Documentation

LeanPool.ParameterFreeGradient.V7.Proofs.Stage1E03.Machine

Dependency-pure causal machine for the frozen V7 Euclidean trial.

@[instance_reducible]
noncomputable def V7.Stage1E03.e03PropDecidable (q : Prop) :

Classical proposition decisions used locally to evaluate observable guards.

Equations
Instances For
    noncomputable def V7.Stage1E03.CheckHolds {d : ℕ} (p M : ℝ) (check : ObservableGuardCheck d) :

    The selected observable inequality evaluated solely from its recorded observations.

    Equations
    • One or more equations did not get rendered due to their size.
    Instances For
      noncomputable def V7.Stage1E03.upperCheck {d : ℕ} (oy ox : Observation d) :

      An upper-model guard formed from the query and proposed next-point observations.

      Equations
      Instances For

        A Euclidean interpolation guard formed from two observations.

        Equations
        Instances For
          noncomputable def V7.Stage1E03.terminalCheck {d : ℕ} (on ov : Observation d) :

          A terminal-descent guard formed from the final query and gradient-step observations.

          Equations
          Instances For
            noncomputable def V7.Stage1E03.allInterpolationChecks {d : ℕ} (n : ℕ) (obsAt : ℕ → Observation d) :

            The ordered list of interpolation checks for every pair through horizon n.

            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              Pure inspection: the error branch is exactly the prefix through the first failed current-V7 point-carrying guard.

              Equations
              Instances For
                noncomputable def V7.Stage1E03.horizon (eps M D : ℝ) :

                The ceiling of the Euclidean trial's accuracy-dependent iteration horizon.

                Equations
                Instances For
                  theorem V7.Stage1E03.horizon_real_ge (eps M D : ℝ) :
                  2 * √(M * D / eps) ≤ ↑(horizon eps M D)
                  theorem V7.Stage1E03.one_le_horizon {eps M D : ℝ} (hkappa : 1 ≤ M * D / eps) :
                  1 ≤ horizon eps M D
                  theorem V7.Stage1E03.horizon_gradient_budget {eps M D : ℝ} (heps : 0 < eps) (hM : 0 < M) (hD : 0 < D) (hkappa : 1 ≤ M * D / eps) :
                  have n := horizon eps M D; 2 * √2 * M * D / ((↑n + 1) * (↑n + 1)) ≤ eps
                  noncomputable def V7.Stage1E03.nextEstimateState {d : ℕ} (M : ℝ) (x0 : Point d) (k : ℕ) (state : O3.EuclideanEstimateState d) (oy : Observation d) :

                  The next accelerated estimate state after receiving a query observation.

                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For
                    noncomputable def V7.Stage1E03.estimateQuery {d : ℕ} (M : ℝ) (x0 : Point d) (k : ℕ) (state : O3.EuclideanEstimateState d) :

                    The estimate-sequence query formed from the current accelerated point and quadratic minimizer.

                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For
                      noncomputable def V7.Stage1E03.ogmgStep {d : ℕ} (n : ℕ) (M : ℝ) (i : ℕ) (state : O3.OGMGExecutionState d) (oi : Observation d) :

                      One OGM-G state update using the observed gradient and prescribed momentum coefficients.

                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For
                        noncomputable def V7.Stage1E03.terminalProgram {d : ℕ} (eps M : ℝ) (n : ℕ) (obsAt : ℕ → Observation d) (exec : O3.OGMGExecutionState d) (guards : List (ObservableGuardCheck d)) :

                        The final one-query program checking descent and selecting the terminal outcome.

                        Equations
                        • One or more equations did not get rendered due to their size.
                        Instances For
                          noncomputable def V7.Stage1E03.phaseBProgram {d : ℕ} (eps M : ℝ) (n i : ℕ) (exec : O3.OGMGExecutionState d) (obsAt : ℕ → Observation d) (guards : List (ObservableGuardCheck d)) (fuel : ℕ) :
                          Program d (fuel + 1)

                          The finite OGM-G query program followed by interpolation and terminal-descent checks.

                          Equations
                          • One or more equations did not get rendered due to their size.
                          Instances For

                            The query bound for the remaining estimate phase and the following OGM-G phase.

                            Equations
                            Instances For
                              noncomputable def V7.Stage1E03.phaseAProgram {d : ℕ} (eps M : ℝ) (x0 : Point d) (n k : ℕ) (estimate : O3.EuclideanEstimateState d) (last : Option (Observation d)) (guards : List (ObservableGuardCheck d)) (fuel : ℕ) :

                              The estimate-phase query program, which checks each upper model before continuing.

                              Equations
                              • One or more equations did not get rendered due to their size.
                              Instances For
                                noncomputable def V7.Stage1E03.euclideanLocalTrial {d : ℕ} (eps : ℝ) (x0 : Point d) (n : ℕ) :

                                The Euclidean local trial with a fixed planned horizon.

                                Equations
                                • One or more equations did not get rendered due to their size.
                                Instances For