New York Journal of Mathematics New York J. Math. 3 (1997) 103{118. Harnack Inequalities For Curvature Flows Depending On Mean Curvature Knut Smoczyk Abstract. We prove Harnack inequalities for parabolic ows of compact orientable hypersurfaces in Rn+1 , where the normal velocity is given by a smooth function f depending only on the mean curvature. We use these estimates to prove longtime existence of solutions in some highly nonlinear cases. In addition we prove that compact selfsimilar solutions with constant mean curvature must be spheres and that compact selfsimilar solutions with nonconstant mean curvature can only occur in the case, where f = Ax with two constants A and . Contents 1. Introduction 2. Some relations and evolution equations 3. Homothetic solutions 4. The Harnack inequality 5. Longtime existence for some highly nonlinear ows References 103 106 107 107 114 118 1. Introduction Assume that M n is a compact orientable surface, smoothly immersed into Rn+1 by a smooth family of dieomorphisms Ft : M n ! Rn+1 that satisfy the PDE @ @t F = ;f where denotes the outward pointing unit normal and f is a smooth function depending only on the mean curvature H of the immersed surface, e.g. for f = H we get the well-known mean curvature ow (MCF) and for f = ; H1 we obtain the inverse mean curvature ow. Hamilton 4] proved a beautiful Harnack inequality for the MCF. In 1] Harnack inequalities were derived for convex hypersurfaces in cases where f may depend on the full second fundamental form. The case f = ; H1 Received September 22, 1997. Mathematics Subject Classication. 53C21 (35K15). Key words and phrases. Harnack, mean, curvature, ow, selfsimilar. 103 c 1997 State University of New York ISSN 1076-9803/97 Knut Smoczyk 104 has not been studied yet nor are there results for arbitrary functions f depending only on H . The aim of this paper is to address the following questions: I. What conditions on f guarantee that the ow becomes parabolic so that we have shorttime existence of a solution? II. For which f do we get nice Harnack inequalities? III. Since selfsimilar solutions play an important role in the Harnack inequality for the MCF, we ask: For which f do selfsimilar solutions exist and can we say something about the nature of these solutions? An interesting case is the inverse mean curvature ow f = ; H1 since it is important in General Relativity 6]. There is some hope that one can generalize our results to other target manifolds N n+1 . Throughout this paper we will use the standard terminology, i.e., h i denotes the euclidean inner product, gij dxi dxj is the induced metric on M n, xi coordinates for M n and hij dxi dxj = hri F rj idxi dxj denotes the second fundamental form, r the covariant derivative with respect to gij . Double latin indices are summed from 1 to n and we set jAj2 := hij hij C := hik hlk hil : In this paper we will always assume that f : ! R is a smooth function dened on an open subset in R and we dene Denition. Let F : M n ! Rn+1 be an immersion. F is called admissible, if H (x) 2 8x 2 M n . The answer to question I is