# Full text of "On Kakeya-Nikodym averages, $L^p$-norms and lower bounds for nodal sets of eigenfunctions in higher dimensions"

## See other formats

ON KAKEYA-NIKODYM AVERAGES, L^-NORMS AND LOWER BOUNDS FOR NODAL SETS OF EIGENFUNCTIONS IN HIGHER DIMENSIONS MATTHEW D. BLAIR AND CHRISTOPHER D. SOGGE Abstract. We extend a result of the second author |27l Theorem 1.1] to dimensions d > 3 which relates the size of L*'-norms of eigcnfunctions for 2 < p < '^^^1^^^ to the amount of L'^-mass in shrinking tubes about unit-length geodesies. The proof uses bilinear oscillatory integral estimates of Lee [22] and a variable coefficient variant of an "e removal lemma" of Tao and Vargas [35] . We also use Hormander's i20, oscillatory integral theorem and the Cartan-Hadamard theorem to show that, under the assumption of nonpositive curvature, the L^-norm of eigenfunctions ex over unit- _ 1 length tubes of width A 2 goes to zero. Using our main estimate, we deduce that, in this case, the L^'-norms of eigenfunctions for the above range of exponents is relatively small. As a result, we can slightly improve the known lower bounds for nodal sets in dimensions d > 3 of Colding and Minicozzi |10| in the special case of (variable) nonpositive curvature. In memoriam: Lars Hormander (1931-2012) 1. Introduction and main results Let (M, g) be a smooth, compact boundaryless Riemannian manifold of dimension d > 3. Let Ag be the nonnegative Laplace-Behrami operator and consider eigenfunctions ex satisfying A^ca = A^e^ with A > 0. If 77 denotes the space of unit length geodesies and dz the volume element associated with the metric (7, then our main result is the following generalizations of [27l Theorem 1.1]: Theorem 1.1. Let e\, X > 1, be an eigenfunction and ^'■'^^^^ < Q < "^^d-P • '^^^n there is a uniform constant C < 00 so that given e > we can find a constant so that (1.1) ||e.||l,(,,) <£A'(^)(^-i)||e.||^.(,,)+C||e.||^.(,,^ q-2 + aA'(^)(^-|)||e,||2 supn \ex{z)\'dz] , -yen \ir^_i/2(7) / if (1.2) rA-i/2(7) = {x e M : d,(x,7) < A'^} denotes the X^^ -tube about 7, with dg{ ■ , ■) being the Riemannian distance function. The first author was supported in part by the National Science Foundation grant DMS-1001529, and the second by the National Science Foundation grant DMS-11069175 and the Simons Foundation. 2 M. D. BLAIR AND C. D. SOGGE Corollary 1.2. The following are equivalent for any subsequence of -normalized eigen- functions {ex^^j'^^^^: (1.3) limsup sup / \exj {z)\'^ dz — (1.4) limsup A;^^^^"^ Va,J|lp(M) =0 for any 2 < p < 2{d+2) d converse, observe that Holder's inequality gives ^ <p<^. The remaining values of p then follow from interpolation. For the Proof of Corollary. Given Theorem |1.1[ it is routine to verify that (1.3) implies (1.4) for aining values of p then follow froi s inequality gives |eA(z)pdz<A-(^)(i-i)||eA||L(M) X and the implication follows. □ In the case when {M, g) has nonpositive sectional curvatures, we shall be able to show that ( 1.4 1 holds for the full sequence of eigenvalues and hence extend the two-dimensional results of the second author and Zelditch [32 to higher dimensions: Theorem 1.3. Let {M,g) be a compact boundaryless manifold of dimension d > 2. Assume further that (M, g) has everywhere nonpositive sectional curvatures. Then if = Ao < Ai < A2 < A3 . . . are the eigenvalues of -y/Ag we have (1.5) limsup sup / \sXj\ dx —Q. Aj^oo \ -ten Jt^-i/2{i) Consequently, if 2 < p < ^^jz-p-, we have, in this case, (1-6) limsup A7^^^"^'||eA,llL.(M) =0. Xj -^00 In [5S] the first author showed that ||eA||L9(Af) ~ 0(A^~^2^p)) when 2 < p < "^^jz^j and that these estimates are sharp on the standard sphere S'^ because of the highest weight spherical harmonics. We should point out that for the complementary range p > "^^^^P improved L^-estimates under the above curvature assumptions follow, by interpolation from the aforementioned p = ^'^'^^P and an improved L°°-estimate which is implicit in Berard [T] (see also the second author and Zelditch [53] and [IH])- Hassell and Tacy |18| have recently obtained further results for this range exponents. Improvements for p > '^'"^^P are a bit more straightforward than (|1.6[) due to the fact that everything follows from pointwise estimates, while, to obtain ( |1.5[ ) and consequently ( 1.6 ), we have to use oscillatory integrals and a finer analysis involving the deck transforms of the universal cover. We should point out that there are no general L^-improvements for the endpoint p — '^'"^^P of the results in [53], which on the sphere are saturated by eigenfunctions concentrating at points as well as ones concentrating along geodesies. KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 3 As noted before, the special case of c? = 2 of Theorem 1.3 is in 32J. When d = 3, if one assumes constant nonpositive curvature, (1.5) follows from recent work of Chen and the second author who showed that if ds denotes arc length measure on 7, then (1.7) sup \exfds o(A) as A In dimensions d > 4, Burq, Gerard and Tzvetkov ^ showed that one has the following bounds for geodesic restrictions (1.8) \ex\' ds ^ 0{X'-') Improving this to o{X^ ^) bounds as in (1.7) for d = 3 is not strong enough to obtain (1.5) when d > 4. This comes as no surprise since, in these dimensions, (1.8 1 is saturated on the round sphere not by the highest weight spherical harmonics which concentrate along geodesies, but rather zonal spherical harmonics, which concentrate at points. By our main result, Theorem |1.1[ we know that (1.5) is relevant for measuring the size of L^'-norms in the range 2 < p < ^^j^p-, which are saturated on S'^ by highest weight spherical harmonics. These eigenfunctions saturate the Kakeya-Nikodym averages in (1.5), by which we mean that the left side of (1.5) is ri(l), but they do not saturate the restriction estimates (1.8 1 for d > 4 Fortunately, we can adopt the proof of the aforementioned improvement (1.7) of Chen and the second author |5j to obtain (1.5) in all dimensions under the assumption of nonpositive curvature. Additionally, even for d = 3, unlike the stronger estimate (1.7), our techniques do not require that we assume constant sectional curvature. Let us conclude this section by recording some applications of Theorems |1.1| and |1.3| First, using (1.5) we can improve the lower bounds for L-'^-norms of the first author and Zelditch |30j under the above assumptions: Corollary 1.4. Let (M, g) be a d- dimensional compact boundaryless manifold with d> 2. Then (1.9) liminf A^lleAllLi(M) = 00. \—^ao As pointed out in |30j . no such improvement is possible for the sphere. The proof of (1.9) is very simple. For, by Holder's inequality, if p > 2, whence As a result. (A- meaning that (1.6) implies (1.9) Let us now see how ( 1.9 1, along with an estimate of Hezari and the second author [15] improves the known lower bounds for the Hausdorff measure of eigenfunctions on mani- folds of variable nonpositive curvature. 