Chapter 8

Numerical Solution of Ordinary

Differential Equations

8.1 Introduction
Consider the initial value problems (IVP) of the form

x′ (t) = f (t, x) (8.1)

x(t0 ) = x0 (8.2)

where x′ = dxdt . (8.1) is a first order ordinary differential equation and (8.2) is a particular
point fixed on the solution (in general known as initial condition). The ode together with
the necessary initial conditions is called as an initial value problem. The initial condition
restricts the possible solutions of the given differential equation because a solution of the
initial value problem must satisfy both the ODE and the initial condition.

8.1.1 First Existence Theorem

If f is continuous in a rectangle D centered at (t0 , x0 ), say

D = {(t, x) ∶ ∣t − t0 ∣ ≤ α, ∣x − x0 ∣ ≤ β} (8.3)

then the initial-value problem (8.1) and (8.2) has a solution x(t) for ∣t − t0 ∣ < min(α, M
β
),
where M is the maximum of f (t, x) in the rectangle D.

EXAMPLE 1 : Prove that the initial value problem x′ = (t + sin x)2 , x(0) = 3 has a
solution in the interval −1 ≤ t ≤ 1.

Proof : Consider the rectangle D = {(t, x) ∶ ∣t∣ ≤ α, ∣x − 3∣ ≤ β} the magnitude of f is

bounded by f (t, x) ≤ (α + l)2 = M . We want min(α, M β
) ≥ 1, and so we can let α = 1. Then
M = 4, and our objective is met by letting β ≥ 4. The existence theorem asserts that a
solution of the initial-value problem exists on the interval ∣t∣ < min(α, M
β
) = 1.

67
68 Sanyasiraju V S S Yedida sryedida@iitm.ac.in

8.1.2 Uniqueness
Consider x′ = x2/3 , x(0) = 0 has more than one solution x ≡ 0 and x = 27t 3
.
df
If f and dx are continuous in the rectangle D defined by (8.3), then the initial-value problem
(8.1) and (8.2) has a unique solution in the interval ∣t − t0 ∣ ≤ min(α, M
β
).
In the given example fx is continuous at x = 0.

8.1.3 Second Existence Theorem of Initial-Value Problems

If f is continuous in the strip a < t < b, −∞ < x < ∞ and satisfies there an inequality

∣f (t, x1 ) − f (t, x2 )∣ ≤ L ∣x1 − x2 ∣ (8.4)

then the initial-value problem (8.1) and (8.2) has a unique solution in the interval [a, b].
Inequality (8.4) is called a Lipschitz condition in the second variable.

For a function of one variable, the condition

∣g(x1 ) − g(x2 )∣ ≤ L ∣x1 − x2 ∣ (8.5)

is stronger than continuity because if x2 approaches x1 , then the right-hand side in (8.5)
approaches 0, and this forces g(x2 ) to approach g(x1 ).

The condition (8.5) is weaker than having a bounded derivative. Indeed, if g ′ (x) exists
everywhere and does not exceed L in modulus, then by the Mean-Value Theorem,

∣g(x1 ) − g(x2 )∣ = ∣g ′ (ξ)∣ ∣x1 − x2 ∣ < L∣x1 − x2 ∣

Therefore, the sufficient condition for the existence of a unique solution is : Suppose f (t, x)
is defined on a convex set D ⊂ R2 . If a constant L > 0 exists with
∂f
∣ (t, x)∣ ≤ L, for all (t, x) ∈ D
∂x
then f satisfies a Lipschitz condition on D in the variable x with Lipschitz constant L.

Convex set A set D ⊂ R2 is said to be convex if whenever (t1 , x1 ) and (t2 , x2 ) belong
to D, then ((1 − λ)t1 + λt2 , (1 − λ)x1 + λx2 ) also belongs to D for every λ in [0, 1]. That is,
a set is convex provided that whenever two points belong to the set, the entire straight-line
segment between the points is also belongs to the set.

