SNT CoreX Dynamics

— Research · Develop · Evolve —

Est. 2025 · Antalya, Turkey

SNT CoreX Dynamics is an advanced research organisation building Spectral Nod Theory — a unified mathematical framework spanning Lie algebra, neural dynamics, and large language model inference optimisation. We operate independently, publish rigorously, and develop from first principles.

4+ Active Papers

G₇ Core Algebra

φ★ CoreX Spark

2025 Founded

Our Origin

SNT CoreX Dynamics was founded in 2025 in Antalya, Turkey, with a singular focus: to develop and formalise Spectral Nod Theory (SNT) — a research program that bridges pure mathematics with applied machine learning and biological neural dynamics. The organisation operates with full research independence, selecting problems based on mathematical necessity and scientific relevance rather than institutional mandate.

What We Build

Our work is organised around the SNT framework — seven canonical operators acting on structured state spaces, anchored by the seven-operator Lie algebra G₇ and the CoreX Spark constant φ★ = (1+√5)/2. From this algebraic foundation, we derive results in operator theory, connectome dynamics, and KV-cache optimisation for large language models. Every result is computationally verified and reproducible.

How We Operate

We publish in peer-reviewed venues across mathematics, computer science, and interdisciplinary science. All manuscripts include complete proofs, SymPy-verified computations, and open experimental pipelines. Our papers are interconnected — sharing operator definitions, algebraic constants, and empirical methodology — forming a coherent body of work rather than isolated contributions.

Our Commitment

SNT CoreX Dynamics maintains a strict standard: no results are presented without formal derivation or experimental validation. We do not speculate — we derive, simulate, verify, and publish. The CoreX Spark constant φ★, the G₇ algebra structure, and our LLM benchmarks are all fully documented and independently replicable.

[∂]

Research

Continuously exploring new theoretical horizons with rigorous mathematical foundations.

[</>]

Develop

Transforming theoretical results into verified, reproducible computational systems.

[⟳]

Evolve

Iterating across domains — mathematics, biology, AI — within a unified framework.

[◈]

Autonomy

Independent systems and autonomous research decision-making, unconstrained by institutional agenda.

[≡]

Rigour

Scientific honesty, transparency, and ethical standards in every result we publish.

[→]

Impact

Contributing meaningfully to human knowledge through peer-reviewed, reproducible science.

SNT-CORE · Lie Algebra · Pure Mathematics ◉ Late-Stage Preparation

Structural Obstruction in a Seven-Operator Lie Algebra: Derivation Rigidity and the CoreX Spark Spectral Constant

This paper establishes G₄(φ) ⊕ su(2) as the unique admissible closed Lie algebra structure over a seven-operator system. Three core theorems are formally proven: (T1) Jacobi identities close uniquely — the key commutator [F,N] = −i·tan(φ)·Cyc is derived rather than postulated; (T2) the spectral obstruction tensor is non-vanishing, rigorously ruling out semidirect product decomposition; (T3) the outer derivation algebra is one-dimensional and uniquely realised at the CoreX Spark constant φ★ = (1+√5)/2, identified as the Perron-Frobenius spectral radius of the structure matrix. All results are verified via SymPy with reproducible code in the appendix. Zero self-citations; bibliography contains only standard mathematical references.

Target Journal AIMS Mathematics

Domain Lie Algebra · Operator Theory

Key Constant φ★ = (1+√5)/2

ShrikeSNT · KV-Cache · Machine Learning ◉ Active — Pipeline Complete

Reversible KV Cache Compression via Memory Inertia: A Stability-Oriented Approach to Transformer Inference

ShrikeSNT introduces a stability-first KV-cache compression architecture for autoregressive transformer inference. The method applies SNT's Memory Inertia operator with a frozen-spark threshold anchored at the CoreX constant φ, routing key-value pairs through a Termop → Subspace → Loss pipeline. Evaluated on LongBench-v2 across Qwen2-7B and Qwen2.5-14B, SNT achieves 24.3–27% task accuracy at 2.33× faster decoding compared to StreamingLLM, maintaining 7/7 ablation consistency across all experimental seeds. Memory mass metric: 0.443 ± 0.013, beating StreamingLLM's 0.457. The architecture operates within a sub-10GB GPU footprint in 4-bit quantisation.

