# Full text of "Liouville and Calabi-Yau type theorems for complex Hessian equations"

## See other formats

Liouville and Calabi-Yau type theorems for complex Hessian equations . . . Slawomir Dinew, Slawomir Koiodziej (N ^vj . Abstract ^ , We prove a Liouville type theorem for entire maximal m-subharmonic , ^ ' functions in C" with bounded gradient. This result, coupled with a stan- dard blow-up argument, yields a (non-explicit) a priori gradient estimate for the complex Hessian equation on a compact Kahler manifold. This ter- OO ' minates the program, initiated in [HMW| . of solving the non-degenerate Hessian equation on such manifolds in full generality. We also obtain, us- ing our previous work, continuous weak solutions in the degenerate case *~^ ' for the right hand side in some L^, with sharp bound on p. u r^ ; Introduction cd (N X a The complex Hessian equation ►^1 (w a Kahler form, / > given function, 1 < m < n) can be seen as an inter- V/^ . mediate step between the Laplace and the complex Monge-Ampere equation. 0^ ' It encompasses the most natural invariants of the complex Hessian matrix of a 0^ , real valued function - the elementary symmetric polynomials of its eigenvalues. Its real counterpart has numerous applications and has been thoroughly stud- Cn . ied (see jW]). Some equations of similar type appear in the study of geometric ^^ I deformation flows such as the J- flow ( [SW] '). Thus it can be expected that Hes- sian equations themselves may also have interesting applications in geometry, and indeed they came up recently in jAVj in problems related to quaternionic geometry. Due to their non linear structure, and similarly to the Monge-Ampere equa- V^ , tion, the Hessian equation is considered only over a suitable subclass of functions for which ellipticity is guaranteed. These subclasses, in a sense, interpolate be- tween subharmonic and plurisubharmonic functions and thus they are also inter- esting object to study from potential theoretic point of view ( |B11] . |DK) . |Chi| ). In the Kahler manifold case the corresponding classes of functions, namely co — m subharmonic ones, do not yield, in general, classical positive definite metrics, which distinguishes them from w-plurisubharmonic functions. The real Hessian equatiorp has been extensively studied, both in suitable domains of E", and on Riemannian manifolds. Likewise, the theory of the complex Hessian equation is being developed in domains of C" and on compact Kahler manifolds. The case of domains in C" is to a large extent understood ^A detailed survey of the theory of the real Hessian equation can be found in [W]. ([El], |B11] . [DKj ) ■ In particular the classical Dirichlet problem is solvable in the smooth category ([Li], |B11) ) under standard regularity assumptions on the data and some convexity assumptions on the boundary of the domain. The corresponding problem on a compact n-dimensional Kahler manifold u is uj — m subharmonic (w + dd'^u)™ A w"^" = /w" (0.1) supxu = 0, with / smooth, strictly positive, and satisfying the necessary assumption J^ fuj" = Jy Lu"' remained unsolved in full generality until now. The equation was solved under the assumption of nonnegative holomorphic bisectional curvature ( |Hou| , [j], [Kop or under additional convexity assumptions on u itself ([G], [Z])- The natural approach by continuity method, as in the case of the com- plex Monge- Ampere equation (0), faces an obstacle which does not appear in Yau's solution to the Calabi conjecture: the C^ a priori estimate needed for the closedness part seems to depend on the a priori gradient bound. Such a gradient bound in the Monge- Ampere case was obtained by Blocki and Guan independently ( |B12j . [G]) but their methods seem to fail in the Hessian case. Thus the only remaining piece in the program of solving the complex Hessian equation on compact Kahler manifolds is the a priori gradient estimate. In this note we prove such an estimate by exploiting blow-up analysis as advocated by Hou-Ma-Wu in |HMW| . We obtain a Liouville type theorem for entire maximal m-subharmonic functions in C" and the gradient bound then follows as a corollary. Below we give the precise statements: Theorem 0.1 (Main result). Let u be a globally bounded maximal m subhar- m,onic function in C", 1 < m < n. Suppose moreover that supc^\\\Iu\\ < oo. Then u is constant. Theorem 0.2 (Gradient estimate). Let (A, w) be a compact Kahler manifold of complex dimension n. Suppose that u G C^(A) is an uj — m subharmonic function solving the equation {u + dd^uy" A w"-" = /w" and maxM = 0, with f a C^ smooth positive function. Then supxW^uWui < C, with a constant C dependent on n, m, ||/^'™||c2 and the bound on the bisec- tional curvature of(X,Lu). As a corollary, our bound, coupled with the existent results f |HMW) . see Preliminaries) , yields that the equation (|0.ip is solvable in the smooth category. Theorem 0.3. The equation liO.l}) has a solution on any compact Kahler man- ifold. The questions whether there is an explicit gradient bound (independent of a blow-up argument) and whether the C^ estimate is indeed dependent on the one for gradient remain open. From Theorem l0.3l and the results of our previous paper |DKj we also obtain weak continuous solutions for nonnegative / in some L^ spaces. The result is sharp as far as the exponent p is concerned. Theorem 0.4. Let X be a compact Kdhler manifold. Then Jar p > n/m and f G LP{dV) there exists a unique function u G ST-LmiX^Lo) satisfying {lo + dd^u)"" A w"-™ = /w" and maxM — 0. In ^DKj this theorem was shown under additional assumption of nonnegative holomorphic bisectional curvature. The same proof apphes here once we have smooth solutions from Theorem 10.31 1 Preliminaries We briefly recall the notions and results that we shall need later on. Throughout the note Vt will denote a relatively compact domain in some C", while (X, io) will be a compact Kahler manifold of complex dimension n. Let also d = d + d and d'^ := i{d — d) denote the standard exterior differentiation operators. By /3 := (i(i^||z|p we shall denote the Euclidean Kahler form in C", while Lu will be a fixed Kahler form on a compact Kahler manifold. The standard positivity cones associated to am equations are defined by r,„ = {A G M'>i(A) > 0, • • • , a„(A) > 0}, (1.1) with crj(A) := El<z^<...<z,<nK K ■■■^^r^■ By Garding theorem (see jGa) ') these positivity cones are convex. If fl is a fixed domain in C and u is any C'^{^) function we call it m — /3- subharmonic (?7i-sh for short) if for any z € fl the complex Hessian matrix dzdz (•^) ^^^ eigenvalues forming a vector in the closure of the cone r^. By analogy given a complex Kahler manifold (X, w) a C^ smooth function u is called w — m-subharmonic if the form uj + dd'^u has at each point z eigenvalues (with respect to uj) forming a vector in r,„. Note that as a; depends on z the positivity assumption is dependent on z toqj. In the language of differential forms u is ?7i-subharmonic (respectively uj — m subharmonic) if and only if the following inequalities hold: (dd'^u)'' Auj''-'' >0, k^l, •■• , m, (resp. (w + dd^uY A w""'' > 0, fc = 1, • ■ • , m). It was obsreved by Blocki ([BIT]) that, following the ideas of Bedford and Taylor ( |BT1| . [BT2j ). one can relax the smoothness requirement on u and develop a non linear version of potential theory for Hessian operators. The relevant definition is as