Equivalent definitions of Brownian local time

brownian motionprobabilityprobability theorystochastic-calculusstochastic-processes

To keep this question simple, we only consider the local time of a standard Brownian motion $B_t$ at zero, we denote it as $L_t^0$. Generally, we define $$\tag{1} L_t^0(B)=\lim_{\varepsilon\to\infty}\dfrac{1}{2\varepsilon}\int_0^t1_{(-\varepsilon,\varepsilon)}(B_s)\,ds. $$
Another definition is given by $$L_t^0(B)= |B_t|-\int_0^t\text{sgn}(B_s)\,dB_s\tag{2}.$$
We can approximate $|B_t|$ pointwise by $\sqrt{\varepsilon+B_t^2}$ by taking $\varepsilon\to0$. Applying Ito formula to $\sqrt{\varepsilon+B_t^2}$ we get, $$\sqrt{\varepsilon+B_t^2}=\sqrt{\varepsilon}+\int_0^t\dfrac{B_s}{\sqrt{\varepsilon+B_s^2}}\,dB_s+\dfrac{1}{2}\int_0^t\dfrac{\varepsilon}{(\varepsilon+B_s^2)^{3/2}}\,ds. $$
If we take the limit, the above formula suggests that $$L_t^0(B)=\lim_{\varepsilon\to0}\dfrac{1}{2}\int_0^t\dfrac{\varepsilon}{(\varepsilon+B_s^2)^{3/2}}\,ds\tag{3}.$$
My question is can we show that by direct computation that equation (3) and (1) are the same?

Best Answer

It may depend on what you mean by "the same". By the occupation time formula for the Brownian local time $$ \int_0^t{\epsilon\over (\epsilon+B_s^2)^{3/2}}\,ds = \int_{-\infty}^\infty {\epsilon\over(\epsilon+x^2)^{3/2}}L_t^x\,dx=\int_{-\infty}^\infty {1\over (1+u^2)^{3/2}}L^{\sqrt{\epsilon}u}_t\,du. $$ Because $$ \int_{-\infty}^\infty {1\over (1+u^2)^{3/2}}\,du=2, $$ sending $\epsilon$ to $0$ on the right side of (3) therefore yields $L^0_t$ because of the continuity in $x$ of $L^x_t$.