You are on page 1of 2

where E(t), A(t) , C(t) , D(t) , B(t) are matrix functions of time t,

and they called the coefficients matrices of the system Eq. (2.5) describes the so called linear
time varying DAE. If the matrix coefficients are constant, i.e., time independent, the system (2.5)
is called linear constant coefficient DAE or linear time-invariant (LTI) DAE, which can be
written as

() () = () + () (2.7)

() = () + () (2.8)

where E, A , C , D , B are constant matrix. For the DAEs (2.7),


there is also a concept of the dynamical order, which is defined as the rank of singular matrix E.
Eq. (2.5) and (2.7) are the two basic classes of DAEs. From this point, we restrict ourselves on
the DAEs of the form (2.7) unless stated otherwise.

2.1.2 Transfer Matrix Representation

Matrix Representation In this section, we discuss the transfer matrix representation. This
representation is derived from the state space representation using the Laplace transform. The
transfer matrix representation is commonly used to validate reduced models in the model order
reduction community and is commonly called the transfer function.

Definition 2.1.1.

(Laplace transform) The Laplace transform of a function f(t) in the time domain is the function
F(s) in the frequency domain and it is defined as


() = () = () , = + ,
0

2.2 Linear Constant Coefficient DAEs

In this section, we discuss that analysis of the linear transformation idea for DAEs. For
simplicity, the coefficient matrix D in (2.7) is taken to be the zero matrix unless specified. Thus,
(2.7) simplifies to:
() () = () + () , (0) = 0

() = ()

where E is singular, A , B , C and u(t) is the input vector,


() is the output vector and 0 is the initial value.

2.3 Application

The Flying Wheelset

In 1994, one year after the flying wheelset had taken off, Sebastian Reich contacted Bernd
Simeon to send him the source code so that Sebastian Reich could run some numerical tests with
it. At that time, Bernd Simeon was working on a class of stabilization methods that are related to
Baumgartes approach, jointly with Uri Ascher, H Chin and Linda Petzold [ACPR95]. The
wheelset seemed a perfect example for these methods, but Sebastian Reich discovered some
strange results that contradicted the theory and made him suspicious. We received an email
where him described the results and questioned our Fortran code. This request made Claus
Fuhrer and Bernd Simeon cross-check the code that we had written five years before when
working on the survey paper [SFR91]. And it turned out that the constraints on acceleration level
had a flaw that had been introduced when merging two blocks of output from a computer algebra
program. A simple - sign was false, and after having corrected it, the drift off was drastically
reduced the wheelset had landed.

In conclusion, the cross-checking of numerical results and the analytical results problems
are absolutely.

You might also like