Proposition 1 — is recovered from through bounded weak** nets
ProvedRockafellarMaxMono.Maximality.conj_subdiff_iff_netsLet be a real Banach space with dual and bidual , let be a lower semicontinuous proper convex function on , and let and . Then if and only if there exist
- a nonempty directed partially ordered index set ,
- a net in converging to in the strong (norm) topology, and
- a bounded net in , with the same index set, converging to in the weak** topology,
such that for every . Here is regarded as a subspace of through the canonical embedding, and the weak** topology is the weak topology induced on by :
In a nonreflexive space maps into and is not the inverse of ; this proposition says that is nevertheless completely determined by .
Formalization Note A net is a function on a type carrying a partial order that is directed and nonempty, with convergence along the filter atTop; the index type lives in the same universe as . Weak** convergence of to is stated as pointwise convergence on of the canonical images of the , which is exactly convergence in the weak topology induced on by . Boundedness is for a real constant and all .
import Mathlib import Definitions.Def_RockafellarMaxMono_Shared_ProperConvex import Definitions.Def_RockafellarMaxMono_Shared_Subdiff import Definitions.Def_RockafellarMaxMono_Shared_Conj open Filter Topology universe u
namespace RockafellarMaxMono.Maximality
theorem conj_subdiff_iff_nets {E : Type u} [NormedAddCommGroup E] [NormedSpace ℝ E]
[CompleteSpace E] (f : E → EReal) (hf : Shared.ProperConvex f) (hlsc : LowerSemicontinuous f)
(x' : StrongDual ℝ E) (x'' : StrongDual ℝ (StrongDual ℝ E)) :
x'' ∈ Shared.subdiff (Shared.conj f) x' ↔
∃ (I : Type u) (_ : PartialOrder I) (_ : IsDirected I (· ≤ ·)) (_ : Nonempty I)
(xs' : I → StrongDual ℝ E) (xs : I → E),
Tendsto xs' atTop (𝓝 x') ∧
(∃ C : ℝ, ∀ i, ‖xs i‖ ≤ C) ∧
(∀ y' : StrongDual ℝ E,
Tendsto (fun i => NormedSpace.inclusionInDoubleDual ℝ E (xs i) y') atTop (𝓝 (x'' y'))) ∧
∀ i, xs' i ∈ Shared.subdiff f (xs i) := by sorry
end RockafellarMaxMono.Maximality
Read-back
What the Lean code literally says, in plain math · claude-opus-5-5
Setting. is a real Banach space, is its continuous dual, and is its bidual. The canonical embedding is given by . The function has two hypotheses:
- satisfies the imported predicate
Shared.ProperConvex. It is not shown here; the name suggests "proper convex". - is lower semicontinuous for the norm topology.
Two imported operations appear, not shown here:
- is
Shared.conj f. The name suggests the Fenchel conjugate. - is
Shared.subdiff. The name suggests the subdifferential. Applied to at a point of , it gives a subset of . Applied to at a point of , it gives a subset of .
Fix and .
Claim. holds if and only if there exist the following:
- an index set in the same type universe as , with some partial order on , such that is nonempty and directed (any two indices have a common upper bound);
- a net in and a net in .
These must satisfy four conditions, where limits are taken along in the direction of increasing :
- in the norm of .
- The net is norm-bounded: there is a real with for all .
- For every , in . In other words, pointwise on (weak-* convergence).
- For every , .
Degenerate cases. Because must be nonempty and directed, the limits cannot hold vacuously. The bound may be any real number; it is not required to be positive. Choosing with one element turns the right-hand side into for some with . If , all three spaces are and both sides concern only and .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.