5  DAEs Forms

Author
Affiliations

Enrico Bertolazzi

University of Trento, Department of Industrial Engineering

Many physical systems are governed not by ordinary differential equations (ODEs) but by differential-algebraic equations (DAEs), which combine differential dynamics with algebraic constraints. A general DAE can be written as:

\bm{f}(\bm{x}'(t), \bm{x}(t), t) = \bm{0},

where bm{x}(t) \in \Bbb{R}^n is the state vector, \bm{x}'(t) is its time derivative, and \bm{f} represents a (possibly nonlinear) vector function.

Unlike ODEs, DAEs may not permit direct solution for \bm{x}'(t) due to the potential singularity of the Jacobian:

\dfrac{\partial \bm{f}}{\partial \bm{x}'} = \dfrac{\partial \bm{f}(\bm{x}', \bm{x}, t)}{\partial \bm{x}'}.

When this Jacobian is singular (i.e., not invertible), implicit or indirect techniques are needed. The differential index of a DAE quantifies how many times we must differentiate the equation to isolate \bm{x}'(t).

5.1 Semi-Explicit Form of DAEs

A more structured (and often more tractable) form of DAEs is:

\bm{A}(\bm{x}, t) \bm{x}' = \bm{g}(\bm{x}, t),

where \bm{A}(\bm{x}, t) is singular. When \bm{A}(\bm{x}, t) has a block-diagonal structure:

\bm{A}(\bm{x}, t) = \begin{pmatrix} \bm{I} & \bm{0} \\ \bm{0} & \bm{0} \end{pmatrix},

the DAE is said to be in semi-explicit form.

Why is this useful?

  • The semi-explicit form naturally separates the system into differential and algebraic components:
    • Differential part: Explicit ODEs (e.g., \bm{y}' = \bm{h}(\bm{y}, \bm{z}, t)).
    • Algebraic part: Constraints (e.g., \bm{0} = \bm{k}(\bm{y}, \bm{z}, t)).
  • This separation simplifies index analysis (discussed later).
NoteGeneral DAE to semi-explicit form

A generic DAE:

\bm{f}(\bm{x}',\bm{x},t)=\bm{0}

can always be rewritten in semi-explicit form by introducing the auxiliary variable \bm{x}_{\mathrm{dot}}(t)=\bm{x}'(t):

\begin{aligned} \bm{x}'&= \bm{x}_{\mathrm{dot}} \\[1em] \bm{0} &= \bm{f}(\bm{x}_{\mathrm{dot}},\bm{x},t) \end{aligned}

Important:

  • This transformation increases the differential index by 1.
  • For example, an index-1 DAE becomes index-2 after conversion.

5.2 Index-0 DAEs are ODEs

By definition, index-0 DAEs are simply ODEs of the form:

\bm{x}'(t) = \bm{f}(\bm{x}, t),

where the system is fully explicit with no algebraic constraints.

5.3 Index-1 DAEs in Semi-Explicit Form

An index-1 DAE in semi-explicit form is:

\left\{ \begin{aligned} \bm{x}' &= \bm{f}(\bm{x}, \bm{y}, t) \quad &\text{(Differential part)}, \\[0.5em] \bm{0} &= \bm{g}(\bm{x}, \bm{y}, t) \quad &\text{(Algebraic part)}, \end{aligned} \right.

where:

  • \bm{x} \in \Bbb{R}^n are the differential variables,
  • \bm{y} \in \Bbb{R}^m are the algebraic variables,
  • \bm{f} \in C^1(\Bbb{R}^{n+m+1}, \Bbb{R}^n),
  • \bm{g} \in C^1(\Bbb{R}^{n+m+1}, \Bbb{R}^m).

The Jacobian of \bm{g} w.r.t. \bm{y},

\bm{g}_{\bm{y}}(\bm{x}, \bm{y}, t) = \frac{\partial \bm{g}}{\partial \bm{y}},

must be non-singular. This ensures that the Implicit Function Theorem applies, allowing us to express \bm{y} as a function of \bm{x} and t:

\bm{y} = \bm{y}(\bm{x}, t).

Substituting back into the differential part reduces the DAE to an ODE:

\bm{x}'(t) = \bm{g}(\bm{x}(t),t),\qquad\textrm{where}\qquad \bm{g}(\bm{x},t)= \bm{f}(\bm{x},\bm{y}(\bm{x},t),t)

5.4 The Implicit Function Theorem

The Implicit Function Theorem ensures that the algebraic constraints can be locally solved for \bm{y}.

