Concentration on curves for nonlinear Schrödinger equations Manuel del Pino Universidad de Chile Michal Kowalczyk Kent State University and Universidad de Chile AND Jun-Cheng Wei Chinese University of Hong Kong Abstract We consider the problem !#"$&%'( where ) #* , '+ is a small parameter and is a uniformly positive, smooth potential. Let , be a closed curve, non-degenerate geodesic relative to the weighted " " . We prove the existence of a solution 87 arclength -. 0/ , where 1 3$24 56" concentrating along the whole of , , exponentially small in at any positive distance from it, provided that is small and away from certain critical numbers. In particular this establishes the validity of a conjecture raised in [3] in the two9 dimensional case. c 2000 Wiley Periodicals, Inc. 1 Introduction and statement of main result We consider standing waves for a nonlinear Schrödinger equation in :<; form B6C (1.1) =?>A@ B6DFE @(GIH C =FJLKM6N C of the CPORQ CQ SUTWVXC D_^ TWV D NedfKM6Ng Assumwhere Y[Z]\ , namely solutions of the form K M`N E exp K>bac@ C ing that the amplitude dKM`N is positive and vanishes at infinity, we see that this satisfies (1.1) if and only if d solves the nonlinear elliptic problem (1.2) where iKM6N E function with O S ^ ^ V ^ Elk dmZ k @(G$Hhd=jiKM`Ned d dPno K: ; N O JLKM6N a . In the rest of this paper we will assume that i prq`s tuv`w imKM6NxZ k g Communications on Pure and Applied Mathematics, Vol. 000, 0001–0033 (2000) y c 2000 Wiley Periodicals, Inc. is a smooth 2 M. DEL PINO, M. KOWLCZYK, J.-C. WEI Considerable attention has been paid in recent years to the problem of construction of standing waves in the so-called semi-classical limit of (1.1) @{z k . In the pioneering work [14], Floer and Weinstein constructed positive solutions to this problem when Y El| and } E \ with concentration taking place near a given point M8~ with iKM8~N Ek , i# KM8~NR E k , being exponentially small in @ outside any neighborhood of M8~ . More precisely, they established the existence of a solution dc such that dc (KM6NiPKM8~N where (1.3) TWV KiKM8~N w @ I KM=M8~NN is the unique solution of = O S Elk ^ ^ k dPZ ^ K k N Elk KN Ek g This result has been subsequently extended to higher dimensions to the construction of solutions exhibiting high concentration around one or more points of space under various assumptions on the potential and the nonlinearity by many authors. We refer the reader for instance to [2], [4], [7]-[13], [15, 16, 17, 18, 21, 31, 32, 34, 36, 37, 38]. An important question is whether solutions exhibiting concentration on higher dimensional sets Q exist. In [3], Ambrosetti, Malchiodi and Ni have Q considered the case of i E iPK M N , also treated in [5, 6], and constructed radial Q Q Q Q solutions d K M N exhibiting concentration on a sphere M ER ~ in the form _ d K NWimK ~ N KiPK ~ N w @ TWV K = ~ NXN ^ under the assumption that ~Z k is a non-degenerate critical point of TWV K N ER ; (1.4) where (1.5) E O Y \ Y=\ = i#'K N \ ^ and is again the unique solution of (1.3). The conjecture is raised in [3] that this type of phenomenon takes place, at least along a sequence @ E @#z k , whenever Q Q the sphere M ER ~ is replaced by a closed hypersurface , which is stationary and non-degenerate for the weighted area functional - i . In this paper we prove the validity of this conjecture in dimension } E . For } E , the functional above, defined on closed Jordan curves , has a simple geometrical meaning: itO corresponds to the arclength of measured with re8M G N in : G . Thus we will establish the concentration spect to the metric i K$8M VG G phenomenon on , provided that this curve is a non-degenerate closed geodesic for this metric in : G . We do not prove the result for all small values @ Z k but only for those which lie away from certain critical numbers. More precisely, there is an explicit number a'Z k such that given ¡Z k , if @ is sufficiently small and satisfies the gap condition (1.6) Q¢ GI@G=ja' Q¤£ $@ ^ for all ¢ nL¥ ^ CONCENTRATION ON CURVES 3 then a solution dc with the required concentration property indeed exists. In other words, this will be the case whenever @ is small and away from the critical numbers ¦ §(¨ © , in the sense that for fixed and arbitrarily small «ª¬ a' , @[n® ­ ¬ ¢ a ¢ = ^ ¬ G ¢ a O ¢ ¯ ^ ¢ for all G n¥®g To state our main result, we need to make precise the concept of a curve being stationary and non-degenerate for the weighted length functional - i . Q Q Let be a closed smooth curve in : G and ° E its total length. We consider natural parameterization ±K²³N of with positive orientation, where ² denotes arclength parameter measured from a fixed point of . Let ´K²³N denote outer unit normal to . Points M which are µ$~ -close , for sufficiently small µ$~ can be represented in the form M E (1.7) OFD ±K²³N ^ ´K²³N Q DIQ ^ ªRµ$~ ^ ²nF­ k ^ °N D_^ where map M¶z·K ²³N is a local diffeomorphism. Any curve sufficiently close to can be parameterized as ±¹¸ºK²³N E O» ±¡K²³N K²¹N¼´K²³N » where is a smooth, ½ -periodic function with small ¾«¿ -norm. Call ¸ the curve D^ defined this way. By slight abuse of notation we denote iPK ²³N to actually mean iPKM6N for M in (1.7). Then weighted length of this curve is given by the functional » of À K » NÁ Q Q ÂÄ i Kű¹¸³K²³NXN ± ¸ K²³N º² ~ E i Uà E Q Q  ~ ¢ Ä i K » ^ Q Oj» Oj» Q ´ ³²`g K²³N ³ ² N ± ´ ¢ K²³NA±' where K²³N denotes curvature of , we get that the Since ±' E \ and ´` E above quantity becomes (1.8) À K » N E » ÂÄ i K ~ ^ O[¢³» O K²³N ³ ² N$­XK¼\ N G » K N G ¯ w ³²`g is said to be stationary for the weighted length - i if the first variation of » the functional (1.8) at EÆk is equal to zero. That is, for any smooth, ° -periodic function ÇK²³N À kPE ¯ K k N$­ÈÇ E ÂmÄ ­Ki NeÉAÇ ~ O i ¢ Ç ¯ ^ º² which is equivalent to the relation (1.9) i6ÉK k ^ ²³N E = ¢ K²³NXiK k ^ ²¹N ^ for all ²+nÊK k °N0g 4 M. DEL PINO, M. KOWLCZYK, J.-C. WEI We assume the validity of this relation at . Let us consider now the second variation quadratic form À ^ K k N$­ÈÇ Ç ¯ E \  E \  ~ ~ Q Q O Ä ­i Ç G Ä ­i Q O ­Ki NeÉÉ Q O Ç G ­Ki NeÉÉË= ¢ ¯ Ki NeÉ ¢ i ¯ G ¯ Ç G Ç G ¯ º² º² We say that is non-degenerate if so is this quadratic form in the space of all ^ V functions Çmnxo K k °N with ÇK k N E ÇKÅ°N . This is equivalent to the statement that the differential equation Ki Ç N =Ì­Ki N ÉÉ = has only the ° -periodic solution ÇjÁ problem Ç (1.10) O i TWV icÎIÇ Ï = ­i ^ ÇKÅ°N Ç K k N E ÇK k N E k ¢ i G ¯ Ç EÍk , or using (1.9), that the boundary value TWV i`Éɳ=ÏK ^ Ç KÅ°N TWV O \N ¢ G ¯ EÍk Ç ^ has only the trivial solution. As an example, let us consider the radial case i E iPK N . Then we see that EÑÐÒEl ~¤Ó is stationary precisely if ÔK ~N Ek where K ~INZ k , namely if ~ is a non-degenerate is defined by (1.4). If in addition ] O ¢ local minimizer,À we have that Ki NeÉÉ Ki NeÉ Z k . This makes automatically the ^ ¯ quadratic form K k N$­ÈÇ Ç positive definite, hence non-singular. A non-degenerate stationary curve close to will still be present if the radial potential is modified by small non-radial perturbations. The geometric interpretation allows the construcQ Q V tion of other examples. If i KM6NÕ Ö Ve×Ø t Ø wXÙÅw up to a large value of M then the metric i ºM G represents approximately that of aQ sphere embedded in :<Ú . If eventually Q i increases so that iPKM`N4®\ for very large M , the whole metric will resemble that of a “globe attached to a plane”, the presence of at least two geodesics thus being clear: one on the globe, the other on the connecting