then given by Proposition I. Let F0 : M n ! Rn+1 be an admissible smooth immersion of a compact orientable surface M n and assume that f : ! R is strictly positive. Then the PDE 0 @ @t F = ;f F (x 0) = F0 (x) 8x 2 M n (?) has a smooth admissible solution on a maximal time interval 0 T ) T > 0. Proof. This follows from the fact that the linearization of (?) diers from the linearization for the mean curvature ow only by a factor f which by assumption is strictly positive. Therefore (?) is a (nonlinear) parabolic equation and the compactness of M n and the theory for parabolic equations imply shorttime existence. 0 In view of Proposition I we will always assume that f > 0. Then the main theorem can be stated as follows Theorem 1. Assume that F0 : M n ! Rn+1 is an admissible smooth and convex immersion of an orientable compact M n and that f : ! R is a smooth function such that for all x 2 we have 0 ; f > 0 ff x2 ax ff x 00 00 0 0 0 0 0 ff 00 x + ff 0 ; (f ) 2 x 0 0 Harnack Inequalities 105 where a 2 R is a constant. Then we can nd a small positive constant d such that @ i j @t f + 2hrf V i + hij V V + cf H 0 holds for all tangent vectors V as long as t < T , d + (a + 2)t > 0 and Mt stays convex, where we have set c(t) := d+(a1+2)t Remark. We do have to make these assumptions on f to avoid negative terms in the evolution equation for the basic Harnack expression Z (see below). f = x satises the assumptions in Theorem 1 on = (0 1) with a = ; 1. The following 0 Propositions show that almost all functions satisfying the assumptions in Theorem 1 are of this form. Proposition IIa. Assume that f : (0 a) ! R 0 < a 1 is a smooth function that smoothly extends to x = 0 and satises all assumptions in Theorem 1. Let us set if := minfl 0jf (l)(0) 6= 0g 1. If 0 < if < 1 then f = Aif xif with a positive constant A. Proof. Since f (0) = 0 and f (x) > 0 8x 2 = (0 a) we observe that f (x) > 0 8x 2 . By de l'Hospital's rule we obtain 0 lim f x = if x 0 f 0 ! and f x =i ;1 lim f x 0f 00 0 ! and then f x 0 and f 0 imply (1.1) f x (if ; 1)f 8x Since f > 0 we have by assumption ; f 00 0 0 0 00 0 j ff x + ff 00 and then also (1.2) But (1.1) and (1.2) imply 0 ff x + ff and therefore 00 ; (f )2x = f 2; ff x 0 0 0 0 0 f x if f 8x 0 0 ; (f )2x (if ; 1)ff + ff ; if ff 0 0 ; which implies that f x = 0 f 0 0 f x=i f f 0 and after integration f = Aif xif with a positive constant A (since f > 0). 0 0 0 =0 Knut Smoczyk 106 Proposition IIb. Assume that f : (a 1) ! R 0 < a < 1 is a smooth function that satises all assumptions in Theorem 1 and that g(x) := ;f ( x1 ) smoothly extends to x = 0. If 0 < ig < 1 then f = ;Aig x ig with a positive constant A. Proof. If we set y = x1 and g(y) = ;f (x) then the assumptions for f and x translate into the same assumptions for g and y and as above we can derive the desired result. ; Remark. f (x) = px does not satisfy the assumptions in Proposition IIa since it does not extend smoothly to x = 0. f (x) = x + 1 satises all assumptions in Theorem 1. f (x) = ln x satises almost all assumptions in Theorem 1. The only condition which is violated is that ff x + ff ; (f )2 x = ; x1 < 0 on = (0 1). 