OAI/Probability/InvariantIsing/Magnetic/MagneticFieldHeightHessian.lean

GitHub source file

3 declarations, 0 contain sorry, 0 declared as axioms

The actual constrained finite-height Hessian. Differentiating the physical optimizing bias gives the scalar Schur-complement correction; no twice-differentiable optimizer is assumed.

  1. lemma OAI.InvariantIsing.hasDerivAt_constrainedHeightFieldValue_line_at line 15
  2. lemma OAI.InvariantIsing.hasDerivAt_magneticHeight_gradient line 41
  3. lemma OAI.InvariantIsing.hasDerivAt_deriv_constrainedHeightFieldValue_line line 96

    Source documentation

    The second height derivative is the Hessian at the physical bias minus the rank-one correction from keeping the mean spin fixed.

    lemma hasDerivAt_deriv_constrainedHeightFieldValue_line (h : FieldStep)
        {I : Set (Fin (h.depth + 1) → ℝ)} (F : FieldFiniteFamily (h.depth + 1) I)
        (hI : IsOpen I) (hU : F.U = fieldFiniteValue (fieldHeightFiniteList h))
        {s : ℝ} (hs : |s| < 1) (r d : Fin (h.depth + 1) → ℝ) (hr : r ∈ I)
        (hrs : r ∈ fieldStrictHeightCone h.depth) :
        HasDerivAt (deriv (fun t : ℝ => constrainedHeightFieldValue h s (r + t • d)))
          ((∑ i, ∑ j, F.PP i j (r, magneticHeightBias h s r) * d i * d j) -
            (∑ i, F.PX i (r, magneticHeightBias h s r) * d i) ^ 2 /
              F.XX (r, magneticHeightBias h s r)) 0 := by
    …

    Mentions: OAI.InvariantIsing.FieldStep, OAI.InvariantIsing.FieldFiniteFamily, OAI.InvariantIsing.fieldFiniteValue, OAI.InvariantIsing.fieldHeightFiniteList, OAI.InvariantIsing.fieldStrictHeightCone, OAI.InvariantIsing.constrainedHeightFieldValue, OAI.InvariantIsing.magneticHeightBias