ScalingStacks

4.2. Uniform estimates for the fundamental solutions [03H8]

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

4.2. Uniform estimates for the fundamental solutions

A crucial step in applying the method of separation of variables is to prove the C0C^{0}-regularity of a formal solution obtained from the above separation of variables. Specifically, in our context, to prove such a C0C^{0}-regularity result, first we need to obtain some effective estimates for the fundamental solutions to the linear differential equation (see (4.28))

(4.35) d2​u​(z)d​z2−(j2​z2+λ)​u​(z)=0,\frac{d^{2}u(z)}{dz^{2}}-(j^{2}z^{2}+\lambda)u(z)=0,

which arises from the harmonic functions on the Calabi manifold (𝒞,g𝒞)(\mathcal{C},g_{\mathcal{C}}). In our context, we always require

(4.36) j≥0,h≥0,z>1.j\geq 0,\ h\geq 0,\ z>1.

There are two different cases to analyze.

The first case is much simpler, i.e. j=0j=0 and the ODE becomes

(4.37) d2​u​(z)d​z2=λ⋅u⁡(z).\frac{d^{2}u(z)}{dz^{2}}=\lambda\cdot u(z).

Further, if λ=0\lambda=0, the solutions to (4.37) are linear. If λ>0\lambda>0, the above equation has two linearly independent solutions eλ​ze^{\sqrt{\lambda}z} and e−λ​ze^{-\sqrt{\lambda}z}. All the required estimates in this case are standard and straightforward. Geometrically, the ODE analysis for (4.37) arises naturally from the flat cylindrical geometry and corresponding gluing constructions.

So in our case, we only focus on the case j∈ℤ+j\in\mathbb{Z}_{+} which is substantially much more technically involved. In the case j∈ℤ+j\in\mathbb{Z}_{+}, we have already shown in Section 4.1 that λ\lambda and jj satisfy the relation

(4.38) λ≥j≥1.\lambda\geq j\geq 1.

Hence for each pair of λ\lambda and jj satisfying the above, we choose h≥0h\geq 0 such that

(4.39) λ=(2​h+1)​j.\lambda=(2h+1)j.

From now on, we focus on the differential equation for every j∈ℤ+j\in\mathbb{Z}_{+} and h≥0h\geq 0,

(4.40) d2​u​(z)d​z2=j⁡(j​z2+2​h+1)​u​(z).\frac{d^{2}u(z)}{dz^{2}}=j(jz^{2}+2h+1)u(z).

We will simplify the above equation by the following transformations. Let

(4.41) y=j​z,V⁡(y)=u⁡(yj),y=\sqrt{j}z,\ V(y)=u\Big(\frac{y}{\sqrt{j}}\Big),

then V⁡(y)V(y) satisfies

(4.42) d2​V​(y)d​y2=(y2+(2​h+1))​V​(y).\frac{d^{2}V(y)}{dy^{2}}=(y^{2}+(2h+1))V(y).

Further, we make the transformation

(4.43) V⁡(y)=e−y22​Q​(y),V(y)=e^{-\frac{y^{2}}{2}}Q(y),

then QQ sovles the differential equation

(4.44) d2​Q​(y)d​y2−2​y​d​Q​(y)d​y−2​(h+1)​Q​(y)=0.\frac{d^{2}Q(y)}{dy^{2}}-2y\frac{dQ(y)}{dy}-2(h+1)Q(y)=0.

Notice that equation (4.44) is invariant under the change of variables y↦−yy\mapsto-y. Given y>1y>1 and h≥0h\geq 0, we define the following exponential integral

(4.45) H−h−1​(y)≡∫0∞e−t2−2​t​y​th​𝑑t.H_{-h-1}(y)\equiv\int_{0}^{\infty}e^{-t^{2}-2ty}t^{h}dt.

