Noise-oriented warm start improves quantum optimization for 100 qubits

Researchers at USRA, the Hartree Center, and IBM Quantum have demonstrated a high-performance quantum optimization approach leveraging a 100-qubit Ising Hamiltonian, achieving results comparable to the highest quality at that scale using similar algorithms. Rather than attempting to compensate for naturally occurring noise components such as amplitude attenuation, the research team reports an implementation of Noise-Directed Adaptive Warm Start (ND-AWS) that exploits it through a bit-flip gauge transformation. “The simplicity of the framework opens the door to future enhancements, such as adaptive bias scheduling and integration with classical solvers,” the researchers wrote. ND-AWS improves performance over existing iterative warm-start methods without requiring additional circuit complexity, suggesting a path toward more easily implementable advances in quantum optimization. The results showed approximation ratios ranging from 0.974 to 1.0 for Erdos-Reny graphs with 10% edge probability and from 0.989 to 1.0 for regular graphs. Quantum optimization demonstrations are currently being tested on the scale defined by the 100-qubit Ising Hamiltonian. The team adopted the SABER algorithm to address this challenge, routing a 100-qubit problem graph over a heavy hexagonal topology.

Quantum optimization with limited resources

Quantum optimization demonstrations are currently testing the algorithm at the scale of a 100-qubit Ising Hamiltonian, representing significant progress in pursuing practical benefits in this field. Researchers from USRA, the Hartley Center, and IBM Quantum detailed a new approach, Noise-Directed Adaptive Warm Start (ND-AWS), that clearly improves the performance of these complex systems. The team’s experiments were conducted on the ibm_boston quantum processing unit and the results were obtained. At the heart of ND-AWS is the surprising tactic of proactively exploiting the inherent noise within quantum hardware. Rather than focusing solely on error correction, the algorithm exploits this through the application of a bitflip gauge transformation. This technology adjusts the optimization process to the natural tendencies of quantum devices, effectively turning a liability into an asset. The researchers explain that while many quantum algorithms can be executed in multiple equivalent ways, noise often breaks this symmetry, creating opportunities for exploitation.

The performance improvements achieved by ND-AWS are achieved without increasing the complexity of the quantum circuits themselves. This algorithm performs better than the standard iterative warm start variant that lacks these gauge transformations. This is a very important advantage, as it suggests that improvements can be achieved with existing hardware capabilities, accelerating the path to practical quantum optimization. The results show a range of approximation ratios from 0.974, 1.0 for the Erdős-Reny graph with an edge probability of 10% to 0.989, 1.0 for the regular graph, suggesting high accuracy in finding near-optimal solutions.

Warm start QAOA and iterative approaches

With advances in qubit control and coherence, researchers are increasingly focused on extracting meaningful results from near-term quantum devices, even with their inherent limitations. Iterative quantum optimization algorithms, along with warm start quantum approximation optimization (QAOA) variants, represent good strategies for maximizing performance within these constraints. Rather than simply fixing errors, these approaches improve the optimization process by leveraging existing solutions to simplify the problem. A recent development, Noise-Directed Adaptive Warm-Starting (ND-AWS), builds on this foundation by actively incorporating hardware characteristics into the optimization cycle. The quantum optimization demonstration is currently being tested at the scale of a 100-qubit Ising Hamiltonian. ND-AWS utilizes a bitflip gauge transform to exploit noise components such as amplitude attenuation. This is a tactic that goes beyond traditional error mitigation. Importantly, the performance improvements achieved with ND-AWS do not require more complex hardware.

This is achieved through an iterative procedure that generates a distribution of solutions through repeated preparations and measurements. At each step, Warm Start Ansatz is updated to bias towards the current best solution. The resulting approximation ratios range from 0.974, 1.0 for an Erdős–Reny graph with 10% edge probability to 0.989, 1.0 for a regular graph, making these results one of the highest quality demonstrations of quantum optimization at this scale.

Noise-Directed Adaptive Remapping (NDAR) technology

Filip B. Their research focuses on Noise-Directed Adaptive Warm-Starting (ND-AWS). This is a technology built on the foundation of Noise-Directed Adaptive Remapping (NDAR). The team’s innovation stems from the realization that many quantum algorithms have inherent symmetries that allow them to perform multiple equivalent circuits. However, noise breaks this symmetry and causes mismatches in the behavior of the qubits. “Noise in quantum hardware often breaks this symmetry, such as when two levels correspond to the ground or excited states of the qubit, which are susceptible to decoherence and dissipation effects, respectively,” the researchers explain. This is achieved by a bitflip gauge transformation of the cost and phase separation operators in the QAOA circuit.

