Read more about the announcements made at the 2022 IBM Quantum Summit:
Delivering performant systems is what we’re best known for: this year, we made major advances in the scale, quality, and speed of our quantum systems. But equally as important to us is ensuring that our users have the most cutting-edge capabilities at their fingertips, and know how to use them. Thanks to the Qiskit Runtime primitives unveiled earlier this year, we’re now able to easily push the latest advances to our users, while abstracting away the details unnecessary for exploring real-world use cases. At this year’s Quantum Summit, we unveiled new error mitigation capabilities for the Qiskit Runtime primitives — and our clients showed us how they’re integrating Qiskit Runtime to tackle their important use cases.
Exploring value with new Qiskit Runtime tools
Since quantum circuits are the basis of a quantum computer’s value, we must be able to handle errors that creep in to execute those circuits faithfully. That’s why, at this year's Quantum Summit, we announced that error suppression and mitigation tools are now available as a beta release through the Qiskit Runtime primitives. These abilities will make it easier than ever for users to explore the value of quantum circuits for their applications, while experimenting with the overhead tradeoffs that come with incorporating error handling.
We have a few families of techniques for handling errors on noisy quantum circuits before the era of quantum error correction, called error suppression and error mitigation. Error suppression techniques are those that remove errors by modifying the circuit, such as by adding gates or changing the shape of the pulses we use to control qubits. Error mitigation instead helps us calculate more accurate expectation values by executing ensembles of related circuits, and then combining the outputs and undoing the effects of noise through post-processing. If you are confused about the difference between error suppression and error mitigation, we’ve put together a primer on this topic
What’s the difference between error suppression, error mitigation, and error correction? Learn more.
Error suppression predates quantum computers, with many techniques born out of research in the field of nuclear magnetic resonance (NMR). We can now incorporate many of these techniques into quantum computers — and can turn them on in Qiskit Runtime with just the flip of a switch. The Sampler and Estimator will now automatically apply the dynamical decoupling error suppression technique to your circuit at optimization level 1 and higher.
Error mitigation techniques tend to require some overhead in compilation, execution, and post-processing. The various methods comes with different cost/accuracy trade-offs, so we are introducing a new option to the primitives called a “resilience level” that allows users to dial in the cost/accuracy trade that is suitable to their work. At the first resilience level, we turn on methods specifically designed to address errors in measurement operations. These methods comes with fairly minor overhead; consequently, we are making resilience level 1 the default setting.
At higher resilience levels, we turn on techniques that also address errors in the interior of the circuit. For instance, at resilience level 2, we will enable zero noise extrapolation (ZNE), which reduces error in an Estimator, but does not come with a guarantee that the answer will be unbiased. At resilience level 3, we turn on our most powerful error mitigation technique known as probabilistic error cancellation (PEC).1 Earlier this year, we demonstrated that PEC can deliver unbiased expectation values of quantum operators. This is the most robust accuracy guarantee we can offer, but it does come with substantial overhead in terms of noise model learning and circuit sampling. So, it makes sense that this method would max out cost/accuracy dial.
Introducing these resilience level options is part of our push to make exploring error handling a frictionless process with Qiskit Runtime. However, for those users interested in more detailed control of the specific methods, we are also providing advanced options in order to fine tune the error mitigation process, such as the ability to tweak the parameters of the ZNE and PEC methods.
We hope that users will use this beta launch as a learning opportunity: what does it mean to spend more time to reduce noise for your specific circuit? When error mitigation is enabled, the primitives will provide metadata with your results in order to contextualize how the resilience levels are impacting your calculation, in order to help you better weigh the time-versus-accuracy cost. This data will also help users construct the optimal quantum circuits for applying error mitigation while reducing overhead; a circuit with many repeating elements is much less expensive to run with error mitigation than a circuit with no such structure, for example.
We’re hoping that this release will begin a conversation about how to balance accuracy with runtime and overhead when using error mitigation. For example, the cost of PEC is dependent on the number of layers in a circuit, where a layer is one time-step’s worth of gates. PEC doesn’t need to learn about every layer of gates included in the circuit, just those containing two-qubit gates. Plus, if two layers contain the same gates on the same set of qubits, PEC need only learn them once. The sampling step depends on the total number of layers, however.
