a question on visualizing a state of DAE and a question from a continous time nonlinear dynamics

76 Views Asked by At
  1. Could someone explain me what is the fundamental difference between the dynamical system of the kind $\dot x = f(x)$ and $E \dot x= f(x)$ where $E$ is a singular matrix with real entries. For the first system, a state is $x(t)$ at time $t$; is it the same for the second system? I wish to visualise trajectories of such type of dynamical system to be more precise.

  2. Consider a continuous time dynamical system \begin{align} \dot x= f\left(x\right) \end{align} on a state-space $\mathcal{X}$, where $x$ is a coordinate vector of the state, $f$ is a non-linear vector valued smooth function of the same dimension as its argument $x$.

    Could anyone tell me why assume $f$ to be smooth and what it means by ''function of the same dimension as its argument $x$'', what is a dimension of function, by the way, and why its dimension needs to be equal as $x$?

Thank you for your clarification.

1

There are 1 best solutions below

0
On BEST ANSWER
  1. The first kind is the classic ordinary differential equation (ODE), the 2nd is a generalized version of it. The matrix $E$ is sometimes called the mass matrix (in analogy of Newtonian Mechanism). In the most general form, it could be state and/or time dependent.

a. If the matrix $E$ is invertible/nonsingular, then we can easily convert the 2nd form into the 1st. $$ \dot x = E^{-1}f(x) $$ So mathematically they are the exact same system, but writing it in the 2nd forms could make the numerical solver's life easier. Why? Intuitively if $E$ has a large condition number, i.e. some variables change very fast, some change very slow, then the numerical solver could have a hard time determine step size without this knowledge. Check out Stiff Equation

b. If the matrix $E$ is singular, this is no longer a classic ODE. Consider a vector $v\neq 0$ s.t. $v^TE=0$ and a vector $u\neq 0,s.t. Eu^T=0$, then we actually get a constraint equation along the projection $v$. $$ 0=v^TE\dot x=v^T f(x) $$ With a $d$ dimensional null space of $E$ we get $d$ constraint equations, which defines the $n-d$ dimension manifold which the solution trajectory shall live on.

One simple example $$ E = \begin{bmatrix}1& 0\\0&0\end{bmatrix}\\ E[\dot x,\dot y]^T= [-y, x^2+y^2-1]^T $$ Then the dynamics of $y$ is not explicitly defined, but we got a constraint function of $x,y$ s.t. the trajectory shall live on the unit circle $x^2+y^2-1=0$. Then the dynamics of $y$ could be figured out from the constraint via implicit function theorem.

Also check this out, from computational perspective. How to handle mass matrix


  1. "function of the same dimension as its argument x'', what is a dimension of function, by the way, and why its dimension needs to be equal as x?"

This simply means if $x\in \mathbb R^n$, then $f:\mathbb R^n\to \mathbb R^n$. The simple reason if $x\in \mathbb R^n$, then its time derivative shall also be a vector in $n$ dimension vector space. This is like if your position in in 3d space, then your velocity vector is also a 3d vector. or you have no way adding them up.

Geometrically, each dynamic system/ODE describes a vector field on the space $\mathcal{X}$, these vectors live in the tangent space of the manifold $\mathcal{X}$. In this case if $x$ live in $n$ dim Euclidean space, its tangent space is also $n$ dim.

For smoothness, I feel it's a requirement from technical convenience to prove something down the line.... In real life we don't necessarily use smooth $f$