Boundedness of the Fifth Derivative for the One-Particle Coulombic Density Matrix at the Diagonal

If |\alpha|,|\beta| \ge 1 then take any l \le \alpha and m \le \beta with |l|=|m|=1. We can then write

It is then straightforward to obtain the required bound by Proposition 4.2 with \mu_1 = 0, \mu_2 = \alpha - l and \mu_3 = \beta - m.

We now consider the case where either \alpha or \beta is zero. Without loss, assume \beta = 0. Let l \le \alpha and m \le \alpha-l obey |l|=|m|=1. We then have by (?),

It suffices to bound each term separately. The final three terms have at least one derivative on each x and y, hence we be treated as before. Next, the first term can be rewritten as \partial^{l+m}_{x+y}\partial_x^{\alpha-l-m-r}\gamma_{r,0} for some r \le \alpha-l-m with |r|=1. Use of Proposition 4.2 with \mu_1 = l+m, \mu_2 = \alpha-l-m-r and \mu_3 = 0 then gives the required bound.

It is clear that (?) holds when \boldsymbol\alpha = 0. Therefore, consider |\boldsymbol\alpha| \ge 1. We first prove (?) for k=0. Take \bold{x} \in \Sigma_{\boldsymbol\alpha}^c, then by the Leibniz rule for cluster derivatives we have

in B(\bold{x}, r\lambda_{\boldsymbol\alpha}(\bold{x})). Now, for each \boldsymbol{\beta} \le \boldsymbol{\alpha} there exists some C, independent of \mathbf x, such that

To prove this, we use Lemma 2.4 and Proposition 2.7 with b=0. In addition, if \boldsymbol\beta = 0 we use (?) and if \boldsymbol\beta = \boldsymbol\alpha we use the definition (?). If \boldsymbol\beta \notin \{0, \boldsymbol\alpha \} we use the inequality \lambda_{\boldsymbol{\beta}}(\bold{x}) \lambda_{\boldsymbol{\alpha}-\boldsymbol{\beta}}(\bold{x}) \le \lambda_{\boldsymbol{\alpha}}(\bold{x}), which is obtained from Lemma 2.1. The required bound (?) for k=0 then follows from (Proof) and (?)-(?).

Take \mathbf x \in \Sigma^c_{\boldsymbol\alpha}. By the definition F= F_c - F_s, the equality \nabla \psi = \psi \nabla F + e^F \nabla \phi and the Leibniz rule, we have (?) with

We bound each term in G^{\boldsymbol\alpha}_{\bold{P}} as in (?). Take any \boldsymbol\beta \le \boldsymbol\alpha with |\boldsymbol\beta| \ge 1. Notice that \Sigma_{\boldsymbol\beta} \subset \Sigma_{\boldsymbol\alpha} and hence \lambda_{\boldsymbol\alpha}(\mathbf x) \le \lambda_{\boldsymbol\beta}(\mathbf x). Now, there exists C, independent of \mathbf x, such that

To prove this inequality, we bound derivatives of F using Lemma 2.4 if \boldsymbol\beta \ne \boldsymbol\alpha and (?) if \boldsymbol\beta = \boldsymbol\alpha. The derivatives D_{\bold{P}}^{\boldsymbol\beta}\psi are then bounded using (?) with k=0, which was proven above. The right-hand side of (?) can be bounded as in (?) by using the simple bound \lambda_{\boldsymbol\beta} \le \mu_{\boldsymbol\alpha}. Next, still considering |\boldsymbol\beta| \ge 1 we can find C, independent of \mathbf x, such that

This is proven using Proposition 2.7 to bound the derivatives \nabla \phi_{\boldsymbol\beta}, and Lemma 2.4 and (?) to bound derivatives of e^F. The right-hand side of the above inequality can be bounded as in (?) after use of (?)-(?) and the inequalities \mu_{\boldsymbol\beta} \le \mu_{\boldsymbol\alpha} and \lambda_{\boldsymbol\alpha - \boldsymbol\beta} \le 1. In addition, by Lemma 2.4 there exists C such that

Use of (?)-(?), as before, and the inequality \lambda_{\boldsymbol\alpha} \le \mu_{\boldsymbol\alpha} give the correct bound. Finally, the last term in (?) is readily bounded appropriately using (?). In view of (?), this completes the proof of (?).

To prove (?) for k=1 we use (?), (?) and the following inequality,

which holds for some C as a direct consequence of Lemma 2.4.

The results are trivial in the case of \boldsymbol{\sigma} = 0 therefore we assume in the following that \boldsymbol{\sigma} is non-zero. First, observe for all j=1, \dots, M,

by the definition of \mu_{\boldsymbol{\sigma}}. Now perform the following trivial expansion of the product,

which proves (2.1) for k=0. For k=1, consider that for each \bold{y} we can find l = 1, \dots, M such that \lambda_{\boldsymbol{\sigma}}(\bold{y}) = \lambda_{P_l}(\bold{y}) and \sigma_l \ne 0. Then,

as required.

