Heat kernel upper bound on Riemannian manifolds with locally uniform Ricci curvature integral bounds Christian Rose1 1Technische Universität Chemnitz, Fakultät für Mathematik, Germany 6 June 23, 2016 1 0 2 n Abstract u J This article shows that under locally uniformly integralbounds of the neg- 2 ative part of Ricci curvature the heat kernel admits a Gaussian upper bound 2 for small times. This providesgeneralassumptions on the geometryof a man- ifoldsuch that certainfunction spacesare in the Katoclass. Additionally, the ] G results imply bounds on the first Betti number. D . 1 Introduction h t a m One of the most important invariants of a Riemannian manifold is given by its heat kernel. Becausethereisnoexplicitrepresentation ofitingeneral, animportant and [ interesting topic in geometric analysis is the small-time behavior of this function 2 in terms of the underlying geometry. Especially Gaussian upper bounds are of v 8 particular interest. The article [Gri99] provides plenty examples for Riemannian 3 manifolds possessing a heat kernel with Gaussian behavior. A very prominent one 4 is given by the following. 7 0 . Theorem 1.1 ([Gri99]). Let M be a Riemannian manifold with bounded geometry. 1 0 Then the heat kernel pt(, ) can be bounded by · · 6 1 sup p (x,x) Ct n/2, 0 < t 1, (1) t − : ≤ ≤ v x M ∈ i X where C > 0 depends on the geometry of M. r a Another interesting example would be the case of complete, connected and non- compact Riemannian manifolds with non-negative Ricci curvature. In this case the heat kernel satisfies a bound like (1) for all t > 0. The techniques used in [Gri99] for the proof of such bounds are the volume doubling property and the so-called relative Faber-Krahn inequalities. For another approach see [SC02]. Considering the above theorem leads to the question under which weaker geometric conditions a bound like (1) is satisfied. Of course, the assumption that the Ricci curvature is bounded from below is a quite strong condition. One could think of a 1 manifold withRiccicurvature stayingaround zeroalmosteverywhere but providing a set where the Ricci curvature has a deep well. In this case, the bound (1) gets worse as it should be, because it grows exponentially as the Ricci curvature bound decreases. From an analytic point of view it therefore seems natural to replace the boundedness by an integral condition. A first step was done by Gallot. For a map f: M R let v := sup(0, v) and v := sup(0,v). Denote by ρ: M R the + → − − → function whose values are the smallest eigenvalues of the Ricci tensor. Theorem 1.2 ([Gal88]). Let D > 0, λ > 0 and p > n 3. There are explicitly ≥ computable constants K ,C > 0 such that for any compact Riemannian manifold G G M with dimM = n, diam(M) D and Ricci curvature satisfying ≤ p/2 1 ρ − λ2 dvol KG, (2) Vol(M) n 1 − ≤ Z (cid:18) − (cid:19)+ the heat kernel p (x,y) can be bounded from above by t sup p (x,y) C t p/2, 0 < t 1. t G − ≤ ≤ x,y M ∈ There are several things which should be noted here. On one hand, the heat kernel does not provide the