Chapter 11 · Introduction

Chapter 11 — Numerical ODEs — Picard, Euler, RK

How to compute the solution of an initial value problem that has no formula: Picard's successive approximations and the Taylor series method, Euler, modified Euler and Runge–Kutta methods, the order and the global error of a method, absolute stability and stiffness, and systems of equations.

In the first volume of his Institutiones calculi integralis (1768) Euler faced the question every differential equation eventually raises: if $y'=f(x,y)$ cannot be integrated in closed form, how do we find $y$? His answer was disarmingly simple. Over a short step $h$ the slope hardly changes, so walk along the tangent, $y_{n+1}=y_n+hf(x_n,y_n)$, then recompute the slope and walk again. The polygon this draws approaches the true solution as $h\to0$ — but slowly: halving the step only halves the error. In 1890 Picard turned a different idea into an existence proof, replacing the equation by the integral equation $y=y_0+\int f(t,y)\,dt$ and iterating it; his successive approximations are both the proof of the Picard–Lindelöf theorem of Chapter 1 and a method for computing the first terms of a solution's Taylor series.

The breakthrough in efficiency came from Germany at the turn of the twentieth century. Carl Runge (1895) noticed that sampling the slope at the midpoint of a step, not just at its start, gains a whole power of $h$; Karl Heun (1900) and Wilhelm Kutta (1901) worked out how to combine several such slope samples, and Kutta's classical fourth-order method — four slopes per step, error divided by sixteen when the step is halved — is still the default in many codes, from orbit calculations to computer games. Half a century later a new difficulty appeared. Chemists modelling fast and slow reactions (Curtiss and Hirschfelder, 1952) found that explicit methods produced wild oscillations unless the step was absurdly small, even where the solution was smooth: the problem was stiff, and Dahlquist's theory of stability (1963) explained why. This chapter follows that whole path: from Picard's iterates to Runge–Kutta methods, and from accuracy to stability.

y′ = f(x, y), y(x₀) = y₀→Picard: φₙ₊₁ = y₀ + ∫ f(t, φₙ) dt→Taylor series method→Euler: order 1→modified Euler (Heun): order 2→RK4: order 4→local O(hᵖ⁺¹) → global O(hᵖ)→y′ = λy: |R(hλ)| ≤ 1→stiffness · systems
🎯 By the end of this chapter you can…
  • compute Picard's successive approximations for an initial value problem, relate them to the Taylor polynomials of the solution, and use the Taylor series method of a given order;
  • carry out steps of Euler's method, the modified Euler (Heun) and midpoint methods and the classical fourth-order Runge–Kutta method by hand, and recognise the different methods that go by the name "modified Euler";
  • define the local truncation error and the order of a one-step method, find the order of a given method by Taylor expansion, and prove that a method of order $p$ has global error $O(h^p)$;
  • estimate the error of a computed value by comparing steps $h$ and $h/2$ (Richardson), and choose a step for a required accuracy;
  • find the stability function and the interval of absolute stability of Euler, the second-order methods and RK4 for $y'=\lambda y$, explain stiffness and the advantage of implicit methods such as backward Euler, and apply all the methods to systems and higher-order equations.
Teaser — Step by Step: Euler, Modified Euler, RK4 and Picard

Three methods march from the same initial value with the same step $h$: Euler (one slope per step), modified Euler (Heun: two slopes) and RK4 (four slopes). The solid curve is the exact solution. Press Play to watch the steps, change $h$ and read the errors at the end point. Switch on a Picard iterate $\varphi_k$ to see successive approximations hug the solution near $x_0$ first. Then choose $y'=-8y$ and try $h=0.25$ and $h=0.5$: an accurate method is not always a stable one. Section 11.1 explains every curve.

exact y(X)—
Euler error—
mod. Euler error—
RK4 error—

How this is examined

The numerical solution of ODEs is a dependable source of short computational questions in the CSIR NET Unit 3 syllabus. In Part B expect to compute the second or third Picard approximation for a simple initial value problem, or the coefficient of a power of $x$ in it; to perform one or two steps of Euler's method, the modified Euler method or the classical Runge–Kutta method and report $y_1$ or $y_2$; to give the order or the leading local truncation error of a stated method; to find the largest step $h$ for which Euler's method (or another method) is stable for $y'=\lambda y$ with $\lambda\lt0$; or to rewrite a second-order equation as a system and take one step. Part C, where several options may be correct, tests the theory: how the global error scales with $h$ for methods of different orders, that a method can be consistent yet unstable for a given step, that RK4 reduces to Simpson's rule when $f$ does not depend on $y$, that Picard's iterates may diverge outside the interval where the solution exists, and that different books mean different formulas by "modified Euler". The wrong options are usually one of the counterexamples in the ⚠️ box of Section 11.1. GATE MA asks very similar questions, frequently with numerical answers.

Sections in this chapter

✅ Before you start

From the CSIR NET Unit 1 course you need: Taylor's theorem in one and two variables (every order condition is a comparison of Taylor expansions), the mean value theorem, uniform convergence of sequences of functions, and, for systems, eigenvalues of matrices. From this course: the Lipschitz condition, Picard–Lindelöf theorem, Picard iterates and continuous dependence on initial data of §1.1 — the global error bound is a discrete version of the same Gronwall argument; systems of first-order equations and the reduction of higher-order equations of §1.3; the trapezoidal and Simpson rules and Richardson extrapolation of §10.1, which reappear inside Heun's method, RK4 and the error estimate; and order of convergence and fixed-point iteration from §7.1, used for iterated correctors and implicit methods. If this is your first visit, the welcome page explains how each section is organised.