Page 212 - Numerical methods for chemical engineering
P. 212

198     4 Initial value problems



                          T
                   If (∂G/∂z ) is singular, then as  t → 0, B(x) also becomes singular. To see this, consider
                   a system with two differential and two algebraic equations. Then, as  t → 0, writing
                   b ij = ( t)a ij ,wehave

                                                                      
                                     α −1   [( t)a 12 ][( t)a 13 ][( t)a 14 ]
                                   [( t)a 21 ]        [( t)a 23 ][( t)a 24 ]
                                             α −1                     
                             B ≈                                                   (4.211)
                                  [( t)a 31 ][( t)a 32 ][( t)a 33 ][( t)a 34 ]  
                                   [( t)a 41 ][( t)a 42 ][( t)a 43 ][( t)a 44 ]
                   The determinant of this matrix is
                                           4   4  4  4

                                                             b
                                                                 b
                                                                       b
                                                                    b
                                  det(B) =             ε i 1 ,i 2 ,i 3 ,i 4 1,i 1 2,i 2 3,i 3 4,i 4  (4.212)
                                          i 1 =1 i 2 =1 i 3 =1 i 4 =1
                   As  t → 0, the terms with i 1 = 1, i 2 = 2 dominate the others, and
                               4  4                             4  4
                              	 	                            2
                                                                            b
                                                      b
                                                                         b
                      det(B) ≈      ε 1,2,i 3 ,i 4  (α −1 )(α −1 )b 3,i 3 4,i 4  = α −1  ε i 3 ,i 4 3,i 3 4,i 4  (4.213)
                              i 3 =1 i 4 =1                    i 3 =1 i 4 =1
                                                            T
                                 2
                   Thus, det(B) ≈ α det(B 22 ), and if B 22 = (∂G/∂z ) is singular, so is B(x) and Newton’s
                                 −1
                   method (4.207) fails, even in the limit as  t → 0.
                                     T
                     If, however, (∂G/∂z ) is nonsingular, then B(x) will not be singular as  t → 0, and the
                                                                            T
                   linear system at each Newton update has a unique solution. If (∂G/∂z ) is nonsingular,
                   it is also easy to determine a set of consistent initial conditions, i.e., one that satisfies all
                   algebraic equations. We set y, and then compute z using Newton’s method to solve
                                                  0 = G(y, z)                        (4.214)
                           T
                   If (∂G/∂z ) is nonsingular, we can take a single time derivative,
                       d     ˙ y = F(y, z)        ∂ F        ∂ F        ∂G         ∂G
                                      ⇒    ¨ y =     ˙ y +     ˙ z  0 =     ˙ y +     ˙ z
                      dt  0 = G(y, z)           ∂y  T     ∂z  T        ∂y  T     ∂z T
                                                                                     (4.215)
                   to obtain an equivalent system with a nonsingular mass matrix,
                                                                       
                                              
                                      I    0                 F(y, z)
                                                  ˙ y
                                          ∂G                           
                                                           ∂G                      (4.216)
                                     0            ˙ z  =   −     F(y, z)  
                                          ∂z T              ∂y T
                   Definition The index of a DAE system is the number of differentiations required to convert
                   it into an ODE system with a nonsingular mass matrix.
                                                             T
                     For the example above with nonsingular (∂G/∂z ), the index is 1. High-index DAE
                   systems are not uncommon, and one must perform the necessary reduction in index to 1
                   through differentiations in order to use the BDF method presented above. Also, for high-
                   index problems, it can be challenging to identify a consistent set of initial questions.
                     In MATLAB, ode15s can simulate index-1 DAE systems. In other languages, the package
                   DASSL by Petzold and coworkers is popular and performs well. A further discussion of
                   DAEs may be found in Ascher & Petzold (1998).
   207   208   209   210   211   212   213   214   215   216   217