Finally we prove (2.1). As above, take arbitrary \bold{y} and a corresponding l such that \lambda_{\boldsymbol{\sigma}}(\bold{y}) = \lambda_{P_l}(\bold{y}) with \sigma_l \ne 0. For each 1 \le j \le n we denote the \mathbb{N}_0^3-components of \boldsymbol{\beta}^{(j)} as \boldsymbol{\beta}^{(j)} = (\beta^{(j)}_1, \dots, \beta^{(j)}_M). We know \boldsymbol{\beta}^{(1)} + \dots + \boldsymbol{\beta}^{(n)} = \boldsymbol{\sigma} so in particular, \beta_l^{(1)} + \dots + \beta_l^{(n)} = \sigma_l. Since \sigma_l \ne 0 there exists at least one 1 \le r \le n such that \beta^{(r)}_{l} \ne 0. Hence by the definition of \lambda_{\boldsymbol{\beta}^{(r)}},

The remaining factors \lambda_{\boldsymbol{\beta}^{(j)}}(\bold{y}), for j \ne r, can each be bounded above by one.

Take any j = 1, \dots, M with \alpha_j \ne 0, then we have D^{\alpha_j}_{P_j}(f(x_l)) \equiv 0 for each l \in P_j^c. Therefore, for D^{\boldsymbol{\alpha}}_{\bold{P}} (f(x_l)) to not be identically zero we require that l \in P_j for each j with \alpha_j \ne 0. For such l we have x_l \ne 0 since \bold{x} \in \Sigma^c_{\boldsymbol{\alpha}}, and

for each j with \alpha_j \ne 0 by (?). Therefore, for constant C in (2.2), we have

because |\boldsymbol{\alpha}| \ge k. Similarly, for each j = 1, \dots, M, with \alpha_j \ne 0 we have D^{\alpha_j}_{P_j} (f(x_l - x_m)) \equiv 0 if either l, m \in P_j or l, m \in P_j^c. Therefore, for D^{\boldsymbol{\alpha}}_{\bold{P}} (f(x_l - x_m)) to not be identically zero we require that

For such (l,m) we have x_l \ne x_m since \bold{x} \in \Sigma^c_{\boldsymbol{\alpha}} and

for each j with \alpha_j \ne 0 by (?). Therefore, for some constant C',

because |\boldsymbol{\alpha}| \ge k.

Let \tau be the function defined as \tau(x) = |x| for x \in \mathbb{R}^{3}. Then, by definition (?) we can write

For each r=1,\dots,N, denote by \nabla_{y_r} the three-dimensional gradient in the variable y_r. We then have

We apply the D^{\boldsymbol{\sigma}}_{\bold{P}}-derivative to each of (Proof) and (Proof), using Lemma 2.2 with f = \tau and f = \nabla \tau respectively. This shows that for k=0,1 the functions D_{\bold{P}}^{\boldsymbol{\sigma}} \nabla^{k} F_c exist and are smooth on the set \Sigma^c_{\boldsymbol{\sigma}}. Furthermore, there exists C, depending on \boldsymbol{\sigma}, such that

for all \bold{y} \in \Sigma^c_{\boldsymbol{\sigma}}. In the second inequality we used that \mu_{\boldsymbol{\sigma}}(\bold{y}) \le q_{\boldsymbol{\sigma}}(\bold{y}).

Recall that \nabla F_s is smooth with all derivatives bounded, see (?). Since F = F_c-F_s and \mu_{\boldsymbol{\sigma}} \le 1, we have that for each \boldsymbol{\sigma} \ne 0 and k=0,1, there exists C, depending on \boldsymbol{\sigma}, such that

for all \bold{y} \in \Sigma^c_{\boldsymbol{\sigma}}. By a straightforward application of Lemma 2.1, this proves the first bound in (?), for k=0,1.

We proceed to prove the second bound in (?). For every \boldsymbol{\eta} \le \boldsymbol{\sigma} we have that \Sigma_{\boldsymbol{\eta}} \subset \Sigma_{\boldsymbol{\sigma}} and therefore D_{\bold{P}}^{\boldsymbol{\eta}} \nabla^k F exists and is smooth in \Sigma_{\boldsymbol{\sigma}}^c, for k=0,1. We use the chain rule for weak derivatives to show that D^{\boldsymbol{\sigma}}_{\bold{P}}(e^F) exists in \Sigma_{\boldsymbol{\sigma}}^c and is equal to a sum of terms, each of the form

for some 1 \le n \le |\boldsymbol{\sigma}| and some collection 0 \ne \boldsymbol{\beta}^{(j)} \in \mathbb{N}_0^{3M} for j=1,\dots,n, where \boldsymbol{\beta}^{(1)} + \dots + \boldsymbol{\beta}^{(n)} = \boldsymbol{\sigma}.

