Codes for coding challenge@Song-Shan-Hu Sping School. All results can be found on ans1.txt and ans2.txt
There are three approaches to solve this question.
The first approach is by sampling all vertices of FVS and summing up others to calculate
cd src/; python exact.pyto get results and running time.
The second approach is contracting tensor networks, run
python src/TNchallenge.pythe result is identical to the first one, but the running time will be highly reduced.
The third approach is by running Variational Autoregressive Networks(VAN) on the antiferromagnetic buckyball. Since VAN is basically a vairational method, thus the exact result will be hard to obtain. We put it here just to show that VAN is suitable to solve questions like calculating src/VAN.py.
From the begining, we used VAN to find ground states(there are tons of methods to do the same task, like simulated annealing), collected over 15000 of them, and then analyzed their patterns.
After brainstorming, we found two ways to solve it. The story is long, maybe we will update it later.
The results can both be calculated by
cd src/; python findfactor.pyand
cd src/; g++ -o gs1 gs.cpp; ./gs1Sujie Li, Pengfei Zhou, Feng Pan. All come from ITP, CAS, directed by Prof. Pan Zhang
We want to thank Yifan Qu of many helpful insights and all the staff and students of this spring school. We really enjoy this wonderful week.