| name | fibonacci-quasicryth-compression |
| title | Aperiodic Structures Never Collapse: Fibonacci Hierarchies in Lossless Compression |
| version | 0.0.3 |
| engine | skillxiv-v0.0.3-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2603.14999 |
| keywords | ["Lossless Compression","Quasicrystals","Hierarchical Coding","Mathematical Foundations","Algebraic Number Theory"] |
| description | Introduce Quasicryth, a text compressor using Fibonacci quasicrystal tilings for phrase-level compression. Prove that aperiodic structures never structurally collapse at depth, enabling compression at arbitrary hierarchy levels. Bridge quasicrystal mathematics with practical compression, achieving 22.59% ratio on enwik9 with unbounded scaling advantages. |
Aperiodic Structures in Lossless Compression
Problem Formulation: Hierarchy Collapse
Traditional periodic tilings used in hierarchical compression have a fundamental limitation: they collapse structurally within O(log p) levels, where p is the period. Beyond that depth, one tile type vanishes, making further compression impossible.
Example: Binary hierarchies collapse at depth log₂(W) (word vocabulary size). The codebook loses structural variety, forcing fallback to escapes.
New Paradigm: Aperiodic Fibonacci Tilings
Fibonacci quasicrystals offer a mathematically-proven alternative: structures that never collapse at any depth, enabling compression at arbitrarily deep hierarchy levels.
Founding Mathematical Results
Main Theorem (Aperiodic Hierarchy Advantage):
The Fibonacci tiling uniquely satisfies five properties simultaneously:
-
Non-Collapse at All Depths: Both tile types persist indefinitely
- Unlike periodic structures that collapse within O(log p) levels
- Guaranteed by golden-ratio eigenvalue φ (Pisot-Vijayaraghavan number)
-
Scale-Invariant Coverage: Potential word coverage remains constant across hierarchy levels
- ~0.724W words available at each level regardless of depth
- Proven via Golden Compensation Theorem
-
Maximum Codebook Efficiency: Exactly Fm+1 distinct patterns at level m
- Minimal for aperiodic sequences
- Maximizes phrase reuse probability
-
Bounded Overhead: Flag entropy capped at 1/φ ≈ 0.618 bits/word
- Constant regardless of hierarchy depth
- Periodic structures have unbounded overhead growth
-
Strict Entropy Advantage: Lower per-word coding entropy than periodic alternatives for long-range sources
- Exploits statistical correlations across deeper hierarchy levels
Mathematical Framework:
Substitution Matrix Analysis:
σ: L → LS, S → L (Fibonacci substitution)
Eigenvalue = φ (golden ratio)
Sturmian Sequences:
Minimal factor complexity p(n) = n+1
Maximum phrase reuse potential
Weyl Equidistribution:
Irrational tiling slopes guarantee uniform density
at all scales
Practical Implementation: Quasicryth v5.6
Hierarchy Configuration:
- 10-level hierarchy with phrase lengths: {2, 3, 5, 8, 13, 21, 34, 55, 89, 144} words
- 36 distinct tilings:
- 12 golden-ratio optimized
- 6 non-golden variants
- 18 optimized via greedy search
Encoding Strategy:
- Multi-level adaptive arithmetic coding with order-2 context models
- Separate LZMA-compressed escape stream for out-of-vocabulary words
- Lazy evaluation: activate deep levels only when shallow codebooks saturate
Performance Results
Compression on enwik9 (1 GB English Wikipedia):
- Achieved Ratio: 22.59%
- Deep Level Activation: Levels 9-10 (89-gram, 144-gram phrases) uniquely activate here
- 89-gram phrases: 5,369 hits
- 144-gram phrases: 2,026 hits
- Advantage Over Periodic Baseline: 11+ million bytes saved vs Period-5 periodic tiling
Scaling Property:
As dataset size increases:
- Periodic tilings: compression ratio plateaus (hierarchy collapses)
- Fibonacci tilings: continuous improvement (new patterns activate at deeper levels)
Why This Matters
Bridge Between Mathematics and Practice:
-
Theoretical Grounding: Proves that structural properties from quasicrystal geometry (physics/materials science) solve fundamental CS problems
-
Unbounded Improvement: Unlike fixed-depth hierarchies, Fibonacci structures keep improving with larger datasets—architectural advantage grows
-
Generalization: Same principle applies to:
- Hierarchical tokenization in LLMs (variable-length codes)
- Semantic hierarchies in knowledge graphs
- Phylogenetic tree compression in biology
Opened Research Directions
-
Adaptation to Other Domains: Can Fibonacci tilings improve hierarchical clustering, multi-level hashing, or tree-based indexing?
-
Hybrid Approaches: Combine Fibonacci hierarchies with modern neural compression or learned arithmetic coding
-
Scaling Experiments: Test on larger corpora (10+ GB) to validate unbounded scaling advantage
-
Physical Quasicrystals Inspiration: Other aperiodic structures (Penrose tilings, Thue-Morse sequences) may have complementary properties
Vocabulary & Founding Concepts
- Quasicrystal: Aperiodic ordered structure without translational symmetry (discovered in materials; novel application here)
- Pisot-Vijayaraghavan Number: Algebraic integer with all conjugates <1; ensures stability (φ is PV number)
- Sturmian Sequence: Infinite binary sequence with minimal complexity; encodes Fibonacci tiling structure
- Golden Compensation Theorem (Novel): Exponential growth in phrase length (Fm~φ^m) exactly cancels exponential decrease in position counts (φ^-(m-1)), yielding level-independent reuse capacity