MSE

FTP OrdDiff Problem sheet 5: Numerical methods

1. Consider the ivp 

ẋ(t) = cos(t + x) + sin(t − x)
x(0) = 0
Find x(1)
a) Sketch the vector field
b) Manually calculate two steps of Eulers method with h = 0.5 to find an approximation
to x(1)
c) Calculate an approximation to x(1) with Eulers method (using Matlab) for h = 0.1,
h = 0.01 and h = 0.001
d) Use ode45 from MATLAB to find an approximation for x(1).

2. Consider the IVP 

ẋ(t) = t · x(t)
x(0) = 0.1
a) Solve it!
b) Manually calculate an approximation at t = 2 with step size h = 0.4.
c) sketch the exact solution together with approximations for step-sizes h = 0.4, h = 0.2
und h = 0.1.

3. Consider the IVP y 0 = y, y(0) = 1

a) Compare the Taylor-series y(h) = . . . of the exact solution around x = 0 with the
approximation ŷ1 of y(h) by the explicit Euler method, the implicit Euler method and
Heuns method (with stepsize h).

4. Consider any autonomous IVP y 0 = f (y), y(0) = y0 . Can you show, that one step of Heuns
method will always give the correct first-order approximation to the solution y(h)? (We
would be very impressed!)
5. Consider the IVP

ẋ = 100 · (x2 − x3 ), x0 = x(0) = 1 − ∆ < 1

For a small ∆ > 0. The exact solution looks as follows (the circle marks x(0))
 1

0.8

0.6
x

0.4

0.2

0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
t

The dashed part is not relevant for us. For t > 0, the solution approaches monotonically
the value x = 1. Despite this ”‘boring”’ behaviour, the ODE is not easy for an explicit
solver.
a) Find the maximum step size (depending on ∆) that guarantees, x̂1 < 1
b) Find the equation for x̂1 for the implicit Euler method. (You don’t need to solve it!)
c) Show that the equation for x̂1 for the implicit Euler method has a solution in the interval
]x0 , 1[ for any stepsize h > 0

6. The value of a company grows with a rate proportional to the square root of its actual
value. Two years ago, it was 1 Mio CHF, now it is 1.44 Mio CHF. When will the value be
2 Mio CHF?
7. Approximate Euler’s number e, using numerical solutions of the IVP

ẋ(t) = x(t)
x(0) = 1
for t = 1 with step width h = 1 (1.e. one step only), with the following methods
a) Explicit Euler
b) Midpoint Rule
c) Heun’s method
d) 4th order Runge-Kutta

8. Consider the IVP 

ẋ(t) = t − x
x(0) = 1
We are intersted in x(2).
a) Find x(2) exactly
b) Use two steps of the midpoint rule with h = 1 (by hand)
c) Use four steps of Heun’s method with h = 0.5 (by hand)
d) Use Heun’s method with step-width h = 0.1, h = 0.01 and h = 0.001 (with a computer)
e) What is the maximum step-width for Heun’s method, if the first 6 decimals of the
numerical approximation x̂(2) must be correct?
9. Consider the IVP 
ẋ(t) = sin(tx)
x(0) = 6
We are interested in x(5).
a) Use the classical Runge-Kutta method with step widths h = 0.5 h = 0.2, h = 0.1,
h = 0.05 und h = 0.01 to find approximations x̂(5).
b) Sketch the exact solution together with approximations for h = 0.5 h = 0.2, h = 0.1,
h = 0.05 and h = 0.01 in the same figure
c) Sketch exact solutions and approximations with step-widths h = 0.5 h = 0.2, h = 0.1,
h = 0.05 and h = 0.01 for the initial conditions x(0) = 1, 2, 3, 4, 5, 6 in the same figure.
Answers for problem sheet 5
1. a) Direction field

x(1) seems to be between 0.5 and 1.

b) Two steps with Eulers method h = 0.5, d.h. t0 = 0, t1 = 0.5, t2 = 1 und f (t, x) =
cos(t + x) + sin(t − x):

x0 = 0
x1 = x0 + hf (t0 , x0 ) = 0 + 0.5 · (cos(0 + 0) + sin(0 − 0)) = 0.5
x2 = x1 + hf (t1 , x1 ) = 0.5 + 0.5 · (cos(0.5 + 0.5) + sin(0.5 − 0.5)) ≈ 0.770

This gives the approximation x(1) ≈ 0.770.