Target Journal SN Computer Science

Models Tested Qwen2-7B · Qwen2.5-14B

Key Result 2.33× speedup vs StreamingLLM

BladeRunnerSNT · KV-Cache · Machine Learning ◉ Active — Multi-Model Benchmarks

Drift-Aware Adaptive KV Cache Pruning for Efficient Transformer Inference

BladeRunnerSNT is the performance-oriented companion to ShrikeSNT, introducing drift-aware pruning at 35% full-KV retention. The system targets adaptive keep ratios via token-level entropy signals, query-aware Value boosting through the SNT Nexter operator, and multi-layer attention fusion. Evaluated across Qwen2-7B, Mistral-7B, and Qwen2.5-14B in BF16, experiments reveal Mistral-7B achieving +10 percentage points over full-KV baseline on targeted subtasks. Ongoing work addresses NIAH robustness under aggressive compression and strengthening drift signal reliability via real attention importance scores.

Target Journal SN Computer Science

Models Tested Qwen2-7B · Mistral-7B · Qwen2.5-14B

Key Result +10pp over full-KV (Mistral-7B)

SNT-LIFE · Connectomics · Neural Dynamics ◉ Under Review / Finalising

Seven-Operator Learning and Memory Framework Validated on C. elegans Connectome Data

The SNT-LIFE series applies the seven-operator formalism to biological neural circuit dynamics. Validation against Cook 2019 and Kato 2015 C. elegans connectome datasets achieves trajectory correlation r = 0.986 at 86× parameter reduction relative to Wilson-Cowan models, with Lie algebraic closure verified at 4.82×10⁻¹⁵. Extended work covers predator-prey survival dynamics with a formally defined Survival Viability Condition (SVC), and N=100 population-level coevolutionary simulations producing r = −0.983 coevolutionary correlation, Cohen's d = 4.06 versus random baseline, with 5/5 seed consistency. Frobenius closure proof included in appendix with complete Python code.

Target Journals Neuroscience of Consciousness · Artificial Life (MIT)

Key Result r = 0.986 · 86× param reduction

Data Cook 2019 · Kato 2015

CoreX Spark Foundation

Lie Algebra · Pure Mathematics

The algebraic foundation of the entire SNT program. Establishes the seven-operator Lie algebra G₇ = G₄(φ) ⊕ su(2) and derives the CoreX Spark constant φ★ = (1+√5)/2 as its unique spectral invariant — the Perron-Frobenius radius of the structure matrix. Derivation rigidity proven: the outer derivation space is exactly one-dimensional, with no semidirect product decomposition possible. All downstream projects inherit this constant and the seven-operator structure.

ShrikeSNT Active

Machine Learning · LLM Inference

Stability-oriented KV-cache compression for autoregressive transformer inference. Implements the Memory Inertia operator with a frozen-spark threshold anchored at φ★. The Termop → Subspace → Loss pipeline maintains memory mass stability across long sequences. LongBench-v2 validated on Qwen2-7B and Qwen2.5-14B; 2.33× decoding speedup over StreamingLLM with higher task accuracy. Mass metric 0.443 ± 0.013, 7/7 ablation consistency.

BladeRunnerSNT Active

Machine Learning · LLM Inference

Performance-oriented companion to ShrikeSNT. Drift-aware adaptive KV pruning at 35% full-KV retention with entropy-driven keep ratios and query-aware Value boosting via the Nexter operator. Multi-model evaluation across Qwen2-7B, Mistral-7B, and Qwen2.5-14B in BF16. Mistral-7B achieves +10pp over full-KV baseline on targeted EXP-2 tasks. Multi-layer attention fusion and NIAH robustness under active development.

