black-hole-sim-randomized

# 🕳️ Black Hole Sim — Randomized Dataset (ProCreations)

**ProCreations/black-hole-sim-randomized** is a high-fidelity, randomized simulation dataset of relativistic physics near Kerr and Schwarzschild black holes. Designed to train and evaluate AI systems on **general relativity**, **orbital mechanics**, and **spacetime geometry**—without any visual dependencies.

---

## 📦 Dataset Overview

- **Samples**: 400,000+
- **Format**: JSON Lines (`.jsonl`)
- **Size**: ~583 MB
- **Compression**: Optionally `.jsonl.gz`
- **Language/Structure**: Pure structured JSON (1 per line)
- **Generated on**: Apple Silicon (Metal-accelerated)
- **Use cases**: Pretraining, fine-tuning, RAG, reasoning, QA, simulation

---

## ✨ Features per Sample

Each sample describes a **single randomly sampled scenario**, containing:

### 🕳️ Black Hole

- `mass_solar` – In solar masses (randomized log-uniform from 3 to 1e9)
- `spin` – Kerr spin parameter `a` (0.0 to 0.998)
- `type` – `"Schwarzschild"` or `"Kerr"`

### 👁️ Observer

- 3D position (`r_km`, `theta_rad`, `phi_rad`)
- Orbital velocity vector in spherical coordinates
- Orbital angular velocity `omega`

### 📐 Metrics (GR & physics)

- `time_dilation`, `redshift`, `gamma`
- `orbital_period_s`, `frame_dragging`, `extremeness_score`
- `metric_tensor` – Full 4x4 Kerr metric in Boyer–Lindquist coordinates
- `a_spin_m`, `event_horizon_m`, `delta`, `rho_squared`
- Velocity vector + coordinate positions
- `horizon_proximity_ratio` (how close the observer is to Rs)

### 🧪 Raw Parameters

- Ground truth values like `mass_kg`, `radius_m`, `rs_m`

---

## 📁 Example

```json
{
"step": 2,
"black_hole": {
"mass_solar": 219.051486,
"spin": 0.504345,
"type": "Kerr"
},
"observer": {
"r_km": 1600.060345,
"theta_rad": 0.624667,
"phi_rad": 4.083344,
"v_phi_rad_s": 128884.609375
},
"metrics": {
"time_dilation": 0.771804,
"redshift": 0.295665,
"frame_dragging": true,
"metric_tensor": [[...]],
"horizon_proximity_ratio": 2.654,
"gr_metric_type": "Kerr"
},
"notes": "Spin=0.504, Time dilation=0.772, Redshift=0.296"
}

# 🕳️ 黑洞模拟——随机化数据集（ProCreations）

**ProCreations/black-hole-sim-randomized** 是一套高保真的克尔（Kerr）与史瓦西（Schwarzschild）黑洞附近相对论物理随机模拟数据集，旨在无需依赖视觉模态的前提下，训练与评估人工智能系统在**广义相对论（General Relativity）**、**轨道力学（Orbital Mechanics）**与**时空几何（Spacetime Geometry）**领域的相关能力。

---

## 📦 数据集概览

- **样本量**：40万条以上
- **数据格式**：JSON Lines（`.jsonl`）
- **数据体量**：约583 MB
- **压缩方式**：支持可选的`.jsonl.gz`压缩格式
- **语言与结构**：纯结构化JSON格式，每行一条数据
- **生成平台**：搭载Metal加速的苹果硅芯片（Apple Silicon）
- **应用场景**：预训练、微调、检索增强生成（Retrieval-Augmented Generation，RAG）、推理、问答、模拟任务

---

## ✨ 单样本特征

每个样本对应一个随机采样的物理场景，包含以下内容：

### 🕳️ 黑洞参数
- `mass_solar`：以太阳质量为单位的黑洞质量，采用对数均匀分布随机采样，取值范围为3至10^9
- `spin`：克尔自旋参数`a`，取值范围为0.0至0.998
- `type`：黑洞类型，可选值为`"Schwarzschild"`或`"Kerr"`

### 👁️ 观测者参数
- 三维位置：采用球坐标表示，包含`r_km`（径向距离，单位：千米）、`theta_rad`（极角，单位：弧度）与`phi_rad`（方位角，单位：弧度）
- 球坐标下的轨道速度矢量
- 轨道角速度`omega`

### 📐 广义相对论与物理度量
- `time_dilation`（时间膨胀）、`redshift`（红移）与`gamma`
- `orbital_period_s`（轨道周期，单位：秒）、`frame_dragging`（参考系拖拽）与`extremeness_score`（极值评分）
- `metric_tensor`：采用Boyer-Lindquist坐标表示的完整4×4克尔度规张量
- `a_spin_m`、`event_horizon_m`（事件视界半径，单位：米）、`delta`与`rho_squared`
- 速度矢量与坐标位置
- `horizon_proximity_ratio`：观测者与事件视界的相对距离比例

### 🧪 原始真值参数
包含质量（`mass_kg`，单位：千克）、半径（`radius_m`，单位：米）与史瓦西半径（`rs_m`，单位：米）等真实物理真值

---

## 📁 示例
json
{
"step": 2,
"black_hole": {
"mass_solar": 219.051486,
"spin": 0.504345,
"type": "Kerr"
},
"observer": {
"r_km": 1600.060345,
"theta_rad": 0.624667,
"phi_rad": 4.083344,
"v_phi_rad_s": 128884.609375
},
"metrics": {
"time_dilation": 0.771804,
"redshift": 0.295665,
"frame_dragging": true,
"metric_tensor": [[...]],
"horizon_proximity_ratio": 2.654,
"gr_metric_type": "Kerr"
},
"notes": "Spin=0.504, Time dilation=0.772, Redshift=0.296"
}