THE HESSIAN OF QUANTIZED DING FUNCTIONALS AND ITS ASYMPTOTIC BEHAVIOR 7 1 RYOSUKETAKAHASHI 0 2 n Abstract. WecomputetheHessianofquantizedDingfunctionalsand a giveanelementaryprooffortheconvexityofquantizedDingfunctionals J alongBergmangeodesicsfromtheviewpointofprojectivegeometry. We 4 study also the asymptotic behavior of the Hessian using the Berezin- 2 Toeplitz quantization. ] G D . 1. Introduction h t a Let X be an n-dimensional Fano manifold and k a large integer such that m kK is very ample. Let ( K ) be the space of smooth fiber metrics φ X X [ o−n K with positive curvHat−ure ω := (√ 1/2π)∂∂¯φ and the space of 2 herm−itiXan forms on H0(X, kKX).φFor a m−etric φ ( KBXk), we denote − ∈ H − v the Monge-Amp`ere volume form by 7 3 ωn φ 0 MA(φ) := , n! 6 0 and the canonical volume form by . 1 2 0 ∂ ∂ e φ := (√ 1)n2 dz dz dz¯ dz¯ . 7 − 1 n 1 n − ∂z ∧···∧ ∂z ∧···∧ ∧ ∧···∧ 1 (cid:12) 1 n(cid:12)φ (cid:12) (cid:12) : v This expression is re(cid:12)adily verified to(cid:12)be independent of the local holomor- (cid:12) (cid:12) i X phic coordinates (z1,...,zn) and hence defines a volume form on X. We normalize e φ to be a probability measure r − a e φ − µ := . φ e φ X − The space ( K ) admits a naturaRl Riemannian metric, called Mabuchi X metric, defiHned−by the L2-norm of a tangent vector f C (X,R) at φ: ∞ ∈ f 2 := f2MA(φ). | |φ X Recall that Ding functionals on the space of infinite dimensional Rie- R D mannian manifold ( K ) is uniquely characterized (modulo an additive X H − 2010 Mathematics Subject Classification. 53C25. Key words and phrases. Dingfunctional, quantization, K¨ahler-Einstein metric. This work was supported by Grant-in-Aidfor JSPS Fellows Number16J01211. 1 2 constant) by the property µ n! φ = , ∇D|φ MA(φ) − ( KX)n − where denotes the gradient and ( K )n is the top intersection number. X ∇ − TheDing functionals are important in the study of Ka¨hler-Einstein metrics. For instance, Donaldson [5] recently gave a “moment map” interpretation of the Ding functional. More precisely, he showed that the ratio of volumes µ /MA(φ) n!/( K )n arises as the moment map for a suitable infinite φ X − − dimensional symplectic manifold and the Ding functional can be viewed as the Kempf-Ness function. This interpretation provides us the direct link between the existence problem of Ka¨hler-Einstein metrics and stability in Geometric Invariant Theory. On the other hand, quantization of the Ding functionals is also studied: given φ ( K ), we define a hermitian form Hilb (φ) by X k k ∈ H − ∈ B s 2 := s 2e kφMA(φ). k kHilbk(φ) | | − ZX Conversely, for a given H , we define a metric FS (H) ( K ) by k k X ∈ B ∈H − 1 s 2 FSk(H) := log k−n sup | | . k H(s,s) s H0( kKX) 0 ! ∈ − \{ } In what follows, we fix some H and a reference H -ONB s := 0 ∈ Bk 0 0 (s1,...,sNk), which defines an embedding ιs0: X ֒→ CPNk−1, where Nk := dimH0( kK ). For g GL(N ,C), let H be a hermitian form such X k g k − ∈ ∈ B that s g is an H -orthonormal basis. Then the map g H gives an isomor0ph·ism GL(Ng,C)/U(N ) , and the tangent spac7→e of g can be k k k k ≃ B B identified with √ 1u(N ). Thus the space admits a natural Riemannian k k − B structure defined by the Killing form tr(AB) (A,B √ 1u(N )) at each k ∈ − tangent space. Write M: CPNk−1 √ 1u(Nk) for the map: → − Z Z α β M([Z ; ;Z ]) := . 1 ··· Nk Z 2 (cid:18) i| i| (cid:19)αβ We define the center of mass M(g) by thePformula M(g) := (M g)µ , ◦ FSk(Hg) ZX where we identify X with ι (X) and the measure µ with its push s0 FSk(Hg) forward (ι ) µ . Quantized Ding functionals (k) on the space of in- s0 ∗ FSk(Hg) D finite dimensional Riemannian manifold ( K ) is uniquely characterized X H − (modulo an additive constant) by the property Id (k) = k 1 M(g) . ∇D |Hg − (cid:18) − Nk(cid:19) 3 There is also a finite dimensional moment-map picture for this setting: the space of all bases (k) GL(N ) is equipped with a Ka¨hler structure in- k Z ≃ duced from the Berndtsson metric and U(N ) acts on (k) isometrically k Z with a moment map which is essentially M(g) Id/N . Critical points of k − the quantized Ding functionals M(g) = Id/N are called anti-canonically k balanced metrics. There is a strong connection between the existence prob- lemofanti-canonically balancedmetricsandKa¨hler-Einsteinmetrics(cf. [1, Theorem 7.1]). In this paper, we study the Hessian of quantized Ding functionals and its asymptotic behavior as raising exponent k . We first give a formula → ∞ for the Hessian of the quantized Ding functional 2 (k): ∇ D Theorem 1.1. The Hessian of the quantized Ding functional is computed by 2 (k) (A,B) = k 1 Re(ξ ,ξ ) µ k 2 H(A)H(B)µ ∇ D |H0 − ZX A B FS FSk(H0)− − ZX FSk(H0) + k 2 H(A)µ H(B)µ , − FSk(H0)· FSk(H0) ZX ZX where ξA denotes the holomorphic vector filed on CPNk−1 corresponding to R A, and H(A) is the Hamiltonian for the Killing vector Jξ , (, ) is the A · · FS Fubini-Study inner product on tangent vectors. As a corollary, we will show the following: Corollary 1.1. We have 2 (k) (A,A) 0 for any A √ 1u(N ), and ∇ D |H0 ≥ ∈ − k theequalityholdsifandonlyifA Lie(Aut(X, kK )), where Aut(X, kK ) X X ∈ − − denotes the group of holomorphic automorphisms of the pair (X, kK ), X embedded into GL(N ,C) by means of the reference basis s . − k 0 Although Corollary 1.1 is a direct consequence of Berndtsson’s convexity theorem [2, Theorem 2.4] (see also [1, Lemma 7.2]), our proof is completely independent, based on the viewpoint of projective geometry, and somewhat elementary. Next, we fix a reference metric φ ( K ) and set H := Hilb (φ ). 