For each j=1,\dots,n, we write \boldsymbol{\beta}^{(j)} = (\beta^{(j)}_1, \dots, \beta^{(j)}_{M}) to denote the \mathbb{N}_0^3-components of \boldsymbol{\beta}^{(j)}. Similarly, the weak derivative D^{\boldsymbol{\sigma}}_{\bold{P}} \nabla(e^F) exists in \Sigma_{\boldsymbol{\sigma}}^c and the gradient of the general term (?) is equal to

We will now find bounds for the above expressions. To do this we will use (?) to bound cluster derivatives of F. This, along with (2.1) and (2.1) of Lemma 2.1, allows us to bound the following product

where C is some constant, depending on \boldsymbol{\sigma}, and all \bold{y} \in \Sigma^c_{\boldsymbol{\sigma}}. We used that for each l=1,\dots,M, we have |\beta^{(1)}_l| + \dots + |\beta^{(n)}_l| = |\sigma_l| as a consequence of \boldsymbol{\beta}^{(1)} + \dots + \boldsymbol{\beta}^{(n)} = \boldsymbol{\sigma}. We now bound the following product in a similar manner using (?) and Lemma 2.1. For any r=1, \dots, n,

for some constant C, depending on \boldsymbol{\sigma}, and all \bold{y} \in \Sigma^c_{\boldsymbol{\sigma}}. Using the fact that e^F and \nabla F are bounded in \mathbb{R}^{3N}, it is now straightforward to bound (?) and (?) appropriately. This completes the proof of (?).

Finally, we prove the inequality (?). It can be shown that the following Leibniz rule holds

in \Sigma^c_{\boldsymbol{\sigma}}. For every \boldsymbol{\beta} \le \boldsymbol{\sigma} we can bound D^{\boldsymbol{\beta}}_{\bold{P}} \nabla F by a constant if \boldsymbol{\beta} = 0, and by (?) otherwise. This gives some constant C, depending on \boldsymbol{\sigma}, such that

for all \bold{y} \in \Sigma^c_{\boldsymbol{\sigma}}. The required bound is obtained after an application of (2.1) of Lemma 2.1.

Take some \bold{x} \in \Sigma_{\boldsymbol{\sigma}}^c. By (?), for each \bold{y} \in B(\bold{x}, \nu\lambda_{\boldsymbol{\sigma}}(\bold{x})) we have

for each j=1, \dots, M with \sigma_j \ne 0. The final inequality above uses the definition of \lambda_{\boldsymbol{\sigma}} in (?). By rearrangement, we have

Therefore,

for all \bold{y} \in B(\bold{x}, \nu\lambda_{\boldsymbol{\sigma}}(\bold{x})). We now prove an analogous inequality. First, we see that \lambda_{\boldsymbol{\sigma}}(\bold{x}) = \lambda_{P_l}(\bold{x}) for some 1 \le l \le M with \sigma_l \ne 0 which will depend on the choice of \bold{x}. We also note that \lambda_{\boldsymbol{\sigma}}(\bold{y}) \le \lambda_{P_l}(\bold{y}) for all \bold{y} \in \mathbb{R}^{3N} which follows from \sigma_l \ne 0 and the definition of \lambda_{\boldsymbol{\sigma}}. Therefore,

for all \bold{y} \in B(\bold{x}, \nu\lambda_{\boldsymbol{\sigma}}(\bold{x})). In the second step we applied (Proof). The required bounds then arise from Lemma 2.3 followed by either (Proof) or (Proof) as appropriate.

We prove by induction, starting with |\boldsymbol{\alpha}| = 1. Let \mathcal{L}(\cdot,\cdot) be the bilinear form corresponding to the operator acting on \phi_{\boldsymbol{\alpha}} on the left-hand side of (2.5), as defined in (?). Since \phi \in H^{2}_{loc}(\mathbb{R}^{3N}) we have \phi_{\boldsymbol{\alpha}} \in H^1_{loc}(\mathbb{R}^{3N}). We use integration by parts to show that, for any \chi \in C^{\infty}_c(\Sigma^c_{\boldsymbol\alpha}),

Now, \mathcal{L}(\phi, D_{\bold{P}}^{\boldsymbol{\alpha}}\chi) = 0 since \phi is a weak solution to (?), hence (2.5) holds. By Lemma 2.3 and that \phi \in C^1(\mathbb{R}^{3N}) we see that g_{\boldsymbol\alpha} \in L_{loc}^{\infty}(\Sigma^c_{\boldsymbol\alpha}). Hence, by Theorems 1.5, \phi_{\boldsymbol\alpha} \in C^1(\Sigma^c_{\boldsymbol\alpha}) \cap H_{loc}^2(\Sigma^c_{\boldsymbol\alpha})

Assume the hypothesis holds for all multiindices 1 \le |\boldsymbol{\alpha}| \le k-1 for some k\ge 2. Take some arbitrary multiindex |\boldsymbol{\alpha}| = k. We will prove the induction hypothesis for \boldsymbol\alpha. First, we state the useful fact that for any \boldsymbol\sigma \le \boldsymbol\alpha we have \Sigma_{\boldsymbol\sigma} \subset \Sigma_{\boldsymbol\alpha}. Take some \boldsymbol{\eta} \le \boldsymbol\alpha with |\boldsymbol{\eta}| = 1. Notice, then, that \phi_{\boldsymbol\alpha-\boldsymbol{\eta}} \in H^2_{loc}(\Sigma^c_{\boldsymbol\alpha}) by the induction hypothesis, and hence \phi_{\boldsymbol{\alpha}} \in H^1_{loc}(\Sigma^c_{\boldsymbol\alpha}). This allows \mathcal{L}(\phi_{\boldsymbol\alpha}, \chi) to be defined for any \chi \in C_c^{\infty}(\Sigma^c_{\boldsymbol\alpha}). Integration by parts is then used on the D^{\boldsymbol{\eta}}_{\bold{P}}-derivative, in a similar way to (Proof). Along with the induction hypothesis and further applications of integration by parts, we find that (2.5) holds for \phi_{\boldsymbol\alpha} as a weak solution. Regularity for \phi_{\boldsymbol\alpha} is shown in a similar way to the |\boldsymbol\alpha|=1 case. Indeed, the induction hypothesis can be used to show g_{\boldsymbol\alpha} \in L^{\infty}_{loc}(\Sigma^c_{\boldsymbol\alpha}).

