ScalingStacks

4.5. Proof of the Liouville theorem [03I0]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

4.5. Proof of the Liouville theorem

With the above technical preparations, we complete the proof of the main result in this section, Theorem 4.3. We need the following lemma which is an immediate corollary of the Bochner formula and the maximum principle.

Lemma 4.17.

Let (Mn,g,p)(M^{n},g,p) be a complete non-compact manifold with Ricg≥0\Ric_{g}\geq 0. Let ω\omega be a harmonic 11-form on (Mn,g)(M^{n},g), i.e., ΔH​ω=0\Delta_{H}\omega=0 and assume that

(4.168) limdg​(x,p)→+∞|ω⁡(x)|=0,\lim\limits_{d_{g}(x,p)\to+\infty}|\omega(x)|=0,

then ω≡0\omega\equiv 0 on MnM^{n}.

Proof.

Since ω\omega is harmonic, by Bochner’s formula,

(4.169) 12​Δg​|ω|2=|∇ω|2+Ricg⁡(ω,ω)≥0,\frac{1}{2}\Delta_{g}|\omega|^{2}=|\nabla\omega|^{2}+\Ric_{g}(\omega,\omega)\geq 0,

then |ω|2|\omega|^{2} is subharmonic. Given the asymptotic property (4.168), applying the maximum principle to the above subharmonic function |ω|2|\omega|^{2}, we have ω≡0\omega\equiv 0 on MnM^{n}.

∎

Proof of Theorem 4.3.

Let uu satisfy Δg​u=0\Delta_{g}u=0 (X4,g)(X^{4},g). We also assume that uu satisfies the asymptotic behavior

(4.170) u=O⁡(eℓ0​z)u=O(e^{\ell_{0}z})

for some ℓ0∈(0,1)\ell_{0}\in(0,1). The main part of the proof is to determine a positive number ℓ0>0\ell_{0}>0 such that if (4.170) holds, then uu has at most linear growth at infinity, which enables us to apply Lemma 4.17.

By assumption, there is a diffeomorphism

(4.171) Φ:X4∖K⟶[102,+∞)×Y3\Phi:X^{4}\setminus K\longrightarrow[10^{2},+\infty)\times Y^{3}

such that for all k≥0k\geq 0

(4.172) ‖g−Φ∗​g𝒞‖Ck≤C​e−δ​z.\|g-\Phi^{*}g_{\mathcal{C}}\|_{C^{k}}\leq Ce^{-\delta z}.

To obtain an accurate growth order of uu, we will study the equation of uu in terms of the metric g𝒞g_{\mathcal{C}} on the model space [102,+∞)×Y3[10^{2},+\infty)\times Y^{3}.

First, we will show that a harmonic function on (X4,g)(X^{4},g) with exponential growth is well behaved in terms of the model metric g𝒞g_{\mathcal{C}} near infinity. Preciesly, we will prove the following claim.

Claim 4.18.

Assume that (X4,g)(X^{4},g) is δ\delta-asymptotically Calabi. Let δ^∈(0,δ/10)\hat{\delta}\in(0,\delta/10) such that uu satisfies

(4.173) Δg​u=0u=O⁡(eδ^​z),\displaystyle\begin{split}\Delta_{g}u&=0\\ u&=O(e^{\hat{\delta}z}),\end{split}

then for every fixed k∈ℤ+k\in\mathbb{Z}_{+}, let 𝐱0∈[T0(k),+∞)×Y3\bm{x}_{0}\in[T_{0}(k),+\infty)\times Y^{3} with T0​(k)≥100k3>0T_{0}(k)\geq 100^{k^{3}}>0 and denote z0≡z⁡(𝐱0)z_{0}\equiv z(\bm{x}_{0}), we have

(4.174) ‖∇kΔg𝒞​u​(𝒙0)‖≤C⁡(k,g)⋅e−δ​z02.\|\nabla^{k}\Delta_{g_{\mathcal{C}}}u(\bm{x}_{0})\|\leq C(k,g)\cdot e^{-\frac{\delta z_{0}}{2}}.
Proof.

Denoting ϕ≡(Δg−Δg𝒞)​u\phi\equiv(\Delta_{g}-\Delta_{g_{\mathcal{C}}})u, then Δg​u=0\Delta_{g}u=0 implies

(4.175) Δg𝒞​u+ϕ=0.\Delta_{g_{\mathcal{C}}}u+\phi=0.