follows: Definition 1.1. Let u he a subharmonic function on a domain 51 G C". Then u is called m- subharmonic (m-sh for short) if for any collection of C^ -smooth m-sh functions Vi, •■• , fm-i the inequality dd^u A dd^vi A • ■ • A dd%rn-i A /S""™ > holds in the weak sense of currents. The set of all m-sh functions is denoted by SUmi^)- ^It can be proved f |Hou| ) that the positivity cones associated to any point z are invariant under parallel transport with respect to the Levi-Civita connection of ui. Similarly weak tLi-m-subharmonicity is defined on compact Kahler m.anifolds (see [DK] and [Chi)). Using an approximating sequence of functions one can follow the Bedford and Taylor construction from |BT2] of the wedge products of currents given by locally bounded m-sh functions. They are defined inductively by dd'^ui A • • • A dd'^up A ;3"-" := dd^im A • ■ • A dd'^Up A ;3"^"'). It can be shown (see [Bllj ) that, in analogy to the pluripotential setting, these currents are continuous under monotone or uniform convergence of their potentials. Moreover the mixed Hessian measures are also positive (!B11!): Proposition 1.2. Let ui,--- ,Um be bounded m-subharmonic functions then the measure dd^ui A ■ • • A dd^um A /3"-™ is nonnegative. Here we list some basic facts about TTi-subharmonicity. Proposition 1.3. Let fi C C" 6e a domain. Lf SHk{^) denotes the class of k-subharmonic functions in Q then 1. SHin) = SHiin) c SH2i^) c • • ■ c SHn{n) = vsH{n), 2. Lf u ^ ST-irn{^) and 7 : K — > M is a C^ -smooth convex, increasing function then J on € ST-Lm{^), 3. Given any convex model domain P the P-average of a m-sh function u defined by up{z) :— -r-^ Jp u{z -\- w) P"' (w) function is again m-sh on the set where it is defined. Note that, in particular, if u is nonnegative m-sh, and B{r) is the ball {z e C"|||z|| < r} then [u^]riz) := /- ^ „„ f„, .u'^{z + w)/3"(w), is stiU m- subharmonic. The following two theorems, known as comparison principles in pluripoten- tial theory, follow essentially from the same arguments as in the case m = n: Theorem 1.4. Let u, v be continuous m-sh functions in a domain fl G C". Suppose that lim inf 2_j.(3si (^ — ''^)(-^) ^ then idd%r A ^"""' < / (dd'^u)™ A Z^"-". {u<v} J {u<v} Theorem 1.5. Let u, v, wi,--- ,Wm~i be continuous m-sh functions in a domain fJ G C". Suppose that livaird z^dn{u — ^)(^) ^ then {dd^v) A dd^wi A • • • A dd^wm-i A /3""" {u<v} < / (dd^u) A dd%i A • • • A dd^Wm-i A /?" ' {u<v} On a compact Kahler manifold {X, to) the solvability of the Hessian equation (for 1 < TO < njl {u is id ~ ni subharmonic; (cj + dcf^u)™ A cj"-™ = /w" (1.2) supxu = 0, with a strictly positve smooth function / satisfying the necessary conditon J-^ fuj'^ = J^ w" has attracted recently a lot of interest. The equation was solved under a curvature assumption ( [Houj , [J] , |Ko] ) or under additional con- vexity assumptions for u itself ([H]: [Z])- The continuity method applied in the solution of the Calabi conjecture by S. T. Yau ([Y]) is already known to work, up to a point, without any extra assumptions. Let us state the relevant ingredients: 1. The openness part for the continuity method works ( [Hon] ) : 2. The uniform estimate needed for the closedness part was furnished by Hou ( [Hou| ) . Alternative approaches were found in |DK) and [Chi] : 3. The C^"*"" a priori estimate for u follows from standard Evans-Krylov the- ory as it was observed by Hou f jHouj ): 4. The C^ a priori estimate holds provided there is an a priori gradient bound (HMW]). In particular the only remaining part to accomplish the program was the gra- dient a priori estimate. In fact the Hou-Ma-Wu |HMW) result yields even more and it will be used in our argument below. Theorem 1.6. If u E C'^{X) solves the equation il.