4 M. D. BLAIR AND C. D. SOGGE To this end, for a given real eigenfunction, e\, we let Za = {x e M : ex{x) = 0} denote its nodal set and T-L'^~^{Zx) its {d — l)-dimensional HausdorfF measure. Yau [55] conjectured that 'H'^-^{Zx) « A. This was verified by Donnelly and Fefferman |14j in the real analytic case and so, in particular, if (Af , g) has constant sectional curvature. The lower bound 7i'^^^{Z\) > cA was verified in the C°° case when d = 2 by Briining [5] and Yau, but much less is known in this case. An upper bound 7i'^^^{Zx) = 0{X^) is also known by Dong [T3] and Donnelly and Fefferman 15 when d ~ 2, but the best known upper bounds for d > 3 are 'H'^-\Zx) = 0((cA)('=^)), which are due to Hardt and Simon [T7] . Until recently, in higher dimensions for the C°° case, the best known lower bounds for H'^^^{Zx) were also of an exponential nature (see [16]). Recently, Colding and Mini- cozzi pill and the second author and Zclditch [30] proved lower bounds of a polynomial nature. Specifically, the best known lower bounds for d > 3 in the C°° case are those of Colding and Minicozzi [TU] who showed that (1.10) cx^-"^ <n'^-\Zx). Subsequent proofs of this using the original approach of the second author and Zeldtich [30] were obtained by Hezari and the second author [19] and the second author and Zelditch [31] . The latter works and the earlier one [5^ were based on a variation of an identity of Dong J3J. The proof of (1.10) in [T^] was based on the following lower bound (1.11) cA ex\dx] <W'-\Z, Indeed, simply combining ( [l.llj ) and the i^-lowcr bound of the second author and Zelditch [3D] (1.12) c\- < yields (1.10 1. Similarly, by using the improvement (1.91 of (1.12), we can improv^the known lower bounds (1.10 1 under our assumptions: Corollary 1.5. Let {M,g) he a compact boundaryless Riemannian manifold of dimension d > 3 with nonpositive sectional curvatures. Then ■ " n''-\Zx) = oo. (1.13) liminfA A— T-oo In particular, when d = 3, H^i^Zx) becomes arbitrarily large as A —>■ oo. By a simple argument (see [27]) one always has ( 1.3 ) and consequently ( 1.4 1 as A ranges over a subsequence of eigenvalues {Aj,, } if the resulting eigenfunctions form a quantum ergodic system (i.e. \exj^ \ '^dx converges in the weak* topology to the uniform probability measure dx /Wo\g{M)) . Consequently, by the above proof, we also have the following ^An alternate approach, which yields the same sort of results, would be to use | |1.6| l to improve the conclusion of 10, Lemma 4] under the assumption of nonpositive curvature. KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 5 Corollary 1.6. Let {ex^^} be a quantum ergodic system on a compact Riemannian man- ifold of dimension d > 3. We then have (1.14) lim A fc— >C30 3 k In particular, if the geodesic flow is ergodic, we have (1.14) as {Xj^} ranges over a subsequence of eigenvalues of density one. The last part of the corollary follows from the quantum ergodic theorem of Snirelman |23j / Zelditch f39! / Colin de Verdiere yj] (see also [28]). This paper is organized as follows. In the next three sections we shall present the proof of our main result, Theorem |1.1| In §2 we shall go through the essentially routine step of reducing matters to proving certain bilinear estimates, and this step is very similar to the argument for the two-dimensional case of one of us j27j . It gives partial control of the left side of (1.1) by the last term in the right. The needed bilinear estimates, which lead to the first term in the right side of ( |1.1[ ) are then presented in §3 and §4. In §3 we show the bilinear estimate we require follows, up to an e loss, from one of Lee |22l Theorem 1.1]. We then are able to remove this loss in §4 using a variable coefficient version of the "e- removal lemma" of Tao and Vargas ^Hl Lemma 2.4] (see also Bourgain [2 ). Then, in the final section, §5, we prove Theorem 1.3 which says that we have o(l) bounds for L^-norms over shrinki ng t ubes under the assumption of nonpositive curvature, and consequently, by Theorem 1.1 improved LP(M)-norms for 2 < p < '^'"^^P of the estimates in |25) . 2. Reduction to oscillatory integral estimates In this section, we begin the proof of Theorem |1.1[ reducing matters to estimates on oscillatory integral operators. Let xx denote the operator x{\/^g ~ X), where x is a smooth bump function with x(0) = 1 and sufficiently small compact support. Hence X\e\ — e\- Recall (see Sogge, Chapter 5 [26]) that the kernel of this operator can be written as XA/(2) = x(v^-A)/(z) = A^ / e'^^^^'-'y^ax{z,y)f{y)dy + Rxf{z) Jm where ax{z, y) is supported in (5 < dg(z, y) < 26 for some S > sufficiently small and less than half the injectivity radius of {M,g). Moreover, ||i?A/||L'!(M) ^ ll/||L2(Af)- Using a sufficiently fine partition of unity, we may assume that the support of a> is sufficiently small. In particular, we may assume that supp(aA) C {|z — zo| + |y — ?/o| ^ £a} for some points zo,yo S M with \zq — yol 6. Let 70 denote the geodesic connecting zq, yo and suppose that E is a suitable codimension 1 submanifold passing through yo such that 7o is orthogonal to E. Now let {t, s) € x M denote Fermi coordinates for E with (0,0) = yo, (0, s) parameterizing 70, and (t,0) parameterizing E. This means that for any fixed to, {to, s) locally parameterizes the geodesic passing through (to, s) orthogonal to E. 6 M. D. BLAIR AND C. D. SOGGE It suffices to prove that |/(z)|«-2rf2 ds < £ (a^(^-|)||/||^,(,,,)'||/||«; -2 (M) + C,A'(^)(|-i)|j/| L^(M) sup ',en \Jr^_i/2(7) \f{z)\^dz Indeed, using Young's inequality for products applied to the Holder conjugates |, we may absorb the contribution of e^''~^^'''||/||^7^M) fro™ the first term into the left hand side, for e sufficiently small, yielding (1.1) when f — e\. Moreover, it suffices to prove that for each s the expression in parentheses on the left hand side is bounded by the right hand side. For convenience, we will show this for s = as the argument below works for any value of s and does not use the structure of S once Fermi coordinates are given. Fix A and let Th{z) = / axiz,{t,0))h{t)dt where ip{z,t) = dg{z,{t,0)). We will show that (2.1) J |A^rMz)n/(z)r2 dz < s (A^(^-i)||/.L.)' 11/11^7 + Ce\'^\\h\\l2 sup jen Jr. -i/2h) \f(z)r'dz. Holder's inequality with conjugates ^p2i "^ill then imply that A" r,_i/2(7) |/(z)r-2dz< A^-(^)(2-|) r._i/2(7) \f{z)\'dz and it is verified that the exponent of A on the right is the same as the one in (1.1 ). Observe that iTh{z)f = [ e'^'^'f'^''''^+'''^'''''^^ax{z,t)ax{z,t')h{t)h{t') dtdt' . Suppose Eq is a small dyadic number such that supp(Q;;^(z, •)) C [— Eoj^o]'' for all z. Let > be a sufficiently large dyadic number (which will essentially play the same role as the integer TV in [371 (2-5)]) and let jg be the largest integer such that 2~^" > A~2. Take a Whitney-type decomposition of [— eoi^o]'' x [—£ot£o]'^ away from its diagonal D into almost disjoint cubes -'eo,eof X [-eo,eof\D I u U X ^i' U Qi° X Qi' KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 7 where each Ql has sidelength 2 ^ and is centered at a point v E 2 ^1/ ^ . Set hi{t) — Iqj {t)h{t) where the first factor denotes the indicator of the cube Ql- Hence (2.2) {Th{z)f= E Thiiz)Thi,{z)+ Thiiz)Thi,{z) where 2^ denotes the collection of {v, v') indexing the cubes satisfying d{Ql, Qi,) ~ (or < N2-^'> when j = jo). Theorem 2.1. Suppose T — T\ is the oscillatory integral operator defined by Th{z) := / e'^'''^'''-*'^axiz,s,t)h{t)dt where ax is smooth and supp{a\) is contained in a sufficiently small uniform compact set and whose derivative bounds can be taken uniform in A. Assume further that (f)(x,s,t) satisfies a Carleson-Sjolin type condition that V^^^ is invertible and that if is a unit vector for which Vf (V(a;,s)0, ^) = 0, then (2.3) (V(^ 0) has eigenvalues of the same sign. Then (2.4) < 2?'