How is the inductive hypothesis used here in order to estimate this integral?

118 Views Asked by At

In Sandra Cerrai, Normal deviations from the averaged motion for some reaction–diffusion equations with fast oscillating perturbation (Journal de Mathématiques Pures et Appliquées, Volume 91(6), 2009, pp.614-647) the author proves Lemma 4.2 (see p. 629) by induction. The thesis of the lemma is that for all $j$ positive integer there exists $C_{T,j}>0$ s.t. $$\left|\int_{[s, t]^j} \mathbb{E} \prod_{i=1}^j \vartheta_{\alpha, \beta}\left(t-r_i\right) \Psi_{\epsilon, h}\left(r_i / \epsilon\right) d r_1 \cdots d r_j\right| \leqslant c_{T, j}\left(1+|x|^j+|y|^j\right) \epsilon^{\frac{j}{2}}\left(\frac{(t-s) \alpha \wedge 1}{\alpha}\right)^{\frac{(1-2 \beta) j}{2}}|h|_H^j,$$ where $H$ is a Hilbert space, $\alpha>0$ and $\beta \in[0,1 / 3)$. For any $x, y, h \in H, r \geqslant 0$, $\epsilon>0$, and $\Psi_{\epsilon, h}(r):=\left\langle F\left(\bar{u}^x(\epsilon r), v^y(r)\right)-\bar{F}\left(\bar{u}^x(\epsilon r)\right), h\right\rangle_H, \quad \vartheta_{\alpha, \beta}(r):=e^{-r \alpha} r^{-\beta}.$ Here $\bar u^x(t)$ and $v^y(t)$, $t \leq T$ are some stochastic processes in $H$ following some stochastic differential equation starting from $x \in H$ and $y \in H$ respectively.

At p.642 the author estimates the term $I_{2,\epsilon}$ using the inductive hypothesis (see 641) and gets: $$I_{2, \epsilon} \leqslant c_{T, n}\left(1+|x|_H^{2 n}+|y|_H^{2 n}\right)|h|_H^{2 n} \epsilon^{-n} \alpha^{-n(1-2 \beta)}[(t-s) \alpha \wedge 1]^{n(1-2 \beta)}.$$ Note that $I_{2,\epsilon}$ was defined at p. 641 as $$\begin{align} I_{2,\epsilon}& = \sum_{j=1}^{n-1} \int_{\frac{s}{\epsilon}}^{r_{2 j}} \cdots \int_{\frac{s}{\epsilon}}^{r_2} \prod_{i=1}^{2 j} \vartheta_{\alpha, \beta}\left(t-\epsilon r_i\right)\left|\mathbb{E} \prod_{i=1}^{2 j} \Psi_{\epsilon, h}\left(t, r_i\right)\right| d r_1 \cdots r_{2 j} \\ & \quad \quad \quad \quad \times \int_{\frac{s}{\epsilon}}^{\frac{t}{\epsilon}} \int_{\frac{s}{\epsilon}}^{r_{2(n-j)}} \cdots \int_{\frac{s}{\epsilon}}^{r_2} \prod_{i=1}^{2(n-j)} \vartheta_{\alpha, \beta}\left(t-\epsilon r_i\right)\left|\mathbb{E} \prod_{i=1}^{2(n-j)} \Psi_{\epsilon, h}\left(t, r_i\right)\right| d r_1 \cdots r_{2(n-j)}. \end{align}$$ However how can the author do that? The inductive hypothesis (hence the thesis of the lemma with $j<2n$) is with the modulus outside the integrals and the terms inside the integral can be negative or positive being scalar products (a part from this it is ok as with a change of variable we get rid of $1/\epsilon$ in the intervals of integration and we get $\epsilon^{-2n}$ and then with the inductive hypothesis, if it can be used, we get $\epsilon^{n}$ so in turn we have $\epsilon^{-n}$). Does anyone has any ideas?