00 0 0 2. Some relations and evolution equations By assumption we have @ @t F = ;f (2.1) In 7] we formally derived the evolution equations for various geometric objects on M n , these are: @ @t gij = ;2fhij @ @t d = ;Hfd (2.2) (2.3) (d denotes the volume form) (2.4) (2.5) (2.6) (2.7) @ @t = rf @ l @t hij = ri rj f ; fhi hlj @ 2 @t H = f + f jAj @ 2 @t jAj = 2hhij ri rj f i + 2fC In addition we have the well-known Gau-Weingarten-Codazzi-Mainardi equations (2.8) (2.9) (2.10) (2.11) (2.12) (2.13) ri rj F = ;hij F = gij ri rj F = ;H rihjk = rj hik ri = hil rl F ri rj = rlhij rlF ; hil hlj = rH ; jAj2 Harnack Inequalities 107 Note that in these equations we assume that F are sets of n + 1 functions on M n. We also have the Simons identity (2.14) ri rj H = hij ; Hhilhlj + jAj2hij 3. Homothetic solutions A homothetic solution Ft is a family of dieomorphisms such that the surfaces given by the rescaled dieomorphisms Fet := F are stationary in Rn+1 , where denotes a function depending only on time t. The assumption that Fet represents a stationary surface means that the normal velocity must be zero. So we have @ F @ hF i ; f e 0 = h @t ei = @t Let us dene c := ; @t@ ln . Then we have shown that for a homothetic solution of (?) we have (H.1) f = ;chF i Taking covariant derivatives of f and using (2.11), (2.10), (2.8) we obtain with e Vi := chF ri F i (H.2) (H.3) (H.4) (H.5) (H.6) ri f = ;hil Vel ri Vej = cgij + fhij ri rj f = ;rl hij Vel ; chij ; fhilhlj f + f jAj2 = ;hrH Ve i ; cH Vej = f rj H + hji ri f 4. The Harnack inequality In the sequel we have to calculate many evolution equations. To avoid too complicated formulas it is most convenient to work with coordinates associated to a moving frame. We use similar moving frame coordinates as in 4]. Here the moving frame fEa ga=1:::n evolves according to @ i i j @t Ea = fh j Ea and we denote the coordinates of a vector V with respect to the moving frame by yai . The following calculations closely follow the procedure in 4]. As in this paper (4.1) we have (4.2) (4.3) (4.4) rba := yai @y@ b i @ k i Da := ya ( @xi ; ;ij ybj @y@ k ) b ab := gac rcb ; gbc rca Knut Smoczyk 108 In addition we dene @ + fh b ra Dt := @t a b (4.5) Then straightforward computations give the commutator relations r (4.12) ; r ; @ D ] = D (fh d ) + D (fh d ) ; Dd (fh ) @t a a b ab b a ; (4.9) (4.10) (4.11) r Da Db ] = Rdeba ed a Dc ] = I c Da a D ] = I a D b b b c c b bc Da ] = Iba Dc Ica Db (4.6) (4.7) (4.8) rd b Dt Da ] = Db (fhac )bc + fhab Db Da ] = Db (Rdlab rld) + Rdlab rldDb Dt ; f Da ] = f Ra bcd Db cd + f Dahbc hbd cd + ff Da f 00 0 0 0 +(f ; f H )ha 0 (4.13) Dt ] = ( ff 0 0 lD l+f 0 han hnl Dl ; H )DnfDn + 2fhabDbDa + 2hanDafDn + Da(Db (fhac)bc) In the moving frame the evolution equations reduce to (4.14) Dt gab = 0 gab = Iab (4.15) Dt hab = Da Db f + fhan hnb (4.16) Dt f = f (f + f jAj2 ) Let us now dene the following tensors, where we assume that Va is an arbitrary tangent vector on M n and c a smooth function to be determined later and only depending on time t. 0 Xa := Da f + hal Vl