# What is the difference between an implicit ordinary differential equation and a differential algebraic equation?

I'm rather confused on this particular point. What is the difference between an implicit ordinary differential equation of the form:

x' = f(x',x,t);


and a differential algebraic equation of the form:

f(x',x,t) = 0;


I've read some documentation on them online, and while some places mention that implicit ODEs are a special class of DAEs, they're not very clear on the point. Also, is there any specific consideration that needs to be made in regards to numerical solutions of such problems?

-
– lhf May 24 '13 at 23:49
Thanks. I was reading "Solving Ordinary Differential Equations" by Hairer and their explanation wasn't very clear, however, the book you recommended cleared things up. – Kimusubi May 25 '13 at 19:47

I think you have done some mistakes when writing down your equations. An implicit ordinary differential equation can be written: $$f(t,x,x') = 0$$ and a DAE can be written: $$f(t,x,x') = 0$$ and since these two equations are syntactically equal, it is very easy to be confused about the distinction.

First, I want to make one thing clear. When talking about ODEs it is quite alright to talk about just one equation, e.g. the equation $$y'(t) + y(t) + t = 0$$ is an ordinary differential equation.

But when talking about a DAE, you always, in a non-trivial case, talk about a system of equations. If your DAE contains only one equation it will either be a differential equation or an algebraic equation (in this context, algebraic means not containing any derivatives). Thus, in the rest of the post, $f$ and $y$ will be vectors of functions.

Thus, the difference between an implicit ODE system and a DAE system is, in a way, that the DAE system can contain purely algebraic equations and variables. The more technical and correct criterion is that the Jacobian $$\frac{\partial f(t,x,x')}{\partial x'}$$ needs to be non-singular for the system $f(t,x,x') = 0$ to be classified as an implicit ODE.

To make the distinction more clear between a DAE and an implicit ODE, you can split the vector $x$ in two parts, $x_D$, containing the $x$ for which derivatives occur in the DAE and $x_A$, containing the algebraic $x$, i.e. the $x$ for which no derivative occur in the DAE, and we write the DAE on the form $$f(t,x_A,x_D,x_D') = 0.$$ we can also split the function $f$ into two parts: $f_D$ containing the equations containing derivatives and $f_A$ not containing any derivatives.

A classic example of a DAE is the following formulation for the motion of a pendulum: \begin{align} 0 &= x' - u \\ 0 &= y' - v \\ 0 &= u' - \lambda x \\ 0 &= v' - \lambda y - g\\ 0 &= x^2 + y^2 - L^2 \end{align} where $L$ (the length of the pendlum) and $g$ (gravitational acceleration) are constants. Classifying the variables as differential (belonging to $x_D$) and algebraic (belonging to $x_A$), we see that $x,y,u,v$ are differential and $\lambda$ is algebraic. All equations except the last (the length constraint) are differential.

We can calculate the Jacobian of this system. We order the functions as above and the variables as follows: $(x,y,u,v,\lambda)$. Then the Jacobian will be: $$\begin{pmatrix} 1 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 & 0 \end{pmatrix}$$ which is singular since the last row ad the last column is zero, hence the system is a DAE.

One often wants to apply techniques to a DAE to transform it to a semi-explicit DAE of index 1, which can be written as follows: \begin{align} x_D' &= g_D(t,x_D,x_A) \\ 0 &= g_A(t,x_D,x_A) \end{align} because then $g_A$ can, in theory, be solved for $y$, which can then be inserted into $g_D$, which can then be integrated numerically.

The pendulum example might look as it is on this form, but its index is not 1.

-