CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)

 zhanglei January 5, 2001 10:19

how to compute the gradient of a function

the value of the function f is known on each grid point, to approximate the gradient of the function f, i.e. (df/dx, df/dy) which is the good choice, central difference? or one-sided difference? or some more complicate form?

 John C. Chien January 5, 2001 13:02

Re: how to compute the gradient of a function

(1). It depends on the nature of this function. (2). In principle, the central difference is more accurate than the one-sided difference, based on the Taylor series expansion. (2). Most numerical analysis books have a chapter on the numerical interpolation. And if you are dealing with test data, with some random distributions, then you will have to create a new approxmate smooth function first, in order to derive the general trend or gradient.

 peter.zhao January 10, 2001 09:11

Re: how to compute the gradient of a function

For finite differnece method,the central difference is usually used to calculate the gradient on account of the accuracy.Meanwhile,for finite volume method,it is better to use Gauss integral theorem over a control unit to calculate the gradient than central differnce.For example,when you calculate the derivative of velocity on the cell interface for viscous term in the N-S equations,the central differnce only takes account of one direction,i.e.you deal with the thin layer N-S eq or PNS,but the Gauss integral considers all direction,i.e.you deal with the full N-S eq.

 All times are GMT -4. The time now is 14:01.