Straightforward calculations show that for each given h≥0h\geq 0, the functions H−h−1​(y)H_{-h-1}(y) and H−h−1​(−y)H_{-h-1}(-y) are linearly independent solutions to (4.44). In fact, the above solutions coincide with the usual Hermite functions up to a constant (see [Leb72] for more details). Eventually, we obtain two solutions to (4.40),

(4.46) ℱ⁡(z)=e−j​z22​H−h−1​(−j​z)=e−y22​∫0∞e−t2+2​t​y+h​log⁡t​𝑑t\mathcal{F}(z)=e^{-\frac{jz^{2}}{2}}H_{-h-1}(-\sqrt{j}z)=e^{-\frac{y^{2}}{2}}\int_{0}^{\infty}e^{-t^{2}+2ty+h\log t}dt

and

(4.47) 𝒰⁡(z)=e−j​z22​H−h−1​(j​z)=e−y22​∫0∞e−t2−2​t​y+h​log⁡t​𝑑t.\mathcal{U}(z)=e^{-\frac{jz^{2}}{2}}H_{-h-1}(\sqrt{j}z)=e^{-\frac{y^{2}}{2}}\int_{0}^{\infty}e^{-t^{2}-2ty+h\log t}dt.

The lemma below shows that ℱ\mathcal{F} and 𝒰\mathcal{U} are two linearly independent solutions.

Lemma 4.5.

The Wronskian is a constant given by

(4.48) 𝒲⁡(ℱ,𝒰)=2−h​j​π​Γ​(h+1)>0.\mathcal{W}(\mathcal{F},\mathcal{U})=2^{-h}\sqrt{j\pi}\Gamma(h+1)>0.

In particular, ℱ\mathcal{F} and 𝒰\mathcal{U} are linearly independent.

Proof.

First observe that 𝒲⁡(ℱ⁡(z),𝒰⁡(z))=ℱ′​(z)​𝒰​(z)−ℱ⁡(z)​𝒰′​(z)\mathcal{W}(\mathcal{F}(z),\mathcal{U}(z))=\mathcal{F}^{\prime}(z)\mathcal{U}(z)-\mathcal{F}(z)\mathcal{U}^{\prime}(z) is a constant. In fact,

dd​z​𝒲​(ℱ⁡(z),𝒰⁡(z))\displaystyle\frac{d}{dz}\mathcal{W}(\mathcal{F}(z),\mathcal{U}(z)) =(ℱ′′​(z)​𝒰​(z)+ℱ′​(z)​𝒰′​(z))−(ℱ⁡(z)​𝒰′′​(z)+ℱ′​(z)​𝒰′​(z))\displaystyle=\Big(\mathcal{F}^{\prime\prime}(z)\mathcal{U}(z)+\mathcal{F}^{\prime}(z)\mathcal{U}^{\prime}(z)\Big)-\Big(\mathcal{F}(z)\mathcal{U}^{\prime\prime}(z)+\mathcal{F}^{\prime}(z)\mathcal{U}^{\prime}(z)\Big)
(4.49) =j⁡(j​z2+2​h+1)​(ℱ⁡(z)​𝒰​(z)−ℱ⁡(z)​𝒰​(z))=0.\displaystyle=j(jz^{2}+2h+1)(\mathcal{F}(z)\mathcal{U}(z)-\mathcal{F}(z)\mathcal{U}(z))=0.

Hence 𝒲⁡(ℱ⁡(z),𝒰⁡(z))\mathcal{W}(\mathcal{F}(z),\mathcal{U}(z)) has to be a constant. Now we evaluate it at z=0z=0, we get

𝒲⁡(ℱ,𝒰)=2​ℱ′​(0)​𝒰​(0).\mathcal{W}(\mathcal{F},\mathcal{U})=2\mathcal{F}^{\prime}(0)\mathcal{U}(0).

Now

ℱ′​(0)=2​j​∫0∞e−t2​th+1​𝑑t=j​Γ​(h2+1),\mathcal{F}^{\prime}(0)=2\sqrt{j}\int_{0}^{\infty}e^{-t^{2}}t^{h+1}dt=\sqrt{j}\Gamma(\frac{h}{2}+1),

and similarly

𝒰⁡(0)=12​Γ​(h2+12).\mathcal{U}(0)=\frac{1}{2}\Gamma(\frac{h}{2}+\frac{1}{2}).

Applying Legendre duplication formula

(4.50) Γ⁡(t)​Γ​(t+12)Γ⁡(2​t)=π22​t−1,t>0,\frac{\Gamma(t)\Gamma(t+\frac{1}{2})}{\Gamma(2t)}=\frac{\sqrt{\pi}}{2^{2t-1}},\ t>0,

we have

𝒲⁡(ℱ,𝒰)=2−h​j​π​Γ​(h+1)>0.\mathcal{W}(\mathcal{F},\mathcal{U})=2^{-h}\sqrt{j\pi}\Gamma(h+1)>0.

