Abstract:Advances in quantum annealing technology make it possible to obtain high quality approximate solutions of important NP-hard problems. With the newer generations of the D-Wave annealer, more advanced features are available which allow the user to have greater control of the anneal process. In this contribution, we study how such features can help in improving the quality of the solutions returned by the annealer. Specifically, we focus on two of these features: reverse annealing and h-gain. Reverse annealing (RA) was designed to allow refining a known solution by backward annealing from a classical state representing the solution to a mid-anneal point where a transverse field is present, followed by an ordinary forward anneal, which is hoped to improve on the previous solution. The h-gain (HG) feature stands for time-dependent gain in Hamiltonian linear ($h$) biases and was originally developed to help study freezeout times and phase transitions in spin glasses. Here we apply HG to bias the quantum state in the beginning of the annealing process towards the known solution as in the RA case, but using a different apparatus. We also investigate a hybrid reverse annealing/h-gain schedule, which has a backward phase resembling an RA step and whose forward phase uses the HG idea. To optimize the parameters of the schedules, we employ a Bayesian optimization framework. We test all techniques on a variety of input problems including the weighted Maximum Cut problem and the weighted Maximum Clique problem. Our results show that each technique may dominate the others depending on the input instance, and that the HG technique is a viable alternative to RA for some problems.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to improve the quality of solutions to NP - hard problems through more advanced annealing paths in quantum annealing technology. Specifically, the paper explores two new techniques - Reverse Annealing (RA) and time - dependent gain with linear bias (h - gain, HG), as well as their combined usage methods, in order to improve the quality of solutions returned by D - Wave quantum annealers.
### Background and Objectives of the Paper
With the progress of quantum annealing technology, it is now possible to obtain high - quality approximate solutions to NP - hard problems using these technologies. The new - generation D - Wave annealers offer more advanced functions, enabling users to have more fine - grained control over the annealing process. This paper studies how these new functions can help improve the quality of solutions returned by annealers, with a particular focus on two techniques: Reverse Annealing (RA) and time - dependent gain with linear bias (HG).
### Technique Introduction
1. **Reverse Annealing (RA)**
- **Definition**: Reverse annealing allows starting from a known classical state (representing the initial solution), annealing backward to an intermediate point, and then performing normal forward annealing, hoping to find a better solution in the quantum phase.
- **Purpose**: To improve the known (sub - optimal) solution.
2. **Time - dependent gain with linear bias (HG)**
- **Definition**: HG allows adjusting the linear terms in the Ising model in a time - dependent manner during the annealing process, thus biasing towards the known initial solution at the beginning of annealing.
- **Purpose**: To guide the system close to the known initial solution during the annealing process, and then gradually reduce the bias to explore the nearby solution space.
### Experimental Methods
- **Bayesian Optimization**: Used to optimize the parameters of the annealing path, including the total annealing time, annealing path parameters, etc.
- **Experimental Problems**: Tested the Weighted Maximum Cut Problem and the Weighted Maximum Clique Problem.
### Experimental Results
- **Maximum Cut Problem**
- **RA**: It works well for low - density graphs.
- **HG**: It works well for high - density graphs.
- **RA + HG**: It combines the advantages of both, but is not significantly better than using RA or HG alone.
- **Maximum Clique Problem**: Similarly, different techniques perform differently on graphs with different densities.
### Conclusions
Each technique may perform best on different input instances, and the HG technique is a viable alternative to RA for some problems. Through Bayesian optimization, the optimal annealing path parameters can be found, thereby improving the quality of solutions.
### Formula Summary
- **Ising Model**
\[
Q(x_1,\ldots,x_n)=\sum_{i = 1}^n h_i x_i+\sum_{i < j}J_{ij}x_i x_j
\]
- **Ising Model with Time - dependent Gain**
\[
H_{HG}(s)=-\frac{A(s)}{2}\sum_{i = 1}^n\hat{\sigma}_x^{(i)}+\frac{B(s)}{2}\left(\sum_{i = 1}^n g(t)h_i\hat{\sigma}_z^{(i)}+\sum_{i > j}J_{ij}\hat{\sigma}_z^{(i)}\hat{\sigma}_z^{(j)}\right)
\]
- **Weight Function of Linear Terms**
\[
h(x)=\sum_{i = 1}^n(-x_0^i)x_i
\]
Through these techniques and methods, the paper shows how to use advanced annealing paths in quantum annealing to improve the quality of solutions.