Yab := Da Vb ; fhab ; cgab Z := Dt f + 2hDf V i + hab V a V b + cf H Wab := Dt hab + Dl hab Vl + chab W := Dt f + hDf V i + cf H 0 0 By (H.1){(H.6) these tensors vanish on homothetic solutions if we take Va = Vea and the induced c. On the other hand we have @ (chF D F i) + fh d rc Ve Dt Vea = @t a c d a @ (lnc)Ve ; chF D (f )i + fh d Ve = @t a a a d @ (lnc)Ve ; chF iD f = @t a a Harnack Inequalities 109 and therefore @ (ln c)Ve + f h b h cVe (Dt ; f )Vea = @t a a b c (4.17) 0 0 and on a homothetic solution @ (ln c)Ve ; f h Db f (Dt ; f )Vea = @t a ab (H.7) 0 0 In view of (H.7) we dene @ (ln c)V + f h Db f Ua := (Dt ; f )Va ; @t a ab 0 0 which also vanishes on a homothetic solution. We want to calculate (Dt ; f )Z . We do this in several steps. Using (4.12) and (4.16) we obtain 0 (Dt ; f )Da f = ff Da f f + (f ; f H )hab Db f + f han hnl Dl f + Da(ff jAj2 ) 00 0 0 0 0 0 and therefore (4.18) (Dt ; f )Da f =f jAj2 Da f + 2ff hbcDa hbc + f han hnl Dl f 0 0 0 0 + (f ; f H )hab Db f + (ff )2 Da fDtf 00 0 0 Further, we use Simons identity to rewrite (4.15) Dt hab = Da Db f + fhan hnb = f (hab ; Hhan hnb + jAj2 hab ) + (ff )2 Da fDb f + fhan hnb 00 0 0 This gives (Dt ; f )hab = f jAj2 hab + (f ; f H )han hnb + (ff )2 Da fDbf 00 (4.19) 0 0 0 0 (4.18), (4.19) and the denition of U give (Dt ; f )Xa = f jAj2 Da f + 2ff hbc Da hbc + f han hnl Dl f + (f ; f H )hab Db f 0 0 0 0 0 + (ff )2 Da fDtf + V b (f jAj2 hab + (f ; f H )han hnb + (ff )2 Da fDbf ) 00 00 0 0 0 0 @ (ln c)V ; f h Dc f ) ; 2f Dc h b D V + hab (Ub + @t b bc a c b = f jAj2 Xa + (f ; f H )hab Xb + hab Ub ; 2f Da hbc(Y bc + cgbc) 0 0 0 0 0 @ (ln c)(X ; D f ) + f D f (W ; cf H ) + @t a a (f )2 a 00 0 0 Knut Smoczyk 110 and nally (4.20) (Dt ; f )Xa =f jAj2 Xa + (f ; f H )hab Xb + hab Ub ; 2f Da hbc Y bc 0 0 0 0 @ (ln c)X + f D fW ; (2c + @ (ln c) + c f H )D f + @t a a (f )2 a @t f 00 00 0 0 Next we compute (Dt ; f )Dt f =Dt f ]f + Dt (ff jAj2 ) =f Dt ]f + Dt f f + Dt f f jAj2 + f jAj2 Dt f + ff Dt jAj2 =(f ; f H )jDf j2 + 2ff hab Da Db f + 2f hab Da fDb f 0 0 0 0 0 0 0 0 0 0 0 + (ff )2 (Dt f )2 + f jAj2 Dt f + 2ff hab Dt hab 00 0 0 0 giving (4.21) (Dt ; f )Dt f =f jAj2 Dt f + 4ff hab Dt hab ; 2f 2f C + (ff )2 (Dt f )2 00 0 0 0 0 + (f ; f H )jDf j 0 0 2 + 2f 0 h ab Da fDb f Furthermore @ (ln c)cf H + c f HD f + cD f (Dt ; f )(cf H ) = @t t t f 00 0 0 0 ; cf (H (f 0 H + f jDH j2 ) + f H + 2f jDH j2 ) @ (ln c)cf H + c(f H + f )(f + f jAj2 ) = @t ; c(f H + f )(f H ) ; cf (Hf + 2f )jDH j2 @ (ln c)cf H + cf (f + f H )jAj2 + c((f )2 H ; f f = @t 0 00 000 0 00 00 0 0 0 00 0 0 0 0 000 00 00 00 0 00 ;f f 0 000 H )jDH j2 which gives @ (ln c)cf H + cf (f + f H )jAj2 ; c; f H jDf j2 (4.22) (Dt ; f )(cf H ) = @t f 00 0 0 0 0 00 0 0 Harnack Inequalities 111 (4.21), (4.22) and (4.18) give (Dt ; f )W =f jAj2 Dt f + 4ff hab Dt hab ; 2f 2f C + (ff )2 (Dt f )2 + (f ; f H )jDf j2 00 0 0 0 0 0 0 @ (ln c)cf H + cf (f + f H )jAj2 ; c; f H jDf j2 + 2f hab Da fDb f + @t f 00 0 0 0 00 + V a (f jAj2 Da f + 2ff hbc Da hbc + f han hnl Dl f 0 0 0 0 0 + (f ; f H )hab Db f + (ff )2 Da fDt f ) 00 0 0 @ (ln c)V ; f h Db f ) ; 2f Da