neck. Non-degeneracy of these geodesics is not true in general, but may be generically expected in a suitably strong ÛÜ -topology. We need a further element to describe the gap condition (1.6). Let denote the unique positive solution of problem (1.3). We consider the associated linearized eigenvalue problem, (1.11) Ç`Ý Ý=jÇ O Y SUTWV Ç E a'Ç ^ ^^ in : ÇKN Elk g It is well known that this equation possesses a unique positive eigenvalue ac~ in V o K:N , with associated eigenfunction even and positive Þ which we normalize so that - v Þ G E \ (this follows for instance from the analysis in [30]). In fact, a simple computation shows that (1.12) ac~ E O \ ßKàY=á\N$KàY | N ^ Þ E Xã \ â - v S×V w g CONCENTRATION ON CURVES 5 We define the number a' as a' E (1.13) \ ac~ ß8ä  Gå ^ ~ Ä iPK k G ²³N w º²æ g Now we can state our main result. Theorem 1.1. Let be a non-degenerate, stationary curve for the weighted length functional - i , as described above. Then given ÒZ k there exists @(~{Z k such that for all @{ª@(~ satisfying the gap condition Q (1.14) @ G ¢ Q'£ G =a' ^èç¢ Ë@ ^ n¥ where acZ k is the number in (1.13), problem (1.2) has a positive solution d4é which near , for M given by (1.7), takes the form d é KM6N E (1.15) i Kk P ^ _ ²³N ^ i Kk P 6ëê D ² N w ³ For some number ~ Z k , d satisfies globally, pró3ô d KM6NÒîÌï$ð¹ñfK¤={ ~ @ KU\ @'ì TWVò KM ^ O®í K¼\NN0g fNºN¡g To explain in few words the difficulties encountered in the construction of these ä solutions, let us assume for the moment that iÑÁõ . Then in \ on and that ° E ^÷ _ D ^ TWV K ²³N the equation would look near terms of the stretched coordinates Kö N E @ the curve approximately like øù3ù O ä øúbú O ø S ÷ TWV = ø Ek Kö ^÷ ^ NfnL: G where ø is @ -periodic in the variable. The effect of curvature and of variations of i are here neglected. The linearization of this problem around the profile Kö¤N thus becomes û û û û ùù O û with û V ä E TWV @ ú úeú O ÷ Y STWV = ERk Kö û -periodic of the form ó3prq in . Functions ó Kö¤N$­ ü ¢ @ ÷<O[ý¹þ$ÿ ¢ @ ÷ ¯ ^ G E Þ KöUN$­ ü ^÷ ^ NnL:¡G óprq ¢ @ ÷0OÊý¹þ$ÿ ¢ ó ¢ @ ÷ ¯ ^ ¢ are eigenfunctions associated to eigenvalues respectively = G @ G and a6~0= G @ G . Many of these numbers are small and “near non-invertibility” of the linear operator thus occurs. Therefore the use of a fixed point argument after inverting the linear operator in the actual nonlinear problem is a very delicate matter. Worse than this, these two effects combined, in principle orthogonal because of the ¾ G orthogonality of Þ and ú , are actually coupled through the smaller order terms û neglected. In [1, 19, 20, 33] related singular perturbation problems involving the V Allen-Cahn equation in phase transitions exhibiting only the translation effect have been successfully treated through successive improvements of the approximation and fine spectral analysis of the actual linearized operator. The principle is simple: the better the approximation, higher the chances of a correct inversion of the linearized operator to obtain a contraction mapping formulation of the problem. 6 M. DEL PINO, M. KOWLCZYK, J.-C. WEI û In [24, 25, 28, 35] resonance phenomena similar to the “ G -effect” has been faced in related problems. In [24, 25] a Neumann problem involving whole boundary concentration, widely treated for point concentration after the works [23, 29, 30]. Recently in [27, 26] this boundary concentration on a geodesic of the boundary in the 3d-case has been treated via arbitrarily high order approximations. Our method, closer in spirit to that of Floer and Weinstein, provides substantial simplification and flexibility to deal with larger noise and coupling of the two effects inherent to this problem. The solution to the full problem in the above idealized situation is roughly decomposed in the form û ø Kö ^÷ N E ä Kö= K@ ÷ NXN O @ºK@ ÷ NXÞ Kö¡= ÷ K@ NXN O Kö û ^÷ N ^÷ Kö N is ¾ G K³öUN where and are ÷ -periodic functions left as parameters, while ÷ ÷ û K@ NXN and to Þ Kö= K@ NXN . Solving first in orthogonal for each both to ú Kö= a natural projected problem where the linear operator is uniformly invertible, the resolution of the full problem becomes reduced to a nonlinear, nonlocal second ^ order system of differential equations in K (N which turns to be directly solvable thanks to the assumptions made. This approach is familiar when the parameters ^ K ¤N lie in a finite-dimensional space, corresponding this time to adjusting infinitely many parameters. To stress out the difference with the radial case: the parameter is not present, and is just a single number. The analysis we make takes special advantage through Fourier analysis of the fact that the objects to be adjusted are one-variable functions, while we still believe that the current approach may be modified to the higher dimensional case. We also believe the gap condition may be improved to size ö(@ , any ZÆ\ . In the rest of the paper we carry out the program outlined above which leads to the proof of Theorem 1.1. 2 The set up near the curve Let be the curve in the statement of the theorem. We shall use the notation introduced in the previous section. Stretching variables, absorbing @ from Laplace’s operator replacing dfKM6N by dK@M6N , equation (1.2) becomes (2.1) Let Kö @ TWV ^÷ N E O S ^ ^ V Elk dZ k H d=FiLK@M6Ned d dnPo K:<G³N<g TWV D_^ @ K ²³N be natural stretched coordinates associated to , now defined for ÷ (2.2) n[­ k ^ @ TWV °N ^ ö{n[Kb={@ TWV µ$~ ^ @ TWV µ$~Ng Equation (2.1) for d expressed in these coordinates becomes (2.3) d ù3ù O d ^ ÷ O úeú O V KdWN`=FiPK@(ö @ Ned d S Elk the curve E CONCENTRATION ON CURVES 7 in the region (2.2), where V KdWN E d ùù ­ \= O K¼\ \ ¢ @ ¯ O ÷ K@ @ N¼öUN G \ ¢ K@ O @ ÷ Ned ¢ K@ ú ÷ O N¼ö K¼\ ¢ @ G ö O ¢ K@ @ K@ ÷ ù Ned ÷ g N¼öUN Ú For further reference, it is convenient to expand this operator in the form (2.4) V KdÕN E ¢ K@ ÷ K@ ¢ N`=x@(Gö ÷ GK@ ú O NXNed ^ ~8KdÕN where (2.5) ^ ÷ ù O @ Ned @(Göü V K@(ö ~8KdÕN E @¤öü ^ D_^ K@(ö G ^ ÷ ù3ù O @ Ned @ Ú öG$ü ^ ^ K@(ö Ú ^ ÷ ú @ Ned | . Observe that all terms in the operfor certain smooth functions ü8K ²³N E \ ator V have @ as a common factor. We consider now a further change of variables in equation (2.3) with the property that replaces at main order the potential i by \ . Let «K²³N E (2.6) ^ iK k ²³N ^ K²³N E ^ iPK k ²³N w ^ ÷ and fix a twice differentiable, ° -periodic function K²³N . We define ø K N by the relation dKö (2.7) ^÷ ÷ K@ N E ø N K ^U÷ ^ N E ÷ K@ N$Kö= K@ ÷ NXN0g We want to express equation (2.3) in terms of these new coordinates. We compute: ú d (2.8) ù d (2.9) d ùù E (2.10) ø ^ E ø O @ E d úbú E G ø ù O ø K K ö= NXN øù ù O ø ù ¯ ø O ø @ G @ ­ K K ö= NXN O Q Q O ø ù ù K K ö= ø K K ö¡= XN N G O ø ùù O øUù ù ¯ K Kö¡= XN N g NXN ù We also have K Kö= NXN ù E @¹­ Kö= ¯ ^ N³= K Kö= NXN ù3ù E @G8­ Kö= N³= = ¯ g In order to write down the equation it is convenient to also expand (2.11) iPK@(ö ^ ÷ @ N E iK k for a smooth function üºK by (2.7) solves: (2.12) K ø NÁ ^ ÷ @ D^ O N i6ÉK k ^ ÷ @ Nb@(ö O ^ ÷ O \ i ÉÉK k @ Nb@ G ö G ` ü8K@(ö ^ ÷ @ Nb@ Ú ö Ú . It turns out that d solves (2.3) if and only if ø defined ²¹N Ú K ø N O¡T G øù3ù O ø O ø S = ø Ek ^ 8 M. DEL PINO, M. KOWLCZYK, J.