Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
NCTM or The Math Forum.



Re: ? influence of temporal discretization
Posted:
May 30, 2012 10:12 AM


On Fri, 25 May 2012 09:41:02 +0000, AMX wrote:
> On Sat, 19 May 2012 05:02:55 0700 (PDT), Cosine <asecant@gmail.com> > wrote: >> Hi: >> >> Suppose we have a diffusion eqn: dT/dt = a*div(grad(T))+Q in non >> dimensional form. >> > [..] > >> Can we discretize the time using dt as 0.1 sec or should we use a dt >> even smaller? How do we estimate the appropriate value of dt? > > It mostly depends on method of integrating over time. > > AMX
For any method of time integration, you can measure the order wrt dt and get an estimate of the error by setting the spatial intervals very small, and trying with dt, dt/2 and dt/4, choosing some resulting quantity as the test, say q1, q2, q4, resp. Then q1  q4 is close to the error, and log_2 of the ratio (q1  q2) / (q2  q4) gives you the order wrt dT.  Dieter Britz



