Numerical Integration: Presented By: Amir Patel PHD (Mechatronics) Cape Town

Lecture 1:
Numerical Integration
13 Feb 2023
Presented by:
Amir Patel
PhD (Mechatronics) Cape Town
Electrical Engineering Department EEE4119F
University of Cape Town Mechatronics II
Overview
• First Order
• Second Order
• Midpoint Method
• Runge-Kutta
Method
Autonomous drifting
Simple First-Order
Example:
Solve this analytically:

• First re-arrange
• Integrate:
Simple First-Order
Assume initial condition:
• This means solution is: 𝑦 = 𝑒 −𝑡
• Initial condition is crucial:

• Sometimes called an Initial Value Problem
Discrete Derivatives
Lecture 1:
Position
Numerical
Integration
time
• Admin
• First Order • Remember, derivative of function→

• Second Order
slope of function
• Eg: Car on highway
• Midpoint Method
• Runge-Kutta
• On computer, represent continuous
Method
function using discrete
Computer Implementation
On a computer we represent as:
• How do we code that for our example?

Comparison
Euler Method
Lecture 1:
Numerical
Integration
• Admin
• First Order • We now have a way of calculating the

• Second Order
initial slope
• We can use this to extrapolate into
• Midpoint Method
the future 𝑑𝑦
• Runge-Kutta = −𝑦 𝑤𝑖𝑡ℎ 𝑦 𝑡 = 0 =1
• Eg. 𝑑𝑡
Method
• →Slope @t=0 is -1
Extrapolation
• Extrapolation is
good for t <0.2
• But horrible
thereafter!
Car Analogy
• Imagine you are driving…

• See a sign that next town
is 10 km away
• Your speedo says 60
km/h
• Predict 10mins journey
• But…
• Traffic lights?
• Slow-moving trucks?
Euler Algorithm
𝑑𝑦
= −𝑦 𝑤𝑖𝑡ℎ 𝑦 𝑡 = 0 =1
𝑑𝑡
• We generally want to take

small increments (∆𝑡), so:
𝑑𝑦
𝑦1 = 𝑦0 + ቤ ∆𝑡
𝑑𝑡 𝑡=0
• Specifically for this case:
𝑦1 = 𝑦0 − 𝑦0 ∆𝑡
Example (∆𝑡=0.5)
1. Start at y =1 with
slope = -1
2. Extrapolate to
(0.5), slope now
-0.5
3. Extrapolate to
(0.25), slope now
-0.25
4. …and so on!
Code
Taylor Series
Lecture 1:
Numerical
Integration
• Admin • Consider an arbitrary function (y)

• First Order • We can make an approximation:
• Second Order
• Midpoint Method
• Solve for coefficients
• Runge-Kutta
• Simplest approximation would be to
Method
use derivative:
Taylor Series
• We can make this better with second derivative:
• Add more derivatives:
• Called Taylor Series

Example: 𝒚 = 𝒆 −𝒕
exp(-t)
1-t
• Taylor series:
• All derivatives of y are 1 at origin

Error Analysis
• Initial value problem:
Euler method
Error term
• Only need to care about first term
• Why?
• Error scales with ∆𝒕

• If we half ∆𝒕 , then we half the error!
Euler Error
Code
Mass spring system
Lecture 1:
Numerical
Integration
• Admin • From Newton II:

• First Order
• Second Order
• Midpoint Method
• Convert to two 1st order equations:
• Runge-Kutta
Method
Euler (mass spring)
• Two 1st order DE’s:
• Euler:
Euler (mass spring)
• Write as vector:
• Then:
• Assuming spring constant/mass = 1

Code (euler solver)
Code (derivative)
Compare to analytical
A better derivative
Lecture 1:
Numerical
Integration
• Admin
• First Order
• Second Order
• We want a better estimate of the slope
between samples
• Midpoint Method
• Euler assumed the initial derivative
• Runge-Kutta
was held throughout step
Method
• This caused instability!
Approximate derivatives (y=e-x)
• Plot derivatives:
• Initial
• Final
• Middle
• Middle is the best!

Midpoint Method Error
• Extrapolation based on
midpoint gradient:
• Taylor series:
• Substitute:
• Exactly Taylor approximation of y for first three terms

• Much more accurate than Euler!
Problem…
• We don’t have the slope at the midpoint!

• Let’s use Euler:
∆𝒕
• Then using 𝒚𝑵+𝟏/𝟐 and , can calculate the derivative
𝟐
• Last step in algorithm is then:
𝒅𝒚
Exercise = −𝒚
𝒅𝒕
• Use large time-step (∆𝒕) = 1s

• with y(t=0) = 1 then 𝒅𝒚/𝒅𝒕0=-1
• Step1: Euler
• Step2: Deriv
• Step3: Shoot to end: 𝑦 ∆𝑡

𝑁+1 = 1 + ∆𝑡(−1 + )
2
Plot
Code
time+dt/2,yH)
Compare (Mass-spring)
𝒅𝒚
Error Plot ( = −𝒚)
𝒅𝒕
Code
Runge-Kutta Method
Lecture 1:
Numerical
Integration
• Admin
• First Order
• Second Order
• Most popular method
• 4th Order Runge-Kutta Method
• Midpoint Method
• Instead of shooting to midpoint,
• Runge-Kutta
shoot to 4 specific points
Method
RK4 Algorithm
Assume:
Then:
𝒅𝒚
Error Analysis ( = −𝒚)
𝒅𝒕
t^4
The end.
• Next lecture

Numerical Integration: Presented By: Amir Patel PHD (Mechatronics) Cape Town

Uploaded by

Copyright:

Available Formats

Numerical Integration: Presented By: Amir Patel PHD (Mechatronics) Cape Town

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Numerical Integration: Presented By: Amir Patel PHD (Mechatronics) Cape Town

Uploaded by

Copyright:

Available Formats

Lecture 1:

Solve this analytically:

Assume initial condition:

• This means solution is: 𝑦 = 𝑒 −𝑡

• Initial condition is crucial:

• First Order • Remember, derivative of function→

On a computer we represent as:

• How do we code that for our example?

• First Order • We now have a way of calculating the

• Imagine you are driving…

• We generally want to take

• Specifically for this case:

• Admin • Consider an arbitrary function (y)

• We can make this better with second derivative:

• Add more derivatives:

• Called Taylor Series

• All derivatives of y are 1 at origin

• Initial value problem:

• Error scales with ∆𝒕

• Admin • From Newton II:

• Two 1st order DE’s:

• Assuming spring constant/mass = 1

• Middle is the best!

• Exactly Taylor approximation of y for first three terms

• We don’t have the slope at the midpoint!

• Use large time-step (∆𝒕) = 1s

• Step3: Shoot to end: 𝑦 ∆𝑡

You might also like