-C. WEI where K Ú Ú K ø ø N is a linear differential operator defined by ¡TWV ¢ N E @ ¢ =x@G G ê O" T G$# @ G&% % O G <T #@ = @ G ê % Gi`É ê % = @ G ø O % % O ø @ G % ø ì! = % O O @ G O @ ì øù ì ø ì ø O @ G <T Gi`ÉÉ O = ê = = ê øù ì ê O G' ø O G ì ø K N and (2.13) ø K G K( N E TWV G N O ~KdÕN K( G N TWV ü8K@(ö ^ ÷ @ Nb@ Ú ö Ú ~KdÕN is the operator in (2.5) where derivatives are expressed in terms of formulas TWV O . (2.8)-(2.10), ü is given by (2.11) and ö is replaced by Let KWN denote the unique positive solution of (1.3). Then, taking KÕN as a first approximation, the error produced is @ -times a function with exponential decay. Let us be more precise. We need to identify both the terms of order @ and those of order @ G : K N E K Ú TWV ¢ N E @ O" T G)# @ G % % O % @ G = # G T O G @¤i6É ê ê % G O % % @ ê O ì O ì! G % @ G = % ¢ =x@ G O = @ G @ G T G i6ÉÉ ê ê ì = = ì O O ì G' K N G ' CONCENTRATION ON CURVES G K N ?K 9 turns out to be of size @Ú . Gathering terms of order @ and @ G we get N E TWV ¢ @ = ¢ G ={@G # O @G G O ¢ K N O G @ G G O G Q G 4G G = \ O Ý Ý O+ G Q Q G i6ÉÉ, G G O+ @ G . Ú ^ ÷ @ N G i É Kk ' Q O Ú T =R@ i`ÉÉ O- N* O G O @ ÷ @ O G # = @ V G ¡T O E ^ i É Kk O O \ K G N³g \ i ÉÉ G ` = O ' ^ Let us observe that grouped this way, the quantities V are odd functions of Ú while , . are even. We want now to construct a further approximation to a G solution which eliminatesû the terms of order û @ in theû error. We û see that O K û where O N û ¡T N E ¾ ~ K (2.14) N E } ~8K (2.15) û We write K O G û N E û ­@³K/ V O @ G N N OG û O O O N SUTWV Y } ~8K û = Y ^ = û SUTWV g û K N G N û S = K Ú S TWV O Y U = O O T + @ G . Ú S O N û û O K O ¾~K 3ù ù O û and (2.16) ?K N E û û G ¯ 3ù ù O+ û K Ú N O û } ~8K Ng V in order to eliminate the term between brackets in the above We choose E ÷ expression. Namely for û fixed , weû need a solution ofû û = O SUTWV = Y @³K/ V E O G N ^ KN ERk As it is well known, this problem is solvable provided that (2.17)  ¿ T K/ V O ¿ N G 0 Elk g Furthermore, the solution is unique under the constraint û (2.18)  ¿ T ¿ 1 Elk g g 10 M. DEL PINO, M. KOWLCZYK, J.-C. WEI We compute  ¿ T O K/ V 0 E N G ¿ ¿  V T ¿ TWV ¢ 0 E G =Fi TWV i6É( ¿ ¿  T ¿ T ¢ ¿ Ò TWV g i6É where is the The assumption that is stationary (1.9) amounts to E ={i constant that precisely makes the amount between brackets identically k . In fact, G . The solution has the form we have the validity of the identity û - v û G 1 E û - v û where VXV E (2.20) Function V N ü V ^ V G E G @ ¡T = ^ V û E G O VXV V KWN <TWV3¢4^ ü VXV E (2.21) ÷ @(ü VXV K@ with ÷ K@ ^ Gi`É3K k Nbü V G ²³N E ÷ K@ ^ N TWV¢ G g the unique odd function satisfying = (2.22) and V E (2.19) O V32 SUTWV V = Y V E O \ ^  v V 1 Elk ^ is the unique even solution satisfying G = (2.23) O 2 G = G SUTWV Y E G g In fact, we can write (2.24) û û Substituting K (2.25) O E G @ G V ûN E \ = Y+=\ g in (2.16), we can compute the new error K û û V E û \ = O O @ G . Ú K N G O T O V N ù3ù G K Ú K û O V N O V N û , V Ng } ~K Observe that since V is of size 4LK@UN all terms above carry @ G in front. We compute û û û for instanceû (2.26) Ú K V N E @ ¡TWV ¢ K V N = ¡T ^ ÷ @ N* V65 =l@ ^÷ are expressed in K N Gi6ÉK k Observe that all functions involved domain for those variables is the infinite strip 9 E ^ ª:PªR Ð = k ª ÷ ¡T Gi6É3K k ^ ÷ @ N V O @ Ú ü87 variables and the natural ªá°;¤@Ó8g 9 We now want to measure the size of the error in ¾ G K N norm. A rather delicate term in the cubic remainder is the one carrying since in reality we shall only û assume a uniform bound on < <>= w Ö ~ 2 Ù . Observe that a similar one arises from the computation of K û V N ù3ù . Both of thoseÄ terms have similar form. For instance the ÷ ÷ ^ one arising from K V N ù3ù can be written as ? E @ Ú K@ Nbü V K@ N 2 KÕN with ü V G G G smooth. Observe that Â@ Q ? Q GîRÛ@ 7 B Q Â+A ~ K@ ÷ Q ÷ E N G @C0< < = w Ö~ 2 Ù g Ä CONCENTRATION ON CURVES Hence 11 @ <>?D<>= w Ö Ù îlÛ@FwE < <>= w Ö ~ 2 Ù g Ä While in total ¾ G -norm we have û <û K Ú @ V N<>= w Ö û Ù îRÛ@FGw g Q Q Let usQ consider the term } ~K V N . Since V can be bounded by Û@ G KÕN for Q large we obtainû that û û Q } Q V N ~8K Q E E O K YfK ûV N OFD hence Q S STWV û = Y V = S T Q Q ORQ Q S V N V G G îRÛ@GK¼\ N S û KWN S @ <} ~8K V N<>= w Ö Ù îlÛ@HGw g In summary, we have <K (2.27) û O @ V N<>= w Ö Ù î@ Gw g To improve the approximation for a solution still keeping terms of order @ G we need to introduce a new parameter, additional to . We let Þ KÕN be the first eigenfunction of the problem Þ O SUTWV Y E ÞÊ=FÞ ac~IÞ ^ Þ KN Elk g Then, as it is well known, ac~+Z k , Þ KWN is one-signed and even in . We now consider our basic approximation to a solution to the problem near the curve to be û O I E (2.28) O V @ºK@ ÷ NXÞõg In all what follows, we will assume the validity of the following constraints on the parameters and : O ^ < =8L Ö ~ 2 Ù < <>= w Ö ~ 2 Ù îÔ@ w Ä Ä Ä O O Ö Ö Ö <NM<>OÁl@ G <N <>= w ~ 2 Ù @P<N <>= w ~ 2 Ù N< Q< =8L ~ 2 Ù î @ w g (2.30) Ä Ä Ä In reality, a posteriori, these parameters will turn û out to be smaller than stated here. O N Ek , which can be expanded in We set up the full problem in the form RK I < (2.29) the following wayû ?KRI O <>J«ÁK< O < =ML Ö ~ 2 Ù û T N E O ùù O G KRIcN û V E KRI6N E O O K O O @³­ YKXK û O O V K û û = û O Ú The new error of approximationû is S < K û SUTWV û û YPI O O S S SUTWV N RK I N =DI = YHI O Elk O V N @¾~UK,¤Þ+N ¯ O SUTWV SUTWV û N V N = N$û K,(Þ+N K,(Þ+ Ú S O S O SUTWV ^ V N = f V N @(Þ+N =ÏK Y K @¤Þ 12 O where K M. DEL PINO, M. KOWLCZYK, J.-C. WEI û V N is given at (2.25). We decompose S where S (2.31) S V E T VXV Ál@¾~8K,(Þ+N E and S (2.32) V S E G S O VXV V G O G @ Ú Þ S V = ^ a6~$@'Þ VXV û û û In summary the problem takes the form near the curve, (2.33) where S V ¾ V K O N K Ú û was just described, and û ¾ V K (2.34) N E û V K } (2.35) T O KRI N O V û ùù O û G N E S O N V K } N ERk û O YHI SUTWV S S =DI YPI = û = STWV ^ û g We recall that the description here made is only local. We will be able however to reduce the problem to one qualitatively similar to that of the above form in the infinite strip. 3 The gluing procedure Let IKM6N denote the first approximation constructed near the curve in the coorªÆµ$~;¹\ kk , be a fixed number. We consider smooth cut-off dinate M in : D G . Let µ D D D D function T1UUK N where n : such that T1U¤K N E \ if ªµ and EÏk if Z µ . Denote as Q Q well T U Kö¤N E T1UUK@ ö N , where ö is the normal coordinate to . We define our first global approximation to be simply ^ w KM6N E T U KöUNVI Ú û extended globally as k beyond the ûWµ;¤@ -neighborhood of . Denote KdÕN û E h H d{= O O+ OX S iPK@M6Ned d for d E w , now globally defined in : G . Then K w N Ek if and only if û û where S û û }K ¾ K N E K S O N E û ^ ?K w N E and We further separate ¾ K (3.1) w û N E û H S O N = ^ }RK O N û STWV Y w S û =Fi û SUTWV w = Y w ^ g in the following form: û û where, in coordinates K that we want ^÷ N E T U Ú OjC û , we assume that ¾ KT U û N O C ¾{K N E S O is defined in the whole strip }KT U û Ng 9 so CONCENTRATION ON CURVES 13 û C{^ N satisfies the following nonlinear coupled system: We achieve this û if the pairû K OFC O SUTWV C{^ S O ¾ÒK N E T U }lK N T U T UY w (3.2) C CRO SUTWV C S O û û =YT U N Y w E K¼û \=ZT U N =i K¼\ @[\T U [ Ú O OC (3.3) @ G KH]T U N K¼\=ZT U N }RKT U N³g Ú Ú 9 Notice that the operator ¾ in the strip may be taken as any compatible extension outside the Wµ ;¤@ -neighborhood of the curve. û What we want to do next is to reduce the problem to one in the strip. To do this, C H O we solve, given a small , Problem (3.3) for . This can be done in elementary way: Let us observe first that since i is uniformly positive and w is exponentially Q Q TWV small for ö Zµ@ , where ö is the normal coordinate to , then the problem C H (3.4) has a unique bounded solution C < û Assume now that STWV C N =ÏKiR=ÏK¼\=ZT U N Y