A new library called eigenkit now links a single codebase to 12 different quantum simulators and actual quantum hardware platforms, including IBM Heron and IQM Emerald. Researchers at EigenOrbit Sàrl in Luxembourg developed the system to address the challenges of moving algorithms between diverse quantum computing environments and integrating them with standard scientific tools like NumPy, PyTorch and scikit-learn.
Experiments reveal that the best processor for a given algorithm isn’t solely determined by hardware limitations, but also by the characteristics of the algorithm itself. This backend portability, the team reports, is important for both computational efficiency and quantum execution quality.
Provider-Neutral Design Enables Algorithm Portability
This finding highlights the importance of software tools that facilitate algorithm portability across diverse quantum computing platforms, as a mismatched topology can substantially degrade execution quality. Experiments on IBM Heron and IQM Emerald show that which processor is better suited to a given algorithm depends not only on hardware constraints, but also on how well the topology of the quantum algorithm matches the physical connectivity of the processor.
The newly developed eigenkit library addresses this challenge by providing a common interface for executing quantum algorithms across a range of simulators and actual quantum hardware. This approach separates the algorithm’s definition from the specific backend used for execution, enabling systematic comparison and optimization. The library currently supports 12 execution providers, streamlining the process of testing and deploying algorithms on different systems. This design choice allows researchers to focus on algorithm development without being constrained by the peculiarities of individual hardware implementations.
Eigenkit’s architecture is built around a provider protocol, a compile/run split, and consistent estimator interfaces, reusing established designs rather than imposing a new model onto existing qubit-circuit representations. This compatibility extends to standard scientific computing tools, with the library offering backend-independent gradient computation that integrates smoothly with NumPy, PyTorch and scikit-learn. “Its goal is not to replace existing quantum software frameworks, but to provide an abstraction layer that makes quantum algorithms easier to move, compare, and integrate across them,” the developers state.
The library’s utility extends beyond mere convenience. The team’s experiments reveal that the most suitable QPU is contingent on both the hardware itself and how well the quantum circuit maps onto that hardware. Additional routing, required when logical and physical connectivity mismatch, can significantly diminish execution quality.
eigenkit’s Circuit Representation and Layered Architecture
The eigenkit library defines an intermediate representation (IR) of a quantum circuit as data, decoupling circuit construction from execution and storing only the ordered list of gates, qubits and parameters, not numerical values, to be supplied at each call. This IR functions as a circuit schema, enabling a unified interface for building and running them across 12 supported simulators and quantum hardware platforms. The library’s architecture facilitates performance comparisons between providers, revealing that no single backend consistently outperforms others across all workloads.
Experiments demonstrate substantial variation in simulator performance based on circuit size, structure, batching and the application being run, highlighting the importance of workload-specific optimization. The team evaluated the practical relevance of this abstraction across both simulation and quantum hardware, showing that no single backend is consistently preferable.
Output ONLY the corrected text: This finding suggests that provider switching can be a valuable strategy for optimizing performance in real-world quantum computing applications. The library’s modular design also allows for extension, enabling the incorporation of new providers and functionalities as the field evolves.
Supported Backends: 12 Simulators and Quantum Hardware
The eigenkit library currently supports execution across 12 distinct providers, encompassing both quantum simulators and commercially available quantum hardware, a breadth of initial compatibility notable for a new software framework. This range includes simulators like NumPy and Qulacs alongside access to quantum processors such as IBM’s ‘ibm_kingston’ and IQM’s ‘emerald’ systems, enabling direct comparison of algorithm performance across diverse platforms, according to the company. Listing 1 demonstrates a provider-neutral execution of an IQP circuit, showing the library’s ability to run the same code on multiple backends with minimal modification.
The library’s architecture allows developers to define models at varying levels of abstraction, compiling them into an intermediate representation before execution. Performance variations across these backends are substantial, and not simply a matter of computational power. Experiments reveal that no single simulator consistently outperforms others, with results heavily influenced by circuit size, structure, batching techniques and the specific application workload.
Performance of Simulation Varies by Workload and Size
Performance variations across quantum simulators and hardware are substantial, extending beyond simple computational power, as experiments with the eigenkit library reveal a complex relationship between workload and execution speed. At qubit counts common for algorithm development, simulators offer a rapid feedback loop, but their performance diverges significantly depending on the specific task assigned to them. For instance, a simulator optimized for single circuit executions may prove slow when handling the large batches of circuits required for end-to-end hybrid quantum neural network training.
This disparity prompted researchers to investigate which simulator best suits which task, finding no single solution consistently outperforms others. The team’s findings suggest that selecting an SDK, and therefore a specific set of simulators, can significantly impact development time, a critical factor in quantum software creation.
IBM Heron and IQM Emerald Topology-Algorithm Matching
The suitability of quantum hardware extends beyond raw qubit count and coherence times. These findings reveal that the topology of a quantum algorithm, the pattern of interactions between qubits, must align with the underlying hardware for optimal execution, a nuance often overlooked in early-stage quantum computing assessments. This discrepancy in gate requirements directly affects execution quality, as two-qubit operations remain a primary source of error on current quantum processing units.
The implications extend beyond simply selecting the “best” processor; they highlight the importance of considering topology when designing quantum algorithms and compiling them for specific hardware. This is particularly relevant as quantum computing moves from isolated algorithm demonstrations toward practical experimentation, where software is the critical link between algorithm design, classical benchmarking and execution on diverse hardware.
Backend-Independent Gradients for Quantum Models
This architecture allows developers to define quantum circuits once and then run them across a diverse range of platforms without significant code modification, streamlining the process of benchmarking and comparison. The library achieves this through a common gradient interface, using native differentiation when available and parameter-shift methods when it is not, ensuring consistent optimization workflows regardless of the underlying hardware. A key design element of eigenkit is its integration with established scientific computing tools; quantum models can directly interface with NumPy and PyTorch, while higher-level estimators follow scikit-learn-style fit/predict workflows.
This allows for the smooth incorporation of quantum components into existing classical machine-learning pipelines, potentially accelerating the development of hybrid quantum-classical algorithms. The library’s modular dependency structure and extensible design further contribute to its adaptability, allowing researchers to easily add support for new backends or algorithms as they emerge.
The ability to compute gradients independently of the chosen backend is a significant feature, as it removes a major bottleneck in algorithm development and deployment, the company says. Currently supported providers include PennyLane, Qiskit, Qiskit Aer, Qulacs, qsim, TensorCircuit, Amazon Braket, CUDA-Q and Perceval, demonstrating a commitment to broad compatibility. Whenever a provider offers its own differentiation method, such as PennyLane’s, the library uses it for improved performance. Otherwise, it relies on its in-house backend-independent parameter-shift rule.
The library’s architecture centers on compiling models into a backend-independent intermediate representation, enabling execution on any of the 12 supported platforms. This approach simplifies the process of porting algorithms and facilitates systematic performance comparisons across different hardware configurations. The library’s documentation details the full API and implementation, offering a comprehensive resource for developers seeking to integrate quantum computing into their workflows.
The team’s work addresses a critical need in the quantum computing community, namely the difficulty of moving algorithms between different execution environments. By providing a common interface and backend-independent gradient calculation, eigenkit lowers the barrier to entry for researchers and developers seeking to explore the potential of quantum computing across a variety of platforms. The library’s focus on integration with the classical scientific stack further enhances its utility, enabling the development of hybrid algorithms that use the strengths of both quantum and classical computing., IBM reports.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
