Systems of Wave Equations on Asymptotically de Sitter Vacuum Spacetimes in All Even Spatial Dimensions

Using (?), (?), and the lower order estimates in Propositions 3.2 and 3.3, we notice that in order to establish Theorem 1.1 it suffices to prove the estimate for all \tau\in(0,1]:

For the rest of the proof we show this estimate. The first step is to combine the high frequency regime estimate in Proposition 3.9 with the low frequency regime estimate in Proposition 3.8. We denote \xi_{0}=\big(\nabla^M\Phi_{0}\big)_J and \xi_i=\nabla^M\Phi_i for 1\leq i\leq I. Using the high frequency regime estimate in Proposition 3.9, we get that for k\geq0 and \tau\in\big[2^{-k-1},1\big]:

We use the low frequency regime estimate in Proposition 3.8 and the bound (?) for the singular component. Thus, we get for k\geq0 and \tau\in\big[2^{-k-1},1\big] the following high frequency regime estimate:

The second step is to prove a bound for the sum of the non-negative frequencies. We define the following energy for all k\geq0:

Using the singular component high frequency regime estimate (?), we get that for all k\geq0:

As a result, the above high frequency regime estimate implies the bound:

The low frequency estimates in Proposition 3.3 and (?) imply the bound:

As a result, we completed the second step of the proof and showed that:

The final step of the proof is dealing with the negative frequencies k<0. We remark that we can repeat the proof of Proposition 3.3 for m=M and use (?) in order to deal with the singular component. Thus, to get:

We notice that by [geometricLP], we have that \|P_k\nabla F\|_{L^2}\lesssim2^k\|F\|_{L^2} for any k<0. We obtain the following bound for the negative frequencies:

To conclude the proof of Theorem 1.1, we combine the estimates proved for non-negative frequencies and negative frequencies. We then apply Gronwall to obtain:

We use the estimates for the regular quantities in Section ? and the top order estimate for the singular quantities (?) to get:

We also recall our notation F_m'=\psi\sum_{i=1}^I\nabla^{m+1}\Phi_i+F_{m}^{0}. Using again the estimates for the regular quantities proved in Section ?, we conclude.

We prove (Notation) by induction, while considering separately the cases m\leq2n and m>2n. The case m=1 is proved in [geometricLP]. We assume at first that (Notation) holds up to m<2n, and prove it for m+1. We notice that for any tensor \phi we have:

As a result, we get:

Since \|\slashed{Riem}\|_{L^{\infty}}\lesssim1 by the bound on the background spacetime in (?), the second term is controlled using the induction hypothesis. For the first term, we write using [\Delta,U(z)]F=0:

by using (?) again. This completes the first induction argument, and establishes (Notation) for all m\leq2n. Next, for some m\geq2n we assume that (Notation) holds up to m, and prove it for m+1. As before, we have:

The second term in the above can be written as:

For each terms of the sum, we bound the factor with the most number of angular derivatives in L^2. The other factors have at most (m-1)/2 derivatives, so we bound them in L^{\infty} and apply the Sobolev inequality to get:

This completes the proof of (Notation). Next, we prove (Notation). By Duhamel’s formula as in [duhamelLP] we have:

Using the standard estimates for the heat kernel in [geometricLP], we get (Notation). The proof of (Notation) is similar. We also have using [\Delta,\nabla^m]=\nabla[\Delta,\nabla^{m-1}]+[\Delta,\nabla]\nabla^{m-1}:

This implies (Notation) using the bounds in (Notation), and by estimating the commutator term as in the proof of (Notation). The proof of (Notation) is similar. The proofs of (Notation) and (Notation) follow, since according to [geometricLP]:

By Duhamel’s formula as in [duhamelLP] we have:

Using (2.2) and the standard estimates for the heat kernel in [geometricLP], we get (Notation). The proof of (Notation) is similar. Finally, to prove (Notation):

We note that the proof of the last statement uses parts of Lemma 2.3. We included its statement in the heat equation bounds section for future convenience.