Fix any \bold{x} \in \Sigma_{\boldsymbol{\alpha}}^c and denote \lambda = \lambda_{\boldsymbol{\alpha}}(\bold{x}). The proof proceeds via a rescaling. Define

for all \bold{y} \in B(0,1). Then by Lemma 2.5 we have that w is a weak solution to

in B(0,1). By (?) and (?), and that \lambda \le 1 by definition, we obtain

for some K, dependent only on E, Z and N and in particular is independent of our choice of \bold{x}. Therefore, by Theorem 1.5, with \theta=0, we get some C, dependent only on N, K, r and R, such that

Finally, we use \nabla w(\bold{y}) = \lambda \nabla \phi_{\boldsymbol{\alpha}}(\bold{x} + \lambda \bold{y}) by the chain rule, and rewrite (?) to give (?).

Suppose |\boldsymbol\alpha| = 1 and k=1. Let l be the index 1 \le l \le M such that \alpha_l \ne 0. Then \mu_{\boldsymbol\alpha} = \lambda_{P_l} and \Sigma_{\boldsymbol\alpha} = \Sigma_{P_l}. The bound then follows from Corollary A.3 with P=P_l. Indeed, for each s>0 we can find some constant C_s such that \log(t) \le C_st^s for all t >0. We use this fact along with the simple inequality \lambda_{P_l} \le \nu_{P_l}.

For |\boldsymbol\alpha| \ge 2 we prove by induction.

Assume the hypothesis holds for all multiindices \boldsymbol{\alpha} with 1 \le |\boldsymbol{\alpha}| \le m-1 for some m \ge 2. Take any \boldsymbol{\alpha} with |\boldsymbol{\alpha}| = m. We prove the k=0 and k=1 cases in turn. It is useful, here, to state the following fact: for all \boldsymbol{\sigma} \le \boldsymbol{\alpha} we have \Sigma_{\boldsymbol{\sigma}} \subset \Sigma_{\boldsymbol{\alpha}} and hence \lambda_{\boldsymbol{\alpha}}(\bold{x}) \le \lambda_{\boldsymbol{\sigma}}(\bold{x}).

The k=0 case is a straightforward application of the induction hypothesis and is described as follows. Firstly, for each \mathbf x \in \Sigma_{\boldsymbol\alpha}^c it is clear that there exists some 1 \le l \le M such that \lambda_{\boldsymbol\alpha}(\bold{x}) = \lambda_{P_l}(\bold{x}) and where \alpha_l \ne 0. Therefore we can find some \boldsymbol{\eta} \le \boldsymbol\alpha with |\boldsymbol{\eta}| = 1 and \eta_j = 0 for each j \ne l where 1 \le j \le M. Now, from the definition of cluster derivatives (?),

We can then use the induction assumption to show existence of C such that

where we can then use \lambda_{P_l}(\bold{x}) = \lambda_{\boldsymbol{\alpha}}(\bold{x}) and \mu_{\boldsymbol\alpha-\boldsymbol{\eta}}(\bold{x}) \le \mu_{\boldsymbol\alpha}(\bold{x}) to complete the required bound for our choice of \boldsymbol\alpha in the case of k=0.

The k=1 case follows from the induction hypothesis and Lemma 2.6. Let r' = (r+R)/2. Firstly, by the definition of g_{\boldsymbol\alpha}, (2.5), along with (?) and Lemma 2.4 we can find C, depending on \boldsymbol\alpha and r', such that

It follows by the induction hypothesis, using b=0, along with the definition of f_{\infty}, (?), that we can find some C, depending on \boldsymbol\alpha, r' and R, such that

for all \bold{x} \in \Sigma_{\boldsymbol\alpha}^c. Next, we apply Lemma 2.6 to obtain some constant C, depending on r and r', such that

We use the k=0 case, proven above, to bound the L^{\infty}-norm of \phi_{\boldsymbol\alpha} in the first term on the right-hand side of the above inequality. To the second term, we apply (?) and the simple inequality \lambda_{\boldsymbol\alpha}(\bold{x}) \le \mu_{\boldsymbol\alpha}(\bold{x}). Together, these give the required bound for k=1 and our choice of \boldsymbol\alpha. This completes the induction.

From the definitions (?) and support criteria it is straightforward to verify that

for all z \in \mathbb{R}^3. It follows that on \mathbb{R}^3,

For any A,B \subset \{2, \dots, N \} with A \cap B = \emptyset we define

for \bold{x} = (x_1, \dots, x_N) \in \mathbb{R}^{3N}. Therefore,

for all \bold{x} \in \mathbb{R}^{3N}.