^ '- r<l/2 It can be verified that setting z — (x, s) e M x M, the phase function in question (l){x,s,t) :— dg{{x, s), {t,0)) — 'ip{{x, s),t) satisfies the Carleson-Sjolin condition given here. Moreover, our assumption that q < "^^'^^P ensures that the exponent of 2^ in (2.4) is positive. Hence this estimate yields E E Thi Thl d-1 d-1 , h\\h LV Since Holder's inequality with conjugates |, and the triangle inequality yield ^'"'/|E E ThiTKi,'^\fr^dz<x''-^Y. E T^Thi, I/I 9-2 LI LV the contribution of this sum is bounded by the first term on the right hand side of \2.1 by taking N suitably large. This estimate can be considered as analogous to H?! (2.6)] Our main tool in proving (2.4) will be a bilinear estimate due to Lee [221 Theorem 1.1] along with a refinement of arguments in §3 of that same work. Indeed, the estimate ( |2.4[ ) should be compared with Lemma 3.3 and (3.3)]. In [52], the author proves bilinear estimates which can be thought of as a variable coefficient versions of bilinear restriction estimates due to Tao [34] for elliptic surfaces (inspired by prior work of Wolff [37 and Tao- Vargas-Vega [36]). Lee then showed that these bilinear estimates in turn implied linear estimates on oscillatory integral operators whose phase function satisfies the Carleson- Sjolin type condition (2.3) (more generally called the "Hormander problem"). However, his estimates suffer losses when compared to the optimal estimate predicted by scaling. 8 M. D. BLAIR AND C. D. SOGGE In the present work, we cannot afford such losses. Hence one of the central tasks in this work is to prove a variable coefficient version of the e-removal lemma for bilinear estimates in [33 Lemma 2.4] (see also Bourgain [2]) and refine the almost orthogonality arguments in [22l §3]. We now turn to the second sum in (2.2); since 2 A 2 it will be treated essentially the same way as in i2Tj p. 527-9]. Observe that Thi{z)Thi,{z) E <N^-'J2\Thio{z)\' The main estimate for this term is then (2.5) \\'^Thio{z)\'\fiz)r'dz<x'^\\hio\\] r2 sup A-l/2 (7) \f{z)r^dz Since Y.AK\\ll = \\h is bounded by the last term in (2.1 ) we may sum in v to see that the contribution of these terms To see (2.5), we will use geodesic normal coordinates centered at the point on M corresponding to (i^, 0) in the Fermi coordinates (recall that v E 2^^°l/^^) and let x ^ k{x) denote the diffeomorphism which makes this change of coordinates. We may assume that k{i',s) = (0, s) (parameterizing the geodesic orthogonal to S through (i^, 0)). We now let {uJi}i denote a A~ 2 -separated collection of points in a neighborhood of (0, ... , 0, 1) on S"*"^ indexed by a subset of Z''^^ so that Now let Si z FT " ^' and observe that the left hand side of (2.5 1 can be dominated by 2 L~(K-i(Si)) I ll/llL9-2(K-i(s,)nK) < sup I k I where the zi are chosen to maximize \T{h^°){z)\ as z ranges over k~^{Si) and if is a small set containing the a;-support of a\(x,y). It thus suffices to see that for some suitable bump function -0, E e'^^^'-'^axizu (0, t))^0 (t - v))K{t) dt After a translation in i, it suffices to assume that v — and the desired — P estimate follows from the one dual to (2.7) below Theorem 2.2. Suppose ip(z,t) is as defined above and p{z,t) is a smooth bump function satisfying \d"p{z,t)\ <q A^^ and supp{p{-,z)) C {\t\ < A^s}. Assume also that p vanishes when z is outside of a small neighborhood M of (sq, 0) with sq ^ S with 6 > KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 9 (in the Fermi coordinates described above). Let zi be a collection of points in M indexed by Z"^^^ such that whenever \l — k\ is sufficiently (2.6) Then (2.7) (Ki(zi), . . . ,Kd-i(z;)) (i^lizk), ■ ■ ■,Kd-lizk)) IE > A-5|?-fc|. j\i>{zi.t) p{zut)ai dt <y^^ ai\ The proof of (2.7) is the same as the one in pTl Prop. 2.3], once it is observed that \Vt^{zi,o) -Vti^{zk,o)\ > x---\i ~ k\. But since the pushforward of d/dzd under z i— > k(z) is itself, this is a consequence of (2.6) and the identity "0(2,0) ^ {Ui,K{z)/\K{z)\), I = 1 where Vi is the pushforward of d/dzi. 3. Almost Orthogonality In this section, we begin the proof of Theorem 2.1 We first appeal to [22, Lemma 3.1] (which follows results of Bourgain [4j and Hormander ^D]) and the ensuing remark, which states that after a change of coordinates and multiplying Th, h by harmless functions of modulus one, we may assume (3.1) where (3.2) {x, s,t) = X • t + 2^\^?' + '^(•^' E{x,s,t)^0{{\x\ + \s\f\t\^)+0{{\x\ + \s\)W). Let be a smooth bump function supported in [—1, l]** ^ satisfying '}2keV-^ ip {x — k) ^ 1 and set A^{x) = -02(2-'(x - with ^ e 2'^''-^ Lemma 3.1. Suppose 1 < p < 2 and that T is as in Theorem \2.1\ There exist amplitudes au,fj., both with x-support contained in supp{A^) and satisfying derivative bounds of the form (3.3) |a>,,^(:r,s,i)| 2^'l"l such that if T^^^ is the oscillatory integral operator with phase 4> o,nd amplitude a^.^t T,,^(/i)(x, s) = / e'^*(^'^'*)a,,^(x, s, t)h{t) dt then LP( 10 M. D. BLAIR AND C. D. SOGGE Proof. For a given s, consider the slice of T{h) at s T^(/i)(x) = T{h){x,r)\r=s- It suffices to show that p A, E T%hi)T%gi,: LP( and hence we shaU assume that s is fixed throughout the proof. Now let ^{x^t,t') 4>{x,s,t) + (j){x,s,t') and observe that A^T'^ {hl)T'' (gl,) can be written as M^) I e'^'^'-^'^'^'^aix, s,t)aix, s,t')hi{t)gl,{t') dtdt\ Treating — —i^x as a vector- valued differential operator we want to write (3.4) (1 + {\-^Vf\\^x^{^,v,v') - DxlTA.T^hDT'^igi,) = nJhi)T^,(gi,) for some N large based on d and each operator on the right satisfies (3.3). It thus suffices to see that this can be done for any monomial of which in turn will follow by induction. To this end, observe that products of functions satsifying (3.3 1 satisfy the same condition as do weighted derivatives {cd^)" of such functions provided |c| < 2^^. On supp(74^) x Ql x Q^, we have that satisfies (3.3). Moreover, since A ^2^ < 2 it is seen that for any a, (A 2^dx)"A 1/2 satisfies (3.3). The claim then follows. It now suffices to see that if P^^,y' is the Fourier multiplier PuyiDx) = (1 + iX-'2^f\XVxH^^,,y,ly')^Dx\Y'', then for any sequence of {f,jy} of Schwartz class functions defined on M^"-"^, < E wf'^yW^^ The latter follows from the triangle inequality and Young's inequality for convolutions, so it suffices to treat the former. But V:r$(^, i^, z^') = 2\7x4'{fJ-,s,i^) + 0(2^-'), so the invertibility of V^0a:,t gives Recall that for each i/, the number of ly' such that {v,v') G is 0(1). Therefore since the V range over a regularly spaced 2~^ lattice, the desired bound follows from a routine computation using Plancherel's identity. □ KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 11 Returning to the proof of Theorem 2.1 fix a pair {i',!^') G Ej. Set hi{t) ~ hl{2^H), o-j.i^.fj-ix, s,t) — a^^^{x, s,2~H), (j)j{x,s,t) — 2^ (j){x, s,2^H) so that rescahng variables t !