right local dimension n, which depends on the tech- nique. Gallot obtains a global isoperimetric inequality of dimension p controlling the negative part of the Ricci curvature. This in turn yields the above estimate for the heat kernel by a symmetrization procedure. On the other hand, he assumes that M is compact and the Ricci curvature is globally Lp-bounded. Yang noticed in [Yan92] that it is possible to get local isoperimetric inequalities assuming a local Lp-bound on the negative part of the Ricci curvature. Theorem 1.3 ([Yan92]). Let M be a complete connected Riemannian manifold of dimension n, p > n/2, r > 0 and v > 0. Assume that for all x M, Vol(B(x,r)) ∈ ≥ v. There are explicit constants K ,C > 0 such that if Y Y p ρ < K , Y ZB(x,3r) − then we have for all Ω B(x,r) with smooth boundary ⊂ A(∂Ω) C . Vol(Ω)n−n1 ≥ Y Unfortunately, this theorem provides a result for the heat kernel only in the non-collapsing case, where one assumes that the volumes of balls cannot approach zero. The purpose of this article is to show that on complete connected manifolds, either compact or not, one can generalize Yang’s result to the collapsing case under the assumption that one has locally uniform Lp-bounds on the negative part of Ricci curvature. A precise definition will be given in the next section. The main obser- vation which is necessary for the proof was made in [PW01]. Assuming a locally 2 uniform Lp-bound on the Ricci curvature yields a local volume doubling condition for balls. This is sufficient to generalize Theorem1.3 to the collapsing case, which is one of our main results. The result implies a local version of Lemma 7.16 in [Gri99], which provides lower bounds for fractions of volumes of balls. We then improve Theorem 1.3 to the collapsing case with explicit dependence on all parameters. The local uniformity of the curvature yields local isoperimetric inequalities in every ball of thesameradius. A covering technique andaresultof [Gri09]then imply theheat kernel bound with the right local dimension in a quantitative way. In particular, in the compact case we cover Theorem 1.2 and obtain the local dimension n. Upper bounds of the heat kernel for small times allow to characterize certain func- tion spaces. In [GP13] the authors show that under certain conditions of the heat kernel, Lp-spaces are in the so called Kato-class. Let us recall the statement briefly. Let M be a Riemannian manifold of dimension n and denote by (P ) the heat t t 0 ≥ semigroup onM andp (, ) its minimal heat kernel. For anon-negative measurable t · · function V : M R and β > 0 let + → β b (V,β) := sup P (V n) dt. Kato t n∈NZ0 k ∧ k∞ We say that a function V : M C is in the Kato class (M) if → K lim b (V ,β) = 0. Kato β 0 | | → Theorem 1.4 ([GP13]). Let M be a Riemannian manifold of dimension n 3 and ≥ p > n/2. Assume that there are t ,C > 0 such that 0 sup p (x,x) Ct n/2, 0 < t t . (3) t − 0 ≤ ≤ x M ∈ Then we have Lp(M)+L (M) (M). (4) ∞ ⊂ K Notethatthelocaldimensionnoftheheatkerneldoesnotplayasignificantrole to obtain qualitative