V b D D f + Da f (Ua + @t a ab a b 0 0 ; ff H )jAj2 + @t@ (ln c)W ; @t@ (ln c)Dtf + (ff )2 Dt fW ; c ff HDt f + (f ; f H )hX Df i + f (4fhab ; 2Da V b )Dt hab + 2ff Da V b han hnb ; 2f 2f C ; c; ff H jDf j2 + 2ff V a Dahbchbc + hDf U i + f habX aDbf @ (ln c)W =f jAj2 W ; c((f )2 H ; ff ; ff H )jAj2 + @t ; (c ff H + @t@ (ln c))Dt f + (ff )2 DtfW + (f ; f H )hX Df i + f (4fhab ; 2Da V b )Wab ; f (4fhab ; 2Da V b )(Dl hab Vl + chab ) ; + 2ff Da V b han hnb ; 2f 2 f C ; c ff H jDf j2 + 2ff V a Da hbc hbc + f hhab Da fXb i + hDf U i =f jAj2 W ; c((f )2 H ; ff 0 0 00 00 0 0 0 00 0 0 0 00 0 0 0 0 0 0 0 0 00 00 00 0 0 0 0 0 00 0 0 0 0 0 0 and after rearranging terms we conclude (4.23) (Dt ; f )W =f jAj2 W + f (4fhab ; 2Da V b )Wab + f hab Xa Db f + hDf U i 0 0 0 0 @ (ln c))W + 2f (fhan hnb + Dl hab Vl )Y ab + (2c + @t 0 ; c((f )2 H ; ff ; ff H )jAj2 ; (c ff H + @t@ (ln c) + 2c)Dtf ; + (ff )2 Dt fW + (f ; f H )hX Df i ; c ff H jDf j2 + 2cf hab Y ab 00 0 0 00 0 00 00 0 0 Eventually the evolution equation for Z is given by 0 0 0 Knut Smoczyk 112 (Dt ; f )Z =f jAj2 W + f (4fhab ; 2Da V b )Wab + f hab Xa Db f + hDf U i @ (ln c))W + 2f (fhan hnb + Dl hab Vl )Y ab + (2c + @t 0 0 0 0 0 ; c((f )2 H ; ff ; ff H )jAj2 ; (c ff H + @t@ (ln c) + 2c)Dtf ; + (ff )2 Dt fW + (f ; f H )hX Df i ; c ff H jDf j2 + 2cf hab Y ab + V a (f jAj2 Xa + (f ; f H )hab Xb + hab Ub ; 2f Da hbc Y bc @ (ln c)X + f D fW ; (2c + @ (ln c) + c f H )D f ) + @t a a (f )2 a @t f @ (ln c)V ; f h Db f ) ; 2f D V (W ab + han Y b ) + X a(Ua + @t a ab a b n 2 ab ab an b =f jAj Z + 2hX U i ; 4f (Y + cg )Wab ; 2f h Yn (Yab + cgab ) @ (ln c) + 2c)(D f + hDf V i) + (f ; f H )jX j2 + (ff )2 W (Dt f + hV Df i) ; (c ff H + @t t @ (ln c)hX V i + (2c + @ (ln c))W ; c((f )2 H ; ff ; ff H )jAj2 + 2 @t @t ;f ; c f H jDf j2 + 2cf habY ab =f jAj2 Z + 2hX U i ; 2f han Ynb Y ab ; 4f Wab Y ab ; 4cW @ (ln c) + 2c)(W ; cf H ) + (f ; f H )jX j2 + (ff )2 W (W ; cf H ) ; (c ff H + @t @ (ln c)hX V i + (2c + @ (ln c))W ; c((f )2 H ; ff ; ff H )jAj2 ; c; f H jDf j2 + 2 @t @t f 00 0 0 00 0 00 00 0 0 0 0 0 0 0 0 00 00 0 0 0 0 0 0 0 00 00 0 0 0 0 00 0 0 00 0 0 0 0 0 00 00 0 0 0 0 0 00 0 0 00 0 0 So we nally arrive at (4.24) ; (Dt ;f )Z = f jAj2 + (ff )2 (Z ; 2hX V i) ; 2(c ff H + 2c) Z 00 0 00 0 0 0 @ (ln c) + 2c)hX V i + 2hX U i + (f ; f H )jX j2 + (ff )2 hX V i2 + 2(c ff H + @t 00 00 0 0 0 ; 2f han YnbY ab ; 4f Wab Y ab + (c ff H + @t@ (ln c) + 2c)cf H ; + c(ff H + ff ; (f )2 H )jAj2 ; c ff H jDf j2 00 0 0 0 0 00 00 0 0 0 0 Now we are able to prove Theorem 1. Proof of Theorem 1. From (4.24) and the assumptions in Theorem 1 we conclude, with c := d+(a1+2)t (Dt ; f )Z f jAj2 + (ff )2 (Z ; 2hX V i) ; 2(c ff H + 2c) Z ; 2f han Ynb Y ab ; 4f Wab Y ab 0 ; 00 00 0 0 0 0 0 @ (ln c) + 2c)hX V i + 2hX U i + (f ; f H )jX j2 + (ff )2 hX V i2 + 2(c ff H + @t 00 00 0 0 0 Harnack Inequalities 113 Now choose d so small such that Z > > 0 for t = 0 and for all tangent vectors V . This is possible since the initial surface is convex. On any compact time interval 0 t0 ] with t0 < T we can therefore estimate (Dt ;f )Z ;bZ ; 2f han Ynb Y ab ; 4f Wab