ASTR 596: Modeling the Universe

Welcome to ASTR 596: Modeling the Universe! This course takes you on a journey from Python fundamentals to neural networks and JAX, building transparent “glass box” models that reveal the physics within astronomical phenomena.

Our projects follow a deliberate progression from deterministic simulation → stochastic methods → statistical inference → machine learning, mirroring how computational astrophysics evolved as a field. Each project builds on previous code and concepts, reducing cognitive load through scaffolding.

Act I: Deterministic simulation (Python → OOP → N-body dynamics)
Act II: Stochastic methods & statistical inference (Monte Carlo → Bayesian/MCMC)
Act III: Machine learning & autodifferentiation (Gaussian Processes → Neural Networks → JAX)

Course Philosophy: “Glass Box” Modeling

Earned Complexity: We implement fundamental algorithms from first principles before leveraging advanced frameworks. Every line of code serves understanding — no black boxes allowed until you’ve built the glass box yourself.

Core Principles:

“Think first, code second” - An hour of planning saves a week of debugging
“Understanding > Performance” - Slower code you understand beats fast code you don’t
“Physics first, Python later” - Debug with physics intuition, not print statements
“Productive struggle is the point” - Being stuck means you’re learning something worth knowing

Growth Mindset: Cognitive science shows that working at the edge of your ability — where you’re challenged but not overwhelmed — optimizes learning. This is where learning happens. Expect to iterate: professional code is rarely right the first time. When you’re confused or stuck, that’s not a bug in your learning process, it’s a feature.

Project Timeline & Core Concepts¶

Project	Duration	Core Concepts	Key Connection
🐍 Project 1: Stellar Populations	1.5 weeks	Object-oriented design, vectorization, simple plotting	Builds `Star` & `StellarPopulation` classes
🪐 Project 2: N-Body Star Clusters	2 weeks	Numerical integration, Monte Carlo sampling, realistic stellar dynamics	Uses P1’s `StellarPopulation` to simulate actual clusters
☀️ Project 3: Monte Carlo Radiative Transfer	3 weeks	Radiative transfer physics, Monte Carlo methods, photon transport	Builds on stellar physics from earlier projects
🌌 Project 4: Cosmological Inference	3 weeks	Bayesian inference, MCMC, parameter estimation	P2’s leapfrog → HMC sampler!
📊 Project 5: Gaussian Processes	3 weeks	Surrogate modeling, kernel methods, uncertainty quantification	Emulates/analyzes P2’s N-body simulations
🧠 Final: Neural Networks + JAX	4 weeks	Deep learning, automatic differentiation, GPU acceleration	Learns from N-body dynamics + accelerates everything

The Scaffolding Architecture¶

Friday Class Structure¶

Every Friday follows a consistent rhythm designed for active learning:

Interactive Lecture

📚 11:00-11:40 AM

Concept review & clarification
Q&A on required readings
Live coding demonstrations
Project requirement discussion

Hands-On Implementation

💻 11:40 AM-1:40 PM

Collaborative work time with project partners
Discuss approaches and debug together
Everyone writes their own implementation
Instructor support rounds

Course Highlights & Features¶

🔬 Glass Box Methodology

Build every algorithm from scratch before using libraries

👥 Collaborative Learning

Project partners for discussion & debugging (individual code)

🚀 Modern Frameworks

Bridge from NumPy to JAX ecosystem by course end

🌟 Real Astrophysics

Every project addresses genuine astrophysical problems

🤖 AI Integration

Strategic three-phase approach that builds authentic AI literacy for modern computational research careers

📈 Growth Mindset

Progress over perfection, productive struggle as pedagogy

Responsible AI: Three-Phase Scaffolding¶

Strategic three-phase approach that builds authentic AI literacy for modern computational research careers:

AI as a Learning Partner, Not a Replacement

Phase	Weeks	Permitted Usage
Phase 1	Weeks 1-7	Foundation building with limited AI (30-min rule before debugging help)
Phase 2	Weeks 8-13	Strategic integration with critical evaluation (refactoring after baseline works)
Phase 3	Weeks 14-16	Professional practice (acceleration while maintaining understanding)

📋 Syllabus

Complete course policies & expectations

🎯 Projects

Six progressive computational projects

🤖 AI Guidelines

Detailed scaffolding framework

🛠️ Setup Guide

Environment setup & tools

Expected Learning Outcomes¶

By course end, you will have:

✅ Built every algorithm from scratch – No black boxes remain
✅ Discovered algorithm generality – Tools transfer across domains
✅ Developed computational intuition – Know why methods fail/succeed
✅ Created a portfolio – 6 working projects demonstrating progression
✅ Mastered modern tools – JAX, autodiff, GPU acceleration potential
✅ Practiced research skills – Git, testing, documentation, scientific writing

Instructor & Support¶

Instructor	Dr. Anna Rosen
Office	Physics 239
Email	alrosen@sdsu.edu
Student Hacking Hours	Wednesdays 1-2 PM (P-239) or by appointment
Course Website	https://astrobytes-edu.github.io/astr596-modeling-universe/
Issues & Questions	GitHub Issues

AI Transparency Statement

Course Development: This course was developed with substantial assistance from Claude AI as a thought partner and content accelerator. All materials underwent rigorous fact-checking and refinement through domain expertise in computational astrophysics.

Reality Check: Through extensive use, I’ve encountered AI’s critical limitations firsthand—frequent generation of incorrect physics content, non-existent functions, faulty code, and confident misinformation. Every piece of content required verification against documentation, textbooks, and primary literature. My workload increased, but became far more productive.

Pedagogical Impact: These experiences fundamentally shaped the course design. The three-phase AI scaffolding policy, glass-box methodology, and emphasis on first-principles understanding all emerged from recognizing that students need robust foundations to identify when AI is wrong (which occurs frequently). This isn’t about avoiding AI—it’s about thoughtful integration that builds genuine expertise.

Why This Matters: As AI tools become standard in academia and industry, the differentiator isn’t who can prompt AI, but who can verify outputs, catch errors, and guide AI toward meaningful solutions. Human creativity and expertise remain irreplaceable. This course prepares you for a reality where deep understanding, strong BS detectors, and depth in technical skills are mandatory for meaningful contributions.

✨ Welcome to a transformative journey from Python basics to the frontiers of computational astrophysics. Together, we’ll model the universe, one algorithm at a time. ✨