The first formula follows since \nabla_t=2^{-k}\nabla_{\tau}=2^{-k+1}\tau\nabla_4=2^{-2k+1}t\nabla_4. In order to prove (2.3), we notice that using the bound in (Notation) we have that:

Similarly, we also have that:

Next, we have that:

Similarly, we also have that:

The proofs of (2.3) and (2.3) follow from (Notation) and (Notation).

In order to prove (2.4), we compute the following:

We can bound the second term in L^2 by:

Next, we write the first term as:

The second term in (Proof) can be bounded in L^2 by:

We write the first term in (Proof) as:

The latter term in the above is bounded in L^2 by:

As a result, we proved that:

The proof of (2.4) follows from (2.4), since \underline{\widetilde{P}}_k^2=\widetilde{P}_k.

We now prove (2.4). We notice that using (2.2) we can write:

The second term can be bounded in L^2 by:

The first term in the above expression can be written as:

To complete the proof of (2.4), we need to write this expression as 2^{-2k}\chi\nabla^2\widetilde{P}_k\nabla F+O\big(2^{-k}\|F\|_{H^1}\big). Indeed, we can bound the last two terms in L^2 by:

Finally, the leading term can be written as:

where in the last step we used the same argument as in the proof of (2.4). This completes the proof of (2.4). Moreover, the proof of (2.4) follows from (2.4) as before.

In the previous proof we obtained the identity:

As a result, we get from Lemma Notation:

A similar proof also gives:

Let \dot{P} be the projection operator with symbol given by \dot{m}(z)=-\int_z^{\infty}m(z')dz'. According to the proof of [geometricLP, Theorem 5.5 (v)] we have that 2^{2k}P_kF=\Delta\dot{P}_kF. This implies the Poincaré inequality:

Also, for all l\geq k\geq 0 we have the following estimate, according to the proof of [geometricLP, Theorem 5.5 (ii)]:

We use these two bounds, together with the other usual bounds for LP projections, to get:

We square this inequality and use Cauchy-Schwarz in order to conclude.

We first notice that we have the inequality:

Using this, we can also bound the following:

Because of (?), we can express the LHS of (2.8) using derivatives at \{\tau=0\}:

Using the finite band property of [geometricLP] we have:

Combining these two estimates, we also have that:

So far we proved (2.8), (2.8), and (2.8). In order to prove (2.8) and (2.8), we notice that for any horizontal k-tensor \Phi we have the formula:

Using this, we get:

Finally, we have:

The bound follows using Duhamel’s formula as in [duhamelLP]:

where we also used the expansion of \slashed{g}_{\tau} at \tau=0. The last term is bounded by \tau^2\big\|F\big\|_{H^{s+2}}.

We denote v=\sqrt{\tau}, and compute that \partial_{\tau}=2\tau\partial_v and e_4=\partial_v. Since \chi=\mathcal{L}_{e_4}\slashed{g}, we use the standard formula for \frac{d}{dv}\int_{S^n}|F|^2(v)dVol_{\slashed{g}} to get:

It suffices to prove that for any K>0 there exists a unique solution of (3.1) such that the above expansions hold in H^K(S^n). We introduce the quantity \widetilde{\Phi}=\Phi-\Phi^0\log({\tau})-\Phi^1, which satisfies the equation:

where F_1 and F_2 are bounded functions of \Phi^0,\Phi^1, and their angular derivatives. To obtain this equation we use the fact that \Phi^0 and \Phi^1 are Lie transported in time, so by (Proof) we have \nabla_{\tau}\Phi^0=\tau\chi\cdot\Phi_0 and \nabla_{\tau}\Phi^1=\tau\chi\cdot\Phi_1. For any k\geq0 we obtain the commuted equation:

where we denote the error terms:

We remark that using the smoothness of \Phi^0,\Phi^1, and the background metric \slashed{g}, we obtain that:

We define \widetilde{\Phi}_{\epsilon} to be the solution to (Proof) on [\epsilon,1]\times S^n with initial data \widetilde{\Phi}_{\epsilon}|_{\tau=\epsilon}=\nabla_{\tau}\widetilde{\Phi}_{\epsilon}|_{\tau=\epsilon}=0. Contracting the commuted equation for \widetilde{\Phi}_{\epsilon} with \nabla^k\nabla_{\tau}\widetilde{\Phi}_{\epsilon}, we obtain the standard energy estimate:

We notice that in the above estimate we dropped the bulk term with a favorable sign. Also, it is essential that \widetilde{\Phi}_{\epsilon} vanishes to sufficiently high order at \tau=0 in order to not have any initial data contribution. Moreover, due to Lemma 3.1, we also have the terms \int_{\epsilon}^{\tau}\tau'\|\nabla\nabla^k\widetilde{\Phi}_{\epsilon}\|_{L^2}^2 and \int_{\epsilon}^{\tau}\tau'\|\nabla^k\nabla_{\tau}\widetilde{\Phi}_{\epsilon}\|_{L^2}^2 on the RHS, but these can be bounded using Gronwall.

Using the higher order version of the commutation formula (2.2), the smoothness of the background metric \slashed{g}, and the Gronwall inequality, we obtain the estimate:

where the implicit constant depends on k\geq0. On the other hand, using Lemma 3.1, Gronwall, and the commutation formulas, we also have:

We use our previous two estimates and Gronwall to obtain that for any \tau\in[\epsilon,1]:

By taking the supremum on [\epsilon,\tau] in the above inequality for each \tau\in[\epsilon,1], we obtain that:

We use this and a similar argument for \big\|\widetilde{\Phi}_{\epsilon}\big\|_{H^{k}} to obtain the bound:

Using the Banach-Alaoglu theorem and compactness, we obtain that for K\ll k there exists \widetilde{\Phi} a solution of (Proof) on \mathcal{M} such that:

As a result, we obtain that \Phi=\widetilde{\Phi}+\Phi^0\log({\tau})+\Phi^1 is a solution of (3.1) and satisfies the desired expansions. Finally, we remark that uniqueness follows by using our standard energy estimate on [0,\tau] for the difference of two solutions with the same asymptotic expansion.

We start with the estimate for \nabla^m\Phi_{0Y}^2. Using the standard \nabla_{\tau} multiplier in (?), we get for \tau\in(0,1]:

We notice that due to Lemma 3.1, we also have the terms \int_0^{\tau}\tau'\|\nabla\nabla^m\Phi_{0Y}^2\|_{L^2}^2 and \int_0^{\tau}\tau'\|\nabla_{\tau}\nabla^m\Phi_{0Y}^2\|_{L^2}^2 on the RHS, but these can be bounded using Gronwall. By the commutation formula (2.2), Lemma 2.7, and Gronwall, we get:

On the other hand, we also have the estimate:

Combining the last two bounds, we obtain the desired estimate for \big(\nabla^m\Phi_{0}\big)_Y^2.

We notice that \nabla^m\Phi_{0Y}^1 satisfies the equation:

We use \nabla_{\tau} as a multiplier to obtain the estimate for any \tau\in(0,1/2]:

We point out that the error term \frac{1}{\tau'|\log\tau'|}\big\|\nabla_{\tau}\nabla^m\Phi_{0Y}^1/\log\tau\big\|^2_{L^2}\cdot\textbf{1}_{[1/10,1/2]} appears on the right hand side because for \tau\in[0,1/10] we have 1+2/\log\tau\gtrsim1, so the bulk term has a favorable sign. The error terms are estimated as usual, and we use Gronwall to obtain:

We also have the estimate:

Combining the last two bounds, we obtain the desired estimate for \big(\nabla^m\Phi_{0}\big)_Y^1 on \tau\in(0,1/2].

In the case of \nabla^m\Phi_{0Y}^1 for \tau\in[1/2,1], we use the same estimates as for \nabla^m\Phi_{0Y}^2, but with data at \tau=1/2. This allows us to obtain the desired estimates for \big(\nabla^m\Phi_{0}\big)_Y^1 on \tau\in[1/2,1].

