Methods for spline fitting for transcendental functions? How to place the knots?

85 Views Asked by At

I was thinking about the following problem the other day:

Fit a spline $s(t)$ to some transcendental function $f(t)$, so that:

$$s(t) = \cases{P_k(t), \text{ if } t_k \leq t \leq t_{k+1}}$$

For polynomials $P_k(t)$:

$$P_k(t) = \sum_{i=0}^{N} c_{ik}t^i$$

Now we seek $c_{ik}$ so that $s(t)$ approximates $f(t)$ well on interval $t\in [t_0,t_{max}]$.

How can we do this? How to choose the $t_k$ knot points?

Which boundary conditions should be obeyed there?

If the knot points were fixed then the problem could be approached like some least squares norm minimization:

$$\sum_l\|s(t_l)-f(t_l)\|_2^2 + \epsilon(\text{boundary terms})$$

All would be linear. But how to tackle the non-linearity that is introduced with not knowing where knot points should be placed?

1

There are 1 best solutions below

0
On BEST ANSWER

If your question is about convergence, then interpolating splines converge to the initial function if the interpolation step converges towards zero. See for example here.