We will show that for each k∈ℕk\in\mathbb{N} we have

(4.176) ‖∇kϕ​(𝒙0)‖≤C⁡(k,g)⋅e−δ​z02,\|\nabla^{k}\phi(\bm{x}_{0})\|\leq C(k,g)\cdot e^{-\frac{\delta z_{0}}{2}},

where C⁡(k,g)>0C(k,g)>0 depends only on k∈ℕk\in\mathbb{N} and the curvature bound of the cutoff region (X4∖K,g)(X^{4}\setminus K,g).

The higher order derivative estimate will be proved by the Wk,pW^{k,p}-estimate for harmonic functions on the complete space (X4,g)(X^{4},g). Since the metric gg is collapsing near the infinity, the standard elliptic estimate cannot be directly applied. To overcome this difficulty, we will scale up the metric g~=λ2​g\tilde{g}=\lambda^{2}g such that B1​(𝒙𝟎)B_{1}(\bm{x_{0}}) is non-collapsing for g~\tilde{g} which guarantees the elliptic estimate holds in terms of the rescaled metric g~\tilde{g}. For fixed 𝒙0∈[T0(k),+∞)\bm{x}_{0}\in[T_{0}(k),+\infty), we take

(4.177) λ=z012\lambda=z_{0}^{\frac{1}{2}}

and hence there is some constant v0>0v_{0}>0 which is independent of the zz-coordinate such that

(4.178) Volg~⁡(B1​(𝒙𝟎))≥v0>0.\Vol_{\tilde{g}}(B_{1}(\bm{x_{0}}))\geq v_{0}>0.

By explicit calculation on the model space 𝒞\mathcal{C} using (4.4) one easily sees that curvatures are uniformly bounded in a ball of definite size of radius, i.e.

(4.179) supB2​(𝒙𝟎)‖Rm‖g~≤Λ0,\sup\limits_{B_{2}(\bm{x_{0}})}\|\Rm\|_{\tilde{g}}\leq\Lambda_{0},

where Λ0>0\Lambda_{0}>0 is independent of the zz-coordinate. It follows that for every k∈ℕk\in\mathbb{N} and 1<p<∞1<p<\infty, there exists C⁡(k,v0,Λ0,p)>0C(k,v_{0},\Lambda_{0},p)>0 such that under the rescaled metric g~\tilde{g},

(4.180) ‖u‖Wg~k+2,p​(B1​(x0))≤C​‖u‖Wg~k,p​(B1+1k2​(x0)),\|u\|_{W_{\tilde{g}}^{k+2,p}(B_{1}(x_{0}))}\leq C\|u\|_{W_{\tilde{g}}^{k,p}(B_{1+\frac{1}{k^{2}}}(x_{0}))},

which implies that for every k∈ℤ+k\in\mathbb{Z}_{+},

(4.181) ‖u‖Wg~k,p​(B1​(x0))≤C​supB3​(𝒙0)|u|.\|u\|_{W_{\tilde{g}}^{k,p}(B_{1}(x_{0}))}\leq C\sup\limits_{B_{3}(\bm{x}_{0})}|u|.

Therefore, for every k∈ℤ+k\in\mathbb{Z}_{+} and sufficiently large p∈(1,∞)p\in(1,\infty), applying the Sobolev embedding on (B4/3​(𝒙0),g~)(B_{4/3}(\bm{x}_{0}),\tilde{g}), there exists C⁡(k,p,v0,Λ0)>0C(k,p,v_{0},\Lambda_{0})>0 such that

(4.182) supB1​(𝒙0)|∇ku|g~≤C​‖∇k+1u‖Lp​(B4/3​(x0)).\sup\limits_{B_{1}(\bm{x}_{0})}|\nabla^{k}u|_{\tilde{g}}\leq C\|\nabla^{k+1}u\|_{L^{p}(B_{4/3}(x_{0}))}.

By (4.181) and the growth assumption on uu, there is some constant C>0C>0 such that

(4.183) supB1​(𝒙0)|∇ku|g~≤C​supB2​(𝒙0)|u|≤C​eδ^​z0.\sup\limits_{B_{1}(\bm{x}_{0})}|\nabla^{k}u|_{\tilde{g}}\leq C\sup\limits_{B_{2}(\bm{x}_{0})}|u|\leq Ce^{\hat{\delta}z_{0}}.

