General Noise Channels and Mid-Circuit Measurements for Pre-Trajectory Sampling with Batched Execution
This paper introduces a generalized pre-trajectory sampling with batched execution (PTSBE) method that enables fast, exact simulation of noisy quantum circuits with mid-circuit measurements and general Kraus channels on GPUs, achieving speedups of up to 83x over conventional methods while revealing significant inaccuracies in Pauli-twirled noise approximations.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer
To build a computer that can solve problems beyond the reach of today's machines, scientists are turning to the strange laws of quantum physics. These quantum computers use tiny particles, like atoms or electrons, that can exist in multiple states at once, allowing them to process information in ways that classical computers cannot. However, these delicate systems are easily disturbed by their environment, causing errors that corrupt the calculation. To fix this, researchers use a technique called quantum error correction, which constantly checks the system for mistakes and repairs them without destroying the information. This process requires taking thousands of measurements to track what is happening inside the machine. To design these systems and train the software that will eventually run them, scientists need to simulate how these computers behave under realistic conditions. The challenge is that simulating the exact, messy reality of these quantum errors is incredibly slow and computationally expensive, while the faster methods available often rely on simplifications that change the results in subtle but important ways.
A team of researchers at NVIDIA, MIT, and the University of Oxford has developed a new method to bridge this gap, allowing for fast and exact simulations of noisy quantum circuits. Their work focuses on a specific type of simulation called trajectory sampling, which tracks the path a quantum system takes as it evolves. In the past, simulators had to choose between speed and accuracy. The fastest tools could only handle simplified models of noise, effectively ignoring the complex, real-world interactions that occur when a quantum computer measures its own state mid-calculation. The more accurate tools could handle these complexities but were so slow that generating the massive datasets needed for modern machine learning was impractical. The researchers created a technique called generalized pre-trajectory sampling with batched execution. This approach allows them to simulate the exact behavior of quantum circuits, including those with mid-circuit measurements and general types of noise, at speeds that were previously thought impossible for such detailed models.
The core of their innovation lies in how they handle the randomness inherent in quantum mechanics. In a quantum system, the outcome of a measurement or the effect of noise is not predetermined; it is a matter of probability. Traditional simulators would run a single simulation, make a random choice for every event, and then start over for the next run. This meant that if two runs happened to follow the same path for a while, the computer would waste time calculating the exact same steps twice. The new method changes this by deciding some of the random outcomes in advance, before the simulation even begins. It identifies parts of the process where the probabilities are fixed and independent of the system's current state, and it pre-selects those outcomes. For the parts that do depend on the state, it waits until the simulation reaches that point to make the decision.
This strategy allows the researchers to group many potential simulation runs together. If a group of runs shares the same pre-selected outcomes, they can travel through the circuit together as a single unit. Only when they reach a point where the state-dependent decision is needed does the group split into smaller branches, each following a different path. Crucially, because the researchers have already calculated the state for the group, they do not need to recalculate it for every single branch. They can reuse the cached state, saving a tremendous amount of time. Once the simulation reaches the end, they can generate thousands of measurement records from that single final state in a single batch, rather than running the entire circuit thousands of times. This is particularly powerful for generating the large, labeled datasets required to train artificial intelligence models to act as decoders for quantum error correction.
The team tested their method on various quantum circuits, including those designed to store information in topological codes, which are a leading candidate for building fault-tolerant quantum computers. When they compared their new simulator to the standard, high-performance tools currently in use, the results were striking. For circuits involving a specific type of noise called amplitude damping, which mimics how energy leaks out of a quantum system, their method was up to 24 times faster than the conventional approach. For more complex random circuits that included mid-circuit measurements, the speedup was even more dramatic, reaching up to 83 times faster. In a specific mode designed for generating massive datasets, where the same simulation path is used to produce many different measurement records, their method was nearly 70,000 times faster per trajectory than the standard method. This leap in efficiency means that researchers can now generate the millions of data points needed to train sophisticated machine-learning decoders in a fraction of the time it used to take.
Beyond just speed, the researchers demonstrated that their method produces more accurate results than the common shortcuts used in the field. A widely used technique called Pauli twirling simplifies complex noise into a form that is easier to simulate, but the team showed that this simplification introduces significant errors. In their tests on a quantum memory system, the simplified model overestimated the rate at which errors were detected by 21 percent. More importantly, when they used the data from this simplified model to train a decoder, the resulting system made 8 percent more logical errors than a decoder trained on the exact, unsimplified data. This finding highlights a critical flaw in relying on approximations: while they might be fast, they can lead to the design of error-correction systems that are less effective than they could be. By providing a way to simulate the exact noise without sacrificing speed, the new method ensures that the data used to build future quantum computers reflects reality.
The implications of this work extend to the broader effort to build a practical quantum computer. As these machines grow in size and complexity, the ability to simulate them accurately becomes a bottleneck. The researchers' method removes this bottleneck for a wide range of realistic noise models, including those that involve measuring the system while it is running. This capability opens the door to studying complex protocols that rely on mid-circuit measurements, such as those used to create special quantum states or to perform error correction in real-time. The team has made their software available as an extension of an existing open-source platform, ensuring that other scientists can immediately apply these speedups to their own research. By making exact, high-throughput simulation practical, this work provides a vital tool for the development of fault-tolerant quantum computers, helping to ensure that the next generation of these machines will be built on a foundation of accurate, realistic data.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.