This is an old revision of the document!


Spline

Modified in 0.9.1

f = InterpolatingFunction.Spline(x, y) or f = InterpolatingFunction.Spline(x, y, yp1, ypN) makes a cubic spline through the points $(x_i, y_i)$, an interpolating function of type “Spline”. On every interval $[x_i, x_{i+1}]$ it is a cubic polynomial, and the function, its first and its second derivative are continuous. With two arguments the second derivative is 0 at both ends (a natural spline); with four, the slopes at the ends are yp1 $= f'(x_1)$ and ypN $= f'(x_N)$.

Input

  • x : a table of $N \geq 2$ different real numbers, the points. They need not be sorted; the points are sorted together with their values.
  • y : a table of $N$ real numbers, the values at the points.
  • yp1, ypN : (optional) Reals, the first derivative at the first and at the last point (after sorting).

Output

  • f : an interpolating function of type “Spline”. f.x and f.y are the sorted points and values, f.knots the points, f.degree is 3. Modified in 0.9.1

The natural spline is, of all functions through the points with a continuous second derivative, the one with the least curvature (the smallest integral of $f''^2$). Its error between the points decreases as $h^4$ with the distance $h$ between the points when $f''$ of the sampled function is 0 at the ends; otherwise the error near the ends decreases as $h^2$. Give the end slopes when they are known. Through two points the natural spline is a straight line. A spline reproduces straight lines exactly, and with the exact end slopes every cubic polynomial.

Each data point influences the whole spline: near a jump in the data the spline overshoots and ripples, see the example on the interpolating function page. Use InterpolatingFunction.HermiteCSpline() for data that must not overshoot.

Example

$\cos(x)$ sampled at 7 points on $[0, \pi]$. The natural spline sets $f''(0) = 0$, but $\cos''(0) = -1$; with the correct end slopes $f'(0) = f'(\pi) = 0$ the error is 70 times smaller.

Input

Example.Quanty
-- cos(x) sampled at 7 points on [0, pi], interpolated by cubic splines
x, y = {}, {}
for i = 0, 6 do x[i+1] = i*math.pi/6; y[i+1] = math.cos(x[i+1]) end
natural = InterpolatingFunction.Spline(x, y)              -- f'' = 0 at both ends
clamped = InterpolatingFunction.Spline(x, y, 0, 0)        -- f'(0) = 0, f'(pi) = 0
 
-- the largest error between the data points, of the function and of its derivative
for _, c in ipairs({{"natural", natural}, {"clamped", clamped}}) do
  local f, e0, e1 = c[2], 0, 0
  for i = 0, 600 do
    local t = i*math.pi/600
    e0 = math.max(e0, math.abs(f(t) - math.cos(t)))
    e1 = math.max(e1, math.abs(f(t, 1) + math.sin(t)))
  end
  print(string.format("%s: max |f - cos| = %.1e, max |f' + sin| = %.1e, f''(0) = %6.3f",
                      c[1], e0, e1, f(0, 2)))
end
print(string.format("integral from 0 to pi/2: %.6f (exact 1)", InterpolatingFunction.Integrate(clamped, 0, math.pi/2)))
Example.out
natural: max |f - cos| = 1.4e-02, max |f' + sin| = 1.5e-01, f''(0) =  0.000
clamped: max |f - cos| = 2.0e-04, max |f' + sin| = 1.2e-03, f''(0) = -1.023
integral from 0 to pi/2: 0.999885 (exact 1)

Table of contents

Print/export