8.1.4 Example 2
Show that the function g(x) = ∑ni=1 ai ∣x − wi ∣ satisfies a Lipschitz condition with the constant
L = ∑ni=1 ∣ai ∣.
Lecture Notes MA5470 Numerical Analysis 69

Solution :
n n
∣g(x1 ) − g(x2 )∣ = ∣ ∑ ai ∣x1 − wi ∣ − ∑ ai ∣x2 − wi ∣ ∣
i=1 i=1
n
= ∣ ∑ ai {∣x1 − wi ∣ − ∣x2 − wi ∣} ∣
i=1
n
≤ ∑ ∣ai ∣ ∣ {∣x1 − wi ∣ − ∣x2 − wi ∣} ∣
i=1
n n
≤ ∑ ∣ai ∣ ∣ {x1 − x2 } ∣ = L ∣x1 − x2 ∣, where L = ∑ ∣ai ∣
i=1 i=1

8.1.5 Example 3
Show that f (t, x) = t ∣x∣ satisfies a Lipschitz condition on the interval D = {(t, x) ∶ 1 ≤ t ≤
2 and 3 ≤ x ≤ 4}.
Solution : For each pair of points (t, x1 ) and (t, x2 ) in D we have

∣f (t, x1 ) − f (t, x2 )∣ = ∣ t ∣x1 ∣ − t ∣x2 ∣ ∣ = ∣t∣ ∣x1 ∣ − ∣x2 ∣ ∣ ≤ 2 ∣x1 − x2 ∣

Thus f satisfies a Lipschitz condition on D in the variable y with Lipschitz constant 2. The
smallest value possible for the Lipschitz constant for this problem is L = 2, because, for
example, ∣f (2, 1) − f (2, 0)∣ = ∣2 − 0∣ = 2∣1 − 0∣.

8.1.6 Example 4
Show that there is a unique solution to the initial value problem

x′ = 1 + t sin(t x), 0 ≤ t ≤ 2, x(0) = 0

Solution : Keeping t constant and applying the Mean Value Theorem to the function
f (t, x) = 1 + t sin(t x), we see, whenever x1 < x2 , a number ξ in (x1 , x2 ) exists with

f (t, x2 ) − f (t, x1 ) ∂f
= (t, ξ) = t2 cos(ξ t)
x2 − x1 ∂x
Thus, ∣f (t, x2 ) − f (t, x1 )∣ = ∣x2 − x1 ∣ ∣t2 cos(ξt)∣ ≤ 4∣x2 − x1 ∣. That is, f satisfies a Lipschitz
condition in the variable x with Lipschitz constant L = 4. Additionally, f (t, x) is continuous
when 0 ≤ t ≤ 2 and −∞ < x < ∞ therefore, a unique solution exists to the given initial-value
problem.

8.1.7 Well posed problem

The initial-value problem

x′ (t) = f (t, x), a ≤ t ≤ b, x(t0 ) = α

is said to be a well-posed problem if:

70 Sanyasiraju V S S Yedida sryedida@iitm.ac.in

• A unique solution, x(t), to the problem exists, and

• There exist constants 0 > 0 and k > 0 such that for any , with 0 >  > 0, whenever
δ(t) is continuous with ∣δ(t)∣ <  for all t in [a, b], and when ∣δ0 ∣ < , the initial-value
problem z ′ (t) = f (t, z) + δ(t), a ≤ t ≤ b, z(a) = α + δ0 has a unique solution z(t) that
satisfies ∣z(t) − y(t)∣ < k for all t in [a, b].

The problem specified above is called a perturbed problem associated with the original IVP.
It assumes the possibility of an error being introduced in the statement of the differential
equation, as well as an error δ0 being present in the initial condition. Numerical methods
will always be concerned with solving a perturbed problem because any round-off error in-
troduced in the representation perturbs the original problem. Unless the original problem is
well-posed, there is little reason to expect that the numerical solution to a perturbed problem
will accurately approximate the solution to the original problem.

If f satisfies, Lipschitz condition then the given problem is well posed.

8.1.8 Problems
1. Find the two solutions of x′ = x1/3 , x(0) = 0.
√
2. Show that x = t4 and x = 1−t are solutions of the initial-value problem 2x′ = t2 + 4x−t,
2

x(2) = 1. Is this contradicting the theorem on the uniqueness of solution to the initial-
value problems?

3. Solve the IVP x′ = f (t, x), x(0) = 0 when

(a) f (t, x) = t3
(b) f (t, x) = (1 − t2 )−1/2
(c) f (t, x) = (1 + t2 )−1/2
(d) f (t, x) = (1 + t)−1
√
4. Show that the IVP x′ = ∣x∣, x(0) = 0 has a solution on the entire real line.

