Quantitative unique continuation for the heat equation with Coulomb potentials

Let us now choose

with the same constant \mathcal C_1 given in Lemma 4. It is easy to check that 0<\lambda_0<T. According to Lemma 4 (where \lambda=\lambda_0), it holds that

Which implies

Notice that

This, along with (?), leads to

Which implies (1) and completes the proof.

Firstly, by Lemma 6 and constructing a sequence of balls chained along a curve, we claim that, for any compact sets K_1 and K_2 with non-empty interior in \Omega, there are constants N_1=N_1(K_1,K_2,k,n)\geq 1 and \alpha_1=\alpha_1(K_1,K_2,k,n)\in(0,1) such that

Indeed, let

and for each B_{r}(x_i) with 1\leq i\leq p, there exists a chain of balls B_{r}(x_i^j), 1\leq j\leq n_i, such that

where 0<r=r(\Omega,K_1,K_2)<1/2 is a constant. By Lemma 6, we get that there are N_i^j= N_i^j(r,k,n)\geq 1 and \theta_i^j=\theta_i^j(r,k,n)\in(0,1) such that

This, together with (Proof), implies that there are N_i=N_i(K_1,K_2,k,n)\geq 1 and \theta_i=\theta_i(K_1,K_2,k,n)\in(0,1) such that

This, along with (Proof) leads to the estimate (Proof).

Secondly, since \partial\Omega is of class C^2, there are a finite set \Lambda\subset \Omega, 0<\delta<1 and a family of positive numbers 0<r_x\le 1, x\in\Lambda, such that

Then we apply Lemma 6 with \Omega\cap B_{(1+2\delta)r_x}(x),x\in\Lambda, and the same arguments as above to get that, when \Gamma is a neighborhood of \partial\Omega and K_3 is a compact set with non-empty interior in \Omega, there are constants N_2=N_2(\Gamma,K_3,k,n)\geq 1 and \alpha_2=\alpha_2(\Gamma,K_3,k,n)\in(0,1) such that

Finally, we derive the desire estimate (2) from the previous two statements with \Omega\subset(\Gamma\cup K_1) and (K_2\cup K_3)\subset\omega.

By the Green formula, we have

which leads to (i).

We now prove (ii). By the integration by parts, it follows that

Meanwhile,

where

Since

it follows from (?) and (?) that

On the other hand, we have by the Green formula,

From (?) and (?), it holds that

Finally, we obtain from (?), (?) and (?) that

Since \Omega is either convex or star-shaped with respect to x_0, it holds that (x-x_0)\cdot\nu_x\geq 0 when x\in\partial\Omega (see for instance [evans, pp. 515]). (Here, \nu_x is the outward normalized vector of \partial\Omega at x.) Consequently, \mathcal{A}\geq 0. By the Cauchy-Schwartz inequality,

which leads to

i.e., (ii) stands.

Recall the well-known Hardy inequality (cf., e.g., [VZ]):

Given \varphi\in H^1_0(\Omega), \lambda>0 and x_0\in \Omega, let g=\varphi e^{-\frac{|x-x_0|^2}{8\lambda}}. Clearly, g\in H_0^1(\Omega) and

By (Proof),

By the integration by parts, we have

The last two inequalities imply (2).

Given x_0\in \Omega, \lambda>0 and \varphi\in H^1_0(\Omega), let z=\varphi e^{-\frac{|x-x_0|^2}{8\lambda}}. Then,

and

This, together with (Proof), (Proof), (Proof), and the following improved version of Hardy inequality (cf., for instance, [VZ, (2.15)])

leads to (3).

We carry out the proof into three steps.

Step 1. Choose a suitable multiplier.

Let \psi\in C^\infty_0(\Omega\cap B_{(1+\delta)R}(x_0)) be the cut-off function verifying

for some generic constant C\geq 1 independent of R and \delta. For h>0 to be fixed later, multiplying by e^{-\frac{|x-x_0|^2}{h}}\psi^2u the first equation of Equation (?) and integrating the latter over \Omega\cap B_{(1+\delta)R}(x_0), we have

From

and

it holds that

By the integration by parts,

Hence,

By Lemma 3 (where \varphi=\psi u, \lambda=h and \gamma=1), we have

These last two inequalities indicate that

Which leads to

Step 2. Derive a localized energy estimate for \int_{\Omega\cap B_R(x_0)}|u(x,T)|^2\,dx.

For the simplicity of writing, we set

where C(k) and C are two constants in (?), and

Clearly,

Therefore, by (?), it holds that

Then, we have

Which implies that when T-Dh\leq t\leq T,

Since \psi(x)=1 and e^{-|x-x_0|^2/h}\geq e^{-R^2/h}, x\in \Omega\cap B_R(x_0), we have

In view of (?), (?) and (?), it holds that

and

for some new constant C_1(k).

Step 3. Fix h>0.

By (?), we have

This, along with (?), derives that the inequality

holds when T-Dh\leq t\leq T\leq1.

Fix

to be such that

Let h_0=Dh. Then h_0 satisfies h_0\in(0,T). Hence it follows from (?), (?) and (?) that when t\in[T-h_0,T],

Because

the desired estimate (5) is valid from (?) and (?).

We only need to prove the desired estimate (6) for an arbitrarily fixed solution u to Equation (?) with u(\cdot, 0)\neq 0 in L^2(\Omega). Let \psi\in C_0^2(\Omega\cap B_{R+2\delta R}(x_0)) be the cut-off function such that 0\leq\psi\leq1,

where the generic constant C is independent of R and \delta. Let z=\psi u and

Associated with the triple (z,f,x_0), we set for each \lambda>0 and t\in(0,T],

and

Step 1. For each t\in[T-D_1h_0,T], where h_0 is given by (5) and 0<D_1\leq 1 is to be fixed later, we obtain from Lemmas 2, 3 and the similar argument as in Lemma 4, as well as Lemma 5 (cf. [PWZ, Lemma 4, Step 1] ) that

and

Step 2. It follows from [pw2, Lemma 2.5 ] that

Step 3. By (?) and (ii) of Lemma 1 (where u=z and f is given by (Proof)), we have that for each t\in[T-D_1h_0,T],

Consequently,

It follows from the above inequality that when t\in[T-D_1h_0,T],

Integrating the above inequality with respect to t over [T-D_1h_0,T-D_1h_0/2], we obtain that

On the other hand, we see from (?) and (i) in Lemma 1 (where u=z and f is given by (Proof)) that

Integrating the above inequality over [T-D_1h_0,T-D_1h_0/2], we get that

Notice from Lemma 5 that

Therefore, it follows from (?), (?) and (?) that when 0<\lambda\leq D_1h_0,

with

Step 4. Let

Here \varepsilon\in(0,1) will be choose later. Then

and consequently

Therefore, by Step 4 and noticing that 0<h_0<T\leq1, we have

Step 5. Finally, we choose \varepsilon\in(0,1), which is independent of T, such that

Hence, by Step 2, we get that

which implies that

Recall that

Thereby,

Which, together with (?), completes the proof.