results like (4) and could also bereplaced by another constant, affecting the choice of p. In the recent paper [RS] Peter Stollmann and the author used the so-called extended Kato class to derive bounds on topological invariants for compact manifolds. A measurable function V : M C is in the extended Kato → class if for some β > 0 b (V ,β) < 1. Kato | | Using Theorem 1.2 it was shown that b (ρ ,β) can be bounded in terms of the Kato averaged Lp-norm of ρ , leading to bounds on−the first Betti number via semigroup − domination. The result shows how the upper bound scales with the involved quan- tities, but it is not explicit. The local results of the present article give explicit bounds on the first Betti number with a locally uniform Lp-bound on the negative part of the Ricci curvature. 3 2 Preliminaries In this article M is always a complete connected Riemannian manifold without boundary of dimension n. For x M and r > 0 let B(x,r) be the open ball around ∈ x with radius r. Denote by dvol thevolume measure of M and forx M and r > 0 ∈ let V(x,r) the volume of B(x,r). Moreover, A(H) gives the (n 1)-dimensional − measure of a hypersurface H M. For any x M and r > 0 we define the local ⊂ ∈ isoperimetric constant for B(x,r) to be the quantity A(∂Ω) I(x,r) := inf , n−1 Vol(Ω) n where Ω runs over all relatively compact subset of B(x,r) with smooth boundary. Let ρ: M R the smallest eigenvalue function of the Ricci tensor and write ρ = → − sup(0, ρ). We define for p > n/2 the quantities − p κ(p,x ,s) := ρ dvol, 0 ZB(x0,s) − κ(p,x ,s) 0 κ¯(p,x ,s) := , 0 V(x ,s) 0 κ¯(p,s) := sup (κ¯(p,x ,s))1/p. 0 x0 M ∈ Note that it is convenient to work with the scale invariant quantity R2κ¯(p,R), as explainedin[PW01]. IfthereisalowerRiccicurvaturebound,R2κ¯(p,R)issmallfor smallR. Forconvenience, werecallanimportant factaboutthequantity R2κ¯(p,R), and quote the following statement. Proposition2.1 ([PW01],Theorem2.1). Assumingthatp > n/2, there areexplicit constants ε= ε(n,p) > 0 and D = D(p) >0 such that for any R > 0 with R2κ¯(p,R) < ε we have for all x M and 0 < r < s R ∈ ≤ V(x,s) s n D . V(x,r) ≤ r (cid:16) (cid:17) In particular we can set ε = 2n((2np−1n)()2pp+11) and D = 2−2√2 2p. − − (cid:16) (cid:17) Remark 2.2. The locally uniformbound of theRiccicurvature andthe localvolume doubling property above allow to compare r2κ¯(p,r) for different values r. Assume thattheassumptionsofProposition2.1aresatisfied. Thenitisadirectconsequence that for all 0< r < r R 1 2 ≤ n 2 r2κ¯(p,r ) D 1 r2 p− r2κ¯(p,r ). (5) 1 1 ≤ − r 2 2 (cid:18) 1(cid:19) 4 The other direction looks similar and can be seen by a packing argument. Assume that 0 < r < r and that r2κ¯(p,r ) < ε. Then 1 2 1 1 r2κ¯(p,r ) 2n p1 r2 2r2κ¯(p,r ). (6) 2 2 ≤ D r 1 1 (cid:18) (cid:19) (cid:18) 1(cid:19) 3 Local isoperimetric inequalities Assuming that the Ricci curvature integral bounds are uniformly small provides local isoperimetric inequalities. Theorem 3.1. Let 2p > n 3 and r > 0 and assume that r < diam(M). There ≥ are explicit constants K = K (n,p,r) > 0 and