Y ab @ (ln c) + 2c)hX V i + 2hX U i + (f ; f H )jX j2 + (ff )2 hX V i2 + 2(c ff H + @t 0 0 0 00 00 0 0 0 with a large constant b depending on t0 . Additionally for t > 0 and an arbitrary positive constant (Dt ; f )(ebt Z + t) >ebt ;2f han Ynb Y ab ; 4f Wab Y ab + 2hX U i + (f ; f H )jX j2 0 ; 0 0 0 @ (ln c) + 2c)hX V i + (ff )2 hX V i2 + 2(c ff H + @t 00 00 0 0 If t1 t0 would be the rst time where ebt Z + t would become zero at some point x 2 M n and for some tangent vector V then we must have Xa = 0 since it is the rst variation with respect to V and we can also extend V in spacetime such that Yab = 0. This implies a contradiction and therefore ebt Z + t > 0 8t t0 . Since and t0 are arbitrary we conclude Z 0 whenever t < T and d + (a + 2)t > 0. Now we want to answer the question for which f one can expect selfsimilar solutions of (?). First we remark that if x 2 and x > 0, then the sphere of constant radius nx and with constant mean curvature H = x gives always rise to a selfsimilar solution for a short time. For any subset A M n let us dene H (A) := fx 2 j9p 2 A : H (p) = xg and Pt := fp 2 MtnjrH 6= 0g The answer to question III is then given by Proposition IIIa. If P0 6= and Ft : M n ! Rn+1 is a selfsimilar solution of (?) for a compact connected M n , then we have f = Ax 8x 2 H (Mt ) with nonvanishing constants A and . Proof. Since Ft is selfsimilar we have Pt = P0 8t 2 0 T ). Since Xa W vanish on selfsimilar solutions so must their time derivatives. From (4.20) we conclude that on H (Pt ) and by continuity also on H (Pt ) we must have c ff x + 2c + @t@ (ln c) = 0. Since P0 6= and M n is connected we have H (M n) H (Pt ) and we derive 00 0 @ (ln c) = 0 c ff x + 2c + @t 00 0 for all x 2 H (Mtn). Since f > 0 there can be at most one point x 2 such that f (x) = 0. At all other points we have 0 ff x + ff 00 ; (f )2 x = f 2; ff x 0 0 0 0 Knut Smoczyk 114 Then (4.23) implies that also ; f x = 0 f 0 0 in all points x 2 H (Mt) where f (x) 6= 0. But after integration we obtain that at these points f = Ax with constants A (nonvanishing since we must have f > 0) and again by continuity this also holds on all of H (Mt ). 0 We observe that by (2.9) jF j2 = 2(n ; H hF i) and with (H.1) we obtain on a homothetic solution jF j2 = 2(n ; fH c ) (H.8) This implies Proposition IIIb. If P0 = and M nn is a compact1 orientable selfsimilar solution of (?), then Mt is a sphere of radius H . If f = ; H , then any compact orientable homothetic solution is a sphere of radius Hn . Proof. P0 = and (H.8) imply jF j2 = const. By the assumptions on M n this constant must be zero and then consequently jF j2 = const. This implies that Mt is a sphere. 