Any relationship between number of reads required based on qubits used or a rule-of-thumb to set number of reads?
I would like to ask if there is a guideline on the number of reads required for problems of certain qubit size.
For example, maybe 1000 reads for less than 5 physical qubit problem and 5000 reads for 25 physical qubit utilization problem?
Also, is simulated annealing solver equivalent to the QA solver but without noise? Is the noise of the Advantage System4.1 computable?
Comments
Hello,
The best number of reads value is highly problem specific. Since readout is usually longer than anneal, balancing the programming time per call with the readout time is a good approach. Recommended number of reads values generally fall in the range of 100 to 1000.
Simulated annealing is a classical heuristic, which is different from Quantum Annealing. It's not possible to simulate the dynamics of a noise-free system.
The ICE Documentation is a good place to start to understand the noise/errors present in the system.
Hello David,
Thank you for the reply. How can we try to quantitatively assess the quality of the result produced by the QPU then?
At the current stage, I was comparing the QPU solver (Advantage4.1) to simulated annealing result, does this comparison hold any meaning?
Hello,
This paper provides a means of comparing various algorithms, such as Simulated Annealing, to the QPU:
https://arxiv.org/abs/2305.00883
Please let us know if you have any questions.
I read a bit on the integrated control error and how they affect solution quality which seems rather random as it depends on when the crossing of phase boundary occur during the anneal process. Is there any way to know if phase boundary has been crossed during the anneal process due to ICE, or any method to measure the total ICE that my problem could have?
Can I also check if the below assumption is true for a QA solver?
Greater number of reads = higher chance of better solution quality.
Hello,
Are you able to clarify a little what you mean?
The ICE documentation helps to account for various sources of error and offers a this equation to calculate the margin of error to expect. Many of these sources of error are random in nature, so accounting for all of them would not be trivial.
Increasing num_reads will increase the range of solutions sampled from the underlying distribution, and better-quality solutions are more likely to appear in larger samples. But the rate at which the sample minimum improves grows logarithmically with sample size which can be almost imperceptibly slow between large values of num_reads (ie 115 vs 125 samples), but is generally more readily visible between smaller values of num_reads (ie 15 vs 25 samples).
Please sign in to leave a comment.