Using \nabla_{\tau} as a multiplier in the equations (?) and (?) satisfied by the regular quantities, we get:

In the second estimate, we dropped the bulk term with a favorable sign obtained because \sigma=1 in (?). Using the above two bounds, together with Lemma 3.2 and Gronwall, we obtain:

We can estimate the last two error terms on the RHS as before, which implies the conclusion.

We can write:

Using (2.3) and (2.3) applied to the projection operator P_k^2, we get for all k\geq0:

As a result, we have that:

Moreover, we notice that \|\slashed{Riem}_0\|_{H^{M-1}}\leq C_1\lesssim1 by our bounds on the background spacetime in Theorem 1.1.

Using the standard \nabla_{\tau} multiplier and the previous lower order estimates we get:

As before, we can also estimate \big\|\mathcal{C}\big\|^2_{L^2} in order to obtain the conclusion.

For any k\geq0, we apply P_k to the equation satisfied by \mathcal{A} to get:

We contract this equation with P_k\nabla_{\tau}\mathcal{A} and integrate by parts to obtain the energy estimate:

We use Gronwall:

We use the bounds in Lemma 2.3 and Lemma 2.4 to control the commutation terms:

As a result, we get that:

We multiply by 2^k and sum over all k\geq0. This amounts to taking half of a derivative.

We recall that from the standard lower order estimates we have:

Since \sum_kP_k^2=I, we obtain using our definition of fractional Sobolev spaces in Section ?:

Finally, we obtain the desired conclusion by Gronwall.

We denote \mathcal{D}=\mathcal{B}/\log\tau. Then \mathcal{D} satisfies the equation:

This has the same properties needed to do energy estimates as the equation satisfied by \mathcal{A}. Thus, an analogous proof gives for \tau\in(0,1/2] and any k\geq0:

We point out that the error term \frac{1}{\tau'|\log\tau'|}\big\|P_k\nabla_{\tau}\mathcal{D}\big\|^2_{L^2}\cdot\textbf{1}_{[1/10,1/2]} appears on the RHS because for \tau\in[0,1/10] we have 1+2/\log\tau\gtrsim1. This can be estimated as usual using Gronwall. We multiply by 2^k and sum over k:

We recall that from the standard lower order estimates we have for \tau\in(0,1/2]:

Since \sum_kP_k^2=I, we obtain using our definition of fractional Sobolev spaces in Section ?:

We obtain by Gronwall that for any \tau\in(0,1/2]:

This implies that for any \tau\in(0,1/2]:

In particular, this also implies that:

Using the equation for \mathcal{B} on the time interval \tau\in[1/2,1], we repeat the energy estimate that we did for \mathcal{A}:

We recall that from the standard lower order estimates we have for \tau\in[1/2,1]:

As a result, we have that for \tau\in[1/2,1]:

We combine the estimates for \tau\in(0,1/2] and \tau\in[1/2,1] to obtain the conclusion.

The proof follows from Lemma 3.1, since \partial_{t}=2^{-k}\partial_{\tau}.

We denote \xi_{0}=\big(\nabla^M\Phi_0\big)_J and for 1\leq i\leq I we denote \xi_i=\nabla^M\Phi_i. We introduce the new time variable t=2^k\tau. Equations (?) and (?) with \sigma=1,m=M can be written as follows for all 0\leq i\leq I:

For any k\geq0, we apply P_k to each equation:

We contract each equation with P_k\nabla_t\xi_i and integrate by parts to obtain the energy estimate:

We note that we dropped the bulk term with a favorable sign in the above estimate. We use Gronwall for t\in\big[0,1/2\big]:

We use Lemma 2.3 in order to bound the commutation terms. Thus, we get:

We change variables to \tau and get for all \tau\in[0,2^{-k-1}]:

Next, using Lemma 3.1 and (2.4), we also have the bound for all \tau\in[0,2^{-k-1}]:

Using Gronwall and the previous estimate, we conclude that:

We denote \xi_{0}=\big(\nabla^M\Phi_{0}\big)_J and for 1\leq i\leq I we denote \xi_i=\nabla^M\Phi_i. We introduce the new time variable t=2^k\tau. As before, equations (?) and (?) with \sigma=1,m=M can be written for all 0\leq i\leq I:

