Conformal Scalar Curvature Equation on Sn : Functions With Two Close Critical Points ( Twin Pseudo - Peaks) 7 1 0 2 n a J Man Chun LEUNG & Feng ZHOU ∗ ∗∗ 3 2 National University of Singapore ] P A . h t a m Abstract [ 1 v By using the Lyapunov-Schmidt reduction method without perturbation, we consider 7 existence results for the conformal scalar curvature on Sn (n 3) when the prescribed 7 ≥ 2 function (after being projected to IRn) has two close critical points, which have the same 6 value (positive), equal “flatness” (‘twin’; flatness < n 2), and exhibit maximal 0 − . behavior in certain directions (‘pseudo-peaks’). The proof relies on a balance between 1 0 the two main contributions to the reduced functional - one from the critical points and the 7 other from the interaction of the two bubbles. 1 : v i X r a Key Words: Scalar Curvature Equation; Blow-up; Critical Points; Sobolev Spaces. 2000 AMS MS Classification: Primary 35J60; Secondary 53C21. ∗ [email protected] ∗∗ [email protected] 1. Introduction. As a counterpart of the Yamabe problem [7] [8] [19] [33] [37] [39] (cf. also [2]), the prescribed scalar curvature problem in Sn (n 3 ) asks for a positive solution U to the nonlinear partial ≥ differential equation (1.1) ∆1U c˜nn(n 1)U + (c˜n )Unn−+22 = 0 in Sn (U > 0), − − K where is a prescribed function on Sn. Here c˜ = (n 2)/[4(n 1)] . See §1d for the n K − − rather standard notations we use. Also known as the Nirenberg/Kazdan-Warner problem [36], it can be compared to the classical Minkowski problem on prescribing Gaussian curvature for convex compact surfaces in IR3. The hallmark of equation (1.1) is the critical Sobolev exponent: the injection H1,2(Sn) ֒ Ln2−n2(Sn) is not compact, typified by blow-up gathering at critical → point(s) of . Close to half a century (cf. an early work in 1972 by Dimitri Koutroufiotis [20], K whose thesis adviser is Louis Nirenberg), equation (1.1) serves as a vehicle for sophisticated techniques in nonlinear partial differential equations to be deployed and developed. It can also be branched out to complete manifolds, CR manifolds, Q-curvature, as well as related to mean field equations. See some recent works [1] [14] [12] [13] [15] [21] [30] [31] [32] [27] [34] [35] [40] on the topic, and the references therein. In general, existence results involve symmetry on , or local conditions on the critical points of together with index inequality(ies). The K K following result provides a good picture see [1] [11], and in particular [14] regarding (iv) { below . Assume the following (i)–(iv). } (i) is a smooth Morse function [namely, all its critical points (collected in the set K denoted by Crt) are non-degenerate]. (ii) ∆ (x ) = 0 for all x Crt. 1 c c K 6 ∈ (iii) ( 1)Ind(K, xc) = ( 1)n , here Crt = x Crt ∆ (x ) < 0 . < c 1 c − 6 − { ∈ | K } xc∈XCrt< (iv) is “sufficiently” close to a positive constant. K Then equation (1.1) has a positive solution (precisely, see Theorem 7.1 in [1], pp. 103). Recall that the index of a non-degenerate critical point is the number of negative eigenvalues of the Hessian matrix at that point. We observe that in case Crt contains only one point, say at the north pole N, then < it must be the peak (maximal point), and hence (together with the non-degenerate condition) Ind( , N) = n = ( 1)Ind(K, xc) = ( 1)Ind(K, N) = ( 1)n . K ⇒ − − − xc∈XCrt< Moreover, via the Kazdan-Warner (Pohozaev) identity, if is strictly decreasing from N to K S, measured via the geodesics, then equation (1.1) does not have any positive solution at all (cf. also [18]). 