In terms of the original metric gg, we have

(4.184) |∇ku​(𝒙0)|g≤supB1/λ​(𝒙0)|∇ku|g≤C⋅z02​k​eδ^​z0<C​eδ′​z0,|\nabla^{k}u(\bm{x}_{0})|_{g}\leq\sup\limits_{B_{1/\lambda}(\bm{x}_{0})}|\nabla^{k}u|_{g}\leq C\cdot z_{0}^{2k}e^{\hat{\delta}z_{0}}<Ce^{\delta^{\prime}z_{0}},

where δ′∈(δ^,(1+10−3)​δ^)\delta^{\prime}\in\Big(\hat{\delta},(1+10^{-3})\hat{\delta}\Big).

Next, by (4.172), there is some constant δ>0\delta>0 such that

(4.185) ‖Φ∗​g𝒞−g‖Ck​(B2​(𝒙0))≤Ck​e−δ​z.\|\Phi^{*}g_{\mathcal{C}}-g\|_{C^{k}(B_{2}(\bm{x}_{0}))}\leq C_{k}e^{-\delta z}.

then the elliptic estimate (4.184) and (4.185) imply that

(4.186) |ϕ⁡(𝒙0)|=|(Δg−Δg𝒞)​u​(𝒙0)|≤C​e−δ⋅z02|\phi(\bm{x}_{0})|=|(\Delta_{g}-\Delta_{g_{\mathcal{C}}})u(\bm{x}_{0})|\leq Ce^{-\frac{\delta\cdot z_{0}}{2}}

and similarly

(4.187) |∇kϕ​(𝒙0)|≤Ck​e−δ⋅z02.|\nabla^{k}\phi(\bm{x}_{0})|\leq C_{k}e^{-\frac{\delta\cdot z_{0}}{2}}.

∎

The above error estimate enables us to construct a harmonic function with respect to the model metric g𝒞g_{\mathcal{C}} on [102,+∞)×Y3[10^{2},+\infty)\times Y^{3} which has at most linear growth and is exponentially close to the original function uu. Let

(4.188) ℓ0∈(0,min⁡{δ102,δ¯}),\ell_{0}\in\Big(0,\min\{\frac{\delta}{10^{2}},\underline{\delta}\}\Big),

where δ¯>0\underline{\delta}>0 is the constant in Proposition 4.10. By assumption the harmonic function uu satisfies the asymptotic behavior,

(4.189) u=O⁡(eℓ0​z).u=O(e^{\ell_{0}z}).

Then applying the above claim and Proposition 4.15 on [T0,+∞)×Y3[T_{0},+\infty)\times Y^{3}, there exists a solution to the equation

(4.190) Δg𝒞​v=ϕ\Delta_{g_{\mathcal{C}}}v=\phi

such that

(4.191) v=O⁡(e−ℓ​z)v=O(e^{-\ell z})

for some ℓ∈(−δ/2,0)\ell\in(-\delta/2,0). Therefore, combine (4.175) and (4.190), we have

(4.192) 0=Δg​(u)=Δg𝒞​(u+v),\displaystyle 0=\Delta_{g}(u)=\Delta_{g_{\mathcal{C}}}(u+v),

and u+v=O⁡(eℓ0​z)u+v=O(e^{\ell_{0}z}). Since 0<ℓ0<10<\ell_{0}<1 has been specified in (4.188), now we are in a position to apply Proposition 4.10 to u+vu+v, which shows that

(4.193) (u+v)=a​z+b+O⁡(e−δ¯​z),(u+v)=az+b+O(e^{-\underline{\delta}z}),

and hence in the non-compact part [T0,+∞)×Y3[T_{0},+\infty)\times Y^{3},

(4.194) u=a​z+b+O⁡(e−δ′′​z),δ′′≡min⁡{ℓ,δ¯}.\displaystyle u=az+b+O(e^{-\delta^{\prime\prime}z}),\ \delta^{\prime\prime}\equiv\min\{\ell,\underline{\delta}\}.

The above asymptotics immediately implies that

(4.195) |d​u|g→0​as​z→∞.|du|_{g}\to 0\ \text{as}\ z\to\infty.

Let ΔH\Delta_{H} be the Hodge-Laplacian on (X4,g)(X^{4},g). Since Δg​u=0\Delta_{g}u=0, it holds that

(4.196) ΔH​(d​u)=d​d∗​(d​u)=−d​Δg​u=0.\Delta_{H}(du)=dd^{*}(du)=-d\Delta_{g}u=0.

Since the complete space (X4,g)(X^{4},g) satisfies Ricg≥0\Ric_{g}\geq 0, and |d​u||du| satisfies the decay property (4.195), applying Lemma 4.17 implies that

(4.197) |d​u|g≡0​on​X4.|du|_{g}\equiv 0\ \text{on}\ X^{4}.

Therefore, uu has to be a constant. ∎

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.