0 X 0 k 0 For f C (X,R), we associate with∈thHe −derivative of the Hilbert map in ∞ ∈ the direction f: d Q := Hilb (φ tf) = (kf ∆ f)(s ,s )e kφ0MA(φ ) , f,k dt k 0− |t=0 (cid:18)ZX − φ0 α β − 0 (cid:19)αβ where∆ denotes the(negative) ∂¯-Laplacian with respectto φ , andin the φ0 0 last inequality, we identified Q with a hermitian matrix by means of the f,k reference basis s . With this operation, we can connect 2 (k) to 2 as 0 ∇ D ∇ D follows: Theorem1.2. For any functionsf,g C (X,R), we have the convergence ∞ ∈ of the Hessian (1.1) 2 (k) (Q ,Q ) 2 (f,g) ∇ D |H0 f,k g,k → ∇ D|φ0 4 as k . In particular, lim 2 (k) (Q ,Q ) = 0 implies that the co→ndi∞tion characterizing degk→en∞er∇acyD 2|H0 (ff,,kf) =f,k0 follows. Finally, the above convergence is uniform when ∇f,gDv|φa0ry in a subset of C (X,R) ∞ which is compact for the C -topology. It is also uniform for φ as long as ∞ 0 φ stays a compact set in the C -toplogy. 0 ∞ This is an analogue of Berndtsson’s result [2, Theorem 4.1], but quan- tization schemes are different. Moreover, his argument is based on Hodge theory, whereas an important technical tool we use in our proofs is Berezin- Toeplitz quantization provided by Ma-Marinescu [8]. We should mention as well that J. Fine [7] studied the quantization of the Lichnerowicz operator on general polarized manifolds. Our method follows a strategy discovered by him. Acknowledgements. The author would like to express his gratitude to his advisor Professor Shigetoshi Bando for useful discussions on this arti- cle. This research is supported by Grant-in-Aid for JSPS Fellows Number 16J01211. 2. Foundations 2.1. Functionals on the space of metrics. We have a quick review on several functionals over the space of metrics ( K ) or which play a X k H − B central role in the study of Ka¨hler-Einstein metrics. The standard reference for this section is [1]. We fix a reference metric φ ( K ). We define 0 X ∈ H − the Monge-Amp`ere energy by n 1 (φ) := (φ φ )ωn i ωi , E (n+1)( K )n − 0 φ− ∧ φ0 − X i=1ZX X and the Ding functional by D (φ):= (φ)+ (φ), (φ) := log e φ. − D −E L L − ZX Let h be the Ricci potential of ω : φ φ µ φ h := log . φ MA(φ) A direct computation implies that the derivative of along a smooth curve D φ in ( K ) is t X H − d2 n! dt2D(φt) = − (φ¨t −|∂φ˙t|2φt) ( K )n −ehφt MA(φ) ZX (cid:18) − X (cid:19) 2 + ∂φ˙ 2 µ φ˙2µ + φ˙ µ . | t|φt φt − t φt t φt ZX ZX (cid:18)ZX (cid:19) Hence the Hessian of is D (2.1) 2 (f,g) = Re(∂f,∂g) µ fgµ + fµ gµ . φ φ φ φ φ φ ∇ D| ZX −ZX ZX ·ZX 5 Remark 2.1. We find that the Hessian 2 is non-negative by the modified ∇ D Poincar´e inequality on Fano manifolds (for instance, see [10, Corollary 2.1]). Set H := Hilb (φ ). we also define the quantized Monge-Amp`ere energy 0 k 0 by 1 (k)(H):= logdet(H H 1), E −kN · 0− k and the quantized Ding functional by (k)(H) := (k)(H)+( FS )(H). k D −E L◦ 2.2. Berezin-Toeplitz quantization. The key technical result that we use in the proof of Theorem 1.2 is the asymptotic