4
$\begingroup$

I have a (nonlinear) function which takes as input 4 parameters and produces a real number as output. It is quite complex to compute the function value given a set of parameters (as it requires a very big summation).

I'd like to answer queries on this function efficiently so I was thinking of trying to use some interpolation methods. I have used Chebyshev polynomials to interpolate univariate functions, but I haven't been able to find (or understand) anything on interpolating multivariate functions. I'm not set on using Chebyshev polynomials, I have just had some exposure to them and know they tend to be efficient (in terms of # of necessary coefficients and interpolation error).

I was wondering if anyone could give me (an engineer) any pointers for how to go about interpolating a multi-variate function? Simple examples or sample code would be awesome, but I'll take any attempts to explain how interpolation would work in higher dimensions, including (readable) references.

  • 1
    Have you seen [this](http://dx.doi.org/10.1090/S0025-5718-1992-1122061-0)? One possibility you can try is to take suitably scaled Chebyshev nodes in each variable (essentially, a tensor product) and then perform a method for multidimensional polynomial interpolation, like in the paper I linked to.2011-12-16
  • 0
    @J.M. Thanks for the reference, I'll check it out.2011-12-16

2 Answers 2