ESS101-Modeling and Simulation: Paolo Falcone

Department of Signals and Systems
ESS101- Modeling and Simulation

Lecture 15
Paolo Falcone
Chalmers University of Technology
Gteborg, Sweden
Paolo Falcone (CHALMERS)
ESS101 Modeling and Simulation
October 3, 2015
Today (Chapters 4, 5)
Linear Multistep Methods

Trapezoidal rule
The Adams-Bashforth method
Two-step methods
Consistency
Convergence and Zero-stability
October 3, 2015
Basics on simulation (IVP)
We assume the mathematical model is in the form
x (t) = f (t, x(t),u(t))

By assuming that u(t) is a known function of the time
x (t) = f (t, x(t))
The problem of simulating the model x = f (t, x(t))

can be formulated as follows
the initial condition x(0), finding a sequence of

Given
points x1, x2,xtf approximating
the solution x at the
time instants 0<t1<t2<<tf, i.e., the x(t1), x(t2),, x(tf)

October 3, 2015
FE and TS(p) methods

For FE
xn+1 = xn + hx!n
En = O(h)
For TS(p)
1 i (i)
xn+1 = xn + h x n
i=1 i!
p
En = O(h )
But, TS(p) methods require high order derivative

of the rhs
Impratical for larger ODE and when an analytical
expression of x! is not available
Idea. Achieve accuracy of TS(p) methods by
exploiting previous solutions rather than higher
order derivatives
October 3, 2015
Trapezoidal rule
Idea. Achieve accuracy of TS(p) methods by exploiting

previous solutions rather than higher order derivatives
Calculate the Taylor expansion
x! (t + h) = x! (t) + h!!
x(t) + O(h 2 )
Substitute h!!
x(t) = x! (t + h) x! (t) + O(h 2 ) in
1 2
x(t + h) = x(t) + hx! (t) + h !!
x(t) + O(h 3 )
2!
1
xn+1 = xn + hx!n + h ( x!n+1 x!n )
2
1
= xn + h ( fn+1 + fn )
2
October 3, 2015
Trapezoidal rule
1
Note that in xn+1 = xn + h ( fn+1 + fn )
2
We need to calculate fn+1 = f (tn+1, xn+1 )
That is, xn+1 is on both left and right hand sides
We may not have an explicit expression of xn+1 as function
of the previous samples (implicit method)
October 3, 2015
Implicit and explicit methods
In general, an algorithm for solving a differential equation

(or ODE solver) can be expressed as
x n +1 = G(t, x n k +1, x n k +2 ,, x n , x n +1 )
Implicit method
k-steps method
x n +1 = G(t, x n k +1, x n k +2 ,, x n , x n +1 )
This equation needs to be solved to find xn+1
Explicit method
x n +1 = G(t, x n k +1, x n k +2 ,, x n )
October 3, 2015
2-step Adam-Bashforth method (AB(2))

Instead of
x! (t + h) = x! (t) + h!!
x(t) + O(h 2 )
calculate the Taylor expansion as
x! (t h) = x! (t) h!!
x(t) + O(h 2 )
The resulting iteration is
1
xn+1 = xn + hx!n + h ( x!n x!n1 )
2
1
= xn + h (3 fn fn1 )
2
October 3, 2015
Example. Comparison of AB(2) and TS(2)

Simulate the system
x! (t) = (1 2t)x(t), x(0) = 1
over the time interval 0 t 0.9

Recall the exact solution x(t) = e
2
1 1
t
4 2
The iteration is
1
xn+2 = xn+1 + h (3 fn+1 fn )
2
1
= xn+1 + h 3 (1 2tn+1 ) xn+1 (1 2tn ) xn
2
October 3, 2015

For n=0
1
x2 = 1+ h (1 2h ) x1 h
2
How do we calculate x1?

1. x1 = (1+ h ) x0
Euler (ABE)
1
2. x1 = x0 + h (1 2t1 ) x1 + (1 2t0 ) x0
2
1+ 1 (1 2t0 )
2
x1 =
x0 Trapezoidal rule (ABT)
1 1 (1 2t1 )
2
October 3, 2015
10

h=0.2
103GE@ t=1.2
En h
October 3, 2015
11
Two-step methods. Consistency
More in general
xn+2 + 1 xn+1 + 0 xn = h ( 2 fn+2 + 1 fn+1 + 0 fn )

that is consistent and of order p if
x(t + 2h) + 1 x(t + h) + 0 x(t) h [ 2 x! (t + 2h) + 1 x! (t + h) + 0 x! (t)] = O(h p+1 )
October 3, 2015
12
Two-step methods. Consistency
Definition (Consistency). A two-step LMM method is

consistent if a p>0 exists such that
x(t + 2h) + 1 x(t + h) + 0 x(t) h [ 2 x! (t + 2h) + 1 x! (t + h) + 0 x! (t)] = O(h p+1 )
Otherwise, it is inconsistent.
Example. The Euler method
The n-th iteration xn+1 = xn + hfn stems from
x(t + h) = x(t) + hx! (t) + O(h 2 )

x(t + h) x(t) hx! (t) = O(h 2 )
Hence, the method is consistent of order p=1
October 3, 2015
13
Example. Consistency
Determine the consistency order of the integration method

