allaun/Research-Stack
1
0
Fork 0
mirror of https://github.com/allaunthefox/Research-Stack.git synced 2026-07-31 03:05:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

docs(infra): append bosonic tensor network scaling knowledge to context stream

Browse source 
Added theoretical entropy power laws, physical target scales, RTX 4070 SUPER hardware limits (VRAM & time complexity), and scaling recommendations to the ContextStream blocks in AGENTS.md and GEMINI.md.

Build: 3314 jobs, 0 errors (lake build Compiler)
This commit is contained in:
allaun 2026-06-18 23:49:31 -05:00
parent 50322c88f9
commit fa7511e8a8
2 changed files with 20 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
AGENTS.md
View file

@ -546,5 +546,15 @@ For sessions that need these skills, reference `~/.claude/skills/<name>/SKILL.md
✅ Looking for code by meaning/concept
✅ Need semantic understanding
### Bosonic Tensor Network Simulator Scaling & Limits:
- **Theoretical Entropy Power Law**: Single-mode marginal entropy is \(H_K(N) \approx \log_2(N) - \frac{0.7213}{K}\) bits. Joint Fock-space capacity is \(H_{\text{joint}}(N, K) = \log_2 \binom{N+K-1}{K} \approx K \log_2(N) - \log_2(K!)\) bits.
- **Physical Target Scale**: The core system (Burgers representation graph in `burgers_chaos_game.py`) has exactly \(N = 22\) modes.
- **RTX 4070 SUPER Hardware Limits**:
- **VRAM Ceiling (\(O(N^2)\))**: Usable limit of 10 GB VRAM is reached at \(N = 50000\) modes (\(\text{VRAM}(N) = N^2 \times 4\) bytes).
- **Time Complexity (\(O(N^3)\))**: The dense graph (edge probability \(p=0.4\)) has a spectral radius scaling as \(\lambda_{\max} \approx 0.4 N\). Thus, RK4 step count scales as \(S \propto N\), making total duration cubic (\(T(N) = c \cdot N^3\) with \(c \approx 1.085 \times 10^{-10}\text{ s/mode}^3\)).
- **Max Scale at 1-Hour Limit**: \(N \approx 32000\) modes (\(T \approx 55\) mins).
- **Comparison to Perceval SLOS**: Perceval compiles the full Fock-space and OOMs at \(N \ge 150\). Our GPU solver reaches \(N = 15000\) (verified) and \(N = 50000\) (limit) representing a \(>330\times\) increase.
- **Scaling Recommendation**: To scale beyond \(N = 50000\) or drop time complexity to \(O(N)\), transition the adjacency matrix to a sparse representation (keeps \(S\) constant and matrix mults \(O(N)\)).
---
<!-- END ContextStream -->

10
GEMINI.md
View file

@ -42,5 +42,15 @@ You are working in the **Research Stack** project. This project is a formally ve
✅ Looking for code by meaning/concept
✅ Need semantic understanding
### Bosonic Tensor Network Simulator Scaling & Limits:
- **Theoretical Entropy Power Law**: Single-mode marginal entropy is \(H_K(N) \approx \log_2(N) - \frac{0.7213}{K}\) bits. Joint Fock-space capacity is \(H_{\text{joint}}(N, K) = \log_2 \binom{N+K-1}{K} \approx K \log_2(N) - \log_2(K!)\) bits.
- **Physical Target Scale**: The core system (Burgers representation graph in `burgers_chaos_game.py`) has exactly \(N = 22\) modes.
- **RTX 4070 SUPER Hardware Limits**:
- **VRAM Ceiling (\(O(N^2)\))**: Usable limit of 10 GB VRAM is reached at \(N = 50000\) modes (\(\text{VRAM}(N) = N^2 \times 4\) bytes).
- **Time Complexity (\(O(N^3)\))**: The dense graph (edge probability \(p=0.4\)) has a spectral radius scaling as \(\lambda_{\max} \approx 0.4 N\). Thus, RK4 step count scales as \(S \propto N\), making total duration cubic (\(T(N) = c \cdot N^3\) with \(c \approx 1.085 \times 10^{-10}\text{ s/mode}^3\)).
- **Max Scale at 1-Hour Limit**: \(N \approx 32000\) modes (\(T \approx 55\) mins).
- **Comparison to Perceval SLOS**: Perceval compiles the full Fock-space and OOMs at \(N \ge 150\). Our GPU solver reaches \(N = 15000\) (verified) and \(N = 50000\) (limit) representing a \(>330\times\) increase.
- **Scaling Recommendation**: To scale beyond \(N = 50000\) or drop time complexity to \(O(N)\), transition the adjacency matrix to a sparse representation (keeps \(S\) constant and matrix mults \(O(N)\)).
---
<!-- END ContextStream -->