5. Longtime existence for some highly nonlinear ows In this paragraph we are going to show that one can use the Harnack inequality to prove longtime existence of solutions for some highly nonlinear ows. To be precise Theorem 2. Assume that f : (0 1) ! R is a smooth negative function that satises the assumptions in Theorem 1 with a > ;2 and that limx 0 f (x) = ;1 and F0 : M 2 ! R3 is a smooth convex immersion of an orientable compact surface M 2 . Then (?) has a smooth immortal solution and we have lim H = 0 tlim jFt j2 = 1 t ! !1 !1 ; Remark. The functions f = x with 1 < < 0 satisfy the assumptions in Theorem 2 on = (0 1). These speed functions are not homogenous of degree one and are not included in the class of functions considered in 2] To prove Theorem 2 we need some lemmas that are interesting on their own. Lemma 1. Assume f > 0 and that F0 : M n ! Rn+1 is a smooth immersion of an orientable compact surface and that on M0 = F0 (M n ) we have f 2 minM0 f 2 > 0. Then this is also true on the maximal time interval 0 T ) where a smooth solution of (?) exists. 0 Harnack Inequalities 115 Proof. The evolution equation for f 2 is given by (Dt ; f )f 2 = f 0 0 ;2jrf j2 + 2f 2jAj2 ; and the result follows from the parabolic maximum principle. Corollary 1. With the assumptions in Lemma 1 we have that for negative f H max H M 0 and for positive f H min H M0 We can do even better Lemma 2. Assume = (0 1) f < 0 f > 0 and that F0 : M n ! Rn+1 is an admissible immersion of an orientable compact manifold M n. Then we can nd a positive such that on 0 T ) 0 M0 H max H 1 +max Mt t maxM0 H Proof. Since Ft is admissible on 0 T ) we must always have H > 0. At a point where maxMt H is attained we have H 0 and rH = 0. The evolution equation for H (2.6), Lemma 1 and the fact that jAj2 Hn2 give us @ H ;(max H )2 Mt Mt @t max with = ; maxM0 f n and after integration we obtain the result. (2.7) and Simons identity give (5.1) (Dt ; f )jAj2 = ;2f jrAj2 + 2f jAj4 + 2(f ; f H )C + 2f hij ri H rj H 0 0 0 0 00 Let R = H 2 ; jAj2 be the Scalar curvature of Mt. Now we turn our attention to the case where n = 2. In this case we can decompose C into C = H2 (3jAj2 ; H 2 ) Then (5.1) and (2.6) imply that for n = 2 (5.2) (Dt ; f )R =2f (jrAj2 ; jrH j2 ) + 2f (H jrH j2 ; hij ri H rj H ) ; (2f jAj2 + (f ; f H )H )R 0 0 00 0 0 Knut Smoczyk 116 Lemma 3. Under the assumptions in Theorem 2 we have R > 0 8t 2 0 T ) Proof. Let t0 T be the maximal time such that R > 0 on all of Mt for t 2 0 t0), i.e., the maximal time for which Mt stays convex. If t0 = T we are done. So assume that t0 < T . Since Mt is admissible for t 2 0 t0 ] we must have H > 0 on 0 t0 ]. By denition of t0 we also have R > 0 on 0 t0 ) and R 0 on 0 t0 ]. Since M and 0 t0 ] are both compact and R 0 we can nd a constant c such that (5.3) (Dt ; f )R 2f (jrAj2 ; jrH j2 ) + 2f (H jrH j2 ; hij ri H rj H ) ; cR 0 0 00 Now assume that x is any point on Mt where the minimum of R is attained. At this point we must have f R 0 and rR = 0. Let and denote the two principal curvatures in a neighborhood of x. We have 0 0 = rR = 2r + 2 r Since H > 0 and 0 we must either have > 0 or > 0. Let us assume that > 0. First we compute that H jrH j2 ; hij ri H rj H = jr2 H j2 + jr1 H j2 0 where we choose normal coordinates such that hij = diag( ). Since f x af and H > 0 a > ;2 we can estimate 00 0 f (jr H j2 + jr H j2 ) 2f (H jrH j2 ; hij ri H rj H ) ;4 H 2 1 0 00 On the other hand Codazzi's equation gives us that jrAj2 = jr1j2 + jr2j2 + 3jr2j2 + 3jr1 j2 Combining the last