∎

The regularity of the formal solutions obtained from the above separation of variables requires very precise uniform estimates for the fundamental solutions ℱ\mathcal{F} and 𝒰\mathcal{U}. We will use the Laplace Method, which is inspired by [SS16] in a different context. Again we denote y=j​zy=\sqrt{j}z and define

(4.51) F⁡(t)≡−t2+2​t​y+h​log⁡tU⁡(t)≡−t2−2​t​y+h​log⁡t.\displaystyle\begin{split}F(t)&\equiv-t^{2}+2ty+h\log t\\ U(t)&\equiv-t^{2}-2ty+h\log t.\end{split}

Straightforward computations tell us that both FF and UU are strictly concave when h≥0h\geq 0. For fixed yy, let t0t_{0} and s0s_{0} be the unique (positive) critical points of FF and UU respectively. It is straightforward that

(4.52) t0=y2+h22+y24s0=−y2+h22+y24.\displaystyle\begin{split}t_{0}&=\frac{y}{2}+\sqrt{\frac{h^{2}}{2}+\frac{y^{2}}{4}}\\ s_{0}&=-\frac{y}{2}+\sqrt{\frac{h^{2}}{2}+\frac{y^{2}}{4}}.\end{split}
Lemma 4.6.

The following uniform estimates hold for all j∈ℤ+j\in\mathbb{Z}_{+} and h≥0h\geq 0,

(4.53) ℱ⁡(z)≤(1+π)​e−j​z22+F​(t0​(z)),\mathcal{F}(z)\leq(1+\sqrt{\pi})e^{-\frac{jz^{2}}{2}+F(t_{0}(z))},
(4.54) 𝒰⁡(z)≤(1+π)​e−j​z22+U​(s0​(z)).\mathcal{U}(z)\leq(1+\sqrt{\pi})e^{-\frac{jz^{2}}{2}+U(s_{0}(z))}.
Proof.

By the definition of ℱ\mathcal{F} and 𝒰\mathcal{U}, it suffices to prove

(4.55) ∫0∞eF⁡(t)​𝑑t≤(1+π)​eF⁡(t0),\int_{0}^{\infty}e^{F(t)}dt\leq(1+\sqrt{\pi})e^{F(t_{0})},

and

(4.56) ∫0∞eU⁡(t)​𝑑t≤(1+π)​eU⁡(s0).\int_{0}^{\infty}e^{U(t)}dt\leq(1+\sqrt{\pi})e^{U(s_{0})}.

We only prove the first inequality and the second can be proved in exactly the same way. In fact, the second can be proved exactly the same way. Denote a≡2​ya\equiv 2y. For ϵ∈(−1,1]\epsilon\in(-1,1] we have

F⁡(t0​(1+ϵ))−F⁡(t0)\displaystyle F(t_{0}(1+\epsilon))-F(t_{0}) =−ϵ⁡(ϵ+2)​t02+ϵ​a​t0+h​log⁡(1+ϵ)\displaystyle=-\epsilon(\epsilon+2)t_{0}^{2}+\epsilon at_{0}+h\log(1+\epsilon)
=−ϵ2​t02+h⁡(log⁡(1+ϵ)−ϵ)\displaystyle=-\epsilon^{2}t_{0}^{2}+h(\log(1+\epsilon)-\epsilon)
(4.57) ≤−ϵ2​t02−h⁡(ϵ22−ϵ33)≤−ϵ2​(t02+h6).\displaystyle\leq-\epsilon^{2}t_{0}^{2}-h(\frac{\epsilon^{2}}{2}-\frac{\epsilon^{3}}{3})\leq-\epsilon^{2}(t_{0}^{2}+\frac{h}{6}).