C whenever <Ç&< < îlÛD<Ç&< ¿ E w O ª ¿ ^ Ç . Moreover, g ¿ satisfies the following decay condition û û Q (3.5) [ KM6N QeOlQ Q KM`N T_^ B îY Q Q for ö Z µ ^ @ for certain constant ±mZ k . Since } has a power-like behavior with power greater û than one, a direct application of contraction mapping principle yields that Problem C C û (small) solution û E K N with û (3.3) has a unique C < N< K ¿ O < =ML Ö Ø ú Ø ` U Ù îlÛ@³­N< ¯ ^ < =ML Ö Ø ú Ø ` U Ù Q Q <N[ where with some abuse of notation by Ð ö Zlµ;¤@C Ó we denote the complement of µ;¤@ -neighborhood of . The nonlinear operator satisfies a Lipschitz condition û form û û û û û of the < C K V N`=a< C K G N< ¿ îRÛ@¹­N< V = G O <>=ML Ö Ø ú Ø ` U Ù <N[jK V = G N<>=ML Ö Ø ú Ø ` U ¯ Ù g The full problem has been reduced to solving the (nonlocal) problem in the infinite 9 û û û û strip (3.6) û 9 ¾ G K T U R } K N E OFC K NXN O T U S O SUTWV C T UP Y I K N for a no G K N satisfying condition (3.5). Here ¾ denotes a linear operator that Q Q G U coincides with ¾ on the region ö ªÆ\ k . Q Q U û ¾ for ö ªÔ\ k is given in We shall define this operator next. The operator ^÷ coordinates9 K N by formula (2.34). We extend it for functions defined in the ^÷ û û entire strip , in terms of K û N , as follows: (3.7) b ¾ G K N E ¾ V K N Ob Q Q K@ N Ú K N where K N is a smooth cut-off function which equals 1 for ªÑ\ k µ an vanishes identically for for Z k µ and ¾ V is the operator defined in (2.34). 14 M. DEL PINO, M. KOWLCZYK, J.-C. WEI Rather than solving problem (3.1) directly, we shall do it in steps. We consider 9 û o G K N : given and satisfying bounds (2.29)the following projected problem in ^ ^ 9 û (2.30), find functions no G K û N , n¾ G K k °N such that ¾û (3.8) (3.9) (3.10) K G ^ K ¿  û T N E kû N E K ¿ b S û V K ^÷ O ^ } Kû °;¤@¤N ^G O ù N ^ K KÕN¤0 E N û OFC û  O K@ ÷ k N Eû ¿ T ¿ b Nû ù O 'K@ ÷ N b Þ ^÷ K ^ K û SUTWV C °;¤@¤N ^ KÕN¤0 NÞ in 9 ^ O = ͪ:Pª ^ Elk ÷ k ª ^ ° ª g @ K NXN T U YHI K N . Here } K N E T U }lK G We will prove that this problem has a unique solution whose norm is controlled S S by the ¾ G norm, not of whole V but rather that of V . G After this has been done, our task is to adjust the parameters and in such a way that and are identically zero. As we will see, this turns out to be equivalent to solving a nonlocal, nonlinear coupled second order system of differential ^ equations for the pair K, ¹N under periodic boundary conditions. As we will see this system is solvable in a region where the bounds (2.29) and (2.30) hold. We will carry out this program in the next sections. To solve (3.8)-(3.10) we need to investigate invertibility of ¾ in ¾ G -o G setting under periodic boundary G conditions and orthogonality conditions. 4 Invertibility of ced Let ¾ be the operator defined in o G K G linear problem û (4.1) (4.2) (4.3) ¾û K N E Çû O K@ ÷ N b G ^ ^ ^ K k û N E K °f;¤@UN ^ ÷ ¿ KWN¤1  K N T ¿ b Q 9 ÇnP¾ G K Ng Here K@ û ù E Q 9 N by (3.7). In this section we study the O 4K@ ÷ N û b Þ ^ K  T k N Eû ¿ ù K K ^÷ ^ in 9 ^ ^ °f;¤@UN = ·ªgª NÞ KWN¤1 ^ ERk O ÷ k ª ^ ° ^ ª @ ¿ N is the cut-off introduced in the definition of ¾ for given G in (3.7). Our main result in this section is the following. Proposition 4.1. If µ in the definition of ¾ is chosen sufficiently small then there G û exists a constant ÛõZ k , independent of @ such that for all small @ Problem (4.1)û h KÇ'N , which satisfies the estimate (4.3) has a unique solution EY < @ @ <Ni w Ö Ù R î ÛZ<Ç&<>= w Ö Ù g For the proof of this result we need the validity of the corresponding assertion û for a simplerû operatorû which does notû depend of µ . Let us consider the problem (4.4) (4.5) (4.6) jû O SUTWV 9 ^ E ûÇ N E Ò = Hû = Y û in ^ ^ ^ ù ^ ^ ^ ù K k û N E K ° ;¤@¤N K k N E û K °f;¤@UN ^ ÷ ^÷ ¿ K ÕN¤1 E  ¿  K N K N Þ KÕN¤1 T T ¿ ¿ K Ek O ª:ª =+ ^ k ª ÷ ª ^ ° @ g CONCENTRATION ON CURVES 15 Lemma 4.2. There exists a constant ÛÌZ k , independent of @ such that solutions of (4.4)-(4.6) satisfy the a priori û estimate @ @ < i w Ö Ù îRÛZ<Ç&< = w Ö Ù g < û Proof. Let us consider Fourier series decompositions for Ç and û K ÇK ^÷ k¿ N E ^÷ û k¿ N E þ$ÿ V © KWN ­ ©l ~ ó ê ÷ G ÷ @ ° (4.7) j © O G @ G ì m © N E ~UK Ç with orthogonality conditions  (4.8) ûFm ¿ T © 1 E ¿ û.m We have denoted here j ûFm © N E ~K = V K CÒ^C N E  ¿ T ­ © ^ Û­R< ° ¯ ÷ ì g Elk g SUTWV û.m © g j associated to the operator ~ , namely SUTWV K¼\= Y © N < G w Ö v6Ù = ^A¢ <(K ê @ ì n: © = Y Since (4.8) holds we conclude that û.m ûFm (4.9) ä ¢ ¯ ^ ÷ @ ° ¿ ¿ © < G w Ö v6Ù O = G © ÞP1 K:N QC Q O G ê © KÕN û.m © 2 O Let us consider the bilinear form in o T © KWN û.m ¿ Â Ç of the form ä ¢ ó3prq O Then we have the validity of û.the m equations û.m ¢ ó3prq ì ä ¢ ê û O @ ° ó þ$ÿ ­Ç V © KÕN ©l ~ ä ¢ ¯ î Q C+Q ¯ G 0jg N ûFm K © ^ û.m © N for a constant Û Z k independent of ½ . Using this fact and equation (4.7) we conclude the estimate ûFm û m . m K¼\ O[¢ @ © < G w Ö v6Ù O < = ûFm N< © 2 < G w Ö v6Ù î = ÛD<Ç © < G w Ö v6Ù g = © In particular, we see from (4.7) an equation of the form û.m that û.m satisfies m © 2 O © E ^ Ç © nL:g © v Ù îÛD<Ç © > v Ù . Hence it follows that additionally we have the where < Ç >< = w Ö ` < = w Ö6 estimate û.m m (4.10) <(K © 2 < =G w Ö v`Ù î Z Û <Ç © < G= w Ö v`Ù g m= m Adding up estimates û (4.9), (4.10)û in <>n G @ < G= w Ö Ù which ends the proof. O <>n ¢ and ½ we û conclude that @ O < G= w Ö Ù < @ < G= w Ö Ù î @ ÛZ<Ç&< G= w Ö Ù ^ o 16 o M. DEL PINO, M. KOWLCZYK, J.-C. WEI We consider now^ the following problem: given ÇPnP¾ G K ^ 9 û ¾ G K k °N such that N , n G K û 9 N , find functions n ÷ O 4K@ ÷ XN Þ û in 9 ^ K@ N û ^ ^ ^ ^ ^ O ^ ù Ö ^ k N E K °;¤@¤N K k N E û ù K °f;¤@UN = ª:ª û ^÷ ^÷ ^ ÷ ° ¿ ¿ K WN¤1  K N E  K NÞ KWN¤1 ERk k ª ª g (4.13) T T @ ¿ ¿ û Lemma 4.3. Problem (4.11)-(4.13) possesses a unique solution. Moreover, jû (4.11) (4.12) K Ç û E N O @ @ <Ni w Ö Ù îRÛZ<Ç&<>= w Ö Ù g < Proof. To establish existence, we assume that ÇK ^÷ k¿ N E ó þ$ÿ ­Ç V © KWN ©l ~ ä ¢ © O G @ G and © T 1 E ¿ Ç ì G ê @ ° ¯ ^ ÷ m ì m û.m ¿ Â ä ¢ © KÕN V m K:N , andm constants © , © such that O © Þ © Pnh:g © m nLo © O Ç ó3p q O ûFm ¿  © N E ¾0~K @ ê û.m ° and consider the problem of finding û.m û.m ¢ ÷ © ÞP1 T Elk g ¿ m this problem is solvablem with the choices Fredholm’s alternative yields that m © E - T¿ = ¿ - T¿ k¿ (4.14) Q ©l ~ Finally define û K ^÷ û k¿ N E ­ ©l ~ ¿ Observe in particular that m ÷ 'K ÷ N E N E k¿ ©l ~ k¿ ©l ~ © E - T¿ = - T¿ © ÞÒ0 Ç ¿ g Þ G 0 ¿ m Q l O Q © Q @ © G G îRÛ@P<Ç&< G= w Ö Ù ó þ$ÿ V © KÕN and correspondingly K m 0 ^ G 1 Ç © þ$ÿ ­ V © þ$ÿ ­ V © ÷ ä ¢ ê ó ê ° ä ¢ ê @ ° ä ¢ ó ÷ ° û O ó3prq G ì ÷ ó3p q O ì ÷ O ì ÷ © G ê G © ê ä ¢ ° ä ¢ óprq ä ¢ © KÕN ê ° @ ° ¯ ^ ÷ ì ¯ ^ ÷ ì ¯ ÷ ì 9 g Estimate (4.14) gives that K@ N and 'K@ NXÞ have their ¾ G K N norms controlled û by that of Ç . The a priori estimates of the previous lemma tell us that the series for 9 is convergent in o G K N and defines a unique solution for the problem with the desired bounds. o CONCENTRATION ON CURVES 17 Proof of Proposition 4.1 . We will reduce Problem (4.1)-(4.3) to a small perturbation of a problem of the form (4.4)-(4.6) in which Lemma 4.2 is applicable. We will achieve thisû by introducing a change of variables that eliminates the weight T ù3ù G in front of û . We let K ^÷ ^ The map ürq­ k Nz Ä û then ^ E m p K N ^1s N ­k ù E üÕK ÷ NXN ^ ÷ üÕK N E @ ù TWV  K Nb g ~ is a diffeomorphism, where t½ E - ~ Ä û ^ p ù ùù E Ý G K@ ÷ ù ù O N p Ý Ý @ K@ ÷ ù N p K Nb . We denote Ý while differentiation in does not change. The equation in terms of p