two statements we see that at a point where rR = 0 we must always have 2f (jrAj2 ; jrH j2 ) + 2f (H jrH j2 ; hij ri H rj H ) 0 0 00 and consequently at any point where the minimum of R is attained Dt R ;cR which implies that min R (min R)e M M t ; 0 This proves that t0 = T . We come to the proof of Theorem 2 ct > 0 8t 2 0 t0 ] Harnack Inequalities 117 Proof of Theorem 2. On 0 T ) we must have H > 0: Otherwise (?) is not well- dened, i.e., the surfaces would not be admissible. If T is nite, then this can only happen for two reasons. Either the solutions converge to a surface that is no longer admissible, i.e., the mean curvature would vanish somewhere, or the surfaces must develop a singularity. Since Mt is convex on 0 T ) we can apply Theorem 1 with V = 0 to obtain @ 1 @t H ; d + (a + 2)t H and since by assumption a > ;2 we can estimate @ H ;1H which implies that @t d H e d1 t min H M ; 0 So H cannot become zero in nite time. This estimate and Corollary 1 imply that the surfaces stay admissible in nite time. Therefore we conclude that if T is nite the surfaces must develop a singularity. It is well-known that the second fundamental form jAj2 must then blow up for t ! T (compare 2] and 5]). By Lemma 3 we conclude that in the case where T < 1, H must also blow up which, in view of Corollary 1, proves that T = 1. By Lemma 2 we conclude that limt (maxMt H ) = 0. It remains to prove that jFt j ! 1. An easy calculation gives the following evolution equation !1 (Dt ; f )( hFf i + (2 + a)t) = 2 ff hr( hFf i + (2 + a)t) rf i + a + 1 ; f fH 0 0 0 The assumptions on f imply that limx 0 ff x exists and that limx 0 ff x a. Since limx 0 f = ;1 we can apply de l'Hospital's rule and get limx 0 ffx = ; 1 + limx 0 ff x a + 1. On the other hand the assumption that ffx 0 implies for any x 2 a + 1 ; ffx 0 00 ! 00 ! 0 0 0 ! ! 0 00 ! 0 0 0 This gives (Dt ; f )( hFf i + (2 + a)t) 2 ff hr( hFf i + (2 + a)t) rf i 0 0 and with the maximum principle we conclude hF i c ; (2 + a)t f for a constant c. Using Schwarz' inequality we have proven that jF j (2 + a)t ; c jf j and since limt jf j ! 1 and 2+ a > 0, this can only hold when limt proving the rest of Theorem 2. !1 !1 jF j = 1, Knut Smoczyk 118 References 1] B. Andrews, Harnack inequalities for evolving hypersurfaces, Math. Zeitschrift 217 (1994), 179{197. 2] C. Gerhardt, Flow of nonconvex hypersurfaces into spheres, J.D.G 32 (1990), 299{314. 3] R.S. Hamilton, The Harnack estimate for the Ricci Flow, J.D.G 37 (1993), 225{243. 4] R.S. Hamilton, Harnack estimate for the mean curvature ow, J.D.G 41 (1995), 215{226. 5] G. Huisken, Flow by mean curvature of convex surfaces into spheres, J.D.G 20 (1984), 237{ 268. 6] G. Huisken, T. Ilmanen, Proof of the Penrose inequality, to appear. 7] K. Smoczyk, Symmetric hypersurfaces in Riemannian manifolds contracting to Lie-groups by their mean curvature, Calc. Var. 4 (1996), 155{170. 8] S.T. Yau, On the Harnack inequalities of partial dierential equations, Comm. Anal. Geom. 2 (1994), 431{430. ETH Zurich, Math. Department, CH-8092 Zurich, Switzerland ksmoczyk@math.ethz.ch Typeset by AMS-TEX