Theorem 5.1 Let A\subset\Bbb{R}^{n+m} an open set and \bm{f}\in C^1(A,\Bbb{R}^m), moreover

  1. (\bm{x}_0,\bm{y}_0)\in A;

  2. \bm{f}(\bm{x}_0,\bm{y}_0)=\bm{0};

  3. \dfrac{\partial\bm{f}(\bm{x}_0,\bm{y}_0)}{\partial\bm{y}} non-singular.

Then, there exists two open set

U\subset\Bbb{R}^n\quad\text{and}\quad V\subset\Bbb{R}^m

and a function \bm{\phi}:U\mapsto V such that

  • \bm{y}_0=\bm{\phi}(\bm{x}_0);

  • \bm{f}(\bm{x},\bm{y})=\bm{0} iff \bm{y}=\bm{\phi}(\bm{x}) (for \bm{x}\in U and \bm{y}\in V);

  • \bm{\phi}\in C^1(U,V)

and

\dfrac{\partial\bm{\phi}(\bm{x})}{\partial\bm{x}}=- \left(\dfrac{\partial\bm{f}(\bm{x},\bm{\phi}(\bm{x}))}{\partial\bm{y}}\right)^{-1} \dfrac{\partial\bm{f}(\bm{x},\bm{\phi}(\bm{x}))}{\partial\bm{x}}

Illustrative Example: A Simple Index-1 DAE

Consider the semi-explicit DAE:

\begin{aligned} x' &= -y, \\ 0 &= x + y - \sin(t). \end{aligned}

Step 1: Apply the Implicit Function Theorem

The algebraic equation is:

y = \sin(t) - x.

Step 2: Substitute Back into the ODE

The system reduces to:

x' = -\sin(t) + x,

which is an ODE solvable with standard methods.

DAE of index 2

DAE of index 1 are in practice ODE with implicit parameters. Index 2 DAE are more complex

\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{g}(\bm{x},t) \end{aligned}

where \bm{x}(t)\in\Bbb{R}^n, \bm{y}(t)\in\Bbb{R}^m, \bm{f}\in C^1(\Bbb{R}^{n+m+1},\Bbb{R}^n) and \bm{g}\in C^1(\Bbb{R}^{n+1},\Bbb{R}^m). Moreover

\dfrac{\partial\bm{g}(\bm{x},t)}{\partial\bm{x}} \times\dfrac{\partial\bm{f}(\bm{x},\bm{y},t)}{\partial\bm{y}}

ia a non singular matrix. In this case implicit function Theorem cannot be used to determine \bm{y} as a function of \bm{x}.

To check the index perform one derivation of the constraints

\dfrac{\mathrm{d}\bm{g}(\bm{x}(t),t)}{\mathrm{d}t}= \dfrac{\partial\bm{g}(\bm{x}(t),t)}{\partial\bm{x}}\bm{x}'(t)+ \dfrac{\partial\bm{g}(\bm{x}(t),t)}{\partial t}=\bm{0}

obtaining the new (hidden) constraints

\bm{h}(\bm{x},\bm{y},t)=\bm{0}

where

\bm{h}(\bm{x},\bm{y},t)= \dfrac{\partial\bm{g}(\bm{x},t)}{\partial\bm{x}} \bm{f}(\bm{x},\bm{y},t)+ \dfrac{\partial\bm{g}(\bm{x},t)}{\partial t}

Thus, a new DAE is obtained

\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{h}(\bm{x},\bm{y},t) \end{aligned}

The derived DAE is now of index 1:

\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{h}(\bm{x},\bm{y},t) \end{aligned}

to check it must be

\dfrac{\partial\bm{h}(\bm{x},\bm{y},t)}{\partial\bm{y}}= \dfrac{\partial\bm{g}(\bm{x},t)}{\partial\bm{x}}\times \dfrac{\partial\bm{f}(\bm{x},\bm{y},t)}{\partial\bm{y}}

non singular.