expansion of Bergman functionandtheirgeneralizations. Forφ ( K ),theBergmanfunction 0 X ρ (φ ): X R is defined by ∈ H − k 0 → ρ (φ ) := s 2e kφ0, k 0 i − | | i X where(s )isaHilb (φ )-ONBofH0( kK ). ThecentralresultforBergman i k 0 X − function concerns the large k asymptotic of ρ (φ ), obtained by Bouche [3], k 0 Catlin [4], Tian [9] and Zelditch [11]: Theorem 2.1. We have the following asymptotic expansion of Bergman function: ρ (φ ) = b kn+b kn 1+b kn 2+ , k 0 0 1 − 2 − ··· where each coefficient b can be written as a polynomial in the Riemannian i curvature Riem(ω ), their derivatives and contractions with respect to ω . φ0 φ0 In particular, 1 b = 1, b = S , 0 1 2 φ0 where S(ω ) is the scalar curvature of ω . The above expansion is uniform φ0 φ0 as long as φ stays in a compact set in the C -topology. More precisely, for 0 ∞ any integer p and l, there exists a constant C such that p,l p ρ (φ ) b kn i < C kn p 1. k 0 i − p,l − − (cid:13) − (cid:13) · (cid:13) Xi=0 (cid:13)Cl (cid:13) (cid:13) We can take the c(cid:13)onstant C indepen(cid:13)dently of φ as long as φ stays in a (cid:13) p,l (cid:13) 0 0 compact set in the C -topology. ∞ AnotherimportanttechnicaltoolinourproofsisprovidedbytheBerezin- Toeplitz quantization [8]. For f C (X,R), the Berezin-Toeplitz operator ∞ ∈ T is a sequence of linear operators f,k T : H0( kK ) H0( kK ) f,k X X − → − 6 defined as two steps: first multiply a given section by f, then project to the space of holomorphic sections using the L2-inner product Hilb (φ ). Using k 0 the Hilb (φ )-ONB (s ), we obtain the explicit description of the kernel: k 0 i K (x,y) = f(z)(s ,s ) (z)s (y) s (x)MA(φ )(z). f,k α β kφ0 β ⊗ ∗α 0 α,β ZX X e If we restrict K to the diagonal, we have f,k Kf,k(x):= Kfe,k(x,x) = f(z)(sα,sβ)kφ0(z)(sβ,sα)kφ0(x)MA(φ0)(z). α,β ZX X e Theorem 2.2 ([8]). We have the following asymptotic expansion: ρ (φ ) = b kn +b kn 1+ k 0 f,0 f,1 − ··· for smooth functions b . Moreover, there are the following formula for f,j coefficients: b = f, f,0 1 b = S f +∆ f. f,1 2 φ0 φ0 The expansion is uniform in f varying in a subset of C (X,R) which is ∞ compact for the C -topology. It is also uniform for φ as long as φ stays ∞ 0 0 a compact set in the C -toplogy. ∞ For f,g C (X,R), we also use the kernel of the composition T T : ∞ f,k g,k ∈ ◦ K (x,y) = K (x,z)K (z,y)MA(φ )(z). f,g,k f,k g,k 0 ZX Restricting the diagonal, we obtain a function e e e K (x) := K (x,x) f,g,k f,g,k = f(y)g(z)(s ,s ) (y)(s ,s ) (z)(s ,s ) (x) e α β kφ0 β γ kφ0 γ α kφ0 αX,β,γZX×X MA(φ )(y) MA(φ )(z). 