-> 2^H in the integral defining Ty^^{hl){x, s) yields T,^uAh^){x,s) j e'^^''^^^^''^'^a,,^{x,s,t)h,{t) dt = 2^^''-^^T,^^{hi){x, s). Also set ft-2(t) — h''^,{2~H) and define Tj_i,i ^^{h2){x, s) analogously, noting that (f)j remains independent of v, v' . Moreover, we may assume that aj ,^^^{x, s, ■) (resp. aj^^i ,fj_{x, s, ■)) is supported in a slightly larger cube containing supp(/ii) (resp. supp(/i2))- It is helpful to observe that given (3.1), (|3.2|) 0j (x, s,t)=x-t + 2-^-\s\t\^ + 2' Six, s, 2-H). Lemma 3.2. There exists an amplitude dj ,^^^{x,s,t) satisfying hounds of the form (3.3) such that Proof. Observe that Since A ^2^-' < 1, second term satisfies (3.3). Moreover, by (3.1), (3.2) {dt^ (t>j {x, s,t) - Hk) = Xk - Hk + 0{2^^ ) and thus by the support properties of Oj^jy^^ the first term satisfies lemma then follows by an inductive argument akin to that in Lemma 3.3) as weh. The 3.1 □ Given this lemma we let — P^{Dt) be the Fourier multiplier with symbol P^{C) (1 + 2'^^\\-^2^C + Aip)"^ and observe that by self-adjointness of P^{-Dt), we have aj,^,fj. {x,s,t) {Pf_,hi ) (t) dt r_,>,^(ft.i)(a;, s) Thus if we can show that _ 2d 2j(d + l) (3.5) \\TJ,^,^,{hl)Tjy^^{h2)\\^^^^^^<\ "2 i ||P^/li||i2(iid-i)||P^/l2||i2(Kd-i) taking a sum with respect to ii and applying Cauchy-Schwarz will give 1 2 / \ 2 L2 (Rd) ^ , , 2d 2j(d+l) \ 2 < \^~2^r- \ /J and by almost orthogonality of the P^hi, (^^ ||P;^^i||^2)' ^ ||^i|l22- Rescaling therefore yields (3.6) Hence Lemma 3.1 and Cauchy-Schwarz mean that the left hand side of (2.4) is dominated by L2(Rd-l) 12 M. D. BLAIR AND C. D. SOGGE The desired estimate (2.4) now follows from the embedding £^ ^ We are left to show (3.5). At this stage, (i(supp(ft.i), supp(/i2)) ~ 1, but we want to exhibit the uniformity of the phases and amplitudes. To this end, observe that 1 s (j}{x,s,t + i/) (x + siy) ■ t + -s\t\^ + E{x,s,t + v) + -\v\^ + x ■ v. The last two terms here can be neglected. A Taylor expansion gives S{x, s,t + ly) ~ £{x, s, v) + Vt£{x, s,v) ■ t + ^ ^ d^£{x, s, v)t'^ + i?^(a;, s, v). \a\=2 As observed in [32 (3.9)], we may change variables y = x + siy + WtSixjSjiy) and, neglecting terms which can be absorbed into either T{hi) or hi, we can write 0(y, s,t + i')=yt+ is|tp + £^{y, s, t), where £^{y,s,t) will also satisfy (3.2) (with y replacing x). Hence 0j (y, s, t + Vv) = 2^V)(y, s, TH + v) = yt + 2-'-^s\t\'' + V£,{y, s, ^^H). Also define Us = -'r sv -\- Vjf (/i,s, z^) (recalling that /i is the center of the a;-support of 0'j,v,fi, and observe that linearizing the change of coordinates gives that if |a; - ^1 < 2"^, then |y - (TsI < 2~-''. We next set 4>{y, s, t) = 22^V>,(2-^y + ct„ s, i) = y • < + ^sjtp + 2''^£,{2-'y + a„ s, + v) and define Ti(5i)(2/,s) = 1 e^^2-=^^(^'^-*)&,-,,^(2-^y + a„s,t)gi(t)di and 12(92) ill the same way except with amplitude (ijy ^^{2^^y + as,s,t). The bound (3.5) will then follow from (3.7) ||Ti(gi)r2(.92)ll^j(j,,) < (A2-2^")-^I1.9iI1l^(k-i)I152|1l2(k-i). This estimate in turn follows from one of Lee [221 Theorem 1.1] along with e-removal lemmas in the next section. We state this using his hypotheses. For i = 1, 2, let Ti be oscillatory integral operators T,f[z) = j e'^^^^'-'i'^a,{z,^)f{Od^ z = (a;, s) G M'^-i x M, ^eW"-^ with fli smooth and of sufficiently small compact support. Assume that V^^0i has rank d — 1 and that ^ f->- V x4>i{x, s,S^) is a diffeomorphism on supp(ai). Take qi{x,s,^) — ds4>i{x, s,[Vx(t>iix, s, ■)]~'^{^)) so that dscj^iix, s, £,) = q.i{x,s,Vx(t>i{x,s,^)). Suppose fur- ther that V^^qi{z,Vx(f>i{z,£,i)) is nonsingular for {z,£^i) S supp(ai). Theorem 3.3. For i — 1,2, ai, (pi satisfy the hypotheses outlined in the preceding dis- cussion. Set Ui — VxPiZjCi) and S{z, ^1,^2) ^'^^qiiz,ui) — V^q2{z,U2). Then if (3.8) I {Vl^Piz, ^,)S{z, Ci, 6), [V^5'^(z, ^^)]-'[Vy^{z, u,)]-'S{z, 6, 6)> I > O for i = 1,2, then for any ^-j^ < p (3-9) \\TifiT2f2\\LP(m'i) ^ ll/l||L2(Rd-i)|l/2|lL2(Rd-i)- KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 13 Moreover, if Ti, T2 are members of a family of operators whose phase and amplitude functions satisfy these hypotheses uniformly and are uniformly bounded in C°° with am- plitudes supported in a set of uniform size, then the implicit constant in (3.9) can be taken independent of each operator in the family. We postpone the proof of this theorem until the next section. It is then verified (see [22 (3.14)]) that if one takes — t, z — [x, s), 4>{x, s, t) = 0i(x, s, t) = 4'2{x, s, t) and oi, satisfies 6I+O(eo) + 0(2-^). ^2! ~ 1, the desired bound follows by taking (3.9 1 with A replaced 02 as the amplitudes in Ti , T2 respectively, then the left hand side of (|3 la Therefore since \ti by A2-2J . Remark 3.4. As a consequence of Theorem |3.3| and the almost orthogonality arguments in this section, we obtain the bound d-l (3.10) \\Th\\Mn^)<\—^\\h\\L.( 1) 2{d + 2) rf- when q > and — 1 < q for operators T satisfying the hypotheses of Theorem 2.1 In other words, we obtain Lee's estimate (22 Theorem 1.3] without the e-loss. Indeed, the Whitney-type decomposition of {TKf' in the previous section is essentially the same as that in his work, and the estimate over the {v^ v') S is treated on p. 85 there. Since Holder's inequality gives \\K\\l'^{w above yield the following variation on (|2.4) < 2 2 1.2 p ' 1 1 /ij 1 1 (jfd^ij, (3.61 and the almost orthogonality arguments E Thi Thi, <2.(^-('^-i)(i-|))A-f||/.||i,,,._,, (since it suffices to treat the cases where q> p). Taking a sum in j then yields (3.10) We also note that when p = 00, the estimate in (3.10) is valid for a larger range of q by a recent work of Bourgain and Guth [S] . 4. The e-removal lemma Turning to the proof of (3.9), the estimate (4.1) ||ri/iT2/2||L9(R'i) < CqA ''^"||/l||L2(Hd-l)||/2||i,2(Rd-l) for arbitrary a > and < q is due to Lee [22l Theorem 1.1]. Moreover, as observed in [221 p. 88], the constant is stable under small perturbations in and 0^. In particular, if families of amplitudes and phase functions are considered and these functions are uniformly bounded in C°° then Ca can be taken uniform within the family of operators. The rest of this section will be dedicated to the following lemma, a generalization of Lemma 2.4] which completes the proof of Theorem 3.3 Lemma 4.1. Suppose Ti, T2 satisfy the hypotheses of the previous theorem and that they satisfy the estimate (4.1 ) for some 1 < q < and some a > 0. Assume further that . , 1 / 8a \ 1 4a 14 M. D. BLAIR AND C. D. SOGGE Then the scale-invariant estimate ||Ti/i'72/2||l'-(K'') ^ ll/l||L2(Rci-l)||/2||L2(Rd-l). is also valid for any r > p. The hypothesis (4.2) is stronger than the one appearing in [35] (corresponding to o- = there) but is sufficient for our purposes. Let /i, /2 be unit normahzed functions in By a Marcinkiewicz interpolation argument, it suffices to see that[^ \{x:\TJ,ix)T,h{x)\>P}\<X-^r''- Denote the set on the left by E. Observe that since IIT1/1T2/2II00 ^ li it suffices to assume that P < 1. Hence we may assume that \E\ > X''^ throughout since the desired bounds are guaranteed otherwise. Moreover, we know from (4.1) and Tchebychev's inequality \E\ < X-'^+^T". _ qa Consequently it suffices to assume that /? > A p-i . This gives the a priori bound (4.4) \E\<X-'^+'"'^^+^^l Since f3\E\ < IIIbTi/iTs/zIIli , it suffices to show that UETihT2f2\\L^ <X-i\E\V. We deduce this by showing that for any unit vectors gi, 52 in L'^{M.'^^^) _ d j_ ||l£;Ti.9iT'2g2||Li < A p\E\p' , where it should be stressed that E is dependent on /i,/2 above, but that gi, g2 are completely independent of these functions. Fix g2 and let T — Te^q^ be the linear operator Tgi = lET2g2Tigi. It suffices to show that \\T*F\\L2^^d-i^ < A"p|£;|?^||F||i^(Kd), since duality then implies that ||rgi||^i < X^p\E\p^. We may assume < 1. Set F := \ET2g2F. By a duality argument, we square the left hand side of the previous inequality to see that it suffices to show that (4.5) mTlF,F)\<X-'^\E\V ^X-^'iX^mv, where the inner product on the left is with respect to L?'