^) then the following C^ a priori estimate holds supxlldd^uW^ < C(supx||V-u|p + 1), (1.3) where C is a constant dependent on \\f^'"^\\c'^(x), fn, n, ||m||oo o,nd the bound on the bisectional curvature of {X,uj). In particular, as mentioned by the authors in [HMWj . this type of estimate is amenable to blow-up analysis which reduces the gradient bound to a certain Liouville type theorem for global m-sh functions on C". 2 The gradient estimate for the complex Hes- sian equation - reduction to a Liouville type theorem Although it is pretty standard, we present the details of this reduction for the sake of completeness. ^The case m = 1 corresponds to the standard Laplace equation, while the solvability in the case m = n is guearanteed by the Calabi- Yau theorem. Suppose that on some compact Kahler manifold {X, uj) the gradient estimate fails. By definiton one then finds a sequence of C^ smooth w — m sh functions Uj solving the problems supxUj — 0, (2.1) with some strictly positive functions fj satisfying ||/^^™||c2(x) ^ C for some uniform constant C (in particular the uniform norm of fj is also under control and hence, so is the uniform norm of Uj { [Houj . |DK| )). yet Cj :— supx\^Uj\ — > CX). After passing to a subsequnce one may assume that the maximum of any IIVmjII is attained maximum point, in a fixed coordinate chart and those points converge to a point z which can be taken to be the center of this coordinate chart. We may further assume that to :— i X]fc ,=i Qjkdzj A dz}, is equal to dd'^v for some local smooth plurisubharmonic function (the local potential of w). Changing the coordinates and shrinking the chart if needed one may further assume that a;(0) = /3 and uj{z) — 0(||2;||^) Then the subsequence (which we shall still denote by Uj) still has its gradient maximum points contained in the chart and converging to the center. Without loss of generality we assume that the unit ball in the introduced coordinates is contained in the chart. Consider now the dilation Uj{z) := Uj{-^) relative to the local coordinates introduced above. By construction Uj is a C^ smooth function on the ball 1(0, Cj) := {z G C"|||z|| < Q} with smpb(o.c,)I|Vuj|| = 1 + O(^). Moreover supK(ia,Cj)\\dd'^Uj\\ < -^supx\\dd'^Uj\\^ < C* by Theorem 11.61 Thus for any a e (0, 1) the family {uj}j>fc is relatively compact in the C^'" topology on the domain B(0,Cfc). Then, by a standard diagonalization procedure and Arzela- Ascoli theorem one can extract a limiting function u G C"'^^"(C"), such that |Vu(0)|| = 1. In particular u is not constant. Observe however that in the introduced coordinates (|2.1I) becomes dd'^iiv + Uj){z))"' A (^#-^(2))"-™ = /j(z)(rf#'y(z))". Thus Uj satisfies the equation (dd-(.^(z/Q) ^^^,^^^^^^„^^^ (dd-(.K^/Q) ^„_^ ^ /^■(,/c^.)(( (^^^W(//^^) )». For j large the latter equality can be written as (recall v is smoothly bounded!) (0(l/Q)/3 + dd^uj^ A ((1 + 0(l/C|))/3)"-" = 0(1/C|™). (2.2) More generally for any 1 < fc < m a similar reasoning coupled with the lo — m subharmonicity of Uj yields (0(l/Cj)/3 + dd'uj)'' A ((1 + 0(l/C|))/3)"-'= > 0. These inequalities inductively yield that the limiting function u is m-sh. Then the equalities (|2.2p can be read also in the pluripotential sense and thus one can extract the weak limit satisfying {dd^u)"" A /3"-™ = 0. (2.3) In particular, by a result of Blocki f |Bll) V this means that u is a maximal m-sh function in C". Thus we have constructed uniformly bounded maximal nonconstant 77i-sh function which is, in addition, globally C^'" smooth with all C^'" norms under control. In particular the gradient is uniformly bounded. Thus a Liouville type theorem saying that no such function exists would yield a contradiciton, and hence a (nonexplicit) a priori gradient estimate. 