Hacker News: ynnkhttps://news.ycombinator.com/user?id=ynnkHacker News RSShttps://hnrss.org/hnrss v2.1.1Tue, 30 Jun 2026 21:27:26 +0000<![CDATA[Show HN: Lorentzian CSS Duality in Causal Diamond Quantum Error-Correcting Codes]]>Article URL: https://github.com/ynnk-research/A-Lorentzian-CSS-Duality-in-Causal-Diamond-Quantum-Error-Correcting-Codes

Comments URL: https://news.ycombinator.com/item?id=48718435

Points: 4

# Comments: 0

]]>Mon, 29 Jun 2026 12:33:14 +0000https://github.com/ynnk-research/A-Lorentzian-CSS-Duality-in-Causal-Diamond-Quantum-Error-Correcting-Codesynnkhttps://news.ycombinator.com/item?id=48718435https://news.ycombinator.com/item?id=48718435<![CDATA[New comment by ynnk in "Show HN: NeuroFlow 55.8x video inference speedup for Vision Transformers PyTorch"]]>Yes, but my work on it was minimal so far.

]]>Wed, 27 May 2026 10:48:56 +0000https://news.ycombinator.com/item?id=48292253ynnkhttps://news.ycombinator.com/item?id=48292253https://news.ycombinator.com/item?id=48292253<![CDATA[Show HN: NeuroFlow 55.8x video inference speedup for Vision Transformers PyTorch]]>Article URL: https://github.com/ynnk-research/-NeuroFlow

Comments URL: https://news.ycombinator.com/item?id=48281606

Points: 8

# Comments: 2

]]>Tue, 26 May 2026 16:04:23 +0000https://github.com/ynnk-research/-NeuroFlowynnkhttps://news.ycombinator.com/item?id=48281606https://news.ycombinator.com/item?id=48281606<![CDATA[New comment by ynnk in "Why continuous gradient descent gets stuck on 3-SAT"]]>The core finding is Theorem 5.3: continuous solvers don't fail because the gradient vanishes—they fail because the gradient is an outward normal to the hypercube boundary, and the Θ(N) backbone scaling makes the obstruction diverge in the thermodynamic limit.

The x-space gradient is 20× larger at stuck points but points away from SAT solutions. This resolves the paradox: the Jacobian of the coordinate change x=cos(φ) suppresses the angular gradient near the poles, and the decoded assignment lies on the boundary of the feasible hypercube where the Cartesian gradient is an outward normal—zero tangential escape.

I'm an independent researcher and the work was AI-accelerated, so I built a standalone verification script that mirrors every claim in the paper. You don't have to trust the math; you can run the physics: python verification_phaserelax.py (~15 min on a T4 GPU, ~60 min CPU).

]]>Sat, 11 Apr 2026 16:43:28 +0000https://news.ycombinator.com/item?id=47731996ynnkhttps://news.ycombinator.com/item?id=47731996https://news.ycombinator.com/item?id=47731996<![CDATA[Why continuous gradient descent gets stuck on 3-SAT]]>Article URL: https://github.com/ynnk-research/Landscape-Geometry-and-Algebraic-Obstructions-in-Phase-Space-Gradient-Descent-for-Random-3-SAT

Comments URL: https://news.ycombinator.com/item?id=47731995

Points: 1

# Comments: 1

]]>Sat, 11 Apr 2026 16:43:28 +0000https://github.com/ynnk-research/Landscape-Geometry-and-Algebraic-Obstructions-in-Phase-Space-Gradient-Descent-for-Random-3-SATynnkhttps://news.ycombinator.com/item?id=47731995https://news.ycombinator.com/item?id=47731995