Developer documentation

Documentation for SeQuant contributors: how the documentation itself is built and extended, how expression types are identified and ordered, and the internal architecture of Wick’s theorem, tensor-network canonicalization, the cost-based optimizer, the batched evaluation engine, and the code-generation/export framework. This complements, rather than repeats, the User Guide, which covers the public API a library user calls.

• Documentation Guide
  • Dependencies
  • Directory Structure
  • Building Documentation
  • How It Works
  • Adding New Documentation
  • API Documentation
• Expression Types
  • Type ids
• Wick’s Theorem
  • Configuration
  • The contraction algorithm
  • Topological pruning and degeneracy
  • The extended Wick theorem (Vacuum::MultiProduct)
  • Reducing the result
  • Debugging and tests
• Tensor Network Canonicalizer
  • Graph construction
  • Computing the canonical form
  • Topological vs. lexicographic canonicalization
  • Subtleties for contributors
  • Recorded canonical form
  • Debugging and tests
• Cost Model and Single-Term Optimization
  • The CostModel concept and driver
  • The peak recurrence and frontier
  • Ordered batch contexts and costing
  • Roofline performance cost
  • Outer-product DP pruning
• Batched Evaluation Architecture
  • Two execution strategies
  • Array layout and mode frames
  • The ordered pipeline
  • Loop identity and value identity
  • Placement, escapes, and pass splits
  • Cell forms, reads, and lifetimes
  • Executing the table
  • A correctness caveat for sparse backends
  • Dry runs and diagnostics
  • Current scope
• Export Framework Architecture
  • The export pipeline
  • Preprocessing: scalars, rebalancing, and renaming
  • Reuse vs. common-subexpression elimination
  • GenerationOptimizer: an optional peephole layer
  • Implementing a new backend
  • Debugging and tests