CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   Main CFD Forum (https://www.cfd-online.com/Forums/main/)
-   -   dissipative and dispersive error of finite difference methods (https://www.cfd-online.com/Forums/main/119656-dissipative-dispersive-error-finite-difference-methods.html)

 shubiaohewan June 21, 2013 04:29

dissipative and dispersive error of finite difference methods

Hi all,

We know that for the spatial derivative using a finite difference method, if the leading error term is odd, then the results contain dispersive error, while if the leading error term is even, then it has dissipative error.

I want to compare the degree of such errors. Say, the leading error term is 7th order (FD7), compared with the leading error term is 5th order (FD5), which one has more severe dispersive error? For sure, FD7 is more accurate, what's its performance with regard to dispersive error?

Thanks.

Shu

 FMDenaro June 21, 2013 06:29

Quote:
 Originally Posted by shubiaohewan (Post 435223) Hi all, We know that for the spatial derivative using a finite difference method, if the leading error term is odd, then the results contain dispersive error, while if the leading error term is even, then it has dissipative error. I want to compare the degree of such errors. Say, the leading error term is 7th order (FD7), compared with the leading error term is 5th order (FD5), which one has more severe dispersive error? For sure, FD7 is more accurate, what's its performance with regard to dispersive error? Thanks. Shu
This is a classical numerical analysis task, you can analyze the local truncation error of the discretization formula, it gives much information about the character of the error.
Furthermore, I suggest using the spectral analysis too, that is the modified wavenumber that each formula produces.
It is important to remark that not necessarily a higher order accurate formula performs better than a lower one on a given grid. Accuracy order is only an asymptotic estimation that does not tell you how the formula behaves for the chosen grid. Thus, the modified wavenumber analysis can help to understand the error distribution for the grid you want to use.

 cfdnewbie June 21, 2013 06:31

There is a standard procedure of analysing dispersive and dissipative behavior of FD schemes. Discretize a linear advection equation, plug in a wave with frequency k and amplitude a and check the resulting frequency and amplitude response - that gives you the dissipation and dispersion error of your scheme.

 FMDenaro June 21, 2013 06:46

Quote:
 Originally Posted by cfdnewbie (Post 435251) There is a standard procedure of analysing dispersive and dissipative behavior of FD schemes. Discretize a linear advection equation, plug in a wave with frequency k and amplitude a and check the resulting frequency and amplitude response - that gives you the dissipation and dispersion error of your scheme.
Such analysis can be extended, in a suitable way, also for the non linear equation (e.g., Burgers)

 cfdnewbie June 21, 2013 07:01

Hello Prof. Denaro,
I have only seen this done for linear equations. I assume that Burger's would lead to a non-linear system to solve for? Do you have any good reference on the details of this? that would be a very interesting read!

 sbaffini June 21, 2013 07:15

I don't know if Prof. Denaro is referencing what i have in mind; however, a possible practical (i.e., non analytical) approach is this:

http://www.sciencedirect.com/science...21999111001148

 cfdnewbie June 21, 2013 07:20

Thank you Paolo,
that looks very interesting!

 FMDenaro June 21, 2013 09:12

Quote:
 Originally Posted by sbaffini (Post 435266) I don't know if Prof. Denaro is referencing what i have in mind; however, a possible practical (i.e., non analytical) approach is this: http://www.sciencedirect.com/science...21999111001148

yes, there are some papers as this one that in these years analysed the non-linear equations.
We also worked on the wavenumber-based analysis for the one and multi-dimensional non-linear equations, using an analytical/numerical approach

http://onlinelibrary.wiley.com/doi/1...d.179/abstract
http://www.sciencedirect.com/science...21999111000933

 All times are GMT -4. The time now is 11:58.