14. Scientific Intelligence AI×Science: AI × HPC Convergence

Chapter focus: AI×Science is the convergence frontier of "AI compute × supercomputing": using AI compute and model capabilities to reshape the accuracy and scale of scientific computing. Half of this chapter covers the engineering route of scientific foundation models and AI4S compute infrastructure; the other half covers disciplinary applications from life, weather, fluid, and materials to molecular dynamics, quantum chemistry, and genomics.

Articles

SFM frontier: from single-task models to general scientific models

阅读文章 →

Five differences between AI4S and pure AI clusters: network/precision/storage/scheduling/stack

阅读文章 →

SFM distributed training: data modality, mixed precision, parallel strategies

阅读文章 →

Scientific agents: LLM collaborative reasoning with CFD/DFT/database tools

阅读文章 →

AI4S reproducible workflows: data versioning, container pinning, full provenance

阅读文章 →

ML potentials for molecular dynamics: DFT sampling, training, MD engine integration

阅读文章 →

ML surrogates for quantum-chemistry DFT: energy/density/Hamiltonian routes

阅读文章 →

AI weather modeling: data/architecture/training/inference engineering from FourCastNet to Pangu

阅读文章 →

Parallel genomics: AI×HPC pipeline from alignment to variant calling

阅读文章 →

AI×Life science: From AlphaFold to protein design

阅读文章 →

AI weather frontier: Pangu, GraphCast, FourCastNet

阅读文章 →

AI×Fluid dynamics: physics-informed CFD

阅读文章 →

AI×Materials science: from structure-property to inverse design

阅读文章 →

## Suggested Reading Path

Engineering readers start with AI4S compute infrastructure and distributed training (linking Chapter 11), then scientific agents and reproducible workflows; disciplinary readers jump straight to their field articles.

Key Takeaways

The engineering mainline of AI×Science is "data-driven + physically falsifiable": AI approximates and discovers, scientific validation falsifies, and only their combination yields credible scientific conclusions.