10/8/2019

Ali M. Sahlodin
Department of Chemical Engineering
AmirKabir University of Technology
1397 S.H

 Local truncation errors in DAEs

 Differential index of DAEs and its
implications
 Handling high-index DAEs
 Index reduction

 Next: Advantages of DAE forms

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 2

1
 10/8/2019

 Original DAE form is

 More natural
 Easier to interpret
 Easier to troubleshoot

x
mxT 0
L
y
my mg T 0
L
x 2  y 2  L2 0

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 3

 Plantmodel generated by a simulation

program.
 Model automatically fed into the solver
 Manual modifications and index reduction
may break the automation

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 4

2
 10/8/2019

What is sparsity?
2 8 0 0 1
0 5 0 3 0 •9 nonzero elements
  •16 zero elements
0 0 6 0 0 •64% sparse
 
0 0 0 4 1
0 0 0 0 5 

 Zero elements are identically zero

 Sparsity occurs frequently in physics and
engineering models
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 5

Why are sparse systems common?

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 6

3
 10/8/2019

 Identify sparse structure

 Structural analysis
 Keep a list of structurally zero elements
 Exclude zero elements from calculations
 Reduced computational cost
 Very important in large-scale simulations

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 7

F, C1
Index 2 DAE Index reduction

dC2 F dC2 F
(t )   C1 (t )  C2 (t )   0 (t )   C1 (t )  C2 (t )   0
dt V dt V
C2 (t )   (t )  0 dC1 F V
(t )   C1 (t )  C2 (t )   (t )  0 F, C2
dt V F

Before After
 F F 
f  F F  V V 
 V V   
x    F
0 1  1
 V 
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 8

4
 10/8/2019

 Differentiation replaces algebraic constraints

with differential equations.
 Numerical solution of differential equations
can drift off the algebraic constraints.

 The pendulum example:

x
mxT 0
L
y
my mg T 0
L Physical constraint
x 2  y 2  L2 0 should hold at all times

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 9

 Simulation of index-0 pendulum (ode45)

Simulation goes crazy!

Algebraic constraint violated!

X2+y2=L2

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 10

5
 10/8/2019

 Add the algebraic constraints back to the

system after index reduction is done.
 Resulting system is over-determined.

Index reduction

f (t , x, x )  0 x  f ( x, t )
Introduce
x  f ( x, t ) new variables
x  f ( x, t )  g Tx 
0  g ( x, t )
0  g ( x, t )
Over-determined

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 11

The constraint stabilized system

 has the same analytical solution as the original
DAE.
 is an index 2 DAE (check it!)
 Ensures the physical constraint remains satisfied
despite truncation errors in state calculations.

x  f ( x, t )  g Tx 
0  g ( x, t )

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 12