now reads O S TWV U SUTWV O ù E K p N YfKRI = N p @ p Çt Ú Ý O ÷ O ÷ ^ 9 in t 4 t K@ NXÞ t K@ N ^ ^ ^ ^ ^ ^ O ^ ù ù p K k N EXp K t½V;¤@N p K k N X E p K t½V;¤@¤N = ·ª:ª Ý Ý ^÷ ^÷ ^ ÷ t½ ¿ ¿ N NÞ KÕN¤0 Elk k ª ª p K p K  KÕN¤0 E  g T T @ ¿ ¿ ^÷ ^ TWV ÷ rN E ÇK ü K NXN and the operator t Here Ç is defined by using the above t K ÷ ÷ ÷ Ú TWV ÷ K N formulas to replace the derivatives by derivatives and the variable by ü H STWV p = Y in the operator Ú O p Ob p t . The key point is the following: the operator b ºK p N E is small in the sense that < t Ú O K p N @ O p ù Ý SUTWV YfKRI = SUTWV N p ºK p N < = w Ö @ s Ù îlÛµP< p < i w Ö @ s Ù g Q Q This last estimate is a rather straightforward consequence of the fact that @(ö ª TWV k µ(@ wherever the operator t is supported, and the other terms are even smaller Ú when @ is small. Thus by reducing µ if necessary, we apply the invertibility result û of Lemma 4.2. The result thus follows by transforming the estimate for p into similar one for via change of variables. This concludes the proof. o 5 Solving the nonlinear intermediate problem In this section û we willû solve problem (3.8)-(3.10) û ¾ û û G K N O Ú K N E b S V O } G K O N K@ ÷ N b O 'K@ ÷ N b Þ ^ 9 û û under periodic boundary conditions, and orthogonality conditions in . Here } K N E b OáC G } V K K NXN whenever this operator is well defined, namely for satisfying (3.5). A first elementary, but crucial observation is that the term S VXV E ­@ Ú ¡T G O ¯ @Ua6~ Þ 18 M. DEL PINO, M. KOWLCZYK, J.-C. WEI S ÷ in the decomposition of V , (2.31)-(2.32), has precisely the form 4K@ NXÞ and can therefore be absorbed for now in that term. Thus, the equivalent problem we will look at is û û û ¾ K G O N Ú b S N E K O V } G The big difference between the terms S < V S G O N VXV and ÷ S UK@ V G ^ @ Ö w <>= w Ù R î Û@ G G S K N b O 'K@ ÷ N b Þõg is their sizes. Notice that b S S V . while VXV is a priori only of size 4LK@ w N . We call Á G G For furtherS reference, it is useful to point out the Lipschitz dependence of the on the parameters and for the norms defined in (2.29)-(2.30). term of error G We have the validity of the estimate S < (5.1) V K G ^ V V N6= S V K G @ N<>= w Ö Ù R î Û@ Gw R­ < V = ^ G G O < J G <N V =u G ¯ <>O Let h be the operator defined by Proposition 4.1. Then the equation is equivalent to the fixed point problem û û û EYh (5.2) K S O } G û NXNÁavxK K G Ng Q The operator h has a useful property: Assume Ç has support contained in ~ U G . Then EYh KÇ'N satisfies the estimate Q (5.3) û ^÷ K N QbORQ û [ ^÷ K Q N Q û îw< Q < ~ U T w x B ¿ Q for Q ß k µ Z @ Q î g û since In fact, is supported on ª G and so do the terms involving and , Ú Q Q¤£ ~ U then satisfies for of the formû G û an equation û ¡T with í form p =ÏK¼\ OFí K¼\NXN Elk Q QU£ ~ U û k uniformly as @?z k . For K¼\Nfz G we can then use a barrier of the Ö ( w ( y x ^÷ Q Q T T B Ù 3 ~ U K N E < < w to conclude that for Z we have û û ¿ ^÷ T y(x Kû Nfîw< < B g û G ù3ù O ¿ The remaining inequalities for are found in the same way. The bound forC [ û follows simply by local elliptic estimates. Now we recall that the operator K N satisfies, as seen directly from its definition, û û û < (5.4) C N< = L K î Q Û­N< [ and also the Lipschitz û û condition (5.5) < C K V N6= C K G N< =8L QeOlQ Q ÛL­N< [jK î Q û < =8L Ö Ø Ø ` (w y(B x Ù û V = G N Q¤O]Q û O û V = T xB ¯ ^ Q < =ML Ö Ø Ø ` (w B y Ù G ¯ g These facts will allow us to construct a region where contraction mapping principle applies. As we have said, < S G @ <>= w Ö Ù îRÛ_@FGw g CONCENTRATION ON CURVES 19 for certain constant ÛZ k . We consider the following closed, bounded subset of 9 o G K N : z û ^ @ w Ö û Ù îgnm@ Gw û % < û <Ni no G K N % Q QeOlQ Q @ T x % [ < =ML Ö Ø Ø ` (yB x Ùîw< < i w Ö Ù B % % % we claim that if the constant n is fixed sufficiently large then the map v z }~ E|{ û 9 defined in (5) is a contraction from into itself. Let us analyze the Lipschitz character of the nonlinear operator involved in v û û û for functions in : } û where D V K } ^ b N E K G 9 9 so thatû denoting U E <} G @ K N<>= w Ö Ù î û Q Ð N î û Q ÛÏ­ û Using Sobolev’sû embedding we get S < < = w Ö@ Ù O @ < G= Ö Ù î < Û û ­N< < S O @ K N < = w Ö x Ù < C @ K N <=G Ö x Ù î û It thus follows that <} (5.6) G K NXN û Q û STWV ¯ ^ G z Q ¯ ^ G û û for have that ­N< û C n S @ N< = w Ö x Ù K S û O < i w Ö@ Ù < , O û C < K @ ¯ ^ < Gi w Ö Ù @ N< G= Ö x Ù ¯ g 9 , (5.3), that the area of U is of order while using estimate (5.4), the facts that n 4Kµ;¤@UN and Sobolev’s embedding, we get û û C z K =DI QS O ªÆ\ k µ û ;¤@Ó , we O @ O < = w Ö@ Ù < < G= Ö Ù S ÛÏ­N< } Q V K Q SUTWV N û for n[K k \N . From here it follows that Q OjC V K û OFD Y ­XKRI N E } T Ûu @ N< = w Ö Ù î xB ­¼\ ÛÏK$@ G w û O O < i w Ö@ Ù < O û S @ ¯ <Gi w Ö Ù g < @ Ú Ng Besides, asû for Lipschitz condition, we estimate, û û find after a direct û <} V K V N6=x} V K G @ N<>= w Ö Ù îÍ­N< O SUTWV @ O ûV < = w Ö Ù @ < û V <>= Ö Ù SUTWV ¯ @ O < >< û = Ö Ù û < < = w Ö@ Ùû G G @ O V <>= w Ö Ù ­N< = < G G û = with constants Û independent of the bound a priori assumed on < û û û û û then conclude for } û , <} G K V N6=} G G K G @ N<>= w Ö Ù î<} O Vû OFC V K <} û V K G û K OFC V û NXN`=x} V NXN`=x} K O @ V = û >< = Ö x û Ù G C C îÊ­N< O ­N< á O C < K Kû V N6= V N`= C K Kû G G < N<>= N<>= <Ni w Ö ~ 2 Ù . We Ä @ Ö K û K V û NXN<>= w x Ù G OFC @ ûV K û K NXN< = w Ö x Ù G G @ ¯ V = >< = w Ö x Ù @Ö Ù G x @ ¯ w Ö Ù OFC V @ ¯ ^ V <>= Ö Ù x 20 M. DEL PINO, M. KOWLCZYK, J.-C. WEI where withm û.m SUTWV O < = w Ö@ x Ù < E C û.m < ûFm SUTWV O N < = w Ö @ x Ù K O V E G < SUTWV O < = w Ö@ x Ù C < Arguing as before, the conclusion of these estimates is û û û <} (5.7) Now, let n G V N`=x} K then pjE G K G v K x @ Nû < = w Ö Ù î N Ö SUTWV Ù O ÛLK@HG w ûFm K û @HGw N< @ ^ N<>= Ö x Ù ½ E û V = G ^ \ g @ < i w Ö Ùg satisfies thanks to (5.6), O S S @ < p <Ni w Ö Ù î< h < Ð ÛI@ Gw Ûn @ Gw 8 Ó g ZRÛ < h Choosing any number n On the other hand we have < we get that for small @ @ < p <Ni w Ö Ù îgP n @ Gw g @ @ < p < = L Ö Ù îRÛZ< p <Ni w Ö Ù g z z But p satisfies an equation of the form ¾ K p N E Ç with Ç compactly supported. G Hence p belongs to thanks to the discussion above. v is clearly a contraction mapping thanks to (5.7). We conclude that (5) has a unique fixed point in . S and the operator h itself carry the functions and We recall that the error G as parameters. A tedious but straightforward analysis of all terms involved in the differential operator and in the error yield that this dependence is indeed Lipschitz with respect to the o G -norm (for each fixed @ ). In the operator, consider for : û instance the following only term involving û N E K @(G ÷ K@ Then we have û < Let p K N E - T ¿ ¿ û Q K ó> ^÷ N G 1 ù ñ p K ÷ ù ñ  T . Then û Q ¿ K ¿ û û û Q Q O Q Q Q ù Q N+îÔ@Õ  G û >ó ÷ O ß TWV Q \ ù QG g @  î ù ñ p K N Hence (5.8) ê ~ Q g ó6 Q Q @ K²³N G ³² K N < G= w Ö Ù î @ Ú ÂmÄ ÷ N ÷ p K NfîlÛ@ so that finally < K û TWV < û @ ^ < Gi w Ö Ù @ N<>= w Ö Ù î@Q< < J g ^÷ Q N G 0 ì CONCENTRATION ON CURVES 21 For the other terms the analysis follows in simpler way. Emphasizing the depen dence on what we find for the Linear operator h is the Lipschitz dependence < h h = w <0îlÛ@Q< V = < J g G z û We recall that we have the Lipschitz dependence (5.1). Moreover, the operator } ^ also has Lipschitz dependence on K (N . It is easily checked that for n we have, with obviousû notation, û O @ <} Ö 2 Ù K N6=x} Ö w 2 w Ù K N< = w Ö Ù îRÛ@FwE ­R< V = < J G Hence from û the fixed point characterization we then see that û Ö 