(last in the table)
xn+2 + 4xn+1 5xn = h ( 4 fn+1 + 2 fn )

We have to find the p in
x(t + 2h) + 4x(t + h) 5x(t) h [ 4 x! (t + h) + 2 x! (t)] = O(h p )

Lets expand in Taylor series the terms on the lhs
4 3
2 4 (4)
2
!
!!
!!!
x(t + 2h) = x(t) + 2hx + 2h x(t) + h x (t) + h x (t) + O(h 5 )
3
3
1 2
1 3
1 4 (4)
x(t + h) = x(t) + hx! + h !!
x(t) + h !!!
x (t) + h x (t) + O(h 5 )
2
6
24
October 3, 2015
14
1 2
1 3 (4)
x! (t + h) = x! (t) + h!!
x(t) + h !!!
x (t) + h x (t) + O(h 4 )
2
6
By summing up the and collecting the terms

x(t + 2h) + 4x(t + h) 5x(t) h [ 4 x! (t + h) + 2 x! (t)] =
(1+ 4 5) x(t) + h 2 + 4 ( 4 + 2) x! (t)

4
1
1
+h [ 2 + 2 4] !!
x(t) + h + 4 4 !!!
x (t)
3
6
2
1
1 (4)
4 2
+h + 4 4 x (t)
3
24
6
2
October 3, 2015
15
After simplification
x(t + 2h) + 4x(t + h) 5x(t) h [ 4 x! (t + h) + 2 x! (t)] =
1 4 (4)
h x (t) + O(h 5 )
6
1 4 (4)
h x (t) is dominating and O(h 5 ) can be neglected
6
1 4 (4)
h x (t) = O(h 4 ) and the method is of order 3
6
October 3, 2015
16
Two-step methods. Consistency conditions
Theorem (consistency conditions for two-step LMMs).

Given the two-step LMM
xn+2 + 1 xn+1 + 0 xn = h ( 2 fn+2 + 1 fn+1 + 0 fn )

define the two polynomials
(r) = r 2 + 1r + 0
(r) = 2 r 2 + 1r + 0
The method is consistent if and only if
(1) = 0, '(1) = (1)

October 3, 2015
17
Two-step methods. Necessary condition for

convergence
Theorem (necessary condition for convergence). A

convergent LMM is consistent.
A consistent LMM is not guaranteed to be convergent
October 3, 2015
18
Example
Solve the IVP x! (t) = x(t), x(0) = 1 with the LMM
xn+2 + 4xn+1 5xn = h ( 4 fn+1 + 2 fn )

with h=0.1, 0.01, 0.001
The LMM can be rewritten as
xn+2 + 4xn+1 5xn = h ( 4xn+1 + 2xn )

and rearranged as
xn+2 = 4(1+ h)xn+1 + ( 5 2h ) xn

October 3, 2015
19
Example
No matter how small is h, the solution oscillates
October 3, 2015
20
k-step methods
The two-step LMM
xn+2 + 1 xn+1 + 0 xn = h ( 2 fn+2 + 1 fn+1 + 0 fn )

can be generalized to k-step LMM
xn+k + k1 xn+k1 +!+ 0 xn = h ( k fn+k + k1 fn+k1 +!+ 0 fn )

Given the starting values x0 , x1,, xk1
Any method is required to converge as h0
That is, the approximated solution should be equal to the
actual as h0
Convergence is required also for x1,, xk1, since they are
calculated with a numerical method as well
October 3, 2015
21
k-step methods. Consistency
In order to have convergence, we must have consistency

Define the polynomials
k
(r) = i r i , k = 1 and (r) = i r i

i=0
i=0
The consistency conditions become

k
k
i
i=0
= 0, k = 1, and i i = i
i=0
i=0
October 3, 2015
22
Example
Solve the IVP x! (t) = 0, x(0) = 1 with the LMM
xn+2 + 4xn+1 5xn = h ( 4 fn+1 + 2 fn )

Use x1 = 1+ h
The LMM becomes xn+2 + 4xn+1 5xn = 0
Note that the polynomial (r) = r 2 + 4r 5 has two roots
r=1 and r=5
The solution of the equation is xn = A + B(5)n where A, B
are determined from the initial condition and x1
xn diverges while it should be constant and equal to 1
October 3, 2015
23
Convergence and Zero-stability
Definition (Zero-stability). A LMM is zero stable if all the

roots of
k
(r) = i r i , k = 1
i=0
satisfy the condition r 1 (root condition)

and the ones satisfying r = 1 have multiplicity 1
Theorem (necessary and sufficient condition for
convergence). A LMM is convergent if and only if is
consistent and zero-stable
October 3, 2015
24

ESS101-Modeling and Simulation: Paolo Falcone

Uploaded by

Document Information

Original Description:

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

ESS101-Modeling and Simulation: Paolo Falcone

Uploaded by

Copyright:

Available Formats

Department of Signals and Systems