Let \Xi = \{(j,k) : 2 \le j <k \le N \}. For each Y, Z \subset \Xi with Y \cap Z = \emptyset we define

for \bold{\hat{x}} = (x_2, \dots, x_N) \in \mathbb{R}^{3N-3}. Therefore,

for all \bold{\hat{x}} \in \mathbb{R}^{3N-3}. Overall, for all x,y \in \mathbb{R}^3, \bold{\hat{x}} \in \mathbb{R}^{3N-3},

which is a sum of cutoffs of the form (?).

Take j \in Q^* and k \in Q^c. By the definition of the cluster Q we have f_{j,k} = \theta_{\epsilon} and g^{(1)}_k = g^{(2)}_k = \theta_{\epsilon} in the formula (?) defining \Phi. If we assume \Phi(x,y,\hat{\mathbf x}) \ne 0, then the support criteria for \theta_{\epsilon}(x-x_k), \theta_{\epsilon}(y-x_k) and \theta_{\epsilon}(x_j-x_k) give (3.2). The inequalities (3.2) and (3.2) are proven in a similar way.

By definition, if k \in P^{*} either g^{(1)}_k = \zeta_{\delta} or there exist pairwise distinct j_1, \dots, j_s \in \{2,\dots,N\} with 1\le s \le N-2 such that g^{(1)}_{j_1} = \zeta_{\delta} and f_{j_1,j_2}, f_{j_2, j_3}, \, \dots,\, f_{j_s, k} = \zeta_{\delta}. In the former case, support criteria for \zeta_{\delta}(x-x_k) gives |x - x_k| < (2N)^{-1}\delta. In the latter case, support conditions give |x - x_{j_1}|,\, |x_{j_1} - x_{j_2}|,\, \dots,\, |x_{j_s} - x_k| < (2N)^{-1}\delta and so by the triangle inequality |x - x_k| < \delta/2. Now, |y-x_k| \ge |x-y|-|x-x_k| > \delta/2 by the reverse triangle inequality. The case of k \in S^* is analogous.

Now let k \in Q^*. First, we consider the case where either g^{(1)}_k \ne \theta_{\epsilon} or g^{(2)}_k \ne \theta_{\epsilon}, or both. Without loss, assume g^{(1)}_k \ne \theta_{\epsilon}. Then either g^{(1)}_k = \zeta_{\delta} or g^{(1)}_k = \theta_{\delta}\zeta_{\epsilon}, but both give the inequality

by support criteria. Hence (?) holds in this case.

Now suppose that g^{(1)}_k = g^{(2)}_k = \theta_{\epsilon}. Then there must exist pairwise distinct j_1, \dots, j_s \in \{2,\dots,N\} with 1\le s \le N-2 such that f_{j_1,j_2}, f_{j_2,j_3},\dots,f_{j_s, k} \ne \theta_{\epsilon} and either g^{(1)}_{j_1} \ne \theta_{\epsilon} or g^{(2)}_{j_1} \ne \theta_{\epsilon}. Without loss take g^{(1)}_{j_1} \ne \theta_{\epsilon}, so that, as before, |x-x_{j_1}| < (2N)^{-1}\epsilon. Similarly, we get |x_{j_1} - x_{j_2}|,\, \dots,\, |x_{j_s} - x_k| < (2N)^{-1}\epsilon. Therefore, by the triangle inequality, |x-x_k| < \epsilon/2. Hence completing the proof of (?). Finally, (?) follows from (?) and |x|,|y| \ge \epsilon.

This identity follows from the definitions (?)-(?) and the equality

since for any 2 \le k<l \le N we have f_{k,l} = f_{l,k} by definition.

By the definition of the cluster Q(\Phi), we have g^{(1)}_j = g^{(2)}_j = \theta_{\epsilon} for each j \in Q^c, and f_{k,l} = \theta_{\epsilon} for each k \in Q^* and l \in Q^c. The formula then follows from (?).

Without loss we consider the case of \zeta_{t}, the case of \theta_{t} being similar. Recall \xi = \xi(s) was defined in (?) and we denote by \xi^{(m)} the m-th (univariate) derivative of \xi. Since |\sigma| \ge 1 the chain rule shows that \partial^{\sigma} \zeta_{t}(x) can be written as a sum of terms of the form

where 1 \le m \le |\sigma|, and \sigma_1, \dots, \sigma_m \in \mathbb{N}_0^3 are non-zero multiindices obeying

Since m \ge 1, we have that if \xi^{(m)}(s) \ne 0 then s \in (1,2). Therefore, for any term (Proof) to be non-zero we require that

By the remark preceeding the current lemma, there exists C, dependent on \sigma_1, \dots, \sigma_m, such that

using (Proof). Therefore, the terms (Proof) can readily be bounded to give the desired result.

By the chain rule, each function in the product (?) for \Phi(\,\cdot\,;Q) has zero derivative upon action of D^{\alpha}_{x,y,Q}.