0 0 ∧ Theorem 2.3 ([8]). We have the following asymptotic expansion: ρ (φ )= b kn +b kn 1+ k 0 f,g,0 f,g,1 − ··· for smooth functions b . Moreover, there are the following formula for f,g,j coefficients: b = fg, f,g,0 1 1 b = S f2+2f∆ f + df 2 . f,f,1 2 φ0 φ0 2| |φ0 The expansion is uniform in f,g varying in a subset of C (X,R) which is ∞ compact for the C -topology. It is also uniform for φ as long as φ stays ∞ 0 0 a compact set in the C -toplogy. ∞ 7 3. Proof of the main theorem 3.1. The second variation formula for (k). Before going to the proof, D we define some notations that we will use later. For a hermitian matrix A = (A ) √ 1u(N ), we write ξ for the corresponding holomorphic i,j k A ∈ − vector field on CPNk−1, i.e., the push forward of i,jAi,jZi∂∂Zj via the standard projection CNk 0 CPNk−1. We set P \{ } → H(A) := tr(AM), then H(A) is a real-valued smooth function satisfying √ 1 i ω = − ∂¯H(A), ξA FS 2π where ω c ( (1)) denotes the Fubini-Study metric. Moreover, if we FS 1 decompose ξ∈ = OξR √ 1JξR, we find that H(A) is the Hamiltonian for R A A − − A Jξ : A 1 iJξARωFS = −4πdH(A). For A∈ √−1u(Nk), we set g(t) := e12tA and s(t) := (s(1t),...,s(Nt)k) = s0·g(t). Lemma 3.1. d 1 (k)(H )= tr(A). g(t) dtE kN k In particular, (k) is affine along Bergman geodesics. E Proof. Since (H (s ,s )) = tg(t) 2, the direct computation shows that g(t) i j − d 1 d (k)(H ) = tr (H (s ,s )) 1 (H (s ,s )) g(t) g(t) i j − g(t) i j dtE −kN dt k (cid:18) (cid:19) 1 dg(t) 1 dg(t) 1 = tr g(t)2 − g(t) 1 +g(t) 1 − − − −kN · dt · · dt k (cid:18) (cid:18) (cid:19)(cid:19) 1 1 1 = tr g(t) A g(t) 1+ A − kN · 2 · 2 k (cid:18) (cid:19) 1 = tr(A). kN k (cid:3) Lemma 3.2. (1) we have d (FS (H )) = k 1M(g(t)). k g(t) − dtL (2) The second variation formula is d2 (FS (H )) = k 1 ξ 2 µ k 2 H(A)2µ dt2L k g(t) |t=0 − ZX| A|FS FSk(H0)− − ZX FSk(H0) 2 + k 2 H(A)µ . − FSk(H0) (cid:18)ZX (cid:19) 8 Proof. (1) Direct computation shows that 1 d (FSk(Hg(t))) = e−FSk(Hg(t)) − d FSk(Hg(t)) e−FSk(Hg(t)) dtL − · −dt · (cid:18)ZX (cid:19) ZX(cid:18) (cid:19) d = FS (H )µ , dt k g(t) FSk(Hg(t)) ZX and d 1 d FS (H ) = k 1 s(t) 2 dt k g(t) − · s(t) 2 · dt | i | i| i | Xi (3.1) = k 1tr((M g(t)) A). − P ◦ · Thus we have d (FS (H )) = k 1 tr((M g(t)) A)µ dtL k g(t) − ◦ · FSk(Hg(t)) ZX = k 1M(g(t)). − (2) We compute d2 d (FS (H )) = k 1 tr((M g(t)) A) µ dt2L k g(t) |t=0 − ZX dt ◦ · |t=0 FSk(H0) d + k 1 H(A) µ . − ZX · dt FSk(H0)|t=0 Since d R tr((M g(t)) A) = tr(dM(ξ ) A) dt ◦ · |t=0 A · R R = 4πω (Jξ ,ξ ) − FS A A = ξ 2 , | A|FS the first term is (3.2) k 1 ξ 2 µ . − | A|FS FSk(H0) ZX On the other hand, using H(A) = tr(AM) and (3.1), we have d d µ = FS (H ) µ µ dt FSk(H0)|t=0 (cid:18)ZX dt k g(t) |t=0 FSk(H0)(cid:19)· FSk(H0) d FS (H ) µ − dt k g(t) |t=0· FSk(H0) = k 1 H(A)µ H(A) µ . − FSk(H0)− FSk(H0) (cid:18)ZX (cid:19) Hence the second term is 2 (3.3) k 2 H(A)2µ +k 2 H(A)µ . − − FSk(H0) − FSk(H0) ZX (cid:18)ZX (cid:19) Combining (3.2) and (3.3) gives our conclusion. (cid:3) 9 When we take into account that the Hessian 2 (k) is a symmetric bi- ∇ D linear form, we can easily get Theorem 1.1 from Lemma 3.1 and Lemma 3.2. Proof