SNT-LIFE Active

Connectomics · Neural Dynamics · Ecology

Application of the seven-operator formalism to biological systems. Three interconnected papers: C. elegans connectome validation (r = 0.986, 86× parameter reduction); predator-prey survival dynamics with formal SVC via viability theory; and N=100 coevolutionary population simulations (r = −0.983, d = 4.06). Targets Neuroscience of Consciousness and Artificial Life (MIT Press).

TarrasqueSNT In Reserve

Machine Learning · Next-Generation KV

Next-generation hybrid KV scoring architecture combining SNT operator signals with segment-level retention guarantees and a global KV memory bank. Designed to unify the stability model of ShrikeSNT with the adaptive performance of BladeRunnerSNT into a single coherent inference system. Currently held in reserve pending publication of Shrike and BladeRunner.

SNT Gravity Earlier Work

Theoretical Physics · Quantum Systems

Spacetime emergence from Planck-scale nod networks with four operators acting on joint qubit-nod Hilbert spaces. Decoherence derivations via replica trick and Born-Markov approximation; NV-center T₂ scaling predictions formulated as relative scaling laws. An earlier thread in the SNT program, preceding the Lie algebra and LLM papers, that established the operator formalism foundations.

Background

Durhan Yazır is a researcher working across pure mathematics, computational neuroscience, and machine learning engineering. He founded SNT CoreX Dynamics in 2025 to develop and publish Spectral Nod Theory — a research program he has been building across multiple scientific domains simultaneously.

His approach is characterised by mathematical precision and domain independence: starting from algebraic first principles, he derives results that apply across apparently disconnected fields — from the structure theory of Lie algebras to the inference efficiency of billion-parameter language models.

Research Philosophy

SNT CoreX Dynamics operates on a clear principle: every claim must be formally derived or experimentally verified before publication. This means complete proofs in every mathematics paper, SymPy-verified computations in appendices, and full experimental pipelines with reproducible benchmarks in every machine learning manuscript.

The SNT framework is deliberately cross-domain. The same seven-operator algebra that governs C. elegans neural trajectories also defines the compression architecture of ShrikeSNT's KV-cache system. This is not coincidence — it is the programme. Spectral Nod Theory exists precisely to find the structural invariants that persist across scales and systems.

Research Domains

Lie Algebra Operator Theory Neural Dynamics Connectomics LLM Inference KV-Cache Optimisation Coevolutionary Systems Theoretical Physics Spectral Theory

Direct Contact

All research correspondence is handled directly by the founder. Response time is typically within 48 hours for academic and research-related inquiries.

✉ durhan@sntdynamics.com

Organisation SNT CoreX Dynamics

Location Antalya, Turkey

Founded 2025

Focus Spectral Nod Theory — Mathematics, Neural Dynamics, LLM Systems

Inquiries Research collaboration · Peer review · Academic correspondence

SNT CoreX

— Dynamics —

We develop mathematics that connects. From the rigidity of a seven-operator Lie algebra to the compression of trillion-token transformer caches — Spectral Nod Theory is the bridge.

Research · Develop · Evolve

SNT CoreX Dynamics

Who We Are

Our Origin

What We Build

How We Operate

Our Commitment

Research

Develop

Evolve

Autonomy

Rigour

Impact

Research

Structural Obstruction in a Seven-Operator Lie Algebra: Derivation Rigidity and the CoreX Spark Spectral Constant

Reversible KV Cache Compression via Memory Inertia: A Stability-Oriented Approach to Transformer Inference

Drift-Aware Adaptive KV Cache Pruning for Efficient Transformer Inference

Seven-Operator Learning and Memory Framework Validated on C. elegans Connectome Data

Projects

About

Background

Research Philosophy

Research Domains

Contact

Direct Contact