5. Show that the IVP x′ = tan x, x(0) = 0 has a solution in the interval ∣t∣ < π4 .

6. Prove that the IVP x′ = t2 +ex , x(0) = 0 has a unique solution in the interval ∣t∣ < 0.351.

7. Show that the IVP x′ = 1 + x + x2 cos t, x(0) = 0 has a solution in the interval ∣t∣ < 13 .

8. Show that the IVP x′ = tx2/3 , x(0) = 1 has a solution in the interval ∣t∣ < 2. Dies it has
more than one solution ?

9. Find the interval in which the IVP x′ = sec x, x(0) = 0 has a unique solution?
Lecture Notes MA5470 Numerical Analysis 71

8.2 Limitations of the Numerical Solutions of Differ-

ential Equations
• Numerical solutions are only approximate solutions

• Numerical solutions are generated only at discrete set of points

8.3 Taylor series method

The method assumes that the dependent function x and hence f are very smooth, that
is, required number of derivatives exist (though it is not necessary for the existence of the
solution of the given ivp problem).

In this method, using the information of x and its derivatives at t (assumed to exits), the
value of x at its neighboring point is computed using the truncated Taylor series.

8.3.1 Illustration through an example

Find the solution of the IVP x′ = cos t − sin x + t2 , x(−1) = 3 using Taylor series method.
Solution : The Taylor series of x(t) at its neighboring point is

∆t2 ′′ ∆t3 (3) ∆t4 (4)

x(t + ∆t) = x(t) + ∆t x′ (t) + x (t) + x (t) + x (t) + ⋯ (8.6)
2! 3! 4!
We also know that

x′ = cos t − sin x + t2
x(2) = − sin t − x′ cos x + 2t
x(3) = − cos t − x(2) cos x + (x′ )2 sin x + 2
x(4) = sin t − x(3) cos x + 3x′ x(2) sin x + (x′ )3 cos x so on...

If we decides to stop with x(4) term then the error is committed from the x(5) term on wards
(5) (ξ) where
5
or by remainder theorem the error term (called Local Truncation Error ) is ∆t5! x
ξ ∈ (t, t + ∆t).

starting at t = 1 and fixing the step length of ∆t as 0.01, compute the values of x, x′ ,
x(2) , x(3) and x(4) then substitute them in the Taylor series (8.6) to obtain the value of
x(−1 + .01). Repeating the same until t = 1 gives the required discrete solution of the given
IVP.

k 0 1 2 3 4 5 196 187 198 199 200

t -1 -.99 -.98 -.97 -.96 -.95 .96 .97 .98 .99 1
x 3. 3.014 3.02803 3.04209 3.05617 3.07028 6.36566 6.37977 6.39386 6.40791 6.42194
72 Sanyasiraju V S S Yedida sryedida@iitm.ac.in

If we repeat the exercise with x(1) = 6.42194 and ∆t = −0.01, the result at t = −.99999
can be obtained as 3.00000. The close agreement with the original initial value tells us that
the numerical solution what we have computed is accurate to all the FIVE significant figures.

Error : The error term in the Taylor series approximation is

∆tn (n) ∆t5 (5)
eT S = x (ξ) = x (ξ) for n = 5
n! 120
∆t5 x(4) (t + ∆t) − x(4) (t)
= ( ) approximating x(5) (ξ)
120 ∆t
This can be estimated because the bracketed part is any how computed during the Taylor
series computations. In the present computation, this value bounded by 3.42 × 10−11 .

8.3.2 Important points of the Taylor series method

1. the method depends on repeated differentiation of the given differential equation (for
n > 1). Hence, the function f (t, x) must possess partial derivatives in the region where
the solution curve passes in the tx-plane. Such an assumption is not necessary for the
existence of a solution.
2. It is necessity to do some preliminary analytic work (computing the partial derivatives
of f )
3. but conceptually a simple procedure with very high precision

8.4 Types of Errors

In numerically solving a differential equation, several types of errors arise. These are classified
as follows:
1. Local truncation error (is the error made in one step when an infinite process is replaced
by a finite one)
2. Local roundoff error
3. Global truncation error
4. Global roundoff error (roundoff error is caused by the limited precision of the compu-
tation)
5. Total error
In the Taylor-series method, if we retain terms up to and including hn in the series, then
the local truncation error is the sum of all the remaining terms that we do not include. By
Taylor’s Theorem, these terms can be compressed into a single term of the form
∆tn+1 (n+1)
x (ξ)
(n + 1)!