TorchLean API

FloatLib.Numerics.Automation.Interval.Bounds

Hypothesis bounds and certified input boxes #

Direct and pointwise hypotheses supply rational endpoints for the real atoms discovered by the reifier. The strongest endpoints form the input box, with a proof of each coordinate's membership. Strict hypotheses are conservatively weakened; unresolved indices or guards supply no bounds.

A rational lower or upper bound, carrying a proof of the real inequality.

  • term : Q(ℝ)

    Real term bounded by this hypothesis.

  • value : ℚ

    Rational endpoint, used to select the strongest available bound.

  • rational : Q(ℚ)

    Quoted syntax of that endpoint.

  • lower : Bool

    Whether the endpoint is a lower bound rather than an upper bound.

  • proof : Lean.Expr

    Proof of rational ≤ term when lower is true, and of term ≤ rational otherwise.

Instances For

    Extract direct bounds and instantiate universally quantified bounds at the supplied atoms.

    For example, ∀ i, x i ∈ Set.Icc lo hi supplies the bounds for each used x i. Unused entries are not enumerated. Concrete guards such as i ∈ Finset.range n are proved before applying a guarded hypothesis; an unresolved index or guard supplies no bounds.

    A rational box and the proof that its ordered real coordinates belong to it.

    • box : Q(Box)

      Rational intervals in variable-index order.

    • reals : Q(List ℝ)

      Real terms in the same order as the box coordinates.

    • membership : Q(unknown_1.ContainsReal fun (i : ℕ) => unknown_2[i]?.getD 0)

      Proof that each real term belongs to its corresponding interval.

    Instances For

      Select the strongest endpoints and prove membership, without an intermediate coordinate.

      Instances For