Theorem 2 — Discriminative Kalman Filter asymptotics
ProvedDKFAsymptotics.theorem2Let the state dimension be positive and finite. Let be real and let be symmetric positive-definite matrices with . Set and .
Let be probability densities and let be Gaussian probability densities with positive-definite covariances, for each positive integer . Assume the normalized density is well defined for every . Let be an arbitrary probability measure and . Assume
All densities and distances use Lebesgue measure. Weak convergence means convergence against every bounded continuous real test function. The approximate update is the specific expression ; its validity is not assumed at every index.
Then .
Then .
Then, for all sufficiently large , the approximate update has a measurable weight and a finite, strictly positive normalizer, and is a Gaussian probability density with positive-definite covariance.
Then . The conclusion concerns the same normalized update as C3, including its eventual validity established in that milestone.
Then . Equivalently, the total variation distance between the exact and DKF filtering densities tends to zero.
Intended proof following Appendix B
The proof targets remain open. The intended argument follows the appendix in this order.
- C1 and C2: bound differences of bounded-continuous test-function integrals by the function's supremum norm times the input error. Apply A1–A2 and A3–A4 respectively.
- Lemma 1: convolve the Gaussian transition with the previous Gaussian, combine the full quadratic form and complete the square. For the asymptotic part, use convergence of Gaussian means and covariances, the positive-definite noise covariance, and the appendix's bounded eigenvalue-perturbation argument. Prove that the updated covariance tends to zero and its mean tends to .
- C3 and C4: apply Lemma 1 using C1 and C2. Gaussian identities and density updates must be transported across almost-everywhere equality of representatives.
- C5: retain the five-term decomposition in equation (B.2). A5 and C4 control the four outer reweighting and normalization terms. Product-measure weak convergence gives the common positive limit for the two middle normalizers. The bound , together with A2 and A4, controls the middle term.
One intermediate identity requires care without strengthening the assumptions. The difference in (B.3) can be singular. Use the equivalent identity and , which give the same limits without assuming an inverse of . At finite indices, combine the full precision quadratic form before identifying a Gaussian: the quotient by the stationary prior need not itself be a Gaussian density.
import Definitions.Def_DKFAsymptotics_Assumptions noncomputable section open MeasureTheory Filter Matrix open scoped Topology ENNReal
namespace DKFAsymptotics
theorem theorem2 {d : ℕ} [NeZero d] (M : Model d) (D : Problem M)
(h : Assumptions M D) :
Weakly D.s' D.P ∧
Weakly D.u' (Measure.dirac D.b) ∧
(∀ᶠ n in atTop, UpdateDefined M (D.s' n) (D.u' n) ∧
IsGaussianPDF (D.p' M n)) ∧
Weakly (D.p' M) (Measure.dirac D.b) ∧
Tendsto (fun n => l1Distance (D.p M n) (D.p' M n)) atTop (𝓝 0) := by sorry
end DKFAsymptoticsRead-back
What the Lean code literally says, in plain math · GPT-6 (Codex independent sub-agent)
Read-back of DKFAsymptotics.theorem2
For every natural number , let have its usual finite-product topology and product Lebesgue measure , and let be real matrices such that and are symmetric positive definite and ; here symmetric positive definite means and for every nonzero . The matrix is an arbitrary real matrix subject to that displayed identity. For and a symmetric positive definite real matrix , set
For any function , write for the measure obtained from by weighting it with , where this nonnegative value is regarded as an extended nonnegative real, and define
Here is the extended nonnegative Lebesgue integral, so it is defined even when is not integrable. Call a probability density precisely when it is nonnegative -almost everywhere, is integrable with respect to , and has real integral ; call it a Gaussian probability density precisely when it is a probability density and there exist and a symmetric positive definite real matrix such that for -almost every . These are almost-everywhere conditions, with no pointwise nonnegativity or pointwise equality required on null sets. Define, for arbitrary real functions on ,
The function sends a finite element of to the corresponding real number and sends to . Real division by zero has value zero, and every real integral in this statement is the total Bochner integral, which has value zero if its integrand is not integrable for the indicated measure. In particular, the formula for defines a real function even when its normalization is zero or infinite; in either of those cases it is the zero function. Say that the update for is defined precisely when agrees -almost everywhere with a strongly measurable function and . Let be four sequences of functions , indexed by the positive integers, let , and let be any probability measure on . Assume that, for every positive integer , both and are probability densities, both and are Gaussian probability densities in the exact senses just specified, and the update for is defined. Put and . Assume also all five of the following conditions, where always means along the positive integers and each limit involving is a limit in the usual topology of the extended nonnegative reals :
Then all five of the following conclusions hold simultaneously: (1) for every bounded continuous , ; (2) for every bounded continuous , ; (3) there is a positive integer such that, for every positive integer , the function agrees -almost everywhere with a strongly measurable function, satisfies , and is nonnegative -almost everywhere, integrable with respect to , has integral , and agrees -almost everywhere with for some and some symmetric positive definite real matrix , with these parameters allowed to depend on ; (4) for every bounded continuous , ; and (5) in . Thus the point-mass limits here mean convergence of integrals against every bounded continuous real test function to its value at . The dimension zero and sequence index zero are excluded. The measure need not be given by a density. The defined-update and Gaussian-density assertions in conclusion (3) are required only from some index onward; the formulas and convergence assertions still use the whole positively indexed sequence, including its finitely many earlier terms.
Confirmed by the mission captain (proposal self-audit).