The above computations imply that under the transformation t=t0​(1+ϵ)t=t_{0}(1+\epsilon),

∫02​t0exp⁡(F⁡(t))​𝑑t\displaystyle\int_{0}^{2t_{0}}\exp(F(t))dt ≤∫02​t0exp⁡(F⁡(t0)−ϵ2​(t02+h6))​𝑑t\displaystyle\leq\int_{0}^{2t_{0}}\exp\Big(F(t_{0})-\epsilon^{2}(t_{0}^{2}+\frac{h}{6})\Big)dt
=t0​exp⁡(F⁡(t0))​∫−11exp⁡(−(t02+h6)​ϵ2)​𝑑ϵ\displaystyle=t_{0}\exp(F(t_{0}))\int_{-1}^{1}\exp\Big(-(t_{0}^{2}+\frac{h}{6})\epsilon^{2}\Big)d\epsilon
=t0t02+h6​exp⁡(F⁡(t0))​∫−t02+h6t02+h6exp⁡(−τ2)​𝑑τ\displaystyle=\frac{t_{0}}{\sqrt{t_{0}^{2}+\frac{h}{6}}}\exp(F(t_{0}))\int_{-\sqrt{t_{0}^{2}+\frac{h}{6}}}^{\sqrt{t_{0}^{2}+\frac{h}{6}}}\exp(-\tau^{2})d\tau
(4.58) ≤π​exp⁡(F⁡(t0)).\displaystyle\leq\sqrt{\pi}\exp(F(t_{0})).

In addition, let t>2​t0t>2t_{0}, then

(4.59) F⁡(t)−F⁡(2​t0)≤F′​(2​t0)​(t−2​t0),\displaystyle F(t)-F(2t_{0})\leq F^{\prime}(2t_{0})(t-2t_{0}),

and hence

(4.60) ∫2​t0∞exp⁡(F⁡(t))​𝑑t≤exp⁡(F⁡(2​t0))​∫2​t0∞exp⁡(F′​(2​t0)​(t−2​t0))​𝑑t=exp⁡(F⁡(2​t0))−F′​(2​t0).\displaystyle\int_{2t_{0}}^{\infty}\exp(F(t))dt\leq\exp(F(2t_{0}))\int_{2t_{0}}^{\infty}\exp\Big(F^{\prime}(2t_{0})(t-2t_{0})\Big)dt=\frac{\exp(F(2t_{0}))}{-F^{\prime}(2t_{0})}.

It can be directly computed that

(4.61) F′​(2​t0)=−4​t0+a+h2​t0=−4​t02+h2​t0<0,F^{\prime}(2t_{0})=-4t_{0}+a+\frac{h}{2t_{0}}=-\frac{4t_{0}^{2}+h}{2t_{0}}<0,

then

(4.62) ∫2​t0∞exp⁡(F⁡(t))​𝑑t≤2​t04​t02+h​exp⁡(F⁡(2​t0))≤2​t04​t02+h⋅exp⁡(F⁡(t0))exp⁡(t02+h6)≤exp⁡(F⁡(t0)).\displaystyle\int_{2t_{0}}^{\infty}\exp(F(t))dt\leq\frac{2t_{0}}{4t_{0}^{2}+h}\exp(F(2t_{0}))\leq\frac{2t_{0}}{4t_{0}^{2}+h}\cdot\frac{\exp(F(t_{0}))}{\exp(t_{0}^{2}+\frac{h}{6})}\leq\exp(F(t_{0})).

Combining the above calculations,

(4.63) ∫0∞exp⁡(F⁡(t))​𝑑t≤(1+π)​exp⁡(F⁡(t0)).\int_{0}^{\infty}\exp(F(t))dt\leq(1+\sqrt{\pi})\exp(F(t_{0})).

∎

Apply the same method as in Lemma 4.6, we have the following asymptotic property of ℱ\mathcal{F} and 𝒰\mathcal{U}.

Lemma 4.7.

For fixed j∈ℤ+j\in\mathbb{Z}_{+} and h>0h>0, we have the following asymptotic formula

(4.64) limz→+∞ℱ⁡(z)π​ej​z22​(j​z)h=1\lim\limits_{z\to+\infty}\frac{\mathcal{F}(z)}{\sqrt{\pi}e^{\frac{jz^{2}}{2}}(\sqrt{j}z)^{h}}=1

and

(4.65) limz→+∞𝒰⁡(z)e−j​z22​Γ​(h+1)​(2​j​z)−h−1=1.\lim\limits_{z\to+\infty}\frac{\mathcal{U}(z)}{e^{-\frac{jz^{2}}{2}}\Gamma(h+1)(2\sqrt{j}z)^{-h-1}}=1.
Proof.

For simplicity, we will calculate the asymptotic behavior in yy. For fixed hh and jj, as y→∞y\rightarrow\infty, it is straightforward that

(4.66) t0=y+h2​y+O⁡(y−2)s0=h2​y+O⁡(y−2)\begin{split}t_{0}&=y+\frac{h}{2y}+O(y^{-2})\\ s_{0}&=\frac{h}{2y}+O(y^{-2})\end{split}

which implies that

(4.67) F⁡(t0)=y2+h​log⁡y+O⁡(y−1)U⁡(s0)=−h+h​log⁡h−h​log⁡(2​y)+O⁡(y−1).\begin{split}F(t_{0})&=y^{2}+h\log y+O(y^{-1})\\ U(s_{0})&=-h+h\log h-h\log(2y)+O(y^{-1}).\end{split}

First, we prove the asymptotics for ℱ\mathcal{F}. As in the proof of Lemma 4.6, we get

∫0∞eF⁡(t)​𝑑t\displaystyle\int_{0}^{\infty}e^{F(t)}dt =eF⁡(t0)​∫02​t0eF⁡(t)−F⁡(t0)​𝑑t+∫2​t0∞eF⁡(t)​𝑑t\displaystyle=e^{F(t_{0})}\int_{0}^{2t_{0}}e^{F(t)-F(t_{0})}dt+\int_{2t_{0}}^{\infty}e^{F(t)}dt
(4.68) =t0​eF⁡(t0)​∫−11e−t02​ϵ2+h⁡(log⁡(1+ϵ)−ϵ)​𝑑ϵ+∫2​t0∞eF⁡(t)​𝑑t.\displaystyle=t_{0}e^{F(t_{0})}\int_{-1}^{1}e^{-t_{0}^{2}\epsilon^{2}+h(\log(1+\epsilon)-\epsilon)}d\epsilon+\int_{2t_{0}}^{\infty}e^{F(t)}dt.

Notice that

(4.69) limt0→∞t0​∫−11e−t02​ϵ2+h⁡(log⁡(1+ϵ)−ϵ)​𝑑ϵ=π,\lim_{t_{0}\rightarrow\infty}t_{0}\int_{-1}^{1}e^{-t_{0}^{2}\epsilon^{2}+h(\log(1+\epsilon)-\epsilon)}d\epsilon=\sqrt{\pi},

and

(4.70) limt0→∞e−F⁡(t0)​∫2​t0∞eF⁡(t)​𝑑t=0.\lim_{t_{0}\rightarrow\infty}e^{-F(t_{0})}\int_{2t_{0}}^{\infty}e^{F(t)}dt=0.

Moreover, by (4.66), limy→+∞t0​(y)→∞\lim\limits_{y\to+\infty}t_{0}(y)\to\infty. It follows that

(4.71) limz→∞ℱ⁡(z)π​e−j​z22+F​(t0​(z))=1.\lim\limits_{z\to\infty}\frac{\mathcal{F}(z)}{\sqrt{\pi}e^{-\frac{jz^{2}}{2}+F(t_{0}(z))}}=1.

Combining the above limit and (4.67), the proof of (4.64) is complete.

In the case j∈ℤ+j\in\mathbb{Z}_{+} and h>0h>0, we will prove the asymptotic behavior of 𝒰\mathcal{U} and we write

∫0∞eU⁡(t)​𝑑t\displaystyle\int_{0}^{\infty}e^{U(t)}dt =s0​eU⁡(s0)​∫−1∞e−s02​ϵ2+h⁡(log⁡(1+ϵ)−ϵ)​𝑑ϵ\displaystyle=s_{0}e^{U(s_{0})}\int_{-1}^{\infty}e^{-s_{0}^{2}\epsilon^{2}+h(\log(1+\epsilon)-\epsilon)}d\epsilon
(4.72) =s0​eU⁡(s0)​∫−1∞e−s02​ϵ2⋅(1+ϵ)h​e−h​ϵ​𝑑ϵ.\displaystyle=s_{0}e^{U(s_{0})}\int_{-1}^{\infty}e^{-s_{0}^{2}\epsilon^{2}}\cdot(1+\epsilon)^{h}e^{-h\epsilon}d\epsilon.

