using System; /// /// Simpson's rule with the number of integration steps determined dynamically. /// Used for integrating a smooth function over a finite interval. /// The number of integration points is doubled at each step, while avoiding redundant function evaluations, until tolerance is met. /// public class SimpsonIntegrator { /// /// Simplest interface, returning no diagnostic data /// /// integrand /// left limit of integration /// right limit of integration /// relative error tolerance /// Integral value static public double Integrate(Func f, double a, double b, double desiredRelativeError) { int log2MaxFunctionEvals = 20; int functionEvalsUsed; double estimatedError; return Integrate(f, a, b, desiredRelativeError, log2MaxFunctionEvals, out functionEvalsUsed, out estimatedError); } /// /// Full interface. Gives a posteriori information, namely the mumber of function evaluations and a conservative estimate of the error. /// /// integrand /// left limit of integration /// right limit of integration /// relative error tolerance /// If this value is N, the number of function evaluations will be less than 2^n + 1 /// Actual number of function evaluations used /// Conservative error estimate /// Integral value static public double Integrate(Func f, double a, double b, double relativeErrorTolerance, int log2MaxFunctionEvals, out int functionEvalsUsed, out double estimatedError) { double integral = 0.0; double mostRecentContribution = 0.0; double previousContribution = 0.0; double previousIntegral = 0.0; double sum = 0.0; functionEvalsUsed = 0; estimatedError = double.MaxValue; for (int stage = 0; stage <= log2MaxFunctionEvals; stage++) { if (stage == 0) { sum = f(a) + f(b); functionEvalsUsed = 2; integral = sum * 0.5 * (b - a); } else { // Pattern of Simpson's rule coefficients: // // 1 1 // 1 4 1 // 1 4 2 4 1 // 1 4 2 4 2 4 2 4 1 // ... // // Each row multiplies new function evaluations by 4, and the evalutations from the previous step by 2. int numNewPts = 1 << (stage - 1); mostRecentContribution = 0.0; double h = (b - a) / numNewPts; double x = a + 0.5 * h; for (int i = 0; i < numNewPts; i++) mostRecentContribution += f(x + i * h); functionEvalsUsed += numNewPts; mostRecentContribution *= 4.0; sum += mostRecentContribution - 0.5 * previousContribution; integral = sum * (b - a) / ((1 << stage) * 3.0); // Require at least five stages to reduce the risk of incorrectly declaring convergence too soon. // Note that you can specify fewer stages, but the early termination rule below will not be used. if (stage >= 5) { estimatedError = Math.Abs(integral - previousIntegral); // conservative if (estimatedError <= relativeErrorTolerance * Math.Abs(previousIntegral)) return integral; } previousContribution = mostRecentContribution; previousIntegral = integral; } } return integral; } }