Mon, 27 Aug 2018
[ Epistemic status: uninformed musings; anything and everything in here might be wrong or ill-conceived. ]
Suppose !!V!! is some vector space, and let !!V_n!! be the family of all !!n!!-dimensional subspaces of !!V!!. In particular !!V_1!! is the family of all one-dimensional subspaces of !!V!!. When !!V!! is euclidean space, !!V_1!! is the corresponding projective space.
if !!L!! is a nonsingular linear mapping !!V\to V!!, then !!L!! induces a mapping !!L_n!! from !!V_n\to V_n!!. (It does not make sense to ask at this point if the induced mapping is linear because we do not have any obvious linear structure on the projective space. Or maybe there is but I have not heard of it.)
The eigenspaces of !!V!! are precisely the fixed points of !!L_n!!.
(Wrong! Any subspace generated by an !!n!!-set of eigenvectors is a fixed point of !!L_n!!. But such a subspace is not in general an eigenspace. (Note this includes the entire space as a special case.) The converse, however, does hold, since any eigenspace is generated by a subset of eigenvectors.)
Now it is an interesting and useful fact that for typical mappings, say from !!\Bbb R\to\Bbb R!!, fixed points tend to be attractors or repellers. (For example, see this earlier article.) This is true of !!L_1!! also. One-dimensional eigenspaces whose eigenvalues are larger than !!1!! are attractors of !!L_1!!, and those whose eigenvalues are smaller than !!1!! are repellers, and this is one reason why the eigenspaces are important: if !!L!! represents the evolution of state space, then vectors in !!V!! will tend to evolve toward being eigenvectors.
So consider, say, the projective plane !!\Bbb P^2!!, under the induced mapping of some linear operator on !!\Bbb R^3!!. There will be (typically) three special points in !!\Bbb P^2!! and other points will typically tend to gravitate towards one or more of these. Isn't this interesting? Is the three-dimensional situation more interesting than the two-dimensional one? What if a point attracts in one dimension and repels in the other? What can the orbits look like?
Or consider the Grassmanian space !!Gr(2, \Bbb R^3)!! of all planes in !!\Bbb R^3!!. Does a linear operator on !!\Bbb R^3!! tend to drive points in !!Gr(2, \Bbb R^3)!! toward three fixed points? (In general, I suppose !!Gr(k, \Bbb R^n)!! will have !!n\choose k!! fixed points, some of which might attract and some repel.) Is there any geometric intuition for this?
I have been wanting to think about Grassmanian spaces more for a long time now.