C = C (n,p,r)> 0 such that if L L I I r2κ¯(p,r) <K , (7) L for all x M we have 0 ∈ 1+1/n V(x ,r) 0 I(x ,r) C . (8) 0 ≥ I (1+r2p)rn (cid:18) (cid:19) In order to prove Theorem 3.1 we use Yang’s idea from [Yan92]. To overcome the issue of non-collapsing we use a local volume doubling result which is based on ideas from [Gri99]. Bounding the ratio of the volumes of two intersecting balls in terms of their radii, the ratio will also be bounded from below. Lemma 3.2. Let ε > 0 be as in Proposition 2.1 and R > 0. Assume that M is non-compact or R diam(M)/3. If ≤ R2κ¯(p,R) < ε, then there are explicit constants a = a(n,p) > 0, b = b(n,p) > 0 and η = η(n,p) > 0 such that for all 0 < r s R/3 and balls B(x,s) and B(y,r) such that ≤ ≤ B(x,r) B(y,s)= ∩ 6 ∅ s η V(x,s) s n a b . (9) r ≤ V(y,r) ≤ r (cid:16) (cid:17) (cid:16) (cid:17) Proof. (following [Gri99]) We know that d(x,y) r + s 2R/3 and therefore ≤ ≤ Proposition 2.1 implies V(x,s) V(y,2s+r) V(y,3s) s n b V(y,r) ≤ V(y,r) ≤ V(y,r) ≤ r (cid:16) (cid:17) where b := 3nD 2p. − Assume now s = 3r and x = y, then there is a ξ M with d(x,ξ) = 2r, B(ξ,r) ∈ ∩ B(x,r) = , such that, using the upper inequality for V(ξ,r) ∅ V(x,s) = V(x,3r) V(x,r)+V(ξ,r) 1+b 1 V(x,r). − ≥ ≥ (cid:0) (cid:1) 5 For the general case we know that there is a k N such that ∈ s 3k < 3k+1, ≤ r so we can conclude V(x,s) V(x,3kr) V(x,r) 1+b 1 k b 1(1+b 1)kV(y,r) − − − ≥ ≥ ≥ = b 1(1+(cid:0)b 1) 1(1(cid:1)+b 1)k+1V(y,r) − − − − s η aV(y,r) ≥ r (cid:16) (cid:17) where a := b 1(1+b 1) 1 = (b+1) 1 and η := log (1+b 1). − − − − 3 − Since we want to follow the main steps of [Yan92] it seems convenient to recall some definitions and results from this text. Let x M and denote by S T M 0 ∈ x0 ⊂ x0 the unit sphere in the tangent space at x . By c we denote the volume of the unit 0 m disc in Rm. Given a subset Sˆ S and ρ > 0 let ⊂ x0 Γ(Sˆ,ρ) := y = exp rθ 0 r < ρ,θ Sˆ,d(x,y) = r { x0 | ≤ ∈ } the geodesic cone with base point x and length r. 0 Proposition 3.3 (see [Yan92]). Let M be a Riemannian manifold, x M, Sˆ 0 ∈ ⊂ Sx0, p > n/2, τ > 0. Set δ = 22pp−n1, ωˆ = Vol(Sˆ), − τ1/δ k := ρp, and r := (h(n,p) 1ωˆk 1)1/(2p n). 0 − − − ZΓ(Sˆ,ρ) − 1+τ Then, for every 0 r ρ ≤ ≤ (1+τ)n 1ωˆn 1rn: 0 r r , Vol(Γ(Sˆ,r)) − − ≤ ≤ 0 (10) ≤ (c(n,p,τ)k((1 δ)r+δr0)2p: r r0. − ≥ p 1 Here, h(n,p) := (2−1/p)p 2pp−1n − and c(n,p,τ) := (1+τ−1)2p−12p2(pn−11)h(n,p). − − (cid:16) (cid:17) The proof of the Proposition above requires k > 0 because it appears in the denominator of r . In the case k = 0 the volume bound of the geodesic cone follows 0 directly from the Bishop-Gromov inequality, [Cha84]. Observe that in the case Sˆ = S it is k = κ(p,x ,r) for ρ = r. We are now able to prove Proposition 3.1. x0 x0 0 Proof of Theorem 3.1. For the proof we distinguish between the non-compact and compact case. The non-compact case: Define the constant D 1/p Dp+1rna p1 2 G := (16(b+2)2/η) 1min ε, 4(b+2)1/η r2 , − 2n c(n,p,τ 1) (cid:18) (cid:19) ( (cid:18) − (cid:19) ) (cid:16) (cid:17) 6 with τ = τ(n,p,r) > 0 to be chosen later. Let Ω B(x ,r) have