Lemma 3.6 gives bounds for the partial derivatives of the functions \zeta_{\delta}, \theta_{\delta}, \zeta_{\epsilon}, \theta_{\epsilon}. Considering \theta_{\delta}\zeta_{\epsilon}, we apply the Leibniz rule with \sigma \in \mathbb{N}_0^3, |\sigma| \ge 1, to obtain, for z \in \mathbb{R}^3

since, for each \mu \le \sigma with \mu \ne 0 and \mu \ne \sigma we have

by Lemma 3.6 and that 2\delta \le \epsilon. Now, to evaluate \partial^{\boldsymbol\alpha}\Phi' for a general \Phi' we apply the Leibniz rule to the product (?) using (?) where appropriate. Differentiated cutoff factors \zeta_\delta, \theta_\delta, \zeta_\epsilon, \theta_\epsilon are bounded by (3.6). The indicator function in this bound is not required in the case of \zeta_\epsilon or \theta_\epsilon, whereas in the case of \zeta_\delta or \theta_\delta the indicator function is bounded above by M_t. Any remaining undifferentiated cutoff factors are bounded above by 1. If \Phi' = \Phi(\,\cdot\,;Q,Q^c), all cutoff factors are of the form \theta_{\epsilon} by Lemma 3.5 and therefore we need only use the bounds in (3.6) with t=\epsilon.

First, set \Phi' = \Phi(\,\cdot\,;Q) \, \Phi(\,\cdot\,;Q^c) and \Phi'' = \Phi(\,\cdot\,;Q,Q^c). We then have

which follows from Lemmas 3.4 and 3.7, and that \Phi(\,\cdot\,;Q^c) is not dependent on variables involved in the D^{\alpha}_{x,y,Q}-derivative. By the definition (?), the derivative D^{\alpha}_{x,y,Q} \Phi'' can be written as a sum of partial derivatives of the form \partial^{\boldsymbol\alpha} \Phi'' where \boldsymbol\alpha \in \mathbb{N}_0^{3N+3} obeys |\boldsymbol\alpha| = |\alpha|. Now, by the Leibniz rule and Lemma 3.8 there exists some constants C and C', independent of \delta and \epsilon, such that

completing the proof.

By Young’s inequality,

It therefore suffices to prove (3.11) where only one of b_1, b_2 and b_3 is non-zero. We consider the case where this is b_1, the other cases are similar. We split the integral into one where |x-x_j| \ge 1 and one where |x-x_j| < 1. In the former case, we can bound using (3.10) of Proposition 3.10, whereas in the latter we use (3.10) of the same proposition with G(x-x_j) = \mathds{1}_{\{(4N)^{-1}\delta < |x-x_j| < 1\}}(x-x_j)|x-x_j|^{-b}. The lower bound can be included in the indicator function due to (3.2) of Lemma 3.2.

Consider \mu_1,\mu_2,\mu_3 satisfying either (i) or (ii). By Lemma 4.5 it suffices to prove the required bound for integrals of the form (?). Define a = |\alpha_1|+|\beta_1|+|\sigma_1|, b = |\alpha_2|+|\alpha_3|+|\beta_2|+|\beta_3| and c = |\sigma_2|+|\sigma_3|. Notice that a = |\mu_1| and b+c = |\mu_2|+|\mu_3|. Then, using (?)-(?) and (?) it is possible to bound (?) in absolute value by some constant multiplying

For I_2 we use (?) to bound this by some constant multiplying

for some C, where we used (3.10). This gives (?) for I_2.

For I_1, we first consider when b \ne 3. By (?), this is bounded by some constant multiplying

which is then bounded by (?).

Now we consider I_1 for b = 3. If we are in case (i) then we must have c \ge 1 since b + c = |\mu_1|+|\mu_2| \ne 3. Here, we use (?) followed by (?) to bound I_1 by some constant multiplying

for some C, which is then bounded by (?). Finally, if we are in case (ii) for b=3 then I_1 cannot be bounded appropriately. Instead, we use that the integral (?) that we are attempting to bound must have \alpha_1 = \beta_1 = 0 and \sigma_i = 0, i=1,2,3. This allows us to use Lemma 4.6, which is stated and proven below, to bound (?) directly.

Firstly, by Lemma 3.1 there exists a finite collection of cutoffs, \Phi^{(j)}, j=1,\dots,J, for some J, such that

holds everywhere for all choices of 0< 2\delta \le \epsilon.

Fix any x,y such that 0<|x-y| \le m(x,y)/2. Then set \delta = |x-y|/2 and \epsilon = m(x,y)/2. Index by j_k, where k=1,\dots, K for some K \le J, the j’s such that P(\Phi^{(j)})^* \cap S(\Phi^{(j)})^* = \emptyset. The j_k’s do not depend on the choice of \delta and \epsilon. By Lemma 3.3,

holds for all |x'-y'| \ge |x-y|/2. Then by Lemma 4.1, for each k=1, \dots, K, we have

for all x', y' such that |x-y|/2 < |x'-y'| \le |x-y| and |x'|,|y'| \ge m(x,y)/2. In particular, it holds for x'=x and y'=y. The constant C does not depend on the choice of \delta and \epsilon, therefore the bound holds for all required x and y.

We can prove (i) using the expressions (?) and (?). The bound follows from Lemma 3.2 along with (?) of Lemma 3.3, since P,S \subset Q. For (ii) we use (?) to write

for some C depending on b. The required inequality then follows from (3.10) of Proposition 3.10 for the first term above and from Lemma 3.11 for the remaining terms.