3 Liouville theorem for maximal m-subharmonic functions Any bounded entire m-subharmonic function is in particular subharmonic. While bounded subharmonic functions do exist in C", n > 2, they all must have con- trolled behavior at infinity for almost all values as the next theorem shows. Theorem 3.1 (Cartan lemma, see jHK' ) . Let u be a bounded subharmonic function in C", n >2. Fix any q > 2n ~2. If a ^ supc^u, then where the set A is contained in, at most, countable collection of balls M{xk,rk) such that J2T=ii'\nn\)'^ < °°- In particular if u is a bounded nonnegative m-sh function with 1 = supcnu then for any z G C" limr-j,oo[u'^]r{z) = limr^oo[u\r{z) = 1. This fact could have been proven using more elementary Harnack inequalities, but by referring to Cartan's lemma we make the argument shorter. Theorem 3.2 (Liouville type theorem for 7n-subharmonic functions). Any u bounded, m-sh maximal function in C" with bounded gradien)^ is constant. Proof. We use induction over n. Note that for n — \ and m ^ n the statement is known. Arguing by contradiction assume u is nonconstant and inf u = 0,supw = 1 and ||Vu|| < cq. First suppose that u has the following property: there exists p > 0, a se- quence of mappings Gk , each a composition of a complex rotations and a trans- lation, and a sequence of balls Bk = B{0,rk), rk ^> oo such that [u^ o GkU (0) + [u o Gk]p{0) - 2u o Gfe(O) > 4/3, (3.1) where we use the notation Vol{B{z,r)) Jb(,,^^) Vol{E) = ^,3"; and ,lim_/ I^Pr=0. (3.2) 'By this we mean that u is Lipschitz with global Lipschitz constant under control. The functions Uk = uoGk, are maximal m-subharmonic. By Arzela-Ascoli theorem and a diagonahsation argument one can choose a subsequence converging locally uniformly to an entire m-sh function v. This function is also maximal by the convergence theorem [BUj . From (I3.2p . and the bound on ||Vw|| it follows that v is constant along the lines with fixed z' = (z2, •••, ^n)- Indeed, suppose that for some complex numbers a, h and positive c we had v{a, z'q) > v{b, z'q) + 2c. Then, by \Vv\ < cq, for 5 = c/(4co) inf{v(zi, z') : |zi — a\ < S, \z' — Zg| < S} > svLp{v{zi,z') : \zi — b\ < 6, \z' — Zg| < (5} + c. Integrating the directional derivative of v along intervals parallel to [a, 6] with endpoints in the sets appearing on both sides of the above formula, using Cauchy-Schwarz inequality and Fubini's theorem, we get for an interval / of length 8 perpendicular to [a, 6] in z\ plane and E —\a^\)\ x I x B{zq, 6) 4^P/3" > c(n)(52»cV(4|a - b\) ^ c(n)c2("+i)/(4"+ic2'> - b\) E ozi {c{n) - dimensional constant). By the uniform convergence the analogous esti- mate would be true for Uk, k large enough, in contradiction to (13. 2p . Therefore vi{z2, ■■■, Zn) = v{z) is m-sh, maximal in C"^^. By induction hypothesis it is constant, but this is impossible, since: v = c would violate (|3.ip as [uk]piO)-UkiO)>l/3. Therefore the property stated at the beginning of the proof does not occur. This means that there is i? > 0, such that ii r > R and z, p > satisfy (z) + [u]p - 2u{z) > 4/3 [u\i^ then for any vector w, \w\ = 1 1 ,, ,fj,. ,, , l|^P/3">ci=ci(r). (3.3) Vol(B{z,r)) JB{z,r) ow Choose origin, p > and r > R so that u{0) < 1/9, [u]p(0) > 3/4 and [u^]r(0) > 3/4. The choice is possible since for the first inequality one can use inf u = 0, and the second and third inequalities for r, p large enough follow from the Cartan lemma. Then 0eU^{2u< [u\{z) + [u]p - 4/3}. By maximality of w the set U is not bounded. From ddf^u^ — 2{duAd'^u + udd'^u), and (j3.3p it follows that dd^lu^ > 2[udd''u]r + C2/3. (3.4) (with C2 depending on ci,r and the average form [2udd'^u\r defined in analogy to [v]r above). Hence, by ProDOsition ll.2l dd^lu^ A (dd^u)"'-'^ A /3"-™ > C2(dd'=u)™-i A /3"-™+\ (3.5) on the set U. Take g{z) — — e||z|p with positive e so smah that 02/? > ~2d(Pg (and therefore [u'^]r + .9 is m-sh on U (see ^.4^ )) and note that Ug = {2w < [u\{z) + up + g- 4/3} is nonempty. The set Ug is bounded since \\uY\z\^ao 9{z) = —00 and it is con- tained in U . Then using p.4p and the comparison principle {dd'^ur-^ A /?" 