1
$\begingroup$

For the numerical method of second order PDE, we use semi-discrete(finite difference in $x$, continuous in $t$) method to translate PDE to a ODE system $$\dfrac{\textrm{d} U}{\textrm{d} t} = AU,$$ $$U(0) = U_0$$ here $$U(t) = (u_1(t),\dots,u_{J-1}(t))$$ is the vector of discrete numerical solution. $A$ is a constant coefficient matrix

My question is this ODE system can already be solved explicitly, why do we need the full-discrete method again i.e difference on time $t$ $$\dfrac{U^{n+1} - U^n}{k} = A U^{n+1},\quad 0\leq n\leq N-1$$ $$U^0 = U_0.$$ Here is $k = \Delta t.$

$\endgroup$
  • $\begingroup$ I'm not sure what you're asking. You might want to add some more information to your question. $\endgroup$ – mattos Jan 17 '17 at 8:29
1
$\begingroup$

Speaking of numerical methods, even if you have an explicit solution of the ode in question, namely $$ U(t) = \exp(tA)U_0 $$ that does not really help. You have to compute $\exp(tA)$ somehow. As - for general matrices $A$ - computing an eigendecomposition is very badly conditioned, the usual way is a bad idea. You have to approximate $\exp$ by a rational function $R$, giving you $$ U_{n+1} = R(kA)U_n $$ $k = \Delta t$ as above. This is the very same thing as applying a RK-method to your equation.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.