Using Lemma 3.9 and (?), the integral can be bounded by some constant multiplying

Expanding, the first term is bounded by (?) and second is bounded using (3.10) of Proposition 3.10.

For each choice of x and y we define a x- and y-dependent change of variables for the integral \gamma_{l,m}(x,y;\Phi). To start, we define two vectors \bold{\hat{a}} = (a_2, \dots, a_N), \bold{\hat{b}} = (b_2, \dots, b_N) \in \mathbb{R}^{3N-3} by

By a translational change of variables using \bold{\hat{a}} we can write

We will then apply differentiation under the integral. Beforehand, we show how such derivatives will act on each function within the integrand. As an illustration, we take a function f defined on \mathbb{R}^{3N} and any \eta \in \mathbb{N}_0^3 to see by the chain rule that

Analogous expressions arise when the cutoff \Phi(x,y,\bold{\hat{x}}) is differentiated which involve cluster derivatives of the form (?), (?).

We now differentiate (?) in x and y, and use the product rule to get that the function \partial^{\mu_2}_x\partial^{\mu_3}_y \gamma_{l,m}(x,y; \Phi) is a linear combination of terms of the form

for some \alpha_2+\beta_2+\sigma_2 = \mu_2 and \alpha_3+\beta_3+\sigma_3 = \mu_3.

We will now apply a (x+y)-derivative to an integral of this form. However, first we apply a change of variables to replace \bold{\hat{a}} with \bold{\hat{b}} as the translation of the integration variable \bold{\hat{x}}. Then, in the same way as before, we obtain that \partial^{\mu_1}_{x+y}I_{(\alpha_2,\alpha_3,\beta_2,\beta_3,\sigma_2,\sigma_3)} is a linear combination of terms of the form

for some \alpha_1+\beta_1+\sigma_1 = \mu_1. Finally, another translational change of variables removes \bold{\hat{b}} to give the expression (?).

We first consider the more difficult case of |l|=|m|=1. By Theorem 1.3 with b=1/2 we can write

and in these cases we have the estimates

With these preliminary estimates, we can prove the result in the case where |\boldsymbol\alpha|,|\boldsymbol\beta| \ge 1. By (Proof), (?) and F_c = F^{(en)}_c + F^{(ee)}_c we bound the integral (4.6) by

By Lemma 4.7 the final term above can be expanded by terms of the form (4.8), since S^* \subset P^c and P^* \subset S^c. Lemma 4.8 then gives the appropriate bound. We will now show that each of the remaining integrals above can be bounded by a constant multiplying an integral of the form

for some a,b \ge 0, b \le 5/2 and a + b \le 3. This follows from using the estimates (?), (?), (?) and (?). We also use (?) to bound the cluster derivatives of \psi(y,\bold{\hat{x}}) in the first term. The required bound then follows by Lemma 4.3.

Next we consider the case where |\boldsymbol\alpha|=3 and \boldsymbol\beta = 0. The other case where |\boldsymbol\beta|=3 and \boldsymbol\alpha = 0 is similar with obvious modifications. By (Proof) and F_c = F^{(en)}_c + F^{(ee)}_c we bound the integral (4.6) by

The integral I_1 is bounded using (?), from which we can then apply Lemma 4.3. Bounding I_2 is a simple application of (?) and Proposition 3.10. For I_3 we use the expression \psi = e^F\phi, where \phi was introduced in (?). Since F = F^{(ee)}_c + F_c^{(en)} - F_s we can write

We expand I_3 according to this expression. By Lemma 4.7, the first term can be expanded in terms of the form (?) and therefore can bounded by Lemma 4.9. Also by Lemma 4.7, the second term can be expanded in terms of the form (4.8) for \beta = m and |\alpha|=4, and therefore can bounded by Lemma 4.8. It therefore suffices to bound

We expand the derivative of F_c^{(ee)} using Lemma 4.7. A general term in this expansion will have the form

for some j \in P^c and |\alpha| = 3. We write this integral as follows, using integration by parts to remove one derivative from |x-x_j|,

since \partial_{x_j}^l\partial_y^m F^{(en)}_c(y,\mathbf x) \equiv 0. We now use (?), that \nabla F^{(en)}_c \in L^{\infty}(\mathbb{R}^{3N}) and that \nabla^k F_s \in L^{\infty}(\mathbb{R}^{3N}) for any integer k \ge 1, see (?). Therefore, these terms can be bounded by some constant multiplying

which is bounded appropriately by Lemma 4.3 and Corollary 4.4.

Finally, it remains to consider the case where one or both of l, m are zero. The strategy is analogous but simpler. We remark that we can use the following improved version of (?) and (?) in the case where \eta = 0 and k=0. For any \boldsymbol{\nu} = (\nu_1, \nu_2) \in \mathbb{N}_0^6 with |\boldsymbol\nu| \ge 1 there exists some C such that

These inequalities follow immediately from (?) of Theorem 1.3.

We suppose that |\alpha| + |\beta| = 5 otherwise the result follows immediately from (?) and Lemma 4.3.

We first prove the case where j \ne k, where the integration by parts is particularly simple. Suppose without loss that |\alpha| \ge |\beta|. Then take any l \le \alpha with |l|=1 and use integration by parts to obtain

This can then be bounded in absolute value as required using (?) followed by either Lemma 4.3 or Corollary 4.4. This completes the proof where j \ne k.