We claim that

(4.73) lims0→0∫−1∞(e−s02​ϵ2−1)⋅(1+ϵ)h​e−h​ϵ​𝑑ϵ=0.\lim\limits_{s_{0}\to 0}\int_{-1}^{\infty}(e^{-s_{0}^{2}\epsilon^{2}}-1)\cdot(1+\epsilon)^{h}e^{-h\epsilon}d\epsilon=0.

In fact, it is straightforward that for any s0>0s_{0}>0,

(4.74) −1≤e−s02​ϵ2−1≤0-1\leq e^{-s_{0}^{2}\epsilon^{2}}-1\leq 0

and for any fixed h>0h>0,

(4.75) ∫−1∞(1+ϵ)h​e−h​ϵ​𝑑ϵ<∞.\int_{-1}^{\infty}(1+\epsilon)^{h}e^{-h\epsilon}d\epsilon<\infty.

Applying the dominated convergence theorem,

(4.76) lims0→0∫−1∞(e−s02​ϵ2−1)⋅(1+ϵ)h​e−h​ϵ​𝑑ϵ=0.\lim\limits_{s_{0}\to 0}\int_{-1}^{\infty}(e^{-s_{0}^{2}\epsilon^{2}}-1)\cdot(1+\epsilon)^{h}e^{-h\epsilon}d\epsilon=0.

This completes the proof the the claim. Next, by the definition of the gamma function,

(4.77) ∫−1∞(1+ϵ)h​e−h​ϵ​𝑑ϵ=eh​∫0∞e−h​s​sh​𝑑s=eh​h−h−1​Γ​(h+1).\int_{-1}^{\infty}(1+\epsilon)^{h}e^{-h\epsilon}d\epsilon=e^{h}\int_{0}^{\infty}e^{-hs}s^{h}ds=e^{h}h^{-h-1}\Gamma(h+1).

Therefore,

(4.78) lims0→0∫−1∞e−s02​ϵ2+h⁡(log⁡(1+ϵ)−ϵ)​𝑑ϵ=eh​h−h−1​Γ​(h+1).\lim\limits_{s_{0}\to 0}\int_{-1}^{\infty}e^{-s_{0}^{2}\epsilon^{2}+h(\log(1+\epsilon)-\epsilon)}d\epsilon=e^{h}h^{-h-1}\Gamma(h+1).

Since limy→+∞s0=0\lim\limits_{y\to+\infty}s_{0}=0 and UU yields to the asymptotic property (4.67), eventually we obtain (4.65). ∎

Lemma 4.8.

There is an absolute constant C0>0C_{0}>0 independent of j∈ℤ+j\in\mathbb{Z}_{+} and h≥0h\geq 0 such that the following uniform estimate holds for all z≥1z\geq 1,

(4.79) 0<eF^​(z)+U^​(z)𝒲⁡(z)≤C0,0<\frac{e^{\widehat{F}(z)+\widehat{U}(z)}}{\mathcal{W}(z)}\leq C_{0},

where

(4.80) F^​(z)=−j​z22+F⁡(t0​(z))\widehat{F}(z)=-\frac{jz^{2}}{2}+F(t_{0}(z))

and

(4.81) U^​(z)=−j​z22+U⁡(s0​(z)).\widehat{U}(z)=-\frac{jz^{2}}{2}+U(s_{0}(z)).
Proof.

The proof is based on Lemma 4.6. First, we discuss the case h=0h=0 and j∈ℤ+j\in\mathbb{Z}_{+}. Direct computations give that t0=yt_{0}=y and s0=0s_{0}=0, then by definition we have that F^​(z)=j​z22\widehat{F}(z)=\frac{jz^{2}}{2} and U^​(z)=−j​z22\widehat{U}(z)=-\frac{jz^{2}}{2}. Therefore, (4.79) immediately follows.

