Introduction to Direction Fields, Euler's Method, Error, and Stability

1Introduction

This lecture explains how to use direction fields to identify qualitative behavior when an analytic solution is unavailable, construct approximate solutions with Euler's method, and distinguish accuracy from numerical stability.

2Problem Setting

Consider the standard initial value problem

A direction field assigns a line segment of slope $$f(x,y)$$f(x,y) to each point $$(x,y)$$(x,y). Euler's method is a numerical method that advances along a short linear approximation determined by the slope at the current point.

3Rationale for the Method

Differentiation supplies a local linear approximation. Evaluating $$y^{\prime }=f(x,y)$$y'=f(x,y) at the current point therefore indicates the direction in which $$y$$y changes over a short interval. A direction field reveals global qualitative tendencies without solving the equation, whereas Euler's method repeatedly applies the corresponding local approximation.

4Euler Update and Error Order

Given the initial value $$y\left(x_0\right)=y_0$$y(x_0)=y_0 and a step size $$h$$h, Euler's method uses

Fix a finite interval contained in the maximal interval of existence. Assume that the exact solution is twice continuously differentiable there and that $$|y^{\prime }\text{'}|$$|y''| is bounded. The Taylor expansion

shows that Euler's method retains only the constant and linear terms. In this lecture, the discrepancy produced by taking one Euler step from an exact value is called the one-step defect. Under this convention, it is $$O\left(h^2\right)$$O(h^2). Under the alternative convention in which the defect is divided by $$h$$h and called the local truncation error, its order is $$O\left(h\right)$$O(h).

Assume in addition that $$f$$f is Lipschitz continuous in $$y$$y on a region containing both the exact solution and the numerical iterates. Controlling the amplification of successive errors by the Lipschitz constant gives a global error of $$O\left(h\right)$$O(h) on the fixed finite interval. This conclusion must not be asserted without the required smoothness and Lipschitz hypotheses.

5Examples

5.1A Growing Solution

For

and $$h=0.1$$h=0.1, Euler's method gives $$y_{n+1}=1.1y_n$$y_{n+1}=1.1y_n. Hence $$y_1=1.1$$y_1=1.1 and $$y_2=1.21$$y_2=1.21. Comparison with the exact solution $$e^x$$e^x shows that reducing the step size improves the approximation.

5.2A Stability Restriction

Consider

The exact solution decays, whereas Euler's method gives

The numerical solution decays only if $$|1+ah|<1$$|1+ah|<1. For example, if $$a=-10$$a=-10 and $$h=0.3$$h=0.3, then $$1+ah=-2$$1+ah=-2. Although the exact solution decays, the numerical solution oscillates with increasing magnitude.

6Accuracy Versus Stability

Accuracy describes how rapidly the approximation approaches the exact solution as the step size decreases. Stability describes whether the iteration amplifies or suppresses errors. Even a high-order method may be unstable for an unsuitable combination of problem and step size.

7Connection to Implicit Methods

Forward Euler uses the slope $$f(x_n,y_n)$$f(x_n,y_n) at the current point. Backward Euler instead uses the slope $$f(x_{n+1},y_{n+1})$$f(x_{n+1},y_{n+1}) at the next point and requires the equation

to be solved. Because the unknown $$y_{n+1}$$y_{n+1} also appears on the right-hand side, each step entails the additional cost of solving an equation. This construction can nevertheless improve stability for decay problems.

Applied to the model equation $$y^{\prime }=ay$$y'=ay, Backward Euler gives

If $$a<0$$a<0, then $$1-ah=1+\left|a\right|h>1$$1-ah=1+|a|h>1 for every $$h>0$$h>0, and hence $$|1/(1-ah\left)\right|<1$$|1/(1-ah)|<1 for every positive step size. This comparison demonstrates that the selection of a numerical method must account for stability as well as accuracy.

8Information Available from a Direction Field

A direction field does not produce an exact solution, but it is useful for diagnosis before analytic or numerical computation.

9Scope and Limitations

A direction field visualizes candidate solution curves; it does not provide an exact solution or a proof of stability. In a region where uniqueness has not been established, it must not be interpreted as defining a single flow. Euler's method is an essential introductory method, but practical computation may require Runge--Kutta methods, Backward Euler, or adaptive step sizes.

10Previous Lecture

11Next Lecture

12Exercises