2 Ù <Ni w Ö @ Ù îRÛ@FGw ­R< V = w w Ö 2 Ù = < (5.9) O < J G ¯ <N V = <N V = G <>O ¯ G <>O g g We summarize the result we have obtained in the following: Proposition 5.1. There is a number n Z k such that for all sufficiently small @ ^ û û and all K ^ (N satisfying (2.29)-(2.30), problem (3.8)-(3.9) has a unique solution E K ¤N which satisfies û Q û < û Besides QbORQ ^ @ < i w Ö Ù î nmû @HGw < û Q [ @ T x <Ni w Ö Ù B î < < =ML Ö Ø Ø ` (yB x Ù and in the sense of estimate (5.9). depends Lipschitz-continuously on Next we carry out the second part of the program which is to set up equations ^ for and which are equivalent to making identically zero. These equations are obtained by simply integrating the equation (only in ) against respectively andS Þ . It is therefore of crucial importance to carry out computations of the terms S V 1 and - v V ÞL0 . We do that in the next section - v 6 Estimates for projections of the error S S In this section we carry out some estimates for the terms - v V 0 and - v V ÞP1 , S where V , we recall, was defined in (2.31)-(2.32), and is an odd function. InteS gration against all even terms in V therefore just vanish. We have  S v V O  v O We recall E   S V v # @3Yf­K v #K G  û O V N O E û V O v STWV ?K O V N SUTWV ¯ = @(Þ+N û S =ÏK K,¤Þ N O O û V N S @ = Ú YfK K,(Þ+N O ' û V N STWV @(Þ ' g 22 M. DEL PINO, M. KOWLCZYK, J.-C. WEI K û O @ G V N E O O Ú O @ G .û O ­K S V N K G = N S = Y û O û K ûV N ¯ O<T S Ú TWV ù3ù V GUK V N where is an odd function, is an even function and G we see that  ?K v û O V N E TWV> ={@ GM O O  ¡T G TWV6¡TWV TWV ¢ O ê O O  (6.1) ý v m  v Here and below we denote by , ½ E functions of the form m m ý ý ý E ÷`^ K K@ \ ÷ V N S O V N K N G OX T Ú i6Éɤ v Ò ì v û û Ú ^ G O Ý Ý Â v G O ­K K v G  is of order @ Ú . Thus N G v  K G S = @ Ú ý V = Y SUTWV V O û @ Ú ý ¯ cg G , generic, uniformly bounded continuous ^ ÷ ^ ÷ ^ ÷ N º K@ N K@ N @ K@ NXN where additionally V is uniformly Lipschitz in its four last arguments. ÷ The coefficient in front of K@ N in (6.1) can be computed as <TWV¢ T ¡ E ¡TWV¢ 6 G O+¡T Ú i ÉÉ Â G x G Ú i ÉÉ Â G  = v v v ì { ì ê E = V - vû G . û - v where we have used the fact that = - v G E { G` ê v O { S S ¯ SUG TWV û û û V N = V is to main = Y that the term ­K order ofû the form O û VXV V where G G V and it is therefore of the order 4K@ G N . We have V E G G û we recall VXV is odd, V is even, and with their sizes respectively proportional to @ Ö Ù S U S W T V G SUT û and to @ . Thus in the expansionû of the square term G G V asymptotically G X V V V only the mixed product between and gives rise to a nonzero term after the G integration against û . We have û We see S Ö SUTWV Ù U S Tû  v ­K O V N E YKàY+=\N  E @(GIü VXV K@ (6.2) ÷ S S = v Nbü V = G K@ SUT ÷ N Y G V 1 G K@ ¯ V 0 SUTWV û ÷ O @ Ú ý N YfKàY=\N G v SUT G V G 1 O @ Ú ý G g CONCENTRATION ON CURVES Now, let us consider ÷ p K@ N E û  23 v V N K Ú g ÷ All terms in this expression carry in ¾ G -norm as functions of ² E @ powers three or higher with the exception of the terms of size @ in . Thus we find ÷ p K@ N E @ ¢ û Since V E with Ú û ^ ÷ V 5 K V N =  Gi6ÉK k @ N* v û T ¡ ^ ÷ O V ={@ Gi É K k @ N  4LK@ Ú Ng v V KWN p K@ ÷ odd and N E @ G ÷ K@ KÕN Nbü V =Ò@G (6.3) ÷ K@ G ÷ K@ N ¡T O V KWN @ ÷ K@ Nbü V ÷ K@ G N KWN G even we obtain TWV N <T N ÷ @ G ü VXV K@ E ÷ @(ü VXV K@ G û <TWV ÷ K@ G K v ÷ GIü VXV K@ Nbü V ¢  ^ ÷ @ N NXi`É_K k N ÷ K@ = T G i6ÉXK k ^ @ ÷ N* N G N # v V v G 2 O 4K@ Ú N O 5 G \ O G V ' g Note that by differentiating the equation (2.23) and using equation (2.22), we obtain  (6.4) YfKàY=\N v SUT G V E G =m v O V  K v O  v ÷ Nbü V O G E E (6.5) ¡TWV @(G =Ò@ G TWV 2 TWV¢ S V N @ G ü VXV K@ E û O G = ÷ K@ S N O \ TWV ¢  G ­ v  ÷ K@ G = 2 G v Y N # YfKàY=á\N v O ¯ O V G G O û SUTWV V  SUT @ Ú ­ V V  v G E = K Y=á\ O \ N v 2 g G G G OÊý ¯ G N ¯ O @ Ú ­ ý V OÊý ¯ G where we have used (1.9) and the following integral identities 2 G V N K Ú N ' O @ Ú ­ ý V OÊý ¯ O TWV fK ý G û v Adding (6.2) and (6.3) and using (6.4), we have ­K \ G ^ TWV  v G E \ K = Y=\ N v G g 24 M. DEL PINO, M. KOWLCZYK, J.-C. WEI In summary, we have established that û O (6.6)  v K V N 0 E @ G ­ ÷ K@ O N ÷ ± V K@ N O ± ÷ K@ G ¯ O N ý OÊý ¯ @ Ú ­ V G where ± V is given by and ± ± V K²³N E (6.7) G T K G O Ý TWV TWV ÝN E TWV i icÎ ^ is given by ± (6.8) K²³N E G TWV ={i O i`ÉÉ TWV`O K ¢ \N GUg Let us estimate now the term  v  #K v (6.9)  v O E  v û O û V N Ð K @XYKàY+=\N O V O SUTWV V N We find now that Ð @XY­K û O # @3Yf­K O SUT @(Þ+N û S V N Ú ' û O = YfK û O =ÏK S K,(Þ+N V N û O = YfK SUTWV STWV V N @(Þ @(Þ ' g ¯ Ó ST O @ G YfKàY=á\N O @ Ú ý g G G Þ G v G V (Þ G v @ V N Ó³K,(Þ+N S O û O =ÏK O V S SUTWV ¯ = K,(Þ+N @¤Þ+N SUTWV û SUTWV ¯ = Theû second integral vanishes, while in the first only the term carrying the odd part of V is non-zero. Thus we find that this terms equals ÷ @ G ü VXV K@ N4Â Þ v V 0 O @ Ú ý g G Let us compute now @ - v K,¤Þ N¤0jg In this term, we have to consider compo Ú nents of order 4LK@UN in the coefficients of which are odd functions. We obtain @4 v Ú Ú K,¤Þ N¤0 ¢ @ G E ÷ K@ O N @ ­ Þ O V ÕÞ ¯ v ¢ ÷ K@ N v Þ ý V O[ý OÊý ¯ @ Ú ­ V g GV G O Summarizing, we have proven that  (6.10) v S V 0 E @(G­ O O ± V ýV @ Ú ­ V O ± OÊý G GV K@ ÷ N ¯ O @(GU­ ± ¯ OÊý G g Ú K@ ÷ N O @(G$±f ¯ CONCENTRATION ON CURVES 25 The next computations, rather analogous, correspond to the projection onto Þ of S the error. We compute now - v V Þ . The main component in this expression is given by T @¹­ @ G We also have a term like û p K@ ÷ N E ¡TWV @ E ¢8¡TWV @(G O E K v @G ¢8 TWV @ G ÷ ü VXV K@ N O V N ­K v ÷ ü VXV K@ N v ­K K@ G O V N ÷ N V ¯ Â Þ G v û O V>5 Þ <T @ ^ ÷ Gi`É3K k @ N  û V Þ v ¯ V Þ V ^ ÷ Gi É K k @ N G ü V ¢8¡T a6~ <T ^ ÷ V N = GIi6ÉK k @ N* ¢  O G v G ¯ O V Þ Þ ý @ Ú G because of the assumption on . There are also terms of order @ G coming from the second order expansion of ?K N . We recall, from the decomposition (2.31)-(2.32) that the error carries either terms accompanied by @ G as a factor or by @ Ú . The terms with @ Ú produce functions ÷ ^ E k with ¾ G K °N -norms of order @ Ú . The terms of order @ G in the decompoof ² @ S sition of V are either even or odd in the variable . Those which are odd do not contribute to the integral since the function Þ is even. Taking also into account that and are uniformly controlled by @ w we just need to consider û ÷ K@ ÷ C K@ ÷ 78K@ <T N E @ Gci É K k N E @ G ü G VXV K@ N E @ G  ÷ N YKàY+=\N T = v OD TWV = ^ ÷ ÷ @ N K@ N ¢ G \ T G O SUT ­ v V Þ1 v G GV ¯ ÞP1 O T Q Q G G TWV T i`ÉÉ, G G O O TWV T Ú ÞP1xg It is easy to see that also 0 E 4LK@ÚN , The common pattern of and 07 is that ÷ C even though they have size @ G in ¾ G norm, they define smooth functions of ² E @ , which is a very important fact to obtain the desired result. For the term parallel to (6.9) we get û SUT @3YfKàY=\N v G V (ÞÞP1 O @ G YfKàY=\N SUT v G G Þ Ú 1 O @ Ú ý V E We consider now another component: @4 v Ú K,(Þ+NXÞ E ={@GP³K@ ÷ N K@ ÷ N ¡T GUi`É v ÞÒG O @ Ú E 4K@ Ú N<g @ Ú ý V g 26 M. DEL PINO, M. KOWLCZYK, J.