2. a, < / dd''{[u% + [u]p + g)^ (dd'^u)"'-^ A /3"-™ < 2™ /" (dd'^u)" A Z?"-™, and having on the left, we obtain [dd^uY'-^ A /3"-™+i = on Ug. Apply the comparison principle again for the same functions and the background form (dd'^u)"'-^ A /3"-™+2 in place of (dd'^uY"-^ A /3"-"'+^: — / (dd"u)'"-2 /\ pn < [ dd%[u\ + [u]p +g)A (dd^uY'-^ A ^"-™+i < 2™-i / (dd'^w)™-! A ^«-™+i Therefore (dd'^u)"'-^ A /3"-™+2 = on Ug. Repeating the procedure, after m — 2 steps we get /3" = 0, a contradiction. D Acknowledgment The authors were partially supported by NCN grant 2011/01/B/ST1/00879. References [AV] S. Alekser, M. Verbitsky, Quaternionic Monge-Ampere equations and Calabi problem for HKT-manifolds, Israel J. Math. 176 (2010), 109138. [BTl] E. Bedford, B.A. Taylor, The Dirichlet problem for a complex Monge- Ampere equation, Invent. Math. 37 (1976), 1-44. [BT2] E. Bedford, B.A. Taylor, A new capacity for plurisubharmonic functions. Acta Math. 149 (1982), 1-40. [BU] Z. Blocki, Weak solutions to the complex Hessian equation, Ann. Inst. Fourier (Grenoble) 55, 5 (2005), 1735-1756. [B12] Z. Blocki, A gradient estimate in the Calabi- Yau theorem. Math. Ann. 344 (2009), 317-327. [Chi] L. H. Chinh, Solutions to degenerate Hessian equations, preprint arXiv 1202.2436v2. [DK] S. Dinew, S. Kolodziej, A priori estimates for the complex Hessian equa- tions, preprint arXivlll2.3063 [Ga] L. Garding, An inequality for hyperbolic polynomials, J. Math. Mech. 8 (1959) 957-965. [G] P. Guan, On the gradient estimate for Monge- Ampere equation on Kahler manifold. Private notes. [HK] W. K. Hayman, P. B. Kennedy, Subharmonic functions Vol I, London Math. Soc. Monographs, (1976). [Hou] Z. Hon, Complex Hessian equation on Kahler manifold, Int. Math. Res. Not. 16 (2009), 3098-3111. [HMW] Z. Hou, X.-N. Ma, D. Wu, A second order estimate for complex Hessian equations on a compact Kahler manifold. Math. Res. Lett. 17 (2010), 547-561. [J] A. Jbilou, Equations hessiennes complexes sur des varietes kahleriennes, C. R. Math. Acad. Set. Pans 348 (2010), 41-46. [Ko] V. N. Kokarev, Mixed volume forms and a complex equation of Monge- Ampere type on Kahler manifolds of positive curvature, Izv. RAN. Ser. Mat. 74:3 (2010), 65-78. [Li] S.-Y. Li, On the Dirichlet problems for symmetric function equations of the eigenvalues of the complex Hessian, Asian J. Math. 8 (2004), 87-106. [SW] J. Song, B. Weinkove, On the convergence and singularities of the J-flow with applications to the Mabuchi energy, Comm. Pure. Appl. Math. 61 (2008), 210-229. [W] X.-J. Wang, The fc-Hessian equation, Lect. Not. Math. 1977 (2009). [Y] S. T. Yau, On the Ricci curvature of a compact Kahler manifold and the complex Monge-Ampere equation, Comm. Pure Appl. Math. 31 (1978), 339-411. [Z] X. Zhang, A priori estimates for complex Monge-Ampere equation on Hermitian manifolds. Int. Math. Rs. Not. 19 (2010), 381 43836. Rutgers University, Newark, NJ 07102, USA; Faculty of Mathematics and Computer Science, Jagiellonian University 30-348 Krakow, Lojasiewicza 6, Poland; e-mail: slawomir.dinew@im.uj .edu.pl Faculty of Mathematics and Computer Science, Jagiellonian University 30-348 Krakow, Lojasiewicza 6, Poland; e-mail: slawomir . kolodzie j Oim . uj . edu . pi 10