Theorem 1.4 — clock, exact leading constant, optimal ratio and counted memory
OpenPrivateContinualCounting.Kraft.Supplement.clock_sufficesFor every fixed and , the concrete Section 5 square-root hypertree is one infinite causal Gaussian factorization counter with sensitivity exactly one and every finite transcript -GDP. There is a constant such that
For every Gaussian factorization counter on any countable inner index set and every ,
and for the hypertree along all integer times. The norm permits every finite inner dimension and arbitrary factors; it is not restricted to Toeplitz factorizations. The running maximum is retained because pointwise variance can be small at super-epoch starts.
For each fixed , there also exist an accurate buffered linear Toeplitz inner family, its A/B/C Gaussian counter and a finite register-program implementation. After strategy normalization each horizon- factor has decoder squared row norms at most . The counter has sensitivity one and is -GDP, obeys the same all-time variance bound with a constant depending also on , and has the same running-maximum ratio limit. Its program has measurable transcripts with exactly this counter's joint law. It terminates on every arrival, and for one chosen before time, every intermediate runtime state at positive time satisfies
The count includes live coefficients, matrices, inputs, buffers, retained sampled randomness, real and integer registers, program literals and clock/control state. Fresh normals are sampled and discarded or put into counted registers. No arbitrary history encoder or premise supplying a BLT implementation is allowed. Existence of the inner family and program is a theorem obligation. Appendix E.2 supplies the fixed-parameter uniform remainder constant, and E.3 supplies the tolerance-dependent state constant; neither depends on or the stream. No finite-precision or time-complexity claim is asserted.
Differences from the source:
- encoding: positive source time and stream indices → natural times with source input i+1 at coordinate i and zero-time noise zero; reason: the published base conventions are reused and every positive time and interval is retained
- encoding: O(log squared t) real-arithmetic words with fixed tolerance → a finite register program with field arithmetic, exact square root and logarithm for deterministic initialization, all-step joint-law correctness, and a uniform bound on every intermediate register state; reason: an explicit exact-real space model counts stored coefficients and control as well as data; the program has no arbitrary functional or history primitive
- encoding: Appendix E.3 inner factor approximation and the same envelope → existential buffered coefficient families with exact normalized products and error at most S(n)+eta, and a counted implementation of their stitched releases; reason: the external BLT existence result remains a proof obligation and is not made a hypothesis
- generalization: the ratio is explicitly stated for the square-root hypertree; the BLT variant is stated to have the same bound → the running-maximum ratio is also concluded for the BLT implementation; reason: its same all-time upper bound and the same universal gamma-two lower bound imply the ratio as in Appendix E.2
import Definitions.Def_PrivateContinualCounting_Kraft_Supplement_StreamingMemory open MeasureTheory ProbabilityTheory Filter set_option autoImplicit false
namespace PrivateContinualCounting.Kraft.Supplement
/-- Theorem 1.4 in full: the concrete hypertree, countable-dimensional universal
lower bound, running-maximum limit, and an existential BLT family and counted implementation.
There is no premise supplying the BLT theorem or an algorithm with the desired guarantees. -/
theorem clock_suffices (a ξ : ℝ) (ha : 0 < a) (hξ0 : 0 < ξ) (hξ1 : ξ < 1) :
(∃ M : GaussianCounter TreeIndex, IsTree M squareRootFamily a ξ ∧
M.factors.deltaSq = 1 ∧ TranscriptGDP (treeLaw squareRootFamily a ξ) 1 ∧
(∃ C₀ : ℝ, 0 < C₀ ∧ ClockEnvelope M.variance C₀) ∧ OptimalRatio M.variance) ∧
(∀ (I : Type) [Countable I] (M : GaussianCounter I) (t : ℕ), 1 ≤ t →
(matrixGamma2 (prefixMatrix t))^2 ≤ runningMax M.variance t) ∧
(∀ η : ℝ, 0 < η → η < 1 →
∃ F : InnerFamily, BufferedFamily F ∧ InnerAccuracy F η ∧
∃ M : GaussianCounter TreeIndex, IsTree M F a ξ ∧ M.factors.deltaSq = 1 ∧
TranscriptGDP (treeLaw F a ξ) 1 ∧
(∃ C₀ : ℝ, 0 < C₀ ∧ ClockEnvelope M.variance C₀) ∧ OptimalRatio M.variance ∧
∃ A : StreamingRAM, ∃ Cmem : ℝ, 0 < Cmem ∧ Implements A F a ξ Cmem ∧
TranscriptGDP A.law 1) := by sorry
end PrivateContinualCounting.Kraft.Supplement