{W^). The integral kernel of TiTl is K{w,z)^ f e^^('^i(-^«)-^i(-'«))ai(w;,e)ai(z,Orfe ^One sees that this inequaUty imphes the lossless bilinear inequalities for each exponent r > p since if t»;(/3) = \{x : |ri/i(a;)T2/2(x)| > /3}| then = for /3 larger than a fixed constant and / \T1hT2f2rdx = r f}'-''uiif3) dp. KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 15 and satisfies estimates \K{w,z)\ < {1 + X\w - zl)-"^ . This bound follows from the invertibility of V|^9s0i when w — z is inside a small cone about (0, ...,0, 1). Otherwise, stronger estimates result from integration by parts and the invertibility of Wx^4>i- We now let R > be a parameter to be determined shortly and write K{w, z) = K^{w, z) + Kfj{w, z) where K^{w, z) is smoothly truncated to \w — z\ > R and Kii{w,z) is supported in |w — < 2R. Observe that by Stein's generalization of Hormander's variable coefficient oscillatory integral theorem (see [33] or [Si Ch. 2]) \\F\\L^(m.'^)<\E\^^\\T2g2\\ ||i^||L~(E^) < |£^|™A^wy. Thus the contribution of to {TiT^F, F) is bounded by (Ai?)-^|£;|^A-^ = {XRr'-^iX^lEDmx-^d. It is now verified that taking 2 f d+3 2_\ R^ X-'^ {X'^IE])"-'^"^' ''''>X-\ ensures that the contribution of is acceptable towards proving (4.5) (by scaling, this is consistent with the choice of R in [351 Lemma 2.4]). We also remark that another computation reveals that (4.4 1 along with the hypothesis (4.2) ensures that i? < A~2 . It remains to control the contribution of Kji to (4.5). Let {4>k}k be a partition of unity over [~eo,£o]'^ such that supp('0fc) is contained in a cube of sidelength 2R centered at a point Wk G RZ'''. Let Pr be the operator determined by the integral kernel Kji and observe that its contribution to the left hand side of ( |4.5[ ) is dominated by (4.6) ^|(Pfl(V'fci?),^fc'F)|. k,k' Given a fixed k, the number of k' for which {PR{'4'kF),ipk' F) ^ is 0(1) and satisfies (i(supp(V'fc): s,upp{4>k')) ^ R- Hence we will restrict attention to the sum over the diagonal k = k' , as a slight adjustment of the argument below will handle the off-diagonal terms. At this stage it will be convenient to use the semiclassical Fourier transform with (cf. 40, §3.3]) (4.7) J,/,{G){r,)^ e-^'-^Giw)dw, J-i^(g)(^) ^ / e^A^"'G(ry) dry. Since S'l/x is related to the usual Fourier transform by J'l/^ (G) (77) — 3^(6) (Ary), we have the Plancherel identity (27r)''||G||2 ^ = A''|| Ji/a(G)|||2 (cf. (3.3.6)]). We now have (4.8) {2TTf{Pn{i:kF),^kF} = A'^ {^i/xiPRii'kF)) ,J,/^{i;kF)) , and the right hand side can be written as 1 {2ny Jkiv, C)^i/A i^kF) (C) ?i/A i^kF) (7?) dCdrj, 16 M. D. BLAIR AND C. D. SOGGE where for some ipk supported in \z — Wk\,\w — Wk\ ^ R satisfying |92 ^V'fcl i?^'"'. Strictly speaking, one needs to justify the use of Fubini's theorem here, but this can be done by passing to Schwartz class approximations to F and employing crude continuity bounds for Pa. Therefore over supp('(/'fe), |V„0iK.,O - V^(/.i(u.,^)| + |V,0iK,e) - V,./)i(z,OI < i? < (Ai?)-\ where we use that i? ^ A~2 in the second inequality. Hence integration by parts gives for any TV and some uniform cube Q C M''^^ JQ as the domain of integration in (w, z) is of volume R^'^. Let Sl denote the hypersurface {V0i(?i'fc,^) : ^ € Q}. This in turn allows us to deduce that \Jkiv, 01 <A^ {XRf+\l + XRdiv, Sl) + XRdiC, sl) + XR\C - ryl)-^. Consequently, by using Cauchy-Schwarz in (4.8 1 we have that {PRii:kF),iJkF) <nXR [ {1 + XR\d{Tj,Sl)\)-^\J,/^{^PkF){ij)\' drj. Now let Sl I denote the (Ai?) ^2' neighborhood of Sl- We have that k k 1=0 We examine the case I = 0, the other cases are similar and aided by the factor of 2~'^. Let {gi,/c}fc be a sequence of functions with supp(gi^fe) C Sl ^ for each k and k\ r2rc2 1. To finish the proof and show that (4.81 is dominated by the right < X-p{XR)--2\E\V\\F\\j side of (4.5), it suffices to see that ^|(^fci?,A-%\(gi,fc)) k which in turn follows from J2 f \^klEX-'%\{gi,k)T2g2\<X-HXR)---\E\V. k •' Now reverse the roles of gi and g2 from the previous step, treating {gi^k}k as a fixed sequence and redefine T = TE^{g^ ^} by Tg2 = {i^k'i-E^i^x(9'^^k)T2g2}k so that it suffices to show ITII < X''~^iXR)-^\E\V Let {Fk}k be any sequence of functions satisfying supj, ||i^fc||ioo ';/'fel_EA~'^3'^A (gi.fe)-Ffc- By duality, the desired bound on T will follow from < 1 and set Fk <X-piXR)-^\E\V KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 17 or equivalently k,k' Observe that (4.9) J2 WFkh^iR^) < J2 ll^'=IU-(K^) / |^a£A-%\(gi,fc)l 1^ 7, dw I f 2(d + l) \ ^('' + 1) By finite overlap of tlie supp(^/;fc), tlie first factor on tlie riglrt is seen to be bounded by d + 3 \E\ 2(<J+i) . Similar to before, an application of the Stein- Tomas theorem for Sl gives ■'k (cf. the formula for 3"]^/^ in (4.7)) where we use that ^ l/^"^ ^ . Decomposing the integral kernel of T2T2 as a sum + Kj^ as before, we may handle the contribution of by using (4.9 1 to reason analogously to the argument above. We are thus reduced to handling the contribution of Kj^ and denote the corresponding operator as Pr. As before, we restrict attention to the diagonal terms, and are thus reduced to seeing that k Since supp(Ffe) C supp(i/'fc), this analogously reduces to showing that ^$:2-'^||Ji/,(i?,)|li2(,2^) < (A-|(Ai?)-i|£;|i^) , k 1=0 where this time ; denotes the (Ai?)^^2' neighborhood of the hypersurface — {'V4'2{wk,0 ■ C G Q}- We again restrict attention to the I = case, and let {g2,k}k be a sequence such that supp(g2.fc) C S*^ q and J2k W^^MI'l^is^ ) " ^- Observe that ^|(F,.,A-%\(g2,fe))| <Ell^^^lU~(K^) / |A-'VaB?r/A(5i,fc):^rA(ff2,fe)l dw. and it suffices to show that the right hand side is bounded by A p (Ai?) ^ p' . But each term on the right is bounded by \E\^ Rescaling w H> Rw and applying the bilinear estimates (4.1 1 (or even those in shows the preceding term is bounded by \E\VRHxR)-l+''iXR)-'\\g,M\LHslJ~92,k\\LHSl„)- Taking the sum in k and applying Cauchy-Schwarz completes the proof once we observe that \E\VX-I+°'R°' < \E\VX-i. 18 M. D. BLAIR AND C. D. SOGGE 2 { 2_ d-l \ p d+1 > a_( 2___ d — 1 \ , , Recalling that R « (A'*|£:|) ''-i^^+i equivalent to \E\-'^x-i {X'^\E\y which in turn can be rearranged as {X'^\E\)^^^~^^ < (A'^|i;|)'"^ . But since X'^\E\ > 1, this follows once it is observed that (4.3 1 is equivalent to this inequality is 2a d- 1 \p 1 1 1 < . q p Even though we only need the weaker condition (4.31 to conclude the argument, the stronger hypothesis (4.2) is used above in a significant way to ensure that XR^ < 1. 5. ESTIMATES FOR SHRINKING TUBES By Corollary 1.2 (1.6) follows from (1.5). Therefore, if , as before, 77 denotes the space of unit length geodesies, we must show that if (M, g) has nonpositive sectional curvature, then if e > is fixed there is a < oo so that (5.1) / IcAp dx <e, A > Ae, 7 e 77. Here, as before, we are denoting the volume element associated to the metric simply by dx. We shall first fix 7 e 77 and prove the special case (5.2) / \ex\^dx<e, X>A,. •^T^-l/2(7) After doing this we shall see that we can adapt its proof using the compactness of 77 to obtain the estimates (5.1 ) which are uniform as 7 ranges over this space. To prove these estimates, we shall want to use a reproducing operator which is similar to the local one, xa, that was used to prove Theorem |1.1[ This operator was a local one, but to able to take advantage of our curvature assumptions and make use of the method of time-averaging, it will be convenient to use a variant that is in effect scaled in the spectral parameter. To this end, let us fix a real-valued function p G S(U.) satisfying 1 =Oif \t\ > - and = p(-t). (5.3) p{0) = 1, Then for a given fixed T 3> 1 we have p{T{X~ ^g))ex^ex. As a result, we would have (|5.2|) if we could show that there is a uniform constant A — A{M,g) so that whenever T ^ 1 is fixed there is a constant At < 00 so that for A > 1 we have (5.4) |p(T(A-yA;))/||^,(^^_^^^(^^^< [at-^+AtX-'s^ WfWmM)- KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 19 Since p{T{\ - a/A^)) : L'^{M) L^{M) is self-adjoint, by duality, ( |5.4[ ) is equivalent to the following (5.5) \\piTiX^^j))h\\^,^^^^< (^AT-i +AtX-"^) \\h\\mM), it snpp h c Tx-i/2ij). If we now let (5.6) m(r) = (p(r))^ we can square the right side of (5.5) to see that whenever h is supported in the tube 7>-i/2(7) we have \\p{TiX - ^,))h\\l,^^^^^ = {m{T{\ - ^g))h, h) < \\m{T{X - V^))^llL2C7-^_^/^(^))|l/i|li2C7-^_,^,(-y)). Whence we deduce that our desired inequalities ( |5.2[ ), (5.4) and (5.5) would all follow from (5.7) ||m(T(A-yA;))/i||^,(^^_^^^(^^^ < (ct-s + CtA-^) |!