Spline Interpolation: Cubic Splines Explained Simply

A spline is a smooth curve through data points built from a different low-degree polynomial on each interval, joined so smoothly you cannot see the seams. A cubic spline uses cubic pieces matching values, slopes, and curvatures at every joint (knot) — like the flexible drafting strip engineers once bent through pins to draw ship hulls.

Single high-degree polynomials through many points oscillate badly (the Runge phenomenon), while straight-line joins look kinked. Cubic splines give the best of both: they pass through every point, stay smooth, and wiggle as little as possible — the natural cubic spline is the smoothest interpolant in a precise energy-minimizing sense.

Key Concepts

• Knot: a data point (x_i, y_i) where two polynomial pieces meet.
• Cubic piece: on [x_i, x_i+1] the spline is S_i(x) = a_i + b_i(x − x_i) + c_i(x − x_i)² + d_i(x − x_i)³.
• Smoothness conditions: at each interior knot, pieces share value, slope, and curvature (second derivative).
• Natural spline: the second derivative is set to zero at both endpoints, meaning the curve straightens out at the ends — the most common default choice.

Formulas

• Second-derivative form: with M_i = S″(x_i), each piece is S_i(x) = M_i(x_i+1 − x)³/(6h_i) + M_i+1(x − x_i)³/(6h_i) + (y_i − M_ih_i²/6)(x_i+1 − x)/h_i + (y_i+1 − M_i+1h_i²/6)(x − x_i)/h_i, where h_i = x_i+1 − x_i.
• Tridiagonal system: the smoothness conditions give one linear equation per interior knot linking M_i−1, M_i, M_i+1 — solvable in linear time.
• Natural boundary conditions: M₀ = 0 and M_n = 0, which close the system.

Worked Example

Build the natural cubic spline through (0, 0), (1, 1), (2, 4) with spacing h = 1. Natural conditions: M₀ = M₂ = 0. The interior equation: 4M₁ = 6[(4 − 1) − (1 − 0)] = 12, so M₁ = 3. On [0, 1]: S₀(x) = 3x³/6 + (1 − 3/6)x = 0.5x³ + 0.5x. On [1, 2] with t = x − 1: S₁(x) = 0.5(1 − t)³ + 0.5(1 − t) + 4t. Checks: S₀(1) = S₁(1) = 1 (values match); S₁(2) = 4 (endpoint matches); slopes S₀′(1) = 1.5 + 0.5 = 2 and S₁′(0) = −1.5 − 0.5 + 4 = 2 (slopes match). The join is perfectly smooth.

Common Misconceptions

• "A spline is one big polynomial." It is piecewise — a different cubic on each interval. That is precisely why it avoids the wild oscillations of a single high-degree polynomial.
• "Splines only approximate the data points." An interpolating spline passes through every point exactly; the smoothness is in how it travels between them.
• "The natural spline is always the best choice." It is a good default, but if you know the endpoint slopes, a clamped spline (prescribed first derivatives at the ends) is more accurate there.

Real-World Applications

• Computer-aided design: car bodies and aircraft wings are shaped with spline surfaces.
• Font rendering: the curves of every letter on your screen are spline-based outlines.
• Animation: camera paths and character motion are smoothed with splines between keyframes.
• Geography: elevation maps interpolate scattered survey measurements with splines.

Cubic splines join piecewise cubics with matching value, slope, and curvature at every knot, giving a curve that passes through all data points with minimal wiggle. The natural spline sets the endpoint curvatures to zero, and the whole construction reduces to one efficient tridiagonal solve.