-C. WEI Additionally, we also need to consider some higher order terms in . The ones involving first derivative are @ Ú Â G O ÞÒG v Ý @ Ú Ý Â Ú ÕÞ Þg v Only one term (in K,(Þ+N ) involving carries also @ Ú . But this term is also G N G '0 Elk . There also is a term of the form accompanied by - v Þ KÕ T @ Ú ­@ G ± ÷ K@ C ¯ O N ÷ 4K@ G N K@ N with 4K@ G N uniform in @ . ÷ In summary, we have established that, as a function of ² E @ , S  (6.11) ^ ^ v V ÞP1 @ Ú ­\ E O O @ @ Ú ±M7 ^* O ± O ¯ T 4LK@ G N C G ÷ @Ua6~ =x@ G ±MK@ O N 4K@ Ú N³g where ± > E+ gXgXg are smooth functions of their argument. Explicit expression for the coefficient ±87 , which we will need later, is ±M7 E (6.12) G = Ý g Ú 7 Projections of terms û involving û in (3.8)-(3.10) integrated against We will estimate next the terms that involve and Þ . Concerning , we call the sum of them p K N , which can be decom û posed as pjEX¡ Ú l V p below. ÷ û . We make the following observation: all terms ûin p V K@ N E - v K N Let ^ Ú K N carry @ and involve powers of times derivatives of k \ or two orders of . The conclusion is that since has exponential decay then Q û Q < Gi w Ö Ù g ÂÄ p K²³N G ³²îRÛ@ Ú < ~ Hence < p V <>= w Ö ~ 2 Ù îRÛ@ Ú g û Ä In K N we single out two less regular terms. The one whose coefficient depends Ú on explicitly has the form û p V E E @ G û ={@G Â Þ K¼\ OÊ¢ @ K= NXN T G v  v û Ð Þ K¼\ O[¢ @ K= NXN T G¤Ó g ^ Since has Lipschitz dependence on K ¤N in the form (5.9), we see that this is transmitted from Sobolev’s embedding into û û < Ö 2 Ù = Ö 2 Ù < =ML Ö @ Ù îÛ@HGw ­R< V = w w G < J O <N V = G <>O ¯ ^ CONCENTRATION ON CURVES 27 from where it follows ^ < p V 8K V V N6= p V K ^ G × N<>= w Ö ~ 2 Ù îRÛ@ Ú w ­R< V = û G Ä ÷ The one arising from second derivative in for û p V X E  ù3ù v Þ O[¢ ­ \=ÏK¼\ @ G O < J ¯ <N V =u G <>O g is K= ¯ g G 0 T NXN We readily see that ^ < p V X8K V V N6= p V XK N<>= w Ö ~ 2 Ù îÛ@ Ú ­R< V = ^ G G G Ä < J O <N V = G <>O ¯ g The remainder p V = p V = p V X actually defines for fixed @ a compact operator of ^ ^ the pair K (N into ¾ G K k °N . This is a consequence of the fact that weak convergence 9 V 9 o K N , and the same is the case for in o G ^ K N implies local strong convergence in ¯ ^ ^ û V o G K k °N and Û ­ k ° . If and > are weakly convergent sequences in o G K k °N Ö ¢ ¢ V 9 2 Ù constitute a bounded sequence in o K N . In then clearly the functions the above remainder one can integrate by parts if necessary once in . Averaging against which decays exponentially localizes the situation and the desired fact û follows. ÷ We observe alsoû that p K@ N E - v }RK N can be estimated similarly. Using G the definition of }lK N and the exponential decay of we obtain û < p @ < Gi w Ö Ù î <>= w Ö ~ 2 Ù îRÛ@ w < G Û@ Ú g Ä Let us consider now Since I E O p û V O @¤Þ Q @ (Þ Ú ÷ K@ N E û  v SUTWV Yf­ I = û STWV ¯ and V can be estimated as KÕN QbOlQ û V K ^÷ Q N Q Q O îlÛ@³K G \N TØ Ø we easily see that for some xZ k we have the uniform bound Q SUTWV I = SUTWV(QXQ From here we readily find that < p < = w Ö ~ 2 Ù îRÛ@FGw < Ú Ä Q î û Û@ T @ < i w Ö Ù î Ø Ø g Û@ Ú g This terms define compact operators similarly as before. We observe that exactly the same estimates can be carried out in the terms obtained from integration against Þ . 28 M. DEL PINO, M. KOWLCZYK, J.-C. WEI 8 The system for £¥¤§¦¥¨1© : proof of the theorem û In this section we set up equations relating end such that for the solution ÷ of (3.8)-(3.9) predicted by Proposition 5.1 one has that the coefficient K@ N is identically zero. To achieve this we multiply first the equation against and Elk is then equivalent to the relation integrate only in , The equation û û û S  V v 1 O  K} v O K G N Similarly Elk if and only if û  S V ÞÒ1 O v  K} v K G K Ú N O û O Ú K O N STWV ¯ SUTWV Yf­ I N = Yf­ I SUTWV N = STWV ¯ 0 Elk g û NXÞÒ1 Elk g Using the estimates in the previous sections we then find that these relations are equivalent ^ to the following nonlinear, nonlocal system of differential equations for the pair K (N«ª ª V K (8.1) (8.2) G NWÁ O ± V E m O ± E G a ~¬ c ± Ú O O ^ @(G± @ V O O ± @ ±Q I @ G g C G T O O E @ Ú ± 7¬ N m G M ^ K m ¤N can be decomposed m m in the following form: ^ ^ O ­ ^ K N E K (N ( (N K ^ m ^ k K,(NÁl m @ G K The operators ­ where is uniformly bounded in ¾ G K °N for K @UN satisfying constraints (2.29)(2.30) and is m also compact. m The operator is Lipschitz in this region, ^ ^ V V N6=Z K N< = w Ö ~ 2 Ù îÛ@³­R< V = G G ^ ^ ^* Ä ª The functions ±8 > E \ ggg are smooth. <> K G < J O <N V =u G <>O ¯ g We will solve now system (8.1)-(8.2). The first observation is that the operaª tor V is invertible under ° -periodic boundary» conditions. This follows from the ^ assumed non-degeneracy condition (1.10): if nF¾ G K k °N then there is a unique ^ » solution nPo G K k °N of V K N E which is ° -periodic andª satisfies < O < = w Ö~ 2 Ù Ä < O < =8L Ö ~ 2 Ù Ä < » < =ML Ö ~ 2 Ù îRÛZ< < = w Ö ~ 2 Ù g Ä ª deal with invertibility of We use now assumption (1.14) to Ä G . We have: ^ Lemma 8.1. Assume that^ condition (1.14) holds. If Lnj¾ G K k °N then there is a unique solution no G K k °N of K,¤N E which is ° -periodic and satisfies G O O TWV Ö w Ù 2 = L Ö ~ 2 Ù îRÛ@ @ G <N > < = ~ @P<N <>= w Ö ~ 2 Ù N< M< M Ä Ä ^ Ä Moreover, if is in o G K k ° N , then O O @(G8<N <>= w Ö ~ 2 Ù <N <>= w Ö ~ 2 Ù <NM< = L Ö ~ 2 Ù î ÛLR­ < <>= w Ö ~ O Ä Ä Ä Û <®< = Z <®<>= w Ö ~ 2 Ù g Ä 2 Ù O wÄÖ~ 2 Ù g Ä < <>= w Ö ~ 2 Ù Ä ¯ CONCENTRATION ON CURVES 29 ª Let us accept for the moment the validity of this result and let us conclude the proof of the theorem. O We solve first K,~N E @±MK²³N and replace E ~ . Observe that by Lemma G 8.1 we have O @ G <N ~ <>= w Ö ~ 2 Ù <N ~ < = L Ö ~ 2 Ù îRÛ@8g ^ Ä Ä ª the same form as (8.1)-(8.2) ª The system resultingª on K U N has except that now the term @M ± disappears. Let us observe now that the linear operator ª ^ ^ ^ K (N E K V K N6=P± =R@ G ± ,K ¤NXN Ú G »c^ ^ E is invertible with bounds for K (N K ËN given by < < J ª O <NQ<>OîRÛD< » < O @ G TWV <¯< g G It then follows from contraction mapping principle that the problem O ­ K@ V ^ @ G G ¯ ^ $N K ¤N E ^ K »c^ ËN ×.° » is uniquely solvable for K (N satisfying (2.29)-(2.30) if < < ªõ@ w , <®< ª G G ×.° ^ @ Gw for some ±mZ k . The desired result for the full problem (8.1)-(8.2) then follows directly from Schauder’s fixedO point theorem. In fact, refining the fixed point region, we can actually get <NM<>O < < J E 4K@¤N for the solution. o ª Proof of Lemma 8.1. We consider then the boundary value problem (8.3) K,(N E G ^ ºK k N E ºKÅ°N ^ Kk N E ^ KÅ°N We make the following Liouville transformation c.f. [22]: °I~ E ³  ~ K²³N E D MWK N E Ä K²¹Nbº² ^LD E - ~ Î K²³Nb³² ä ^² ac~ E °I~ Î ¡T w ^ \ exp b K =  G±878K²³Nb³²³N ~ ³ ^ D ³ TWV ° G~ ÝÝ K²³Nº K²¹N ` K N E ä [ ³ g G ³ ä ° G~ G a6~ G Observe ª that is ° -periodic thanks to the explicit formula (6.12) for the coefficient ±M7 . Then (8.3) gets transformed into (8.4) G @ G KM KM`N E O D `K NeM`N O- a6~M E ^ MÕK k N E MÕK ä N ^ M Kk N E M K ä N ^ and it then suffices to establish the estimates in Lemma 8.1 for the solution of this problem in terms of corresponding norms of . It is standard that the eigenvalue problem (8.5) M O D ËK NeM O a`M Elk ^ MWK k N E MWK ä N ^ M Kk N E M K ä N ^ 30 M. DEL PINO, M. KOWLCZYK, J.-C. WEI ¢x£ k , with associated orthonormal has an infinite sequence of eigenvalues a © , ^ ä by eigenfunctions. A result in [22] provides asbasis in ¾ G K k N , Ð M © Ó , constituted ¢ O ymptotic expressions as z for these eigenvalues and eigenfunctions, which turn out to correspond to those for ?