The team demonstrated an improvement in the iterative warm-start variant without gauge transformation, achieving approximation ratios ranging from 0.974, 1.0 for Erdős-Reny graphs with 10% edge probability to 0.989, 1.0 for regular graphs. “To our knowledge, our results are among the highest quality results to date for a combinatorial problem solved by a QAOA variant of approximately 100 qubits,” they report. Further simulations suggest that even better performance is possible by increasing the circuit depth, providing a promising path toward more robust and effective quantum optimization.

The technology, tested on a 100-qubit system, represents a shift in perspective that recognizes that noise is not just a barrier to overcome, but a resource to be exploited. However, noise breaks this symmetry and biases certain states. This performance increase is achieved without increasing circuit complexity. The researchers report that ND-AWS generally provides improved performance “at no additional circuit cost” over iterative warm-start variants without gauge conversion. This is a very important advantage, as it suggests a path to enhanced optimization without requiring further advances in hardware capabilities. Experimental implementations in ibm_boston yielded approximation ratios ranging from 0.974, 1.0 for Erdos-Reny graphs with 10% edge probability to 0.989, 1.0 for regular graphs.

The pursuit of quantum advantages in optimization problems often collides with the reality of noisy hardware. But recent detailed approaches suggest that embracing rather than fighting these flaws can yield surprising benefits. This is in contrast to traditional methods that focus solely on error mitigation, and represents a shift toward exploiting unique system characteristics. The team tested the algorithm on 20 random Hamiltonian instances with varying connectivity and achieved approximation ratios ranging from 0.974, 1.0 for an Erdős-Reny graph with 10% edge probability to 0.989, 1.0 for a regular graph when combined with classical post-processing. The algorithm iteratively updates the “warm start” Ansatz to bias towards the current best solution and simultaneously adjusts the cost and phase separation Hamiltonian based on the results of each iteration. This process effectively directs the quantum search towards higher quality regions of the solution space, as shown in their published work.

Exploiting algorithmic bias and hardware noise simultaneously represents an important advance in quantum optimization strategies. Researchers at USRA, the Hartley Center, and IBM Quantum have demonstrated a new approach, Noise-Directed Adaptive Warm Start (ND-AWS), that synergizes iterative warm starts with a technique called Noise-Directed Adaptive Remapping. Rather than simply trying to correct errors, this method actively exploits the inherent properties of quantum noise to improve performance. The team’s work, detailed in recent findings, builds on concepts such as warm-start QAOA and time-blocked QAOA, and provides a path toward more robust quantum algorithms. A central element of ND-AWS is the use of bitflip gauge transformations. Experiments conducted on a 100-qubit system ibm_boston demonstrate the effectiveness of ND-AWS. The quantum optimization demonstration has now been expanded to address problems defined by a 100-qubit Ising Hamiltonian.

Recent advances have made it possible to test quantum optimization algorithms on systems beyond the capabilities of classical simulation. The results showed approximation ratios ranging from 0.974, 1.0 for the Erdős–Reny graph with edge probability 10% to 0.989, 1.0 for the regular graph when considering the best of three independent runs. The team reports and suggests a path towards more robust and efficient quantum optimization strategies.

Researchers are increasingly focused on translating quantum algorithms from theoretical designs to practical implementations on noisy intermediate-scale quantum (NISQ) hardware. A key step in this process, detailed by Maciejewski et al., involves efficiently mapping an abstract problem graph onto the physical connections of available quantum processors. The team employed the SABER algorithm to address this challenge, routing a 100-qubit problem graph through a heavy hexagonal topology on the ibm_boston quantum processing device. This routing is not just a technical detail. It directly affects the fidelity of quantum calculations. In particular, the team achieved approximation ratios in the range 0.974, 1.0 for the Erdos-Reny graph with 10% edge probability, 0.97, 1.0 for the regular graph with 10% edge probability, and 0.989, 1.0 for the regular graph. This level of performance highlights the importance of efficient graph routing in maximizing the potential of NISQ devices. The researchers say further simulations using matrix product states suggest even better results as the depth of the quantum circuit increases, suggesting a promising path toward enhanced quantum optimization capabilities.