Differentiable Graph Neural Network Simulator for the Back-Analysis of Post-Liquefaction Residual Strength from Flow Failure Runout
Yongjin Choi, Jorge Macedo · 2026
Read Differentiable Graph Neural Network Simulator for the Back-Analysis of Post-Liquefaction Residual Strength from Flow Failure Runout by Yongjin Choi, Jorge Macedo (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-11621.txt.
Description
After soil turns to liquid in a quake, crews need to know how strong the failed ground stays. Old ways lean on simple physics and slow hand work. Yongjin Choi and one co-author automate the task with a graph network tool that models grain flows and works back from seen runout to the strength left behind. Gradient tuning drives the search, so no hand steps are needed.
The tool learns from runs shaped by real failure cases, so it can model slopes both small and large. Two well-known dam failures, Lower San Fernando and La Marquesa, serve as tests. In both cases the strengths it finds match past values closely and give runout shapes that make sense. It can also find several linked inputs at once, not just one. For ground work after quakes, that means fast answers from past flows that can be checked and run again.
Catalog record
| CREATOR | Yongjin Choi, Jorge Macedo |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-11621