smooth boundary 0 ⊂ ∂Ω. By b > 0 from Lemma 3.2 we define s:= 4(b+2)1/ηr. Given x Ω, let Sˆ S denote the set of unit tangent vectors v such that the x x ∈ ⊂ corresponding geodesic exp sv,s > 0, is aminimal geodesic joining x tosomepoint x inB(x ,s/3) B(x ,r)andchoosex ΩsuchthatSˆ hasminimalvolume. Croke’s 0 0 x \ ∈ inequality [Cro80, Cha93] tells us that 1+1/n A(∂Ω) c ωˆ n 1 − . Vol(Ω)(n−1)/n ≥ (cn/2)1−n1 (cid:18)cn−1(cid:19) Therefore it suffices to find a lower bound for ωˆ. Since r2κ¯(p,r) < G, (6) gives 1 Darn p s2κ¯(p,s)< min ε,D s2 , (11) c(n,p,τ 1) ( (cid:18) − (cid:19) ) such that Lemma 3.2 and the definition of s yield Vol(Γ(Sˆ ,s/3+r)) V(x ,s/3) V(x ,r) x0 ≥ 0 − 0 s η a 1 V(x ,r) 0 ≥ 3r − (cid:16) (cid:16) (cid:17) (cid:17) aV(x ,r). (12) 0 ≥ On the other hand, (11) and (5) imply r+s/3 np−2(r+s/3)2κ¯(p,r+s/3) Darn p1 s2 (13) s ≤ c(n,p,τ 1) (cid:18) (cid:19) (cid:18) − (cid:19) and therefore aD κ(p,x ,r+s/3) V(x ,r+s/3)rn(r+s/3) n. (14) 0 ≤ c(n,p,τ 1) 0 − − Apply Proposition 2.1 to (12) and use (14) to get n r Vol(Γ(Sˆ ,s/3+r)) aD V(x ,r+s/3) x0 ≥ s/3+r 0 (cid:18) (cid:19) c(n,p,τ 1)κ(p,x ,r+s/3). (15) − 0 ≥ In the following we want to choose τ such that in Proposition 3.3 we can always use the first case in (10) . If one assumes s/3+r r , combining the second case 0 ≥ of (10) and (15) yields c(n,p,τ 1) c(n,p,τ)(s/3+r)2p − ≤ or τ2p 1 (s/3+r)2p. − ≤ 7 2p Therefore, defining τ := s2p−1 > 0 suffices. In this case one can conclude that r+s/3 r , and in turn 0 ≤ ωˆ Vol(Γ(Sˆ ,s/3+r)) (1+τ)n 1 (s/3+r)n. x0 ≤ − n Combining the upper and lower bound for the volume of the geodesic cone we get ωˆ na (1+τ)1 n(s/3+r) nV(x ,r) − − 0 c ≥ c n 1 n 1 − − = na (4(b+2)η1 +1)−n(1+s2p2−p1)1−nV(x0,r). c rn n 1 − By assumption it is 2p > n, and therefore (1+s2p2−p1)n−1 2n(4(b+2)η1 +1)2p(1+r2p). ≤ This yields ωˆ V(x ,r) 0 C , (16) c ≥ N(1+r2p)rn n 1 − where CN := na 2−n(4(b+2)η1 +1)−2p−n. c n 1 − diam(M) The compact case: Let G and s be as above. If r , then s diam(M) ≤ 4(b+2)1/η ≤ and the proof above applies. Otherwise, let diam(M) 2p r˜:= such that s˜:= diam(M), τ˜ := s˜2p−1 4(b+2)1/η and D 1/p Dp+1r˜na p1 2 G˜ := (16(b+2)2/η) 1min ε, 4(b+2)1/η r˜2 . − 2n c(n,p,τ˜ 1) (cid:18) (cid:19) ( (cid:18) − (cid:19) ) (cid:16) (cid:17) Assuming r2κ¯(p,r) < G˜, (5) gives r˜2κ¯(p,r˜) < G˜. The proof above applies again and leads to (16) with r replaced by r˜. Since r2κ¯(p,r) < ε, Proposition 2.1 implies the inequality ωˆ V(x ,r) 0 C , c ≥ c(1+r2p)rn n 1 − where C := DC . We end up by setting c N c CI := (cn/n2−)11−n1 Cc1+1/n and diam(M) G: M non-compact or r , K := ≤ 4(b+2)1/η L (G˜: otherwise. 