Consider the following equation, subject to homogeneous Neumann boundary condition. $$ u_t = \Delta u + f(u). $$ The weak formulation is as follows: $$ (u_t,w) = (\Delta u, w) + (f(u),w), \quad \forall w \in H^1(\Omega). $$ The standard finite element scheme is: $$ (u_{h,t},w_h) = (\Delta_h u_h, w_h) + (f(u_h),w), \quad \forall w_h \in V^r_h(\Omega), $$ where the $V^r_h$ denotes the space of polynomials of degree at most r. And the mass lumping method scheme is $$ (u_{h,t},w_h)_h = (\Delta_h u_h, w_h)_h + (f(u_h),w)_h, \forall w_h \in V^r_h(\Omega), $$


$$ (u_{h,t},w_h)_h = (\Delta_h u_h, w_h) + (f(u_h),w)_h, \forall w_h \in V^r_h(\Omega), $$

where $(\cdot, \cdot)_{h}$ signifies the use of the mass lumping method for integration. The main difference between the above two mass lumping methods is the assembly of the stiffness matrix.

Defined $P_h$$L^2(\Omega) \to V_h(\Omega)$ denote the $L_2$-projection onto the space of $V_h^r$ , and $\Pi_h: C(\bar{\Omega}) \to V^r_h$ denote the interpolation by a function in $V^r_h$ through the nodes of mass lumping integration nodes. What I want to do is construct the error equation $e_h = u_h - \Pi_h u$, then $$ \partial_t \Pi_h u(t) - \Delta_h \Pi_h u(t) = \Pi_h f(u) + \underbrace{\Pi_h \Delta u(t) - \Delta_h \Pi_h u (t)}_{g_h(u)}. $$

$$ \partial_t \Pi_h u(t) - \Delta_h \Pi_h u(t) = \Pi_h f(u) + \underbrace{P_h \Delta u(t) - \Delta_h \Pi_h u (t)}_{g_h(u)}. $$

In the paper https://link.springer.com/article/10.1007/s10915-021-01746-y Lemma 2.4 proof that
$$ (\Pi_h \Delta u - \Delta_h \Pi_h u , w_h) \leq h^{r+1}||u||_{2r+2} ||w_h||_1 $$ Consider the $g_h(u)$ in the second equation: $$ (g_h(u),w_h) = (P_h \Delta u - \Delta_h \Pi_h u, w_h) = (P_h \Delta u - \Delta u,w_h) + (\Delta u - \Delta_h \Pi_h u,w_h). $$

$$ =(\Delta_h R_h u - \Delta u, w_h) + (\Delta u - \Delta_h \Pi_h u,w_h) $$

$$ =(\Delta u - \Delta_h \Pi_h u,w_h). $$ where the $R_h$ is the Ritz projection. These two schemes seem to be equivalent, but when I use $(\cdot,\cdot)_h$ to integrate the stiffness matrix, the convergence order will be reduced in some cases. Did I go wrong somewhere?

  • $\begingroup$ How exactly are $\Pi_h$ and $R_h$ defined? $\endgroup$ Nov 7, 2023 at 17:08
  • $\begingroup$ @WolfgangBangerth I fixed my problem, now $\Pi_h$ is the Lagrange interpolation operator, which interpolation by a function in $V^r_h$ through the nodes of mass-lumping intergrate nodes. For example, in 1D case, it means interpolate a function through the Gauss-Lobatto nodes. And $P_h$ is a standard L2 projection. $R_h$ is the Ritz projection. $\endgroup$
    – Owen Jun
    Nov 8, 2023 at 3:16


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.