For the remainder of the proof we consider the case where j = k. First, suppose j \in P^c \cap S^c. In the procedure that follows, we will use integration by parts to remove all derivatives from |y-x_j| in I_{j,j}.

Since |\beta| \ge 1 we can find some multiindex \beta_1 \le \beta with |\beta_1| = 1. Integration by parts then gives

We leave untouched the second integral above. For the first, if |\beta| - |\beta_1| \ge 1 we can remove another first-order derivative from |y-x_j| by the same procedure. That is, using integration by parts to give two new terms as in (?). We retain the term where the derivative falls on \psi(x,\bold{\hat{x}}) \psi(y,\bold{\hat{x}}) \Phi(x,y,\bold{\hat{x}}). Whereas on the term where the derivative falls on |x-x_j| we repeat the procedure, so long as there remains at least one derivative on |y-x_j|. Through this process, we obtain a formula for I_{j,j}. To express this formula, we first set T=|\beta| and write \beta = \sum_{s=1}^{T}\beta_s for some collection |\beta_s| = 1 where 1 \le s \le T. Furthermore, we define

Then,

where

We can bound |I^{(i)}_{j,j}| for 1 \le i \le T-1 in the same way as (?) since j \in P^c \cap S^c. It remains to bound I^{(T)}_{j,j}, along with the first integral in formula (?).

We begin by expanding |y - x_j| = |x - x_j| + \big(|y-x_j| - |x-x_j| \big) and noticing that

It is then straightforward to show (for example see the comment preceeding Lemma 3.6) that for some C,C',

where in the second step we used formula (?), since j \in P^c, along with (?). This is then bounded as required by Lemma 4.3 and Corollary 4.4.

We now bound the first integral in (?). Once again we use |y - x_j| = |x - x_j| + \big(|y-x_j| - |x-x_j| \big), (?) and (?) to show it suffices to bound the two expressions

The first is readily bounded as required using (?) of Lemma 4.3 with b=4.

It remains to bound (?). We simplify the calculation by denoting \sigma = \alpha + \beta and writing \sigma = \sigma_1 + \dots + \sigma_5 for some |\sigma_s|=1, s=1, \dots, 5. Define (in the same way as \beta_{<i} and \beta_{>i} above)

We now apply the same method used above, that is, we transfer successive first order derivatives via integration by parts. At each step we leave as a remainder the term where the derivative falls on \psi(x,\bold{\hat{x}}) \psi(y,\bold{\hat{x}}) \Phi(x,y,\bold{\hat{x}}). Since |\sigma|=5 is odd, the result after this procedure has occured five times is that (?) is precisely equal to minus the same integral plus remainder terms. This explains the \frac{1}{2}-factor in the following formula,

By (?), each of these integrals can be bounded by the right-hand side of (?) for a new constant C'. This completes the proof in the case where j \in P^c \cap S^c.

Finally, we must bound (4.8) for when j=k but where j \in P^* or j \in S^*. We assume that j \in P^*, the case of j \in S^* is similar. Hence we need only consider |\alpha|=1. To bound the integral, we first apply integration by parts to obtain

By Lemma 3.3 we have |x-x_j| < \delta/2 when \Phi(x,y,\bold{\hat{x}}) \ne 0. Along with (?) and (?) we then get

for some C. The two terms are just (?) and the right-hand side of (?) respectively, and hence can be suitably bounded. This completes all required cases.

We prove the bound for (?), the case of (?) is similar. We use the cutoff function \xi \in C^{\infty}_c(\mathbb{R}), introduced in (?), to define

z \in \mathbb{R}^3. Then \chi_1(z) \ne 0 implies |z| < 1/4 and \chi_2(z) \ne 0 implies |z| > 1/8.

We write the integral (?) as

Expanding this and using, for example, the comment before Lemma 3.6, the integral involving \chi_2 can be bounded in absolute value by some constant multiplying

We then use (?), (?) and that F is uniformly bounded on \mathbb{R}^{3N} to bound this integral by some constant multiplying (3.10) of Proposition 3.10 with, say, R=1/2.

We now consider the \chi_1 term of (?). Firstly, we recall the notation introduced in (?)-(?), namely we can write

Take any \theta \in (0,1). We know that \phi \in C^{1,\theta}(\mathbb{R}^{3N}). In particular, using (?) and (?) there exists a constant C such that when |x-x_j| < 1/4,

The constant C depends on \theta but is independent of x,y and \bold{\hat{x}}.

We now write the \chi_1 term in (?) as the sum of the following two integrals,

The second integral can be bounded by some constant multiplying

using (?) and that F is uniformly bounded. Lemma 4.3 can then be used to obtain the required bound.

It suffices to bound (?). Integration by parts in the variable x_j is used in (?) to remove a single derivative from \partial_{x_j}^{\alpha}|x-x_j|. Take any l \le \alpha with |l|=1. Integral (?) can therefore be rewritten as

We can use (?), (?) and the fact that both F and \nabla F are uniformly bounded, to bound this in absolute value by some constant multiplying

The relevant bound then follows by Lemma 4.3 and Corollary 4.4.