Problem faced between SA and D-wave sampler solution
I have uploaded the image of QUBO matrix. But the solutions of dimod's simulated annealing sampler and D-wave sampler are not same. Simulated annealing sampler found the lower energy states but Dwave sampler gave something else (first 2 energy state are same). I have also checked the exact solver for all possible solutions.
the above image shows the QUBO matrix and Exact solution. where you can see first 3 states are lower energy states.
This above image shows the result of simulated annealing sampler and d-wave sampler. I think some embedding is required.
- But how to perform the embedding for this problem, if embedding can solve the issue?
- I found there are some issue with chain strength also. Is that a problem?
- How to fix this up?
Comments
Hello,
In order to solve problems on the QPU, they generally need to be embedded in order to fit the architecture.
The chimera graph is not fully connected, so in order to work around this, we use a minor-embedding.
When an embedding is used, chains are created.
There needs to be a balance between the problem strength and the chain strength in order to get better quality results.
This recently written white paper might be useful in tuning chain strength:
https://www.dwavesys.com/sites/default/files/14-1041A-A_Setting_The_Chain_Strength.pdf
Some solvers do function better on smaller problems.
With the exact solver, all combinations of inputs are tested, so it is possible to see the exact lowest energy.
With larger problems this is not always the case.
I hope this helps.
Please feel free to reach out with more questions.
If anything needs clarifying or you would like more information, don't hesitate to ask!
Please sign in to leave a comment.