Intuition behind the spectral theorem in infinite dimensions

82 Views Asked by At

I would like to verify my intuition behind the spectral theorem in infinite dimensions. For the moment I am putting bounded/unbounded and domain issues aside.

In finite dimensions the spectral theorem says that for any self-adjoint/Hermitian matrix $A$, $$A = U^{-1}\Lambda U $$ where $\Lambda$ is a diagonal matrix consisting of the eigenvalues of $A$ and $U$ is a change of basis matrix. The interpretation here is that $U$ maps to a basis that is a direct sum of eigenspaces. In each eigenspace the the action of $A$ is multiplication by the appropriate eigenvalue. Thus we may write $$A = \sum_i^n \lambda_i P_i \tag{1}$$ where $P_i$ is the projection onto the eigenspace corresponding to $\lambda_i$.

Moving to infinite dimensions, let $A$ be a self-adjoint operator on some Hilbert space $H$. Unlike the finite dimensional case, in general the spectrum of $A$:

  1. Is no longer countable.
  2. Contains elements that are not eigenvalues (i.e. the continuous part of the spectrum).

To address the first point, the sum in (1) is now made into an integral: $$A = \int_{\sigma(A)} \lambda dP_\lambda. \tag{2}$$ For points in the spectrum of $A$ that are eigenvalues, the integral (2) is to be interpreted similarly to (1). That is, we project onto the eigenspace spanned by $\lambda$ and the action of $A$ on this subspace is to multiply by $\lambda$.

Making sense of (2) for points in the continuous spectrum addresses the second point above. The operator $P_\lambda$ for $\lambda$ in the continuous spectrum is the zero projector. To account for the contribution of the continuous spectrum we make sense of (2) in a limiting sense similar to how the Lebesgue integral is constructed. That is, we consider Borel sets $\Omega$ containing an approximate eigenvalue and interpret $P_\Omega$ as projecting onto an approximate eigenspace where the action of $A$ is approximately multiplication by $\lambda$. Its contribution to (2) is obtained as $\Omega \rightarrow \{\lambda\}$ in some sense, again similar to how we use simple functions to construct the Lebesgue integral and then take a limit.

I know the above is a bit handwav and can be made more precise using distribution functions $P_\lambda = P((-\infty, \lambda])$ to define a Stieltjes integral, but my goal is to build an intuition for now. Is my intuition of the spectral theorem and the integral (2) correct?