Step size h approximation to x(1)

0.1 0.6718
c)
0.01 0.6558
0.001 0.6542

d) With ode45
x(1) ≈ 0.6540

2. a) Separating the variables one finds

2
x(t) = 0.1e0.5t

b) Five steps of Eulers method h = 0.4, i.e. t0 = 0, t1 = 0.4, t2 = 0.8, t3 = 1.2, t4 = 1.6,
t5 = 2 and f (t, x) = t · x:

x0 = 0.1
x1 = x0 + hf (t0 , x0 ) = 0.1
x2 = x1 + hf (t1 , x1 ) = 0.116
x3 = x2 + hf (t2 , x2 ) = 0.15312
x4 = x3 + hf (t3 , x3 ) ≈ 0.226618
x5 = x4 + hf (t4 , x4 ) ≈ 0.37165

(Exact value: x(2) = 0.7389 . . .)

 c) Graphs:

3. a) The exact solution is y(h) = exp(h) = 1 + h + h2 /2 + . . .. The explicit Euler method

gives ŷ1 = 1 + h, i.e. it represents the first order term of the exact solution.

The implicit Euler method gives ŷ1 = 1 + hŷ1 which we can solve explicitely for
1
ŷ1 = = 1 + h + h2 + h3 + . . .
1−h
where we used the geometric series (or equivalently the taylor-expansion of the fraction
1/(1 − h)). The implicit Euler solution gives the correct first order term and contains
(wrong) higher order terms.

Heuns Method gives

h
k1 = 1, k2 = 1 + h, ŷ1 = 1 + (1 + 1 + h) = 1 + h + h2 /2
2
thus it represents the correct approximation to second order.
4. The solution has a taylor expansion
h2 00 h2
y(h) = y0 + hy 0 (0) + y (0) + . . . = y0 + hf (y0 ) + f 0 (y(0))f (y(0)) + . . .
2 2
Heuns method gives
h
ŷ1 = y0 +
(f (y0 ) + f (y0 + hf (y0 ))
2
Expanding the last term to first order gives

f (y0 + hf (y0 )) = f (y0 ) + hf (y0 )f 0 (y0 ) + h2 . . .

And inserting
h h2
ŷ1 = y0 + (f (y0 ) + f (y0 ) + hf (y0 )f 0 (y0 ) + h2 . . .) = y0 + hf (y0 ) + f (y0 )f 0 (y0 ) + . . .
2 2
All terms in the . . . have at least order three in h, so the approximation is indeed correct
to second order in h.
5. a) From
x̂1 = (1 − ∆) + 100h · (∆ − 2∆2 + ∆3 )
one concludes
1 1 1
h< 2
≈
100 1 − 2∆ + ∆ 100
 b)
x̂1 = x0 + 100h · (x̂21 − x̂31 )
c) The function
f (x̂) = x0 + 100 h · (x̂2 − x̂3 ) − x̂
satisfies (for 0 < x0 < 1)

f (x0 ) = 100 h · (x20 − x30 ) = 100 h x20 · (1 − x0 ) > 0

f (1) = x0 − 1 < 0

So there is at least one zero of f in ]x0 , 1[.

6. Let x(t) the value at time t. The ODE must be
√
ẋ = λ x

with λ ∈ R to be found. The general solution is (separation of variables)

λt + C 2
 
x= , C ∈ R.
2
Inserting x(−2) = 106 and x(0) = 1.44 · 106 yields
C 2
= 1.44 · 106


2

−2λ+C 2
2 = 106
So C = 2400 and λ = 200. The solution of the IVP is therefore

x(t) = (100t + 1200)2 .

√
From the equation x(t0 ) = 2 · 106 we find t0 = 2 · 10 − 12 ≈ 2.142.
7. a) Eulers explicit method:

x1 = x0 + h · f (t0 , x0 )
= 1+1·1
= 2

b) Midpoint rule:

 
h h
x1 = x0 + h · f t0 + , x0 + f (t0 , x0 )
2 2
= 1 + 1 · f (0.5, 1.5)
= 2.5

c) Heuns method:

m1 = f (t0 , x0 ) = 1;
m2 = f (t0 + h, x0 + hm1 )
= f (2, 1 + 1) = 2;
h
x1 = x0 + (m1 + m2 )
2
= 1 + 0.5(1 + 2)
= 2.5
 d) Fourth order Runge-Kutta

m1 = f (t0 , x0 ) = 1;
h h
m2 = f (t0 + , x0 + m1 )
2 2
= f (0.5, 1 + 0.5) = 1.5;
h h
m3 = f (t0 + , x0 + m2 )
2 2
= f (0.5, 1 + 1.75) = 1.75;
m4 = f (t0 + h, x0 + hm3 )
= f (1, 3.75) = 2.75
h
x1 = x0 + (m1 + 2m2 + 2m3 + m4 );
6
1
= x0 + (1 + 2 · 1.5 + 2 · 1.75 + 2.75)
6
≈ 2.70833

8. a) closed form solution:

ODE: ẋ = −x + t; homogeneous DGL: ẋ = −x with general solution

x = Ke−t , K ∈ R.

Trial function for xs : xs = at + b with ẋs = a; inserting into the ODE: a = t − (at + b),
thus
0 = 1−a
,
a = −b
a = 1, b = −1 and thus xs = t − 1 the general solution is therefore

x = Ke−t + t − 1, K ∈ R.

With the initial condition x(0) = 1 we get K = 2 and the solution of the IVP is
x = 2e−t + t − 1.
Thus
x(2) = 2e2 − 1 ≈ 1.270670566473225

b) Two steps with the midpoint-rule for h = 1: t0 = 0 , t1 = 1, t2 = 2:

x0 = 1
 
h h
x1 = x0 + h · f t0 + , x0 + f (t0 , x0 )
2 2
= 1 + 1 · f (0.5, 1 + 0.5f (0, 1))
= 1  
h h
x2 = x1 + h · f t1 + , x1 + f (t1 , x1 )
2 2
= 1 + 1 · f (1.5, 1 + 0.5f (1, 1))
= 1.5
 c) Four steps with Heuns method and h = 0.5: t0 = 0 , t1 = 0.5, t2 = 1, t3 = 1.5, t4 = 2:

x0 = 1
0.5
x1 = 1 + (−1 + 0) = 0.75
2
0.5
x2 = 0.75 + (−0.25 + 0.375) = 0.78125
2
0.5
x3 = 0.78125 + (0.21875 + 0.609375) = 0.98828125
2
0.5
x4 = 0.98828125 + (0.51171875 + 0.755859375) = 1.30517578125
2

d) Results for different step sizes

ˆ h = 0.1: x(2) ≈ 1.271644915004169
ˆ h = 0.01: x(2) ≈ 1.270679656916392
ˆ h = 0.001: x(2) ≈ 1.270670656764417
e) h ≈ 0.00235
9. a) Fourth order Runge Kutta:
ˆ h = 0.5: x(5) ≈ 4.730126534759491
ˆ h = 0.2: x(5) ≈ 4.705924274348357
ˆ h = 0.1: x(5) ≈ 4.706729484540293
ˆ h = 0.01: x(5) ≈ 4.706804777341525
ˆ h = 0.001: x(5) ≈ 4.706804785185181

b)

c) Approximate and ”‘exact”’ solution curves