Q7Advanced Engineering Mathematics

Ordinary Differential Equations (ODEs) are ubiquitous in engineering, modeling dynamic systems ranging from electrical circuits to fluid flows. When an exact analytical solution is impossible to find, numerical methods are indispensable. While single-step methods like Euler's or Runge-Kutta calculate the next value using only the immediately preceding point, multi-step methods like Milne's method utilize a history of several past points to achieve significantly higher accuracy. Milne's method is a classic "Predictor-Corrector" algorithm, consisting of two distinct mathematical phases.

1. The Predictor Phase (Extrapolation)

The algorithm begins with the Predictor formula. This formula essentially fits a polynomial (using Newton's Forward Difference interpolation formula integrated over a specific interval) through the previous four known data points and extrapolates forward to predict the function's value at the next step. Let the ODE be $\frac{dy}{dx}=f(x,y)$ with step size $h$.

Milne's Predictor Formula, an open-type integration formula, is mathematically stated as:

To calculate the next value $y_{n+1}$, this formula strictly requires the four preceding function values ($y_{n-3},y_{n-2},y_{n-1},y_n$) and their corresponding derivative values ($f_{n-2},f_{n-1},f_n$). Because an initial value problem only provides one point $(x_0,y_0)$, a multi-step method is NOT self-starting. We must first utilize a high-accuracy single-step method, such as the 4th-order Runge-Kutta method, or Taylor series expansion, to accurately generate the required starting values $y_1,y_2,$ and $y_3$.

2. The Corrector Phase (Interpolation)

The value $y_{n+1}^{(P)}$ generated by the Predictor is an approximation. It is subject to truncation error. To refine this approximation, we transition to the Corrector phase. First, we use the predicted $y$ value to estimate the derivative at the new point: $f_{n+1}^{(P)}=f(x_{n+1},y_{n+1}^{(P)})$.

Next, we apply Milne's Corrector Formula, which is based on Simpson's 1/3 integration rule (a closed-type formula):

This formula "corrects" the predicted value by incorporating the newly estimated derivative $f_{n+1}^{(P)}$. The process is iterative. We take the corrected value $y_{n+1}^{(C)}$, plug it back into the differential equation to get a better estimate of $f_{n+1}$, and apply the Corrector formula again. This iterative loop continues until two successive corrected values converge (meaning their difference falls below a pre-defined tolerance level). Once convergence is achieved, the value is locked in as the final $y_{n+1}$, and the algorithm steps forward to calculate $y_{n+2}$.

3. Application to the Given Problem

The specific problem asks to solve $\frac{dy}{dx}=x-y^2$ with $y(0)=0$ to find $y(0.8)$ and $y(1.0)$. Assuming a step size $h=0.2$, the $x$ values are $x_0=0,x_1=0.2,x_2=0.4,x_3=0.6,x_4=0.8,x_5=1.0$.

Step 1: Use Taylor series or Runge-Kutta to calculate highly accurate values for $y_1(0.2),y_2(0.4),$ and $y_3(0.6)$.

Step 2: Use Milne's Predictor formula (with $n=3$) to predict $y_4^{(P)}$ at $x=0.8$.

Step 3: Calculate $f_4=0.8-(y_4^{(P)}{)}^2$.

Step 4: Use Milne's Corrector formula to refine $y_4$ until it converges.

Step 5: Repeat the Predictor-Corrector cycle with $n=4$ to find the final desired value $y_5$ at $x=1.0$.