Self-intersection local time: critical exponent, large deviations, and laws of the iterated logarithm (Q1769512)

The article is devoted to the moment estimation and investigation of the limit behavior for the process \[ \beta _t=\lim _{\varepsilon \rightarrow 0+}\left[\int_0^t\int_0^s \varphi_\varepsilon (X_s-X_u)\,du\,ds-E\int_0^t\int_0^s \varphi_\varepsilon (X_s-X_u)\,du\,ds\right]. \] Here \(X_t\) is the planar Brownian motion and \(\varphi_\varepsilon\) approximates \(\delta_0-\)function when \(\varepsilon \rightarrow 0+.\) The authors prove that the critical exponent \(\gamma_\beta\) such that \(Ee^{\gamma \beta_1}=+\infty, \gamma >\gamma_\beta,\) and \(Ee^{\gamma \beta_1}<+\infty, \gamma <\gamma_\beta ,\) is equal to the best constant in the Gagliardo-Nirenberg inequality \(\| f\| _4\leq C\sqrt{\| \nabla f\| _2}\sqrt{\| f\| _2}\) for the smooth functions on \(R^2.\) Besides this they establish the large deviations asymptotic for \(\beta _1\) and the law of iterated logarithm for \(\beta _t\) under \(t\rightarrow +\infty.\) To prove these results the authors use the representation \(\beta _t\) as a sum of the summands of the two different types. One of them corresponds to the intersection between the two independent Brownian motions and another one has the same distribution as a scaled \(\beta _1.\) Then they use the Feynman-Kac approach which allows to get the precise asymptotic for the studied probabilities.