Theorem 5.6 — spectral theorem for compact symmetric operators on an inner product space
ProvedTeschlODE.SturmLiouville.spectral_theorem_compact_symmetricLet be a complex inner product space (not necessarily complete) and a compact symmetric operator. Then there are , real numbers and vectors () such that
- and is an orthonormal set (normalized eigenvectors);
- if is infinite dimensional, and otherwise;
- if then ;
- every can be written as
the partial sums converging to in ;
- if is dense, the form an orthonormal basis: the expansion (5.42) holds for every (the book's (5.34)).
Formalization Note. The book writes with and proves the theorem assuming infinite dimensional "without loss of generality", where its construction produces an infinite sequence; in finite dimension the construction stops after steps. The statement makes this explicit with N : ℕ∞ counting the eigenvectors (indices ) and the two clauses on . The series is the limit of as in the norm of . "Orthonormal basis" is the book's (5.32)/(5.34), not Mathlib's HilbertBasis, which needs completeness.
import Mathlib import Definitions.Def_TeschlODE_SturmLiouville_IsCompactOp
namespace TeschlODE.SturmLiouville
/-- Teschl, Theorem 5.6 (spectral theorem for compact symmetric operators), p. 151: let `H₀` be
a complex inner product space and `A : H₀ → H₀` compact and symmetric. There are `N ∈ ℕ₀ ∪ {∞}`,
real eigenvalues `αⱼ` and normalized eigenvectors `uⱼ` (`j < N`) forming an orthonormal set,
with `N = ∞` when `H₀` is infinite dimensional (and `N = dim H₀` otherwise) and `αⱼ → 0` when
`N = ∞`, such that every `f ∈ Ran(A)` satisfies `f = Σ_{j<N} ⟨uⱼ, f⟩ uⱼ` (5.42), the partial sums
converging in `H₀`; if `Ran(A)` is dense, the `uⱼ` form an orthonormal basis, i.e. the same
expansion holds for every `f ∈ H₀` (5.34). -/
theorem spectral_theorem_compact_symmetric {E : Type*} [NormedAddCommGroup E]
[InnerProductSpace ℂ E] (A : E →ₗ[ℂ] E) (hA : IsCompactOp A) (hsym : A.IsSymmetric) :
∃ (N : ℕ∞) (α : ℕ → ℝ) (u : ℕ → E),
(FiniteDimensional ℂ E → N = (Module.finrank ℂ E : ℕ∞)) ∧
(¬ FiniteDimensional ℂ E → N = ⊤) ∧
(∀ j : ℕ, (j : ℕ∞) < N → A (u j) = (α j : ℂ) • u j) ∧
Orthonormal ℂ (fun j : {j : ℕ // (j : ℕ∞) < N} => u j) ∧
(N = ⊤ → Filter.Tendsto α Filter.atTop (nhds 0)) ∧
(∀ f ∈ LinearMap.range A,
Filter.Tendsto
(fun n => ∑ j ∈ (Finset.range n).filter (fun j : ℕ => (j : ℕ∞) < N),
inner ℂ (u j) f • u j) Filter.atTop (nhds f)) ∧
(Dense (LinearMap.range A : Set E) → ∀ f : E,
Filter.Tendsto
(fun n => ∑ j ∈ (Finset.range n).filter (fun j : ℕ => (j : ℕ∞) < N),
inner ℂ (u j) f • u j) Filter.atTop (nhds f)) := by sorry
end TeschlODE.SturmLiouville
Read-back
What the Lean code literally says, in plain math · claude-opus-5-5
This theorem is stated for a complex inner product space , not assumed complete and possibly , and a -linear map , not assumed continuous. The hypotheses are:
- Compactness. Every sequence with bounded set of values has a subsequence along which the images under converge in norm to a point of .
- Symmetry. for all , with the inner product conjugate-linear in the first argument.
The conclusion asserts the existence of:
- ,
- a real sequence ,
- a sequence in ,
such that all of the following hold:
- If is finite-dimensional, then .
- If is not finite-dimensional, then .
- For every with , .
- The family is orthonormal: , and for , both .
- If , then as .
- For every in the range of , the partial sums
converge to in the norm of as . 7. If the range of is dense in , then for every .
The values for are unconstrained. The eigenvalues are not required to be distinct, nonzero, or ordered. In the finite-dimensional case, eigenvectors with eigenvalue are permitted among the .
Degenerate cases:
- If , then is finite-dimensional with dimension , so . Items 3–5 are then vacuous (the orthonormal family is empty), every partial sum is , and the only is , so items 6–7 hold trivially.
- If , the range is , so item 6 is trivial. Item 7 applies only if is dense, that is, only if .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.