/j|U2(m), ifsupp/icr^-i/2(7), with C and Ct being equal to A"^ and A^, respectively. Since, by (5.3), m(T) - (27r)-i(p*p)(T) is supported in |r| < 1, we can write nT m{T{X-^g)) 1 27rT m(r/r)e*^^e-'^V^ dr. After perhaps multiplying the metric, we may assume that the injectivity radius of the manifold is larger than 10. Let us then fix an even bump function /3 £ C^(M) satisfying /3(t) = 1, |t|< ^, and /3(r) = 0, |r| > 2. We then can split m(T(A - ^A;)) =Rx + Wx where (suppressing the T-dependence) ^ - /?(r))m(r/T)e^^-e-"V^ dr, and, if rxir) denotes the inverse Fourier transform of r — > l3{T)rh{T/T), R\h = T-^rriX - ^/\,)h. Clearly, |r-T(T)| < B for some B independent of T > 1, and therefore, \\Rxf\\mM)<BT-^\\n\v^(M), T>1. As a result, we would obtain (5.7 1 if we could show that (5.8) \\Wxh\\L2^^r^_^^^m < [cT--2 +CTX-i)\\h\\L2, if supp /i c rA-1/2 (7). 20 M. D. BLAIR AND C. D. SOGGE By Euler's formula, if toj^ denotes the inverse Fourier transform of T ^(1 — /3(t)) m(r/T), we have Wx^-^ I (1 - /3(T))m(r/T)e^^^ cos{T^Ag) dr + rhriX + ^Ag). 2ttT ,i\r T Since mriX+y^) has a kernel which, for T > 1, is Ot((1+A)"^) for every TV = 1, 2, as iJiT G 5(M), we conclude that we would obtain (|5.8|) if we could prove that (5.9) \\Sxh\\L2^r^_^^^h)) < {cT--2 +CTX--^)\\h\\L2, ii supp h c Tx-U2ij), with (5.10) = ^ / (1 ~ /3(T))m(r/T)e^^- cos(rv/A3) dr. 1 '•^ At , , . ^ T It is at this point that we shall finally use our hypothesis that (M, 17) has nonpositive sectional curvature. By the Cartan-Hadamard theorem (see jS], [H]), for each point P G M, the exponential map at P, expp, sending TpM, the tangent space at P, to M is a universal covering map. For our sake, it is natural to take P to be the center of our unit-length geodesic segment 7. Thus, with this choice, if we identify M'* with TpM, we have that (5.11) K = expp : R'^ ~ TpM M is a covering map. If g ~ K*g denotes the puUback via k of the metric g to M'', it follows that k is a local isometry. We let dg{y,z) denote the Riemannian distance with respect to g of y,z G M"*. By the Cartan-Hadamard theorem there are no conjugate points for either (M,g) or (K'^,5). Also, the image under k of any geodesic in (M'',^) is one in {M,g). If {'j{t) : t G M.} denotes the parameterization by arc length of the extension of our geodesic segment 7 G 77, let 7 = {7(i) : i G K} denote the Hft of this extension, which is the unique geodesic in (M'^jg) that passes through the origin and satisfies K(7(t)) — 7(t), t e K. Next we recall that the deck transforms are the set of diffeomorphisms a : M'' — > M"^ for which KO a — K. The collection of these maps form a group F. Since a*g = g, a £ F, any deck transform preserves angles and distances. Consequently, the image of any geodesic in (M'', g) under a deck transform is also a geodesic in this space. As a result, the collection of all a e F for which a{"f) = 7 is a subgroup of F, which is called the stabilizer subgroup of 7 that we denote by Stab(7). If {j(t) : t G M.} is not a periodic geodesic, i.e., if there is no to > so that ^{t + to) = 7(t), t € IR, then Stab(7) is just the identity element in F. If the extension of 7 G 7T is periodic with minimal period to > then Stab(7) is a cyclic subgroup which we can write as {ai : £ G Z}, where ai is determined by ctiilit)) — l{t + ito), ^ = 0, ±1, ±2, Thus, restricted to 7, ai just involves shifting the geodesic 7(t) by £ times its period, and Stab(7) is generated by ai. Next, let DD^r ^{yeW^ : dg{0,y) < dg(0,a(y)), Va G F, Identity} KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 21 be the Dirichlet domain for {R ,g). We can then add to Djjir a subset of dDjjir — DuirXlnt {Djjir) to obtain a natural fundamental domain I?, which has the property that K'' is the disjoint union of the a{D) as a ranges over F and {y S K*^ : dg{0, y) < 10} C D since we are assuming that the injectivity radius of (M, g) is more than ten. Given x € M , let x d D he the unique point in our fundamental domain for which k{x) = X. We then have (see e.g. j28j §3.6]) that the kernel of cos(ry^A^) can be written as (5.12) {cos{T^/Ag)){x,y) = ^{cosT,/Ag){£,a{y)), where C08{T^y Ag) : L {R ,g) L (R ,g) is the cosine transform associated with g. Thus, if dVg is the associated volume element, we have that when / € C^(M^) ^g){i,S)f{S)dVgCz) is the solution of the Cauchy problem (with = —idt) {Dl - A~g)u = 0, U\r=0 = /, drU\r=0 = 0. Therefore, by Huygens principle, (5.13) (cosr-\/Ag)(5;,z) = if dg[x,z)>\T\. Also, this kernel is smooth when dg{x,z) ^ |r|, i.e., (5.14) sing supp (cosryA^)(- , •) C {{x,i) e M'* x M'^ : dg(i,z) = |t|}. To proceed, we need a result which follows from the Hadamard parametrix and sta- tionary phase: Lemma 5.1. Let ra be as in (5.3 1 and (5.6 1, and, as above, assume that (3 € C5"(M) satisfies I3{t) = 1, |r| < § and /3{t) = 0, |t| > 2. Then if X,T > I and £,y e M'', we have (5.15) 1 2ttT (1 - /3(t)) rh{T/T)e'^'' (cosT^Ag) (5, y) dr = p{x,y)—Y,a±{\T-d~g{i,y))e^'^''~^^^'y^ + R{\T,i,y), where p £ L°°{R'^ x R'^) n C°°{R'^ x R'^), (5.16) a±(A,r;r) =0, r ^ [1,T], \dia±{X,T;r)\ < C,r-^-^ , with constants Cj independent of T, X > I, and (5.17) R{X,T;i,y) = if dgiS:,y)>T, and \R{X,T-S:,y)\<CT,KX'^''^, if £,y £ K <^ R'^ . This lemma is standard and can essentially be found in [3^ or [551 §3.6]. So let us postpone its proof to the end of the section and focus now on using it to help us to prove (5.9). 22 M. D. BLAIR AND C. D. SOGGE If we combine (5.10) and ( |5.12| , we can write the kernel of our operator as (5.18) Sx[x,y)^^Y. r(l-/3(T))m(r/T)e*^-(cosryA^)(i,«(y))dr, with, as in (5.12), x and y being the unique points in our fundamental domain having the property that x = k(x) and y = K.{y), respectively. In view of (5.13) the number of nonzero summands in (5.181 is finite, but, if the sectional curvatures of (M, g) are strictly negative, the number of such terms grows exponentially with T. Therefore, as in |32j and [S], it is convenient and natural to split the sum into the terms in the stabilizer group for 7 and everything else. So let us write Sx{x,y)^Sf'^'°{x,y) + S^^^{x,y), (5.19) where (5.20) 5r^x,2/) = and (5.21) 50^(0:, y) = 1 2^ 1 E a£Stab(7) (l-/3(T))m(r/T) e^' cos T 1 /Ag)(5,Q(?