Title: A novel second-order midpoint approximation formula for Riemann-Liouville derivative: algorithm and application

Abstract: Compared to the the classical first-order Grw\"unwald-Letnikov formula at time $t_{k+1} (\textmd{or}\, t_{k})$, we firstly propose a second-order numerical approximate scheme for discretizing the Riemann-Liouvile derivative at time $t_{k+\frac{1}{2}}$, which is very suitable for the use of the Crank-Niclson technique to the time-fractional differential equations, the formula has the form $$ \begin{array}{lll} \displaystyle \,_{\mathrm{RL}}{{{\mathrm{D}}}}_{0,t}^{\alpha}u\left(t\right)\left|\right._{t=t_{k+\frac{1}{2}}}= \tau^{-\alpha}\sum\limits_{\ell=0}^{k} \varpi_{\ell}^{(\alpha)}u\left(t_k-\ell\tau\right) +\mathcal{O}(\tau^2),\,\,k=0,1,\ldots. \end{array} $$ where the coefficients $\varpi_{\ell}^{(\alpha)}$ $(\ell=0,1,\ldots,k)$ can be determined via the following novel generating function
$$ \begin{array}{lll} \displaystyle G(z)=\left(\frac{3\alpha+1}{2\alpha}-\frac{2\alpha+1}{\alpha}z+\frac{\alpha+1}{2\alpha}z^2\right)^{\alpha},\;|z|<1. \end{array} $$ Next, applying this formula to one- and two-dimensions time fractional Cable equation with Riemann-liouville derivative and get two high-order compact finite difference schemes. The numerical methods are proved to be uniquely solvable, stable and convergent with orders $\mathcal{O}(\tau^2+h^4)$ and $\mathcal{O}(\tau^2+h_x^4+h_y^4)$ by the energy method, where $\tau$ is the temporal stepsize and $h$, $h_x$, $h_y$ are the spatial stepsizes, respectively. Finally, some numerical experiments are provided to show that the algorithms have second- and fourth-order accuracy in temporal and spatial directions, respectively.