We multiply by \sqrt{t} to get for all 0\leq i\leq I:

For any k\geq0, we apply P_k to obtain the equations for all 0\leq i\leq I:

We contract each equation with P_k\nabla_t(\xi_i\sqrt{t}) and integrate by parts to obtain the following energy estimate:

We point out that the bulk term with a favorable sign provides a significant simplification for our analysis. On the other hand, in the analysis of the second model system the corresponding term will create several complications. We use Gronwall for t\in\big[1/2,2^k\big], and the bounds in Lemma 2.4 to get for all 0\leq i\leq I:

Once again we use Gronwall for t\in\big[1/2,2^k\big] to get for all 0\leq i\leq I:

Finally, we sum the above estimates for all 0\leq i\leq I to obtain:

We change variables to \tau in order to obtain the conclusion.

For the purpose of this proof we fix 0\leq m\leq M and we denote \xi_i=\nabla^m\Phi_i. We can rewrite (?) with \sigma=2 for all 1\leq i\leq I:

Preliminary estimates. We first prove the following preliminary estimates for all 1\leq i\leq I:

To prove this, we contract each equation (Proof) by \nabla_{\tau}\xi_i. Using Lemma 3.1, we obtain the energy estimate:

We use Cauchy-Schwarz, Gronwall, and the bulk term to obtain:

On the other hand, we also have the bound:

We sum the last two inequalities for all 1\leq i\leq I:

We use Gronwall to complete the proof of the above preliminary estimate. For the data terms at \tau=1 we also use the commutation formula (2.2) and the assumption on the background spacetime \|\psi\|_{H^{M+1}(S_1)}\leq C_2.

The main estimates. For each k\geq0, we apply P_k to equation (Proof) to obtain:

We contract each equation by P_k\nabla_{\tau}\xi_i, in order to obtain the energy estimate:

We use Cauchy-Schwarz and Gronwall to obtain:

We use Lemma 2.3 and Lemma 2.4 to get:

We multiply each inequality by 2^k and sum for all 1\leq i\leq I:

We then sum for all k\geq0, using the preliminary estimate as well to obtain:

We use Gronwall and we bound the initial data term at \tau=1 as before (using C_2) to obtain the conclusion.

For the purpose of this proof, we denote \xi_m=\nabla^m\Phi_0. We can rewrite equation (?) as:

Using Lemma 3.1, we obtain the energy estimate:

Using Gronwall, we obtain:

Similarly, we have the estimate:

In particular, this implies:

which gives in the above inequality:

By Gronwall, we obtain:

Using this, we also have:

We obtain the conclusion using Gronwall and bounding the initial data term at \tau=1.

Since 0\leq k<x, we have using the preliminary estimate in Proposition 4.2:

We apply P_k to (?) to obtain for any k\geq x:

We contract the equation with \tau P_k\nabla_{\tau}\xi and integrate by parts to obtain the following energy estimate:

We point out that similarly to the proof of Proposition 4.2, we obtain a bulk term with a favorable sign, which we drop. We use Lemma 2.3 to bound the commutation terms:

Using Cauchy-Schwarz, we obtain the bound:

Using the Gronwall inequality for \tau\leq X2^{-k-1}\leq1 we get:

Finally, we use the above estimate (?) to complete the proof.

We introduce the new time variable t=X^{-1}2^k\tau\in[1/2,X^{-1}2^k]. The equation (?) becomes:

We multiply by \sqrt{t} to get:

We apply P_k to the equation:

We contract the equation with P_k\nabla_t(\xi\sqrt{t}) and integrate by parts. We notice that using the analogue of Lemma 3.3 for the new time variable t does not introduce any constants that depend on X. We obtain the energy estimate:

We use Lemma 2.4, and Gronwall for t\in\big[1/2,X^{-1}2^k\big] (to deal with the fifth and sixth error terms):

We use Gronwall for t\in\big[1/2,X^{-1}2^k\big] to get:

We change variables to \tau and we obtain the conclusion.

