Small oscillations
In this notebook we consider a model of a two dimensional linear physical system which conserves energy. There are very many physical systems which can be described by this mathematics. Our linear -dimensional system is governed by Newton's law . We assume that the force doesn't depend on the velocity (for example, a particle in some potential). In our notation, this means Our assumption that the system is linear means that for some matrix . We can choose units where . Thus our system is
Just as we convert an -th order one-dimensional ODE to a -dimensional system of ODEs, we can convert this 2nd order 2-dimensional system into a 1st order 4-dimensional system as follows: let Then define and . It follows that and . In vector notation, we have , so that . Our system can then be written in the form of a 1st order linear system, which is given by a block matrix as follows:
where our matrix is a matrix made up of blocks.
Let's use our last assumption in the problem, that the system conserves energy. Mathematically the corresponds to the fact that the eigenvalues of cannot have non-zero real part, since if this were the case there would be factors in the solutions, which either contract to or escape to , with exponentially increasing or decreasing velocity. It follows that the eigenvalues of are purely imaginary. Since is a real matrix, its eigenvalues must be of the form so that our general solution (let us ignore the issue of Jordan blocks...) is for vectors .
It's clear that we have two decoupled oscillatory systems, one in the and directions with frequency , and another in the and directions with frequency . In hindsight we can say more: our original matrix must have had eigenvalues and because of the possible choices for the second derivative of our solution (1). Thus our original system decouples into a pair of oscillators and , after a linear change of coordinates.
Since we cannot easily visualize the 4-dimensional phase space of the system, let's only plot the -dimensional coordinates . All of this analysis shows that, after a linear change of coordinates to diagonalize , we know the solutions are of the form Shifting the time coordinate we can assume that , while subsequently rescaling , we can assume . So there are really only 2 free parameters, and . We will now plot these solution curves. A few remarks:
These curves are known as Lissajous curves.
Note that in the 4-dimensional system the curves do not intersect (this would violate the uniqueness half of the Existence/Uniqueness theorem), but when we project to two dimensions, we get intersecting solution curves.
Along the same lines, as we vary we can alternatively interpret this literally as a rotation in the -dimensional phases space, or a -dimensional subspace thereof. It is a good exercise to make this precise mathematically. It explains the phenomenon we notice as we vary .
Notice also the following: if is a rational number then the curve eventually returns to itself and repeats. On the other hand, if then the curve will eventually fill the entire rectangle. Why is this?
We can see that it appears to rotate in a higher-dimensional space as we vary , despite our problem initially appearing 2-dimensional. These patterns show up on oscilloscopes (this rotation here comes from slightly irrational , which end up being equivalent to varying .) Regardless we can actually plot a third component of our -dimensional system and see that we really do have a rotating figure in a higher dimensional space. Our 2d picture we started with doesn't tell the whole story.
The picture we have plots the and components of the vectors of our solution. If we plotted the component we would get the 3d rotation but our curve will still intersect itself. So we choose a rather strange slice of the 4d system where we plot the coordinates just to demonstrate that the solution curve actually doesn't intersect itself, and the rotation we see in the 2d picture actually has a physical interpretation in terms of a rotation in the coordinates and .
Since it is hard for sage to update quickly enough to animate this and see rotation as we vary , here is an animation of the 3d rotation for a few values of and the corresponding 2d Lissajous figures.
Ratio | 3d Figure | 2d Figure |
---|---|---|