric analysis. No. 1, 545-658, Adv. Lect. Math. (ALM), 7, Int. Press, Somerville, MA, 


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

Department of Mathematics, McGill University, Montreal, Candada 

Department of Mathematics, Johns Hopkins University, Baltimore, MD 21218 
E-mail address: 
E-mail address: 
E-mail address: