3/11/2021

ECE 204 Numerical methods

4th-order Runge-Kutta method

Douglas Wilhelm Harder, LEL, M.Math.

[email protected]
[email protected]

4th-order Runge-Kutta method

Introduction

• In this topic, we will

– Derive the 4th-order Runge-Kutta method by estimating
and averaging slopes
– Look at the technique visually
– See the error is O(h5) for a single step
– Look at two examples of a single step
– See how to apply this method under multiple steps
• We will implement this in C++
– Look at two examples of multiple steps
– Compare the algorithm with Euler’s and Heun’s methods

1
 3/11/2021

4th-order Runge-Kutta method

Simpson’s rule

• A simple Reimann sum approximates an integral with one value:

b

 f ( x ) dx  f ( a )( b − a )
a

• The value of the function, however, changes across [a, b],

so Simpson’s rule integrates an interpolating quadratic:
 a+b
b f (a) + 4 f   + f (b )
 2 
a f ( x ) d x  (b − a )
6

4th-order Runge-Kutta method

4th-order Runge-Kutta method

• Heun’s method uses two slopes

– We’d like to take more samples of the slope…

y ( 0) = 1
f ( t 0 , y0 )
f ( t0 + h, y0 + hs0 )
y (1) ( t ) = −0.2 y ( t ) − sin ( t ) − 0.1
y ( 0) = 1

2
 3/11/2021

4th-order Runge-Kutta method

4th-order Runge-Kutta method

• Without justification,
4th-order Runge-Kutta says to proceed as follows:

s0  f ( tk , yk )
s1  f ( tk + 12 h, yk + 12 hs0 )

s2  f ( tk + 12 h, yk + 12 hs1 )

s3  f ( tk + h, yk + hs2 )

s0 + 2s1 + 2s2 + s3
yk +1  yk + h
6

4th-order Runge-Kutta method

4th-order Runge-Kutta method

• Visually, we proceed as follows

y ( 0) = 1

y (1) ( t ) = −0.2 y ( t ) − sin ( t ) − 0.1

y ( 0) = 1

3
 3/11/2021

4th-order Runge-Kutta method

One step of the 4th-order Runge-Kutta method

• What is the error? O(h4) or O(h5) or better?

– We will look at two initial-value problems and
approximate y(t0 + h) for successively smaller values of h
– For example, approximate y(0.4) with:
y (1)( t ) = − y ( t )

s0 ← f (0, 1) = –1 y ( 0) = 1
s1 ← f (0.2, 1 + 0.2s0) = –0.8
s2 ← f (0.2, 1 + 0.2s1) = –0.84
s3 ← f (0.4, 1 + 0.4s2) = –0.664
e −0.4 = 0.6703200460356393
s0 + 2s1 + 2s2 + s3
y1  y0 + h
6
= 1 + 0.4
( −1) + 2 ( −0.8 ) + 2 ( −0.84 ) + ( −0.664 )
= 1 + 0.4 ( −0.824 )
6 7
= 0.6704
7

4th-order Runge-Kutta method

One step of the 4th-order Runge-Kutta method

• Let’s approximate the solution at y(0 + h) to

y (1)( t ) = − y ( t )
y ( 0) = 1
n h= 2–n Exact Approximation Error Ratio
1 0.5 0.606530659712633 0.606770833333333 –0.0002402
2 0.25 0.7788007830714049 0.7788085937500000 –0.000007811 0.03252
3 0.125 0.8824969025845955 0.8824971516927084 –0.0000002491 0.03189
4 0.0625 0.9394130628134758 0.9394130706787109 –0.000000007865 0.03157
5 0.03125 0.9692332344763441 0.9692332347234091 –0.0000000002471 0.03141
6 0.015625 0.9844964370054085 0.9844964370131493 –0.000000000007741 0.03133
7 0.0078125 0.9922179382602435 0.9922179382604858 –0.0000000000002423 0.03130
8 0.00390625 0.9961013694701175 0.9961013694701251 –0.000000000000007550 0.03116
9 0.001953125 0.9980487811074755 0.9980487811074757 –0.0000000000000002220 0.02941
10 0.0009765625 0.9990239141819757 0.9990239141819757 0 0

4
 3/11/2021

4th-order Runge-Kutta method

One step of the 4th-order Runge-Kutta method

• Let’s approximate the solution at y(0 + h) to

y (1)( t ) = −0.2 y ( t ) − sin(t ) − 0.1
t
−
−13 + 25cos(t ) − 5sin(t ) + 14e 5
y (t ) =
y ( 0) = 1 26
n h = 2–n Exact Approximation Error Ratio
1 0.5 0.738852315643913 0.738856449702695 –0.000004134
2 0.25 0.8962693342116731 0.8962695046719316 –0.0000001705 0.04123
3 0.125 0.9552271839309132 0.9552271898849072 –0.000000005954 0.03493
4 0.0625 0.9794223225078814 0.9794223227035564 –0.0000000001957 0.03286
5 0.03125 0.9901670100357426 0.9901670100420059 –0.000000000006263 0.03201
6 0.015625 0.9951978758220640 0.9951978758222620 –0.0000000000001981 0.03162
7 0.0078125 0.9976275785667972 0.9976275785668035 –0.000000000000006328 0.03195
8 0.00390625 0.9988209552460877 0.9988209552460880 –0.0000000000000003331 0.05263
9 0.001953125 0.999412269826320 0.999412269826320 0 0

4th-order Runge-Kutta method

Multiple steps of Heun’s method

• Can we see that the error is indeed O(h5)

• As with Euler’s method:
– First, we will implement a function to find the n approximations
by dividing a range [t0, tf] into n sub-intervals
– For two IVPs with the initial condition y(0) = 1,
we will approximate y(5) by using 2n intervals

10

10

5
 3/11/2021

4th-order Runge-Kutta method

Implementation
std::tuple<double *, double *, double *> rk4(
double f( double t, double y ), std::pair<double, double> t_rng, double y0,
unsigned int n
) {
double h{ (t_rng.second - t_rng.first)/n };

double *ts{ new double[n + 1] };

double *ys{ new double[n + 1] };
double *dys{ new double[n + 1] };

ts[0] = t_rng.first;
ys[0] = y0;
dys[0] = f( ts[0], ys[0] );

for ( unsigned int k{0}; k < n; ++k ) {

ts[k + 1] = t_rng.first + h*(k + 1); // ts[k + 1] = ts[k] + h;
double s0{ dys[k] };
double s1{ f( ts[k] + h/2.0, ys[k] + h*s0/2.0 ) };
double s2{ f( ts[k] + h/2.0, ys[k] + h*s1/2.0 ) };
double s3{ f( ts[k + 1], ys[k] + h*s2 ) };
ys[k + 1] = ys[k] + h*(s0 + 2.0*s1 + 2.0*s2 + s3)/6.0;
dys[k + 1] = f( ts[k + 1], ys[k + 1] );
}

return std::make_tuple( ts, ys, dys ); 11

}

11

4th-order Runge-Kutta method

Multiple steps of 4th-order RK method

• Let’s approximate the solution at y(5) to

y (1)( t ) = − y ( t )
y ( 0) = 1
n Approximation Error Ratio
2 0.42047119140625 –0.4137
4 0.00893558527119917 –0.002198 0.005312
8 0.006810674597968526 –0.00007273 0.03309
16 0.006741425022840268 –0.000003478 0.04782
32 0.006738137657266484 –0.0000001907 0.05482
64 0.006737958161994555 –0.00000001116 0.05855
128 0.006737947674390917 –0.0000000006753 0.06050
256 0.006737947040610186 –0.00000000004152 0.06149
512 0.006737947001659729 –0.000000000002574 0.06199
1024 0.006737946999245688 –0.0000000000001602 0.06224