5.5 DAE of index 3

Index 3 DAE semi-esplicit take the form

\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},\bm{z},t)\\[1em] \bm{y}' &= \bm{g}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{h}(\bm{y},t) \end{aligned}

where

\bm{x}(t)\in\Bbb{R}^n\qquad \bm{y}(t)\in\Bbb{R}^m\qquad \bm{z}(t)\in\Bbb{R}^p \\[1em] \bm{f}\in C^1(\Bbb{R}^{n+m+p+1},\Bbb{R}^n)\quad \bm{g}\in C^1(\Bbb{R}^{n+m+1},\Bbb{R}^m)\quad \bm{h}\in C^1(\Bbb{R}^{m+1},\Bbb{R}^n)

To check the index derive the constraints

\dfrac{\mathrm{d}\bm{h}(\bm{y}(t),t)}{\mathrm{d}t}= \dfrac{\partial\bm{h}(\bm{y}(t),t)}{\partial\bm{y}}\bm{y}'(t)+ \dfrac{\partial\bm{h}(\bm{y}(t),t)}{\partial t}=\bm{0}

obtaining the new (hidden) constraints

\bm{a}(\bm{x},\bm{y},t)=\bm{0}

where

\bm{a}(\bm{x},\bm{y},t)= \dfrac{\partial\bm{h}(\bm{y},t)}{\partial\bm{y}}\bm{g}(\bm{x},\bm{y},t)+ \dfrac{\partial\bm{h}(\bm{y},t)}{\partial t}

Thus, a new DAE is obtained

\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},\bm{z},t)\\[1em] \bm{y}' &= \bm{g}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{a}(\bm{x},\bm{y},t) \end{aligned}

The DAE

\begin{aligned} \begin{pmatrix} \bm{x}' \\ \bm{y}' \end{pmatrix} &= \begin{pmatrix} \bm{f}(\bm{x},\bm{y},\bm{z},t)\\ \bm{g}(\bm{x},\bm{y},t) \end{pmatrix}\\[1em] \bm{0} &= \bm{a}(\bm{x},\bm{y},t) \end{aligned}

is index 2 iff

\dfrac{\partial\bm{a}(\bm{x},\bm{y},t)}{\partial\bm{x}}\times \dfrac{\partial\bm{f}(\bm{x},\bm{y},\bm{z},t)}{\partial\bm{z}}+ \dfrac{\partial\bm{a}(\bm{x},\bm{y},t)}{\partial\bm{y}}\times \dfrac{\partial\bm{g}(\bm{x},\bm{y},t)}{\partial\bm{z}}

is non singular, i.e.

\dfrac{\partial\bm{h}(\bm{y},t)}{\partial\bm{y}}\times \dfrac{\partial\bm{g}(\bm{x},\bm{y},t)}{\partial\bm{x}}\times \dfrac{\partial\bm{f}(\bm{x},\bm{y},\bm{z},t)}{\partial\bm{z}}

is non singular.

5.6 DAE in semi-esplicit form

DAE of index 1

\left\{\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{g}(\bm{x},\bm{y},t) \end{aligned}\right. \qquad \qquad \dfrac{\partial\bm{g}(\bm{x},\bm{y},t)}{\partial\bm{y}}\quad \textbf{non singular}

DAE of index 2

\left\{\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{g}(\bm{x},t) \end{aligned}\right. \qquad \underbrace{ \dfrac{\partial\bm{g}(\bm{x},t)}{\partial\bm{x}} \times\dfrac{\partial\bm{f}(\bm{x},\bm{y},t)}{\partial\bm{y}} }_{\textbf{non singular}}

DAE of index 3

\left\{\begin{aligned} \bm{x}' &= \bm{f}(\bm{x},\bm{y},\bm{z},t)\\[1em] \bm{y}' &= \bm{g}(\bm{x},\bm{y},t)\\[1em] \bm{0} &= \bm{h}(\bm{y},t) \end{aligned}\right. \quad \underbrace{ \dfrac{\partial\bm{h}(\bm{y},t)}{\partial\bm{y}}\times \dfrac{\partial\bm{g}(\bm{x},\bm{y},t)}{\partial\bm{x}}\times \dfrac{\partial\bm{f}(\bm{x},\bm{y},\bm{z},t)}{\partial\bm{z}} }_{\textbf{non singular}}

DAE of index \mu (in the Hessemberg form)

The DAE

\begin{aligned} x_1' &= f_1(x_1,x_2,\ldots,x_{\mu},t)\\[1em] x_2' &= f_2(x_1,x_2,\ldots,x_{\mu-1},t)\\[1em] &\vdots&\\ x_k' &= f_k(x_{k-1},x_k,\ldots,x_{\mu-1},t)\qquad k=3,\ldots,\mu-1\\[1em] &\vdots\\ 0 &= f_{\mu}(x_{\mu-1},t) \end{aligned}

with

\dfrac{\partial f_{\mu}}{\partial x_{\mu-1} } \times \dfrac{\partial f_{\mu-1}}{\partial x_{\mu-2} } \times\cdots\times \dfrac{\partial f_{2}}{\partial x_1} \times \dfrac{\partial f_{1}}{\partial x_{\mu} }

non singular is a DAE of index \mu.