Next, we prove the case j∈ℤ+j\in\mathbb{Z}_{+} and h>0h>0. We notice that

(4.82) t0−s0\displaystyle t_{0}-s_{0} =a2\displaystyle=\frac{a}{2}
(4.83) t02+s02\displaystyle t_{0}^{2}+s_{0}^{2} =a24+h\displaystyle=\frac{a^{2}}{4}+h
(4.84) t0​s0\displaystyle t_{0}s_{0} =h2,\displaystyle=\frac{h}{2},

by elementary calculations,

F⁡(t0)+U⁡(s0)\displaystyle F(t_{0})+U(s_{0}) =−(t02+s02)+a⁡(t0−s0)+h​log⁡(t0​s0)\displaystyle=-(t_{0}^{2}+s_{0}^{2})+a(t_{0}-s_{0})+h\log(t_{0}s_{0})
(4.85) =a24−h+h​log⁡(h2)=j​z2−h+h​log⁡(h2).\displaystyle=\frac{a^{2}}{4}-h+h\log(\frac{h}{2})=jz^{2}-h+h\log(\frac{h}{2}).

Immediately we have that

(4.86) eF^​(z)+U^​(z)≤e−h+h​log⁡(h2).e^{\widehat{F}(z)+\widehat{U}(z)}\leq e^{-h+h\log(\frac{h}{2})}.

Combining (4.86) and Lemma 4.5,

(4.87) eF^​(z)+U^​(z)𝒲⁡(z)≤e−h+h​log⁡(h2)j​π​2−h​Γ​(h+1)≤C0.\displaystyle\frac{e^{\widehat{F}(z)+\widehat{U}(z)}}{\mathcal{W}(z)}\leq\frac{e^{-h+h\log(\frac{h}{2})}}{\sqrt{j\pi}2^{-h}\Gamma(h+1)}\leq C_{0}.

This proves the lemma. ∎

A key technical point of this section is to construct a well-behaved solution of the Poisson equation

(4.88) Δ𝒞​u=v\Delta_{\mathcal{C}}u=v

by applying separation of variables and the uniform estimate on the ODE solutions. For this purpose, we need the following monotonicity.

Lemma 4.9.

Let F^​(z)\widehat{F}(z) and U^​(z)\widehat{U}(z) be the function defined in Lemma 4.8, then F^​(z)−η​z\widehat{F}(z)-\eta z is increasing for z>2​ηz>2\eta and U^​(z)+η​z\widehat{U}(z)+\eta z is decreasing for z>2​ηz>2\eta.

Proof.

Let y=j​zy=\sqrt{j}z and a=2​ya=2y, then by definition,

(4.89) F^=−a28−(t0​(a))2+a​t0​(a)+h​log⁡(t0​(a))\widehat{F}=-\frac{a^{2}}{8}-(t_{0}(a))^{2}+at_{0}(a)+h\log(t_{0}(a))

and

(4.90) U^=−a28−(s0​(a))2−a​s0​(a)+h​log⁡(s0​(a)).\widehat{U}=-\frac{a^{2}}{8}-(s_{0}(a))^{2}-as_{0}(a)+h\log(s_{0}(a)).

We show that F^\widehat{F} is increasing in aa and U^\widehat{U} is decreasing in aa. Indeed,

(4.91) d​F^d​a=−a4+t0​(a)+(−2​t0​(a)+a+ht0​(a))​t0′​(a)=h2+a216≥a4.\displaystyle\frac{d\widehat{F}}{da}=-\frac{a}{4}+t_{0}(a)+\Big(-2t_{0}(a)+a+\frac{h}{t_{0}(a)}\Big)t_{0}^{\prime}(a)=\sqrt{\frac{h}{2}+\frac{a^{2}}{16}}\geq\frac{a}{4}.

So the monotonicity of F^​(z)−η​z\widehat{F}(z)-\eta z immediately follows when z>2​ηz>2\eta. Similarly, the monotonicity of U^+η​z\widehat{U}+\eta z follows from the computation

(4.92) d​U^d​a=−a4−s0​(a)+(−2​s0​(a)−a+hs0​(a))​s0′​(a)=−h2+a216≤−a4.\displaystyle\frac{d\widehat{U}}{da}=-\frac{a}{4}-s_{0}(a)+\Big(-2s_{0}(a)-a+\frac{h}{s_{0}(a)}\Big)s_{0}^{\prime}(a)=-\sqrt{\frac{h}{2}+\frac{a^{2}}{16}}\leq-\frac{a}{4}.

∎

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