Error mitigation takes us a step further in quantum computing's future, where runtime, rather than error rates, are of most concern to the users. And, while you can still access any hardware you'd like with Qiskit, error mitigation is only available to users accessing IBM Quantum hardware using Qiskit Runtime primitives.
Integrating quantum circuits into applications today
Value means performance, capabilities, and frictionless. But if the value we’re actually providing is quantum circuits with built-in error mitigation and error suppression from the Qiskit Runtime primitives, then we need to ensure that our clients are able to make use of these quantum circuits for their use cases.
We have theoretical evidence for exponential and superpolynomial speedups for quantum algorithms over classical algorithms in a few domains: notably, simulating nature and processing certain kinds of datasets with complex structure. We also see promise for search and optimization problems, in the form of more modest quadratic speedups. However, bringing about useful quantum computing means more than finding and proving the theoretical existence of speedups. Instead, we must connect with partners to identify high-value use cases for quantum algorithms, map their use cases onto quantum circuits, and run them on quantum hardware with error handling to explore potential areas for quantum advantage.
We’ve identified five overarching industry areas where we expect the most impact from these kinds of quantum circuits, based on alignment of popular use cases. These areas are “Aerospace and Automotive,” “Financial Services,” “High Tech,” “Energy, Environment and Utilities,” and “Healthcare and Life Sciences.” Each of these areas come with their own promising potential areas to explore, such as materials design, fraud detection, catalysts, and drug discovery.
In these industry areas, our collective mission is to integrate Qiskit Runtime into our partner’s applications and services in order to extract value from quantum circuits. Qiskit Runtime is built to encourage integration — offering a high-impact tool and access to unrivaled quantum hardware with a low barrier to entry. We continue to introduce new integration tools like error mitigation to Qiskit Runtime, giving our partners access to increasingly advanced capabilities for applications research and software development.
Chemistry as a case study
Good Chemistry Company and Dow are using Qiskit Runtime to explore workflows for accelerating materials design with the help of quantum circuits. Good Chemistry developed the QEMIST Cloud to simulate molecules with the help of a massively parallel supercomputer, leveraging community data and data produced by simulations to train machine learning models, reducing the time and cost of simulation. They also introduced the Tangelo open source software development kit for end-to-end chemistry simulation on quantum computers. By integrating Qiskit Runtime estimator primitive into Tangelo, Good Chemistry's open source development kit, Good Chemistry and Dow were able to deliver end-to-end electronic structure calculations on IBM's quantum computers.
Meanwhile, QunaSys and JSR are using IBM software to devise new methods for probing the behavior of molecules when irradiated or under an applied voltage, one of JSR’s core concerns. QunaSys had developed Qamuy, a cloud service for quantum chemistry simulations that translates user inputs to quantum algorithms. Qamuy also integrates with Qiskit Runtime, receiving expectation values with the Qiskit Runtime estimators plus integrated error mitigation. With Qamuy + Qiskit Runtime, JSR used a QunaSys-developed quantum algorithm to analyze the minimum energy crossing point for the methylene molecule, a value that’s important for understanding the molecule’s photosynthetic properties.
Finally, physicists at Lawrence Berkeley National Lab integrated Qiskit Runtime into NWChemEx, an open source, high-performance parallel computational chemistry code for modeling molecular systems. NWChemEx performs classical computing, and converts pieces of the problem to quantum circuits in Qiskit when necessary, in order to run on IBM Quantum hardware with the help of Qiskit Runtime primitives. Qiskit Runtime primitives plus NWChemEx are allowing physicists at LBNL to hone tighter, more efficient quantum-classical interactivity for their compute fabric, and find ways to standardize those interactions for future builds.
The beta release of Qiskit Runtime primitives with error mitigation, plus these examples from the Quantum Network, demonstrates how our clients are beginning to integrate Qiskit Runtime capabilities into their workflows in order to extract value from quantum circuits. We plan to continue delivering capabilities over time as we progress along our development roadmap, based on the latest research coming out of IBM Quantum and the broader quantum community, and leveraging those capabilities to produce a frictionless experience. We’re excited to see what kinds of applications our developers will build with these exciting new tools.