8 4 Heat kernel bounds and the Kato class As mentioned in the introduction, it is sufficient to obtain a Gaussian upper bound fortheheatkernelforsmalltimesifwehaveanisoperimetricinequalityforeveryball of the same radius. The reason is that such isoperimetric inequalities yield relative Faber-Krahn inequalities in these balls. Using a covering argument, Theorem 15.14 in[Gri09]impliesanupperboundfortheheatkernelforsmalltimesinthecollapsing case. Theorem 4.1. Assume 2p > n 3, r > 0 and assume that r < diam(M). Let ≥ D =D(n,p) > 0 and K = K (n,p,r)> 0 as above and L L r2κ¯(p,r) < DK . (17) L Then, there exists an explicit C = C(n,p,r) > 0 such that for all x M and ∈ 0 < t r2/4 ≤ C p (x,x) t n/2. t ≤ V(x,r)n+1 − Proof. Let x M. By Proposition 3.1 it follows from our assumption on r2κ¯(p,r) ∈ that 1+1 1+1 V(x,r/2) n (2p+n)(n+1) V(x,r/2) n I(x,r/2) ≥ CI (4p +2 2pr2p)2 nrn =CI2 n (42p +r2p)rn . (cid:18) − − (cid:19) (cid:18) (cid:19) This isoperimetric bound in B(x,r/2) is equivalent to the statement that for all Ω B(x,r/2) with smooth boundary ∂Ω we have ∈ 1+1 VAo(l∂(ΩΩ)) ≥ CI2(2p+nn)(n+1) (4V2p(+x,rr2/p2))rn n Vol(Ω)−1/n. (cid:18) (cid:19) By Cheeger’s theorem, this implies that the first Dirichlet eigenvalue of Ω, λ (Ω), 1 is bounded from below, 2+2/n λ1(Ω) ≥ CI222(2p+nn)(n+1)−2 (4V2p(+x,rr2/p2))rn Vol(Ω)−2/n. (cid:18) (cid:19) That means, for every ball B(x,r/2) we have a relative Faber-Krahn inequality. Since (B(x,r/2)) is a cover for M, Theorem 15.14 in [Gri09] implies that, after x M collecting all the c∈onstants, for all x M and t (0,r2/4) the heat kernel p on M t ∈ ∈ satisfies p (x,x) C C n2n (2p+n)(n+1)(4p +r2p)2n+2r2n(n+1)V(x,r/2) n 1t n/2 t ≤ h I− − − − − where 2√174nn++12+n+2e2 4n+3 C := 10e2 . h 4n+2 1 ≤ 4n+2 1 − − Appling Lemma 3.2 to V(x,r/2) proves the theorem. 9 Corollary 4.2. Assume that for M the assumptions of Theorem 4.1 hold and ad- ditionally T := liminfV(x,r)/rk > 0 x M ∈ for some k N. Then we have for all x M and 0< t r2/4 ∈ ∈ ≤ C p (x,x) t n/2. t − ≤ Tn+1rk(n+1) The corollary holds as well when assuming that the injectivity radius i of M M is positive and r < i /2, see [Cro80], but we need the lower bound only for one M r and not for a whole scale. The same argument works under the non-collapsing assumption of Theorem 1.3. Corollary 4.3. Under the assumptions of Corollary 4.2 we have Lp(M)+L (M) (M). ∞ ⊂ K In the special case that we assume that M is compact Theorem 4.1 and Propo- sition 2.1 enable to prove a version of Theorem 1.2 providing the local dimension n of the heat kernel. Contrary to Gallot’s result we do not use a global isoperimetric inequality. Corollary 4.4. Assume that M is compact and 2p > n 3. There is an explicit ≥ constant C = C (n,p,diam(M),Vol(M)) > 0 such that if G G (diam(M))2κ¯(p,diam(M)) < K , L then for all x M and 0 < t 1diam(M) ∈ ≤ 3 p (x,x) C t n/2. t G − ≤ Proof. We have by (5) (1/3diam(M))2κ¯(p,1/3diam(M)) 3n/p 2(diam(M))2κ¯(p,diam(M))2 < K . − L ≤ Theorem 4.1 and Proposition 2.1 imply for all x M and t (0,1/3diam(M)] ∈ ∈ C(n,p,1/3diam(M)) 3n n+1 p (x,x) t n/2 C(n,p,1/3diam(M)) t n/2. t − − ≤ V(x,1/3diam(M)) ≤ Vol(M) (cid:18) (cid:19) 5 An upper bound for b (M) 1 Asexplained intheintroduction, Katoclasstechniques areapowerfultooltoderive bounds on topological invariants. For compact manifolds M it was shown in [RS] that it is possible to bound the first Betti number b (M) in terms κ¯(p,diam(M)) 1 using Theorem 1.2. Because b (M) = dimH1(M) can be bounded by the trace of 1 10