Normalizing flows for lattice field theory — reading mapNormalizing flow 用于格点场论 —— 阅读地图
A personal reading list / road map for normalizing-flow samplers on the lattice, with short comments on each entry.个人在整理的 normalizing flow 格点采样阅读清单，每条都附带一两句自己的看法。
This page is meant to grow over time — it tracks what I’ve actually read versus what I’ve only skimmed, and which papers I’d recommend to someone starting out. For the broader map (flows, autoregressive models, path-based samplers, and the gates that are still closed), see the working map.
Core papers
- Albergo, Kanwar, Shanahan (2019) — first normalizing-flow sampler for lattice $\phi^4$; the cleanest starting point.
- Kanwar et al. (2020) — gauge-equivariant flows for $U(1)$ and $SU(N)$.
- Boyda et al. (2021) — non-trivial group manifolds; introduces matrix-valued coupling layers.
Tutorials & code
- I’ll add links here as I clean up my own training scripts.
Things I’m tracking
- How autocorrelation behavior scales with lattice size near criticality.
- Whether you can usefully share flow parameters across different bare couplings.
这一页会慢慢长大。我用来记自己实际读过 vs. 只是扫过的论文，以及推荐给入门者的清单。 更宽的地图（flow、自回归、路径采样器，以及还没过的几道关）写在这篇工作笔记里。
核心论文
- Albergo, Kanwar, Shanahan (2019) —— 第一次把 normalizing flow 用到格点 $\phi^4$ 抽样，最适合做起点。
- Kanwar et al. (2020) —— $U(1)$ 与 $SU(N)$ 上的 gauge-equivariant flow。
- Boyda et al. (2021) —— 推广到非平凡群流形，引入矩阵值的 coupling layer。
教程与代码
- 等我把自己的训练脚本整理干净之后会陆续放链接。
自己在追踪的问题
- 临界点附近，自相关时间随格点尺寸怎么 scale。
- 不同 bare coupling 下，flow 参数能不能共享/迁移。