We prove this by induction on k. In the case k=x, we have from (4.1) that u(x,\tau)\leq A(x,\tau) as desired. Next, we assume that (4.1) holds for x,x+1,\ldots,k and we prove it for k+1. We have:

We conclude, since we have:

We apply Lemma 4.1 for the inequality (?), where:

From (4.1) we obtain for \tau\in[X2^{-k-1},1]:

In order to bound the above, we first note that:

In this bound we used the good control of the constant in the definition of b(k), obtained using the smallness of \delta in (?). As a result, we obtain for all \tau\in[X2^{-k-1},1]:

where we bounded the terms with 2^{-j}<\tau using the inequality x+1\leq e^x. Finally, we conclude by noticing that:

As a consequence of the preliminary estimates in Proposition 4.2, we obtain for all 0\leq m\leq M:

so it remains to prove the above bound for \big\|\nabla^{M+1}\mathcal{O}\big\|_{L^2}^2.

For any k\geq x and 0\leq\tau<X2^{-k-1}\leq1, the low frequency regime estimate in Proposition 4.4 implies:

Using the expansion of \nabla_{\tau}\nabla^M\Phi_0 at \mathcal{I}^-, we get that for all k\geq x:

Together with our previous estimate for \|\mathcal{O}\|_{H^M}, we obtain that:

We complete the proof since we already proved in Section ?:

Using (?) as in the proof of Proposition 4.4, we get that \overline{\xi}_k satisfies the equation on \tau\in(0,X2^{-k-1}]:

We contract by \overline{\xi}_k to obtain the energy estimate:

We use Lemma 2.3, Gronwall, and the notation in Section ? to obtain:

We use the preliminary estimates in Proposition 4.2 to get:

Using the finite band property of the LP projections for P_k=\underline{P}_k^2, we have the bound:

Once we established the above estimate for 2^{k}\|\overline{\xi}_k\|_{L^2}, the idea of the proof is to decompose the error term in (?) into its low frequency and high frequency regime parts, in order to use our previous estimates. For each \tau'\in[\tau,X2^{-k-1}] we have:

The first three terms in (?) are in the low frequency regime, and the last term is in the high frequency regime.

We bound the first term in (?) using the L^2 almost orthogonality of the LP projections:

The corresponding term in (?) is bounded using Proposition 4.2:

For the second term in (?) we use Cauchy-Schwarz and we consider a projection operator with \underline{P}_k=\underline{\widetilde{P}}_k^2:

where we used Proposition 4.4 in the second line. The corresponding term in (?) is bounded by:

For the third term in (?), we have similarly from Proposition 4.4:

As a result, the corresponding term in (?) is bounded using Proposition 4.4:

For the high frequency regime fourth term in (?), we have using our notation in Section ?:

The corresponding term in (?) is bounded in Lemma 4.2 at the end of the section.

We combine the previous bounds and we get from (?) that for all \tau\in(0,X2^{-k-1}]:

We take the limit \tau\rightarrow0 and sum for all k\geq x:

For the last three terms, we use the estimates in Section ? and we obtain the conclusion.

We use the estimate (?) as a starting point. We also use the bound on \widetilde{S}_l in Section ?, and the definitions of \widetilde{E}_l and \widetilde{I}_l. Thus, we have for all \tau\in[X2^{-l-1},1]:

We use this bound to obtain a total of 8 error terms that control our main integral. We notice that the first two terms can be dealt with in a straightforward way. For the third and fourth terms we compute that:

where we used the fact that l\geq m and l\geq k in the first line. For the fifth and sixth terms we have:

For the seventh and eight terms we can write:

where we introduce the notation:

We first notice that in the inner integral of I we have \tau''\geq X2^{-m-1}>X2^{-k-1}. Thus, we have the bound:

where in the second inequality we used the definitions of \widetilde{E}_k and \widetilde{I}_k. Next, we have the bound:

where we used again \tau''\geq X2^{-m-1} to compute the inner integral in the first line.

Finally, we notice that the last error term III must vanish as \tau\rightarrow0, since the inner sum would be empty. We prove a bound consistent with this expectation:

Collecting all the bounds proved for the error terms, we conclude the proof.