The idea of a moduli space is that points in the space parameterize isomorphism classes of some objects.
Goal: Given a vector space V, you want to classify all k-dimensional subspaces, where the
isomorphism classes are isomorphisms of sets
(bijections). So two different subspaces of dimension k are considered equivalent if and
only if they're literally the exact same set.
Notation: Grk(V)
is the set of all k-dimensional subspaces of a vector
space V.
Construction: For each subspace, we
choose a basis. Once we have a basis, all other choices of basis for
that particular subspace are just going to be given by actions of the
general linear group GLk.
Choosing a basis for a k-dimensional subspace is the same
as picking a rank k, (n×n) matrix
The problem is that if we use n × n matrices, we not only need to quotient out by the action of GLk, we also need to quotient out by switching things into the (n−k) other positions. For example, if k = 2, n = 10 then having a matrix of all zeros, except in the first two rows should correspond to the same subspace as the matrix that moves the non-zero rows to the bottom.
To avoid this extra redundancy (or gauge freedom) we can use Mat*(k,n), the set of (k×n) matrices of rank k. Then we have
Grk(V) ≅ Mat*(k,n)/GLk.
Instead of working with any basis, we can do the same type of construction, restricting to orthonormal bases:
We first take any orthonormal basis on V, which corresponds to an element of O(n).
We use projections
Grk = {P ∣ P2 = P, P = P†, Tr(P) = k}
Physics: Gr1(ℂn)
is then the space of pure states
The fiber above any point in the Grassmanian is literally the subspace that point represents