/))dr, 2ttT E aer\Stab(7) (1 - /3(t)) m(T/T) e'^^ (cost^A^) {i, a{y)) dr, We shall also call the operator associated with the second term in the right side of (5.19) S^^'^ since we shall be able to use oscillatory integral operator bounds to control it. The other piece is very easy to estimate. We claim that (5.22) U=(r,_v2(7)) < +CTX-^)\\hh2, if supp /icr;,-i/2(7). By Young's inequality, this would be a consequence of the following estimate for the kernel (5.23) since we may restrict to {x,y) £ 7^-1/2(7) x 7^-1/2(7). If our 7 € 77 is not a segment of a periodic geodesic in {M,g) then Stab(7) is just the identity element, in which case (5.23) follows trivially from Lemma 5.1 Otherwise, if the geodesic has period > then as noted before Stab(7) = {a^jfez where ai{'j{t)) = ^{t + Uq). Since dg{a{w),a{z)) is uniformly bounded as w and z range over D and a over F, Lemma |5.1| also yields, in this case, (5.24) ^Stab (x,y)|<CT-i J2 A^(l + £)-'^ +CtA' l<eto<2T using (5.16) (with j = 0) to obtain the first term in the right and (5.171 to obtain the other term. Since d > 2, (5.24) implies (5.23). For later use, note that, since the period to must be larger than 10, in view of our assumption on the injectivity radius of {M,g), the constants in (5.22) can be chosen to be independent of 7 G 7f. In view of (5.22), the proof of (5.9) would be complete if we could show that (5.25) ||^o^^/i|U2(r^_^^^(^)) <CTA-i||/i|U2, ifsupp/icr;,-i/2(7). KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 23 By Lemma |5.1 (5.26) S'^'^x,y)^pix,y) A 2 T a±(A,r;dg(£,a(y)))e±^^^^ ±i\dg{x,a{y)) aGr\Stab(7) + R\{x,y), where, with bounds independent of 7 g 7T, (5.27) \Rx{x,y)\<CTX-^+'^. By invoking Young's inequahty one more time, we find that by ( 5.26 ) and (5.27 ) we would have (5.251 if we could show that (5.28) 7^-1/2 (7) pix, 2/)a±(A, T; dg{x, a(y)))e±^^'^^(-^"(^» Hy) dy dx < C„A"'^"i||/i||i2, aer\Stab(7). Here, to simplify the notation to follow, as we may, we are identifying 7>~i/2(7) with its preimage in D via k. So we have lifted our calculation to K'^, and dy denotes the volume element coming from the metric g. To prove this we shall use the following result which is an immediate consequence of Hormander's i^-oscillatory integral theorem in [50] (see also ^26j Theorem 2.1.1]). Lemma 5.2. Let be real and (t){z;x,y) e C°°(M™ x a{z;x,y) e C^{R"' x ] Assume that the mixed Hessian in the {x,y) variables of (j) satisfies 92 Rank I ^ dx d '^(^'^'^0 = ^~ on supp a. dxjdyk Then there is a uniform constant C so that for A > 1 e^^H^■,-,y)a(^■ a;, y)/(y) dy\^ dx) ' < CX-'-^Wfh^M"), where all the integrals are taken with respect to Lebesgue measure. We also require the following simple geometric lemma so that we can use Lemma 5.2 to exploit the fact that our tubes are only have width A 2 to obtain (5.28). Lemma 5.3. Suppose that a G T\Stab{'Y) and that xq, yo G ^CiD. Then either a{yo) ^ 7 or a^^{xQ) ^ ^ or both. Proof of Lemma \5.3\ Since a G r\Stab(7), it follows that 7 and 0(7) are distinct or intersect at a unique point P ~ -P(7, a) (by the Cartan-Hadamard theorem). In the first case both a{yo) ^ 7 and a~^{xo) ^ 7. We also have the desired conclusion if P 7^ «(yo), for then we must have a{yo) ^ 7 as a(yo) S Ck(7). Suppose that we are in the remaining case where 7na(7) = {a(?/o)}. Since XQ,yQ (z D and D D a{D) — 0, it follows that xq ^ a{yo). Therefore, as xq G 7, we must have that 24 M. D. BLAIR AND C. D. SOGGE a(yo) = ^0 7 C 7 Figure 1. Transversal intersections xq ^ Q-ij)- Thus, in this case, we must have a~^{xQ) ^ 7, meaning that we have the desired conclusion for this case as well. □ To use these two lemmas we require some simple facts about the Riemannian distance function dg{x, z). We recall that {M.'^, g) has no conjugate points. Thus, the dx d Hessian dx dzk ^9^^^ rank identically equal to d — 1 away from the diagonal. With this in mind, let us fix points xq and yo on our unit geodesic segment ^ <Z D. We shall now prove a local version of our remaining estimate (5.281. By Lemma 5.3 for our given a G r\Stab(7), we know that either a(yo) ^ 7 or a'^{x^)) ^ 7. For the moment, let us assume the former, i.e., (5.29) a(yo) i 7- We then have that the geodesic passing through zq = a(?/o) a-nd G 7 C 7 intersects 7 transversally. We may therefore choose geodesic normal coordinates in vanishing at Xq so that 7 is the first coordinate axis, i.e. 7 = {(t,0,...,0) : ieM}, and, moreover, if x' = (xi, . . . , Xd-i) are the first d— 1 coordinates of x in this coordinate system then 92 Rank {^ Q^,^^ dg[{x' ,Qi),z)^ ^ d-1 at x' = and z = zq = "(yo)- By Gauss' lemma this will be the case if the geodesic through the origin and Zq intersects the hyperplane {x : x„ = 0} transversally as shown in Figure 1 below, which can be achieved after performing a rotation fixing the first coordinate axis if needed. Since a : R"^ — )■ R'' is a diffeomorphism it follows that in given our fixed points xq, j/o € 7 C 7 n D, we can find i5 > so that, in the above coordinates, (5.30) Rank ( g^/g^^ d,g({x', Xn),a{y))j = d - 1, if x G ^^(xo) and yo e Bs{yo), with Bs{w) denoting the geodesic ball of radius S about x G M.'^. KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 25 Next, it follows from (5.261 and Lemma 5.2 that, in our coordinates, for each fixed value of Xm we have P{x, a(y)) a±(A, T; dg{x, a(y))) - r X e^^>-ds(xMy)) /^(y) dy dx') ' < C^A-"^ f / \h{y)\'^ dy Since \xn\ ^ A 2 in 7^-1/2(7), from this, we deduce that, under our assumption (5.291 we have that (5.31) r^-i/2(7)nBs{yo) p{x, a{y)) a± (A, T; dg{x, a{y))) X e±*^''«(^^"(^»/i(?;)dy d-1 1 dx < CaX — ^ A" \hiy)\'dy Lemma 5.3 tells us that if we do not have (5.29) then (5.32) a~\xo)ii. We claim that for our fixed points xa,yo & 7 we can find (5 > so that (5.31) remains valid for this case as well. To do this, we just use the fact that our a G r\Stab(7) is an isometry and therefore dg{x,a{y)) = dg{a-^{x),y). Consequently, since a^^ G r\Stab(7), we obtain (5.31) under the assumption (5.32) since it is essentially just the dual version of the case we just handled, and so follows from the above argument after taking adjoints. Since we have shown that (5.31) holds either under assumption (5.291 or (5.32), Lemma |5.2 tells us that given any two fixed points xo,yQ £ 7 we can find a S > so that (5.31) is valid. By the compactness of our unit geodesic segment 7, this implies (5.28), which completes the proof of the estimate (5.2) for our fixed 7 G 77 It is straightforward to see how to obtain the stronger estimate (5.1), which involves uniform bounds over 77, by using the proof of (5.2). We use the fact that if T ^ 1 is fixed and if 7 G 7J is fixed then there is a neighborhood A^(7) of 7 in 77 so that if a G r\Stab(7) and the geodesic distance between our fundamental domain D and its image a{D) is < 2T, then we also have that a ^ r\Stab (70) for any 70 G A/'(7). This follows from the fact that there are only finitely many a G F for which the distance between D and a{D) is < 2T, and if a is not a stabilizer for 7 then it is also not a stabilizer for nearby geodesies. Because of this and the uniform dependence on the smooth parameter z in Lemma [572} if we define S'^'''^''' to be the operator whose kernel is given by (5.21 ), we have the uniform bounds \\Sx'"''''Hl^t^_,^2{i)) < C'tA^J if 70 G 7V(7) and supp C Ta-i/2(7o). If then sf*'^^''^ = Sx — S'^^'^''' is then defined using 7, then the proof of (5.22 ) clearly also yields the following variant 1^' ■Stab, 7 < (CT-5+CtA-2) \\h\\L2, if 70 e A/'(7) and supp h C Tx-mijo)- 26 M. D. BLAIR AND C. D. SOGGE Together these two esthnates imply the analog of (5.1) where, instead of having the geodesic segments range over 77, we have them range over Afi'f) and = A^{J\f{j)) depends on Afij). By the compactness of 77, this in turn yields (5.1 1. To wrap things up, we also need to prove Lemma 5.1 Proof of Lemma 5.1 Since m(r) = when |t| > 1/2 it follows that the left side of (5.151, (5.33) 1 27rr -T (1 - /3(t)) m(T/r) e*^^ (cos r (i, v) dr, vanishes when dg (5, y) > T. Since /3{t) = 1 for |t| < 3/2, by ( |5T4| , it is OAr,T((l + A)-^) for any A*" = 1, 2, 3, . . . if dg{x, y) < 1. Therefore, we need only to prove the assertions in Lemma 5.1 when 1 < dg{x,y) < T. To prove this, we shall use the Hadamard parametrix (see e.g. [H] and 2]). Since (M'',^) has nonpositive curvature, for < r < T we can write Chapter (5.34) COSTa ^){x,y) = p{x,y) (27r) -d cosrjf I + 7?