1 Motivated by this, attention is given to the situation where Crt contains at least two < points. Cf. [6] [4] [10] [41] (a discussion on the existence and non-existence results can be found in §1c). Thus in this article we consider juxtaposed ‘twin’ pseudo-peaks (described in §1a). We note that “side-by-side” is a kind of symmetry condition. To state the local conditions on the Taylor expansions at the two pseudo-peaks, we introduce the stereographic projection ˆ P [see (4.15)] which sends the north pole to infinity. Equation (1.1) is transformed to (1.2) ∆v + (c˜nK)vnn+−22 = 0 in IRn (v > 0). See, for examples, [24] [38]. Here n−2 2 2 (1.3) v(y) := U (ˆ−1(y)) and K(y) = (x) for y = ˆ(x) IRn. P · 1 + y 2! K P ∈ | | (Note that max K is not affected by whichever point we choose as the north pole.) §1a. Close ‘twin’ pseudo-peaks and their key parameters. Consider two critical points q 1 and q of K. Via a translation, one may assume without loss of generality that 2 (1.4) q = 0. 1 Let γ denote the distance (or gap) between the two critical points. Moreover, in this article, we always assume that q and q are close, namely, 1 2 (1.5) γ := q = O(1). 2 | | The two critical points are symmetric (or ‘twin’) in the following sense [(1.6) & (1.8)]: (1.6) K(0) = K(q ) > 0 2 [after a rescaling, we may accept without loss of generality that (1.7) (c˜ K)(0) = (c˜ K)(q ) = n(n 2)], n n 2 · · − in conjunction with (similarity on the Taylor expansions) (1.8) (c˜ K)(y) = n(n 2) + Pℓ(y q ) + Rℓ +1(y q ) for y B (ρ) n· − j − j j − j ∈ qj and j = 1, 2. Here (1.9) ρ = h¯ γ (h¯ is a fixed number less than half), (i) · (1.9) Pℓ is a homogeneous polynomial of degree ℓ 2 (ℓ - the flatness, is the same for (ii) j ≥ j = 1, 2), and 2 (1.9) Rℓ +1 the remainder in the Taylor expansion, satisfying (iii) j (1.10) Rℓ +1(y) C y ℓ+1 for y B (ρ), | 1 | ≤ R1 ·| | ∈ o Rℓ +1(y) C y q ℓ+1 for y B (ρ) . | 2 | ≤ R2 ·| − 2| ∈ q2 Here C and C are positive constants. (1.8) implies that R1 R2 (1.11) (c˜ K)(y) n(n 2) C y ℓ for y B (ρ), | n· − − | ≤ P1 ·| | ∈ o (c˜ K)(y) n(n 2) C y q ℓ for y B (ρ) . | n· − − | ≤ P2 ·| − 2| ∈ q2 Here the positive constant C (j = 1, 2) is linked to the sum of the absolute values of the Pj coefficients of P . Assume that j 1 ℓ is even, and let h = ℓ . ℓ Hence 2 · (1.12) ∆(hℓ)Pℓ(y¯) = ∆ ( [∆ [∆ Pℓ(y¯)]])) = ̟ (j = 1, 2) y¯ j y¯ ··· y¯ y¯ j j h times ℓ ← → is a number. Here y¯ = y q . The key condition for the critical points q and q to be j 1 2 − called pseudo-peaks is the following: (1.13) ̟ < 0 for j = 1, 2. j We add the following “symmetry” condition as well: 1 (1.14) ̟ ̟ C ̟ , 1 2 p 1 C ·| | ≤ | | ≤ ·| | and p (1.15) ̟ C . 1 ω ≥ − In (1.14) and (1.15), C and C are positive constants. p ω Main Theorem 1.16. For 6 n < 10, let ≤ (1.17) ℓ [2, n 2) ∈ − be an even integer Cℓ+1(Sn), and K the projection of to IRn via (1.3). Assume that K ∈ K (1.18) K C¯ in IRn, b | | ≤ and K has twin pseudo-peaks in the sense of (1.7), (1.8) and (1.13), located at q = 0 1 and q IRn. Under the conditions in (1.10), (1.11), (1.14) and (1.15), there is a positive 2 ∈ constant γ so that if o q γ , 2 o | | ≤ then equation (1.1) has a positive C2-solution. Moreover, γ depends only on n, ℓ, C¯ , and o b the parameters of the twin pseudo-peaks (namely, h¯, C , C , C , C , C and C ). R1 R2 P1 P2 p ω 3 Remarks. (1) To gain an idea on the dependence of γ on C [appeared in (1.14)], we have o ω c µ γ . o ≈ 1 Cℓ ω Here the small positive number c depends onthe other parameters in Theorem 1.16. See §5c. µ (2) With the help of Theorem 1.16, one can consider multiple solutions for well-separated multiple twin pseudo-peaks. (3) There is no condition on other critical points. (4) Dimension restriction (n = 6, 7, 8 & 9) mainly due to the process when key information are extracted out of the reduced functional (refer to Proposition 4.1). §1b. Lyapunov-Schmidt reduction method without perturbation. Organization. As described in [1], the elegant Lyapunov-Schmidt reduction method is considered on those K which is a perturbation of a positive constant, that is (after a rescaling), (1.19) (c˜ K) = n(n 2) + ε (c˜ H). n n − · Here ε is “small enough”. A new insight is introduced in [40], where Wei and Yan bring home to the point that when a large number of standard bubbles are arranged near the critical points of K, one can still apply the Lyapunov-Schmidt reduction method, this time without the requirement on ε being close to zero (see also an earlier work of Yan [41]). Thus the number of bubbles replaces the parameter ε. In this article, we show that by “planting” one bubble each near one of the twin pseudo- peaks, the Lyapunov-Schmidt reduction method is also applicable without the need for K being close to a constant (§2 & §3). In this case the “gap” γ take the place of the parameter ε. Moreover, we show that the reduced functional has two main contributions (Proposition 4.1; cf. also [10]), one from the critical point (§4), and the other one from the interaction with the other bubble (§2b). By properly balancing these two effects, we show that equation (1.2) has a solution if the peaks are close enough (§5). This solution can be transferred back to Sn via (1.3) as a solution of (1.1). Moreover, as the two bubbles are highly concentrated near the twin pseudo-peaks, other critical points (if any) do not contribute to the consideration. This is in harmony with a theme in [23] (cf. also [22]) that concentration can be put to good use to find solutions of equation (1.1). 4 §1c. Comparison with some related existence and non-existence results. Our result should be compared to [10], in which the authors use a version of Lyapunov-Schmidt reduction method for ε small enough, when H in (1.19) has two critical points [among other possible critical point(s)], say at q′ = 0 and q′ = (q′ , , q′ ) IRn (not necessarily close), which 1 2 2|1 ··· 2|n ∈ satisfy (1.20) H(y) = H(0) + a y β1 + + a y β1 + O y β1+σ1 for y B (ρ), 1 1 n n i o | | ··· | | | | ∈ (cid:16) (cid:17) (cid:16) (cid:17) H(y) = H(q′ ) + b y q′ β2 + + b y q′ β2 2 1| 1 − 2|1| ··· n| n − 2|n| (cid:16) (cid:17) q′ + O y q′ β2 + σ2 for y B (ρ) ρ < | 2| , | i − 2|i| ∈ q′2 " 2 # (cid:16) (cid:17) where β , β (0, n 2), 1 2 ∈ − n n a = 0, b = 0, a < 0 and b < 0 (i = 1, 2, , n), i i i i 6 6 ··· i=1 i=1 X X then for ε in (1.19) small enough, equation (1.2) has a (two peaks) solution (see Theorem 1.1 in [10] for the precise description). In the above, σ , σ (0, 1) are fixed numbers. Besides 1 2 ∈ the requirement on ε being small enough, we note that in (1.20), there is no cross over terms like y y , which is allowed in our Main Theorem 1.16. 