Á k . We have: ¢O ´ a © E (8.6) 4µ ¢ \ ÚQ¶ g ¢ Problem (8.4) is then solvable if and only if a © @ G E c a ~ for all . In such a case, the solution to (8.3) then can be written as D MWK N E © k¿ D M © K N ©l ~ a6~=ja © @ G with this series convergent in ¾ G . Hence Q © Q G g ©l ~ K a6~=Fa © @ G N G k¿ <3M·< =G w Ö ~ 2 ¸ Ù'E We then choose @ such that Qß ¢ (8.7) G @ G = QÕ£ ac~ Ë@ ¢ for all , where is small. This corresponds precisely to the condition (1.14) in the Q Q£K¹ statement of the theorem. From (8.6) we then find that a6~=a © @ G @ if @ is also TWV G Ö Ö sufficiently small. It follows that <3M·< = w ~ 2 ¸ Ù îRÛ@ < §< = w ~ 2 ¸ Ù . Observe also that <3M < G= w Ö ~ 2 ¸ Ù îRÛ Hence k¿ ©l ~ Q © Q G OlQ © Q a K ac~=Fa © @ G N G \ îÛ k¿ ©l ~ K¼\ OÊ¢ Q Q N © G g O T V W = L Ö ~ 2¸ Ù R @P<3M < = w Ö ~ 2 ¸ Ù 3< M·< 8 î Û@ < ¯< = w Ö ~ 2 ¸ Ù g ^ ä ä ä o G Kk N with 'K k N E 4K N , K k N E K N , then the sum ¡ Besides, if is in is finite and bounded by the o G -norm of . This automatically implies O @G<3M <>= w Ö ~ 2 ¸ Ù and the proof is complete. O <3M <>= w Ö ~ 2 ¸ Ù © ¢ ©G ^ <3M·< = L Ö ~ 2 ¸ Ù îRÛD< ¯N< i w Ö ~ 2 ¸ Ù o Acknowledgment. The first author has been partly supported by grants Fondecyt 1030840 and FONDAP, Chile. He is grateful to the Department of Mathematics of The Chinese University of Hong Kong for its hospitality. The second author has been supported by grant NSF OISE-0405626, and Fondecyt grants 1050311 and 7040142. The research of the third author is partially supported by an Earmarked Grant (CUHK402503) from RGC of Hong Kong. CONCENTRATION ON CURVES 31 Bibliography [1] Alikakos, N.; Chen, X.; Fusco, G. Motion of a droplet by surface tension along the boundary. Calc. Var. Partial Differential Equations 11 (2000), no. 3, 233–306. [2] Ambrosetti, A.; Badiale, M.; Cingolani, S. Semiclassical states of nonlinear Schrödinger equations. Arch. Rational Mech. Anal. 140 (1997), 285–300. [3] Ambrosetti, A.; Malchiodi, A.; Ni, W.-M. Singularly perturbed elliptic equations with symmetry: existence of solutions concentrating on spheres, Part I. Comm. Math. Phys. 235 (2003), 427-466. [4] Ambrosetti, A.; Malchiodi, A.; Secchi, S. Multiplicity results for some nonlinear singularly perturbed elliptic problems on º&» . Arch. Rat. Mech. Anal. 159 (2001), 253–271. [5] Badiale, M.; D’Aprile, T., Concentration around a sphere for a singularly perturbed Schrödinger equation, Nonlin. Anal. T.M.A. 49 (2002), 947–985. [6] Benci, V.; D’Aprile, T. The semiclassical limit of the nonlinear Schrdinger equation in a radial potential. J. Differential Equations 184 (2002), 109-138. [7] Byeon, J.; Wang, Z.-Q. Standing waves with a critical frequency for nonlinear Schrödinger equations. Arch. Rat. Mech. Anal. 65 (2002), 295–316. [8] Cingolani, S.; Lazzo M. Multiple positive solutions to nonlinear Schrödinger equations with competing potential functions. J. Differential Equations 160 (2000) 118–138. [9] del Pino, M.; Felmer, P. Local mountain passes for semilinear elliptic problems in unbounded domains. Calc. Var. Partial Differential Equations 4 (1996), 121–137. [10] del Pino, M.; Felmer, P. Semi-classcal states for nonlinear Schrödinger equations, J. Funct. Anal. 149 (1997), 245–265. [11] del Pino, M.; Felmer, P. Multi-peak bound states for nonlinear Schrödinger equations. Ann. Inst. H. Poincare Anal. Non Lineaire 15 (1998), no. 2, 127–149. [12] del Pino, M.; Felmer, P. Semi-classical states of nonlinear Schrödinger equations: a variational reduction method. Math. Ann. 324 (2002), no. 1, 1-32. [13] Felmer, P.; Torres J.J. Semi classical limits for the one dimensional nonlinear Schrödinger equation. Comm. Contemp. Math. 4 (2002) 481-512. [14] Floer, A.; Weinstein, A. Nonspreading wave packets for the cubic Schrödinger equation with a bounded potential. J. Funct. Anal. 69 (1986), 397-408. [15] Grossi, M. On the number of single-peak solutions of the nonlinear Schrödinger equation. Ann. I. H. Poincare, Anal. non lineaire 19 (2002) 261–280. [16] Gui, C. Existence of multi-bump solutions for nonlinear Schrödinger equations via variational method. Comm. in P.D.E. 21 (1996), 787-820. [17] Jeanjean, L.; Tanaka K., Singularly perturbed elliptic problems with superlinear or asymptotically linear nonlinearities. Calc. Var. Partial Differential Equations 21 (2004), no. 3, 287-318. [18] Kang, X.; Wei, J. On interacting bumps of semi-classical states of nonlinear Schrödinger equations. Adv. Diff. Eqn. 5(7-9) (2000), 899-928. [19] Kowalczyk, M. On the existence and Morse index of solutions to the Allen-Cahn equation in two dimensions. Annali di Matematica Pura et Aplicata, to appear. [20] Kowalczyk, M. Approximate invariant manifold of the Allen-Cahn flow in two dimensions. Partial differential equations and inverse problems, 233–239, Contemp. Math. 362, Amer. Math. Soc., Providence, RI, 2004. [21] Li, Y. Y. On a singularly perturbed elliptic equation. Adv. Diff. Eqn. 2(6) (1997), 955-980. [22] Levitan, B. M.; Sargsjan, I. S. Sturm-Liouville and Dirac operators. Mathematics and its Applications (Soviet Series), 59. Kluwer Academic Publishers Group, Dordrecht, 1991. [23] Lin, C.-S.; Ni, W.-M.; Takagi, I. Large amplitude stationary solutions to a chemotaxis system. J. Differential Equations 72 (1988), no. 1, 1-27. 32 M. DEL PINO, M. KOWLCZYK, J.-C. WEI [24] Malchiodi, A.; Montenegro, M. Boundary concentration phenomena for a singularly perturbed elliptic problem.Commun. Pure Appl. Math. 55 (2002), 1507-1568. [25] Malchiodi, A.; Montenegro, M. Multidimensional boundary layers for a singularly perturbed Neumann problem. Duke Math. J. 124 (2004), no. 1, 105-143. [26] Malchiodi, A. Solutions concentrating at curves for some singularly perturbed elliptic problems. C. R. Math. Acad. Sci. Paris 338 (2004), no. 10, 775-780. [27] Malchiodi, A. Concentration at curves for a singularly perturbed Neumann problem in threedimensional domains. Preprint. [28] Mazzeo, R.; Pacard, F., Foliations by mean curvature sets. Preprint. [29] Ni, W.-M.; Takagi, I. On the shape of least-energy solution to a semilinear Neumann problem. Commun. Pure Appl. Math. 41 (1991), 819–851. [30] Ni, W.-M.; Takagi, I. Locating the peaks of least-energy solutions to a semilinear Neumann problem. Duke Math. J. 70 (2003), 247–281. [31] Oh, Y.-G. Stability of semiclassical bound states of nonlinear Schroedinger equations with potentials. Commun. Math. Phys. 121 (1989), 11–33. [32] Oh, Y.-G. On positive multibump states of nonlinear Schroedinger equation under multiple well potentials. Commun. Math. Phys. 131 (1990), 223–253. [33] Pacard, F.; Ritore, M., From constant mean curvature hypersurfaces to the gradient theory of phase transitions. J. Differential Geom. 64 (2003), no. 3, 359–423. [34] Rabinowitz, P.H. On a class of nonlinear Schrdinger equations. Z. Angew. Math. Phys. 43 (1992) 270–291. [35] Shatah, J.; Zeng, C. Periodic solutions for Hamiltonian systems under strong constraining forces. J. Differential Equations 186 (2002), no. 2, 572–585. % ¼ . Ann. Mat. [36] Sirakov, B. Standing wave solutions of the nonlinear Schrödinger equation in F Pura Appl. 4(181) (2002), no. 1, 73–83. [37] Wang, X. On concentration of positive bound states of nonlinear Schrdinger equations. Commun. Math. Phys. 153 (1993), 229–243. [38] Wang, Z.-Q. Existence and symmetry of multi-bump solutions for nonlinear Schrödinger equations. J. Differential Equations 159 (1999), no. 1, 102–137. MANUEl DEL PINO Departamento de Ingenierı́a Matemática and CMM, Universidad de Chile, Casilla 170 Correo 3, Santiago, Chile. Email: delpino@dim.uchile.cl MICHAL KOWALCZYK Department of Mathematical Sciences, Kent State University, Kent, OH 44242 and Departamento de Ingenierı́a Matemática, Universidad de Chile, Casilla 170 Correo 3, Santiago, Chile. Email:kowalcyk@math.kent.edu JUN-CHENG WEI Department of Mathematics The Chinese University of Hong Kong, CONCENTRATION ON CURVES Shatin, N.T., Hong Kong. Email: wei@math.cuhk.edu.hk Received Month 200X. 33