(T,i,j/), where the leading Hadamard coefficient, p, is smooth and uniformly bounded (by the curvature hypothesis), and if m G N is fixed we can have dlRij, x, y) S L^^, < j < m, and also (5.35) |<s_^9^a±(T,i,y,r)| <Ct,k, 0,j -2-i if r > 1, < r < T, j = 0, 1, 2, . . . , and y e d M''. We also recall (see e.g. j^S]) that we can write the Fourier transform of Lebesgue measure on the sphere in as (5.36) where for each j = 0, 1, 2, . . . , we have (5.37) \dic+{r)\ + \dlc^{r)\ < C^r'^ r > 1. If in (5.331 we replace (cosT-^Ag)(a;, y) by the first term in (5.34|, the resulting expression equals p{x,y) times a fixed multiple of (5.38) 1 E 2ttT 1 2nT T T Jo m(r/T) e'^^ cos(r|^|) e'''5(i,y)?i d^dr m{r/T) e'^^ cos(rr)e±"''^'(^'*)c± (dg(i, y)r) {d-gii^y))"^^ drdr KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 27 minus (5.39) d-l f 2 2 Jo {dg{x,y))' drdr. If we replace cos(rr) by e in the right side of (5.38), the resulting expression equals the sum over ± of m(T(A - r)) c± {dg{x, y)r)e^''^'^sis,v) d-1 7" 2 dr A 2 T a±{X,T; dg{x,y)), where, using the fact that m G S{M.) and (5.37), a± satisfies (5.16). If in ( |5.38[ ) we replace cos(rr) by e*'^'", then this argument also implies that the resulting expression is 07v,t((1 + •^)^^) for any iV = 1, 2, 3, . . . . Thus, modulo such an error p times the terms in (5.38) can be written as the first term in the right side of (|5.15|) with (|5.16|) being valid. Since this argument shows that the same is the i the first term in the right side of (5.34), up to ON,T{{i Since this argument shows that the same is the case for (5.39), we conclude that A)^^) errors, gives us the first term in the right side of (5.15) This argument and (5.35) also shows that if in (5.15) we replace {cosT^/A^){x,y) by the second term in the right side of (5.34 1, then we get a term obeying the bounds in (5.17) the same is true for it Since, as noted we can take the remainder term in (5.34) to satisfy for a given 0, 1, . . . , TO, we also see that if we choose to large enough, □ oo ■ loC J References [1] Berard, P. H. On the wave equation on a compact manifold without conjugate points, Math. Z. 155 (1977), 249-276. [2] Bourgain, J. Estimates for cone multipliers, Operator Theory: Advances and Apphcations 77 (1995), 41-60. [3] Bourgain, J. Geodesic restrictions and -estimates for eigenf unctions of Riemannian surfaces. Linear and complex analysis, Amer. Math. Soc. Transl. 226 (2009), Amer. Math. Soc, Providence, RI, 27-35. [4] Bourgain, J. -estimates for oscillatory integrals in several variables, GAFA 1 (1991), 321—374. [5] Bourgain, J.; Guth, L. Bounds on oscillatory integral operators based on multilinear estimates, Geom. Funct. Anal. 21 (2011), 1239-1295. [6] Bruning, J. Uber Knoten Eigenf unktionen des Laplace- Beltrami Operators, Math. Z. 158 (1978), 15-21. [7] Burq, N.; Gerard, P.; Tzvetkov, N. Restrictions of the Laplace- Beltrami eigenfunctions to subman- ifolds. Duke Math. J. 138 (2007), 445-486. [8] Chavel, I. Riemannian geometry. A modern introduction, 2nd Edition, Cambridge University Press, Cambridge 2006. [9] Chen, X.; Sogge, CD. A few endpoint restriction estimates for eigenfunctions, [arXiv:1210.7520 | [10] Colding, T.H.; Minicozzi, W.P. II Lower bounds for nodal sets of eigenfunctions, Comm. Math. Phys. 306 (2011), 777-784. [11] Colin de Verdiere, Y.: Ergodicite et fonctions propres du laplacien, Comm. Math. Phys., 102 (1985), 497-502. [12] Do Carmo, M. Riemannian geometry, Birkhauscr, Basel, Boston, Berlin, 1992. 28 M. D. BLAIR AND C. D. SOGGE [13] Dong, R.T. Nodal sets of eigenf unctions on Riemann surfaces, J. Differential Geom. 36 (1992), 493-506. [14] Donnelly, H.; Fefferman, C. Nodal sets of eigenf unctions on Riemannian manifolds, Invent. Matli. 93 (1988), 161-183. [15] Donnelly, H.; Fefferman, C. Nodal sets for eigenfunctions of the Laplacian on surfaces, J. Amer. Math. See. 3(2) (1990), 332-353. [16] Han, Q; Lin F.H. Nodal sets of solutions of Elliptic Differential Equations, book in preparation (online at |http://www.nd.edu/qhan/nodal.pdf[ |. [17] Hardt, R.; Simon, L. Nodal sets for solutions of elliptic equations, J. Differential Geom. 30 (1989), 505-522. [18] Hassell, A; Tacy, M. Improvement of eigenfunction estimates on manifolds of nonpositive curvature, [arXiv:1212.2540 [19] Hezari, H; Sogge, CD. A natural lower bound for the size of nodal sets. Analysis and PDF 5 (2012), 1133-1137. [20] Iformander, L. Oscillatory integrals and multipliers on FLP, Ark. Matfi. 11 (1973), 1-11. [21] Iformander, L. The analysis of linear partial differential operators. III. Pseudodifferential operators. Springer- Verlag, Berlin, 1985. [22] Lee, S. Linear and bilinear estimates for oscillatory integral operators related to restriction to hypersurfaces. J. Funct. Anal. 241 (2006), 56-98. [23] Snirelman, A. I.: Ergodic properties of eigenfunctions, Uspenski Math. Nauk 29 (1974), 181—182. [24] Sogge, C.D. Oscillatory integrals and spherical harmonics, Duke Math. J. 53 (1986), 43—65. [25] Sogge, C.D. Concerning the norm of spectral clusters for second-order elliptic operators on compact manifolds. J. Funct. Anal. 77 (1988), 123-138. [26] Sogge, C.D. Fourier integrals in classical analysis, Cambridge Tracts in Math., Cambridge Univ. Press, Cambridge, 1993. [27] Sogge, C.D. Kakeya-Nikodym averages and -norms of eigenfunctions. Tohoku Math. J. 63 (2011), 519-538. [28] Sogge, C.D. Hangzhou lectures on eigenfunctions of the Laplacian, Annals of Math Studies, Prince- ton Univ Press, to appear. [29] Sogge, CD.; Zelditch, S. Riemannian manifolds with maximal eigenfunction growth, Duke Math. J. 114 (2002), 387-437. [30] Sogge, CD.; Zelditch, S. Lower bounds on the Hausdorff measure of nodal sets. Math. Res. Lett. 18 (2011), 25-37. [31] Sogge, CD.; Zelditch, S. Lower bounds on the Hausdorff measure of nodal sets /J, |arXiv:1208.2045[ [32] Sogge, CD.; Zelditch, S. On eigenfunction restriction estimates and L'^ -bounds for compact surfaces with nonpositive curvature, arXiv:1108.2726 [33] Stein, E.M. Oscillatory integrals in Fourier analysis. Beijing lectures in harmonic analysis (Beijing, 1984), Ann. of Math. Stud. 112, Princeton Univ. Press, Princeton, NJ, 1986, 307-355. [34] Tao, T. A sharp bilinear restriction estimates for paraboloids. Geom. Funct. Anal. 13 (2003), 1359— 1384. [35] Tao, T.; Vargas, A. A bilinear approach to cone multipliers I: Restriction estimates. Geom. Funct. Anal. 10 (2000), 185-215. [36] Tao, T.; Vargas, A.; Vega, L. A bilinear approach to the restriction and Kakeya conjectures. J. Amer. Math. Soc. 11 (1998), 967-1000. [37] Wolff, T. A sharp cone restriction estimate. Ann. of Math. 153 (2001), 661-698. [38] Yau, S.T. Survey on partial differential equations in differential geometry. Seminar on Differential Geometry, pp. 3—71, Ann. of Math. Stud., 102, Princeton Univ. Press, Princeton, N.J., 1982. [39] Zelditch, S.: Uniform distribution of eigenfunctions on compact hyperbolic surfaces, Duke Math. J., 55 (1987), 919-941. [40] Zworski, M. Semiclassical Analysis, Graduate Studies in Mathematics 138, Amer. Math. Soc, Providence, 2012. KAKEYA-NIKODYM AVERAGES IN HIGHER DIMENSIONS 29 Department of Mathematics and Statistics, University of New Mexico, Albuquerque, NM 87131, USA E-mail address: blair@math.uiim.edu Department of Mathematics, Johns Hopkins University, Baltimore, MD 21093, USA E-mail address: sogge@jhu.edu