1 2 × ··· In [41], a counterpart to the situation above is considered. There Yan studies the case when K hasapairofstrictlylocalmaximum pointsat m and m , whose distance m m 1 2 1 2 | − | is very large [flatness of these two local maxima is in the range (n 2, n)]. See Theorem 1.1 − in [41] for the complete statements. On the other hand, a non-existence result obtained by Bianchi in [5] suggests that for certain “very sharp” twin peaks with flatness lesser than or equal to n 2, equation (1.1) has − no positive solution. For details, see [4] [5]. Cf. also [27]. Thus the smallness of γ in the Main Theorem cannot be totally removed. §1d. General conditions, assumptions and conventions. Throughout this work, (1.21) Sn = x = (x , , x ) IRn+1 x2 + + x2 = 1 (n 3), { 1 ··· n+1 ∈ | 1 ··· n+1 } ≥ with the induced metric g . ∆ is the Laplace-Beltrami operator associated with g on Sn. 1 1 1 Likewise, ∆ is the Laplace-Beltrami operator associated with Euclidean metric g on IRn, o with coordinates y = (y , , y ) IRn. Moreover, the norm and the inner product 1 n ··· ∈ k k , are defined via Euclidean metric g on IRn. o h i 5 As mentioned earlier, c˜ = n−2 . We observe the practice on using ‘C’, possibly with •1 n 4(n−1) sub-indices, to denote various positive constants, which may be rendered differently from line to line according to contents. Whilst we use ‘c¯’ or ‘C¯’, possibly with sub-indices, to denote a fixed positive constant which always keeps the same value as it is first defined. Denote by B (r) the open ball in (IRn, g ) with center at y and radius r > 0, and 2 y o • ∂B (r) its boundary. Whenever there is no risk of misunderstanding, we suppress dy from the y integral expressions on domains in IRn. §1e. e-Appendix. Some of the preparatory estimates are situational modifications of well- established arguments. We gather those details in the e-Appendix, which is presented from pp. 36 onward. § 2. The Lyapunov - Schmidt reduction scheme sans perturbation : the case of two bubbles. Equation (1.2) is naturally associated with the Hilbert space (2.1) 1,2 = 1,2(IRn) := f Ln2−n2 (IRn) W1,2(IRn) f, f < . D D ∈ loc IRnh▽ ▽ i ∞ (cid:26) (cid:12) Z (cid:27) \ (cid:12) The inner product is defined by (cid:12) (cid:12) (2.2) f , ψ := f , ψ for f , ψ 1,2, and f 2 := f , f . h i▽ IRnh▽ ▽ i ∈ D k k▽ h i▽ Z The functional corresponding to (1.2) is given by 1 n 2 2n (2.3) I(f) = f , f − (c˜ K)fn−2 for f 1,2. 2 IRnh▽ ▽ i − 2n · IRn n· + ∈ D Z (cid:18) (cid:19) Z Here f denotes the positive part of f . See Part I [24] on the regularity of the critical points + of (2.3). Cf. also [9] in relation to equation (1.1). Let (2.4) (c˜ K) = n(n 2) + (c˜ H) (c˜ H) = (c˜ K) n(n 2). n n n n · − · ⇐⇒ · · − − Accordingly, I can be split into two parts (2.5) I(f) = I (f) + G(f), o 1 n 2 2n (2.6) where Io(f) = 2 IRnh▽f , ▽f i − n(n − 2)· 2−n IRnfn−2 , Z (cid:18) (cid:19)Z n 2 (2.7) and G(f) = − 2−n · IRn(c˜n·H)fn2−n2 for f ∈ D1,2. Z 6 Here we pay special attention on the negative sign in G(f). One of the key themes in this article is to expound the interaction between I′ and G′. o Let us present the following flow chart to guide our discussion. I(f) = I (f) + G(f) for f 1, 2. o ∈ D ↓ “Pseudo” Kernel of I′ : = z = V + V [ Refer to (2.11)]. o Zσ { σ λ1, ξ1 λ2, ξ2 } ↓ T [tangent space, cf. (3.2).] zσ Zσ ↓ := (T )⊥ (Write 1, 2 = T .) ⊥σ zσ Zσ D zσ Z ⊕⊥zσ ↓ P : 1, 2 (Projection unto the “normal”.) σ σ D → ⊥ ↓ P I′(z + w ) = 0 The auxiliary equation. “Small ” solution : w . σ◦ σ zσ zσ ∈ ⊥σ (Cancelation along the normal directions.) | ↓ I (z ) := I(z +w ) [Finite dimension functional : (IR+ IR+) (IRn IRn).] R σ σ zσ × × × ↓ I ′(z +w ) = 0 (Critical point ˜z of the reduced functional.) R σ ˜zσ σ ↓ I′(z +w ) = 0 (Full functional.) σ ˜zσ ↓ (˜z +w ) = V + V + w is a solution of equation (1.2). σ ˜zσ λ˜1, ξ˜1 λ˜2, ξ˜2 ˜zσ (Refer to Lemma 3.44.) – Flow Chart of the Lyapunov-Schmidt reduction scheme without perturbation. – §2a. First order property - interaction between two ‘well-separated’ bubbles. For f 1,2, ∈ D a calculation using (2.6) shows that the Fr´echet derivative of I at f is given by o n+2 (2.8) I′ (f)[h] = f , h n(n 2)f n−2 h for h 1,2. o IRn h▽ ▽ i − − + · ∈ D Z (cid:20) (cid:21) 7 The kernel of I′ consists of functions of the type (see [9]) o n−2 λ 2 (2.9) V (y) = for (λ, ξ) IR+ IRn, λ, ξ λ2 + y ξ 2! ∈ × | − | which satisfies the equation (2.10) ∆Vλ, ξ(y) + n(n 2)[Vλ, ξ(y)]nn+−22 = 0 in IRn. − We consider juxtaposition of two bubbles (2.11) z = V + V for (λ , λ ; ξ , ξ ) (IR+ IR+) (IRn IRn). σ λ1, ξ1 λ2, ξ2 1 2 1 2 ∈ × × × §2b. Unit and restrictions. In the following we assume that (2.12) C¯−1 λ < λ < C¯ λ , ξ < c¯ λ and ξ q < c¯ λ. 2 1 2 1 2 2 · · | | · | − | · Here C¯ (> 1) and c¯ ( 0+) are positive constants (to be more precisely described in §5 ). ≈ With (2.11), we define γ q 2 (2.13) λ = λ λ and D = = | | . 1 2 q · λ √λ1 ·λ2 ! These imply (2.14) 1 λ λ √C¯ λ for j = 1, 2, √C¯ · j ≤ ≤ · j ξ ξ 1 1 c¯ 1 2 [ D 2c¯] d := | − | [ D + 2c¯] and = 1 + O . − ≤ √λ λ ≤ d D · D 1 2 (cid:20) (cid:18) (cid:19)(cid:21) · §2c. Weak interaction. We know that I′ (V ) 0, but I′ (V + V ) 0. o λ,ξ ≡ o λ1,ξ1 λ2,ξ2 6≡ In this section we investigate the “interaction” in more detail. From (2.8) and (2.11) we have (2.15) I′o(zσ)[h] = n(n − 2) IRn [Vλ1, ξ1]nn−+22 + [Vλ2, ξ2]nn−+22 − [ Vλ1, ξ1 + Vλ2, ξ2 ]nn−+22 ·h . Z (cid:26)(cid:16) (cid:17) (cid:27) for h 1,2, ∈ D 8 Lemma 2.16 (Weak Interaction Lemma). Assume that n 6, with the notations and ≥ conditions in (2.12) and (2.13), there exists a positive constant D¯ > 1 such that 1 γ (2.17) if D = D¯ , 1 λ ≥ ln D (2.18) then I′ (z ) C¯ . k o σ k ≤ 1 · Dn+2 2 In (2.17) and (2.18), the positive constants D¯ and C¯ can be precisely determined by C¯, c¯ 1 1 [appeared in (2.12)] and n, and they are independent on (λ , λ ; ξ , ξ ) as long as (2.12) 1 2 1 2 is satisfied. The proof can be seen from the proof of Lemma 2.1 in [28], together with Lemma A.5 in the Appendix. §2b. Interaction terms. In the following we describe the interaction between two bubbles via (2.15). We first observe that in a small neighborhood of ξ , V is small when compared to 1 λ2, ξ2 V . Precisely, we let λ1, ξ1 (2.19) ρ = µ ξ ξ . µ 1 2 ·| − | Here µ is a chosen small positive number so that (2.20) µ 0+ (slowly) and µM D when D . → · → ∞ → ∞ Here M is a (fixed) large integer. For most particular purpose one can take 1 1 µ = for D 1, 2 · Dǫ ≫ where ǫ < 1 is any fixed small positive number. Under the conditions in (2.12), we have λ n−22 1 n−22 (2.21) V (y) = 2 = λ1 λ2, ξ2 λ22 + |y − ξ2|2!  λλ21 + |yλ−1·λξ22|2  (cid:16) (cid:17)  n−2 1 1 2 = λn−22 · λ2 + |(y − ξ1) + (ξ1 − ξ2)|2  1 λ1 λ1·λ2  (cid:16) (cid:17)  n−2 1 1 2 = λn−22 · λ2 + |ξ1 − ξ2|2 + |y − ξ1|2 + 2(y − ξ1)∗(ξ1 − ξ2)  1 λ1 λ1·λ2 λ1·λ2 λ1·λ2  (cid:16) (cid:17)  = d2 [ 1; cf. (2.14)], dominating term { ↑ ≫ } 9