0.006737946999085467 12

12

6
 3/11/2021

4th-order Runge-Kutta method

Multiple steps of 4th-order RK method

• Let’s approximate the solution at y(5) to

y (1)( t ) = −0.2 y ( t ) − sin(t ) − 0.1
t
−
−13 + 25cos(t ) − 5sin(t ) + 14e 5
y (t ) =
y ( 0) = 1 26
n Approximation Error Ratio
2 0.1469019038207984 0.008348
4 0.1548307896015398 0.0004188 0.05016
8 0.1552239200410955 0.00002570 0.06119
16 0.1552479334528051 0.000001612 0.06291
32 0.1552494441496338 0.0000001015 0.06294
64 0.1552495392562453 0.000000006371 0.06278
128 0.1552495452276594 0.0000000003991 0.06265
256 0.1552495456018131 0.00000000002498 0.06258
512 0.1552495456252274 0.000000000001563 0.06257
1024 0.1552495456266942 0.00000000000009581 0.06131
0.1552495456267901 13

13

4th-order Runge-Kutta method

Comparison

• We could compare Euler, Heun and 4th-order Runge-Kutta for

1024 intervals
– Issue: This isn’t really fair, as
• Euler’s method uses one function evaluation per interval
• Heun’s uses two
• 4th-order Runge-Kutta uses four
– For 1024 function evaluations, how accurate is each?
Euler 1024 0.006655931188587414 0.00008202
Heun 512 0.006738486441915978 –0.0000005394
RK4 256 0.006737947040610186 –0.00000000004152

Euler 1024 0.152997481619969 0.002252

Heun 512 0.1552516585204115 –0.000002113
RK4 256 0.1552495456018131 0.00000000002498
14

14

7
 3/11/2021

4th-order Runge-Kutta method

Comparison

• Another question is:

– How many function evaluations are required to get the same
accuracy as Euler’s method with 1024 intervals?

y(5) = 0.00682130435141457
Euler 1024 0.006655931188587414 0.00008202 1024
Heun 43 0.006737946999085467 –0.00008336 86
RK4 8 0.006810674597968526 –0.00007273 32

y(5) = 0.1552495456267901
Euler 1024 0.152997481619969 0.002252 1024
Heun 16 0.153866775462848 –0.002317 32
RK4 3 0.153866775462848 0.0013827 12

15

15

4th-order Runge-Kutta method

Summary

• Following this topic, you now

– Understand the 4th-order Runge-Kutta method for approximating
a solution to a 1st-order initial-value problem
– Are aware of a visual interpretation with respect to slopes
– Understand the error is O(h5) for a single step
– Are aware that we must apply this technique multiple times to
estimate the solution on a larger interval
– Know that the error drops in this case to O(h4)
– Have seen a number of examples and an implementation
– Understand how much better the algorithm is,
even when we consider the number of function evaluations

16

16

8
 3/11/2021

4th-order Runge-Kutta method

References

[1] https://en.wikipedia.org/wiki/Runge%E2%80%93Kutta_methods

17

17

4th-order Runge-Kutta method

Acknowledgments

None so far.

18

18

9
 3/11/2021

4th-order Runge-Kutta method

Colophon

These slides were prepared using the Cambria typeface. Mathematical equations
use Times New Roman, and source code is presented using Consolas.
Mathematical equations are prepared in MathType by Design Science, Inc.
Examples may be formulated and checked using Maple by Maplesoft, Inc.

The photographs of flowers and a monarch butter appearing on the title slide and
accenting the top of each other slide were taken at the Royal Botanical Gardens in
October of 2017 by Douglas Wilhelm Harder. Please see
https://www.rbg.ca/
for more information.

19

19

4th-order Runge-Kutta method

Disclaimer

These slides are provided for the ECE 204 Numerical methods
course taught at the University of Waterloo. The material in it
reflects the author’s best judgment in light of the information
available to them at the time of preparation. Any reliance on these
course slides by any party for any other purpose are the
responsibility of such parties. The authors accept no responsibility
for damages, if any, suffered by any party as a result of decisions
made or actions based on these course slides for any other purpose
than that for which it was intended.

20

20

10