Positive continuous additive functionals of multidimensional Brownian motion and the Brownian local time (Q2475053)

Let \(W_{t}\) be a \(N\)-dimensional brownian motion starting from 0, \(p_N\) its transition density, \(\varphi\) a function in the \({\mathcal D}\) of \({\mathbb{R}}^N\), \(\int_{\mathbb{R}^N}\varphi =1\), \(\varphi_{\varepsilon }=\varepsilon^{- N}\varphi (\cdot /\varepsilon )\), \({\mathbf D}_{2}^{s}\) be, for \(s>0\), the space of all \(F\in L^{2}\) for which, in their chaos representation \(F=\sum_{\mathbf{n}\in {\mathbb{Z}}_{+}^N}I_{\mathbf{n}}(f_{\mathbf{n}})\), we have \(\| F\|_{s}^{2}= \sum (1+| \mathbf{n}| )^{s}\mathbf{n}!\| f_{\mathbf{n}}\|^{2}<\infty\) and \({\mathbf D}_{2}^{- s}\) be its dual, where, for \(\mathbf{n}=(n_{1},\dots,n_N)\), \(\mathbf{n}!=\prod n_{i}!\). Let also \(H_{\mathbf{n}}(x_{1},\dots,x_N)=\prod H_{n_{i}}(x_{i})\), \(H_{n_{i}}\) being the Hermite polynomials. Let \(T\in {\mathcal D}'\) be a positive distribution, \(\mu_T\) the corresponding measure (\(\int\varphi d\mu_T=T(\varphi )\)), \[ g_{\mathbf{n}}(s,x)= n!^{-1}s^{-| \mathbf{n}| /2}H_{\mathbf{n}}(xs^{-1/2})p_N(s,x) (1_{[0,s]}\otimes^{| \mathbf{n}| }), \] \[ a_{\mathbf{n}}^{0}(t,x)= \int_0^t \int g_{\mathbf{n}}(s,z-x)\mu_T(dz)ds, \] \(A_T(t,x)=\sum I_{\mathbf{n}}(a_{\mathbf{n}}^{0}(t,x))\). If \(\sup_{y\in {\mathbf R}^N}\mu_T(B(y,r))<\infty\) for all \(r\), where \(B(y,r)=(\|\cdot -y\|\leq r)\), and \(\sup_{z\in B(x,r)}\int | y-z| ^{2-N-\eta }e^{-\delta | y-z| ^{2}}\mu_T(dy)< \infty\) for all \(\delta >0\) and small enough \(\eta >0\),\(r>0\), the first theorem states that, for \(\alpha <1-(N/2)\), in \({\mathbf D}_{2}^{\alpha }\), \[ \lim_{\varepsilon\to 0} \int_0^t (T\ast\varphi_{\varepsilon })(W_{s}+x)ds=A_T(t,x). \] The local times \(L(t,x)\in {\mathbf D}_{2}^{\alpha }\), \(x\neq 0\), are introduced as \(\lim_{\varepsilon\to 0}\int_0^t \varphi_{\varepsilon }(W_{s}-x)ds\) [\textit{P. Imkeller} and \textit{F. Weisz}, Probab. Theory Relat. Fields 98, No. 1, 47--75 (1994; Zbl 0794.60046)] and, if also (condition 3.3), for all \(\delta >0\) and small enough \(\eta >0\), \(\int | y-x| ^{2-N-\eta }e^{-\delta | y-x| ^{2}}\mu_T(dy)<\infty\), we have \(A_T(t,x)=\int L(t,y-x)\mu_T(dy)\). Let now \(A\) be a positive functional of \(W\) in the usual sense, \(\nu_{A}\) be its Revuz measure (defined by \(\int fd\nu_{A}=\int E(\int_{_{0}}^{^{1}}f(W_{t}+x)d_{t}A(t,x))dx)\). If \(E(A(t,x)^{2})<\infty\) and \(\int E((\int_{_{r}}^{^{t}}\chi_{R}(W_{u}+ x) d_{u}A(u,x))^{2})^{1/2}dx<\infty\), where \(\chi_{R}\in {\mathcal D}\), \(0\leq\chi_{R}\leq 1\), \(\chi_{R}(x)=1\) for \(| x| \leq R\), \(=0\) for \(| x| \geq R+1\), and if \(\nu_{A}\) satisfies 3.3 then (second theorem) \[ A(t,x)=\int L(t,y-x)\nu_{A}(dy). \] If also \(T\in {\mathcal D}'\) is positive with \(\mu_T=\nu_{A}\) then \(A=A_T\). The proof of the second theorem goes through four lemmas and two (having also independent interest) propositions.