of Corollary 1.1. For A √ 1u(N ), let ξ be the component of ∈ − k A⊤ ξ which is tangent to X and ξ the component which is perpendicular. A|X A⊥ Then we have √ 1 iξA⊤ωFSk(H0) = 2−π ∂¯ k−1H(A) on X. It follows that k 1 ξ 2 = ξ 2 (cid:0) = k 2(cid:1)∂H(A)2 . Com- − | A⊤|FS | A⊤|FSk(H0) − | |FSk(H0) bining with the formula ξ 2 = ξ 2 + ξ 2 , we have | A|FS | A⊤|FS | A⊥|FS 2 (k) (A,A) = k 2 2 (H(A),H(A))+k 1 ξ 2 µ ∇ D |H0 − ∇ D|FSk(H0) − ZX| A⊥|FS FSk(H0) k 1 ξ 2 µ (by Remark 2.1) ≥ − | A⊥|FS FSk(H0) ZX 0. ≥ Now we assume that 2 (k) (A,A) = 0, then we have ξ = 0, and ∇ D |H0 A⊥ hence A Lie(Aut(X, kK )) as desired. Conversely, the conditiion A X ∈ − ∈ Lie(Aut(X, kK )) implies that ξ = 0 and ξ is holomorphic. Differenti- − X A⊥ A⊤ ating the equation Ric(ω ) ω = √ 1∂∂¯h with respect FSk(H0) − FSk(H0) 2−π FSk(H0) to ξ , we obtain A⊤ ∆ H(A) (∂h ,∂H(A)) H(A)+ H(A)µ = 0. − FSk(H0) − FSk(H0) FSk(H0)− FSk(H0) ZX MultiplingH(A)andintegratingbyparts,weobtain 2 (H(A),H(A)) = 0. Therefore, we have 2 (k) (A,A) = 0. ∇ D|FSk(H0) (cid:3) ∇ D |H0 3.2. Asymptotic of the Hessian 2 (k). Our starting point is the fol- ∇ D lowing: Lemma3.3([6],Lemma18). ForanyHermitianmatricesA,B √ 1u(N ), k ∈ − we have H(A)H(B)+(ξ ,ξ ) = tr(ABM). A B FS By Lemma 3.3, we have 2 (k)(A,B) = k 1 Re(tr(ABM))µ k 1(1+k 1) H(A)H(B)µ ∇ DH0 − FSk(H0)− − − FSk(H0) | ZX ZX + k 2 H(A)µ H(B)µ . − FSk(H0)· FSk(H0) ZX ZX For given functions f,g C (X,R), we set A := Q , B := Q and ∞ f,k g,k ∈ compute the asymptotic of 2 (k)(Q ,Q ) as k . However, in ∇ DH0 f,k g,k → ∞ | the course of the proof, we find that 2 (k)(Q ,Q ) has an asymptotic ∇ DH0 f,k g,k expansion whosecoefficients are also symm|etric and bilinear with respect to 10 f and g. Hence it follows that we may assume f = g to prove Theorem 1.2. We set A := k 1 tr(Q2 M)µ , 1 − ZX f,k Tk(φ0) A := k 1(1+k 1) H(Q )2µ , 2 − − − ZX f,k Tk(φ0) 2 A := k 2 H(Q )µ , 3 − (cid:18)ZX f,k Tk(φ0)(cid:19) and we will compute these terms separately, where := FS Hilb . The k k k T ◦ following arguments are based on [7, Section 2]. Lemma 3.4. The volume form e−Tk(φ0) has the asymptotic expansion e−Tk(φ0) = (1+O(k−2))e−φ0. Proof. Since (φ ) = φ +k 1log(k nρ (φ )), we have k 0 0 − − k 0 T e−Tk(φ0) = (k−nρk(φ0))−1/ke−φ = (1+O(k 2))e φ, − − where we used the asymptotic expansion of ρ (φ ) in the last equality (cf. k 0 Theorem 2.1). (cid:3) Lemma 3.5. The term A has an asymptotic expansion 1 A = f2µ k+ ∂f 2 µ +O(k 1). 1 φ0 · | |φ0 φ0 − ZX ZX Proof. We can write M as X | (s ,s ) M = α β kφ0 . |X (cid:18) ρk(φ0) (cid:19)αβ It follows that A = k 1 Q Q M , 1 − αβ βγ γα α,β,γ X where Q = (kf ∆ f)(s ,s ) MA(φ ), αβ − φ0 α β kφ0 0 ZX (s ,s ) M = γ α kφ0µ . γα ZX ρk(φ0) Tk(φ0) By Theorem 2.1 and Lemma 3.4, we have e−Tk(φ0) = 1+O(k−2) e φ0 − ρ (φ ) b kn+b kn 1+O(kn 2) k 0 0 1 − − = (k n b k n 1+O(k n 2))e φ0. − 1 − − − − − −