Insider Brief
- EigenOrbit’s research team has introduced eigenkit, an open-source Python library that lets developers run quantum algorithms across different simulators and hardware providers through a common interface.
- The study evaluates 12 execution providers, including IBM Heron and IQM Emerald processors, and finds that no single backend performs best across all workloads.
- The researchers found that simulator performance varies with circuit size and structure, while hardware performance also depends on how closely a circuit’s topology matches a processor’s physical connectivity.
PRESS RELEASE — A new study from the EigenOrbit research team examines a growing challenge in quantum computing: how to run and compare the same quantum algorithm across different simulators and hardware platforms without rebuilding the software workflow for each provider.
The paper, “eigenkit: A Provider-Neutral Quantum Computing Library,” introduces and evaluates eigenkit, an open-source Python library designed to separate the definition of a quantum algorithm from the backend used to execute it.
The study evaluates quantum workloads across 12 execution providers , spanning simulators and quantum hardware, and includes experiments on IBM Heron and IQM Emerald quantum processors.
The results show that no single backend is consistently preferable across different workloads .
Simulator performance can change substantially depending on circuit size, structure, batching and application workload. A simulator that performs well for one task may be considerably less suitable for another.
On real quantum hardware, the researchers found that choosing a processor also depends on the relationship between the algorithm and the hardware itself. In experiments on IBM Heron and IQM Emerald, the processor better suited to a particular algorithm depended not only on hardware constraints, but also on how closely the topology of the quantum circuit matched the physical connectivity of the processor.
The findings highlight a broader challenge for quantum software: as the ecosystem grows across multiple hardware architectures, simulators and cloud providers, developers increasingly need the ability to evaluate a workload across different execution environments before deciding where it should run.
A provider-neutral approach to quantum software
eigenkit was developed to address this problem by allowing researchers and developers to define a quantum circuit independently of a specific execution provider.
The same circuit can then be executed across supported simulators and quantum hardware through a common interface, making it easier to compare execution environments without rewriting the surrounding workflow.
The library currently supports eight simulator backends and four hardware execution routes , for a total of 12 execution providers. It also provides backend-independent gradient computation and integrates with widely used scientific computing and machine-learning tools including NumPy, PyTorch and scikit-learn .
The goal is not to hide differences between quantum platforms, but to make those differences easier to evaluate.
The study demonstrates that backend portability can have practical implications beyond software convenience. The ability to move the same workload between execution environments can affect both computational efficiency in simulation and execution quality on quantum hardware .
From open-source software to experimental evaluation
EigenOrbit released eigenkit as an open-source Python library for practical quantum computing workflows.
Developers can write a quantum circuit once and switch between supported execution environments while keeping the circuit definition, training workflow and gradient interface consistent.
The library also includes tools for quantum machine learning, optimization and chemistry, as well as benchmarking functionality for comparing quantum models with classical baselines.
The new research provides an experimental evaluation of the provider-neutral approach and shows why backend choice remains an important technical decision rather than a fixed property of an algorithm.
Read the research
- eigenkit: https://eigenorbit.com/kit/