University of Michigan tool simulates circuits with lost qubits quickly

This represents a performance increase of at least 1000x compared to tools like Clifft and PPVM. The simulator achieves this speed by accurately modeling circuits even with qubit loss and noise, using a “common reference” approach for circuits defined as frame-CSS. The work introduces a near-linear-time procedure to certify whether a circuit can benefit from Ellis’s faster simulation capabilities.

Frame-CSS Circuits Enable Masked Pauli Frame Simulation

The conventional approach, which relies on XORing Pauli flips onto a shared reference measurement record, falters when qubits are lost mid-circuit, as the resulting variation disrupts the common reference needed for accurate simulation. This work demonstrates that specific circuits, those retaining a common reference under loss, can be accurately modeled with a masked Pauli frame, accounting for dropped operations. The researchers define these circuits as frame-CSS, meaning they are CSS (complementary to self) up to local single-qubit basis changes, given a defined loss model specifying where qubits can be lost and reloaded.

Proving that every frame-CSS circuit possesses a common reference is central to their approach, and they’ve developed a procedure to certify frame-CSS status. This certification process is important, as it allows users to determine if a circuit will benefit from the accelerated simulation capabilities offered by their new tool. Unlike methods requiring separate tableau simulations for each shot, a process that sacrifices efficiency, ELLIS uses the shared reference inherent in frame-CSS circuits to maintain speed.

The tool applies each frame update only when the relevant qubits are present, producing the correct variation and fault flips for each shot. “With one shared reference r, its records are uniform on r + ds + Vs, whereas the exact records are uniform on ds + As,” explains the paper, detailing the mathematical basis for this approach.

On BB144, a specific benchmark circuit, the simulator maintains this advantage, achieving three to four orders of magnitude faster performance even when using 128 cores. The ability to simulate circuits exactly under realistic conditions, including both qubit loss and Pauli noise, is a key advancement. The researchers note that losing a qubit changes the detector matrix but does not invalidate the shared reference for CSS circuits, eliminating the need for computationally expensive tableau simulations. This efficiency extends to circuits compiled to other gate sets, further broadening the applicability of the ELLIS simulator.

Tableau and Pauli-Frame Simulation Distinguish Circuit Records

This speed is enabled by a focus on frame-CSS circuits, a specific class of circuits that maintain a common reference point even when qubits are lost during computation. The core principle behind this advancement lies in recognizing that circuits within the frame-CSS definition allow for a single noiseless measurement reference to be computed via tableau simulation, while subsequent shots are represented as Pauli-frame differences from that reference.

This approach bypasses the need for repeated tableau simulations for each shot, a computationally expensive process in traditional methods. The work identifies that the records of these circuits form an affine space, meaning they exist within a structure where all points are equally likely, simplifying the simulation process.

This is because the possible noiseless records form an affine space a + V, where V represents the image of a matrix B. Unlike a Pauli fault which only affects the Pauli deviation, a dropped Clifford gate changes the ideal circuit itself. This distinction allows ELLIS to maintain accuracy and speed even when simulating circuits with qubit loss, a common source of error in real quantum hardware.

The records of a noiseless circuit form an affine space, expressed as a + Bξ, where *ξ* is a vector of independent, uniformly random bits. This foundational understanding of how noiseless records behave underpins the entire simulation framework. This is a key insight that highlights the efficiency gains achieved through this approach. This advancement promises to accelerate the development of more robust and reliable quantum computers by providing a powerful tool for simulating and analyzing the performance of quantum error correction codes under realistic conditions.

Qubit Loss Disrupts Standard Pauli-Frame Sampling

Unlike traditional Pauli-frame sampling which relies on a single, unchanging reference for all shots, Ellis adapts to qubit loss by constructing a reference specific to circuits exhibiting particular characteristics. Every frame-CSS circuit, the researchers prove, possesses a common reference, allowing for efficient simulation under Pauli noise and loss through a Pauli frame with per-shot masks for dropped operations. The ability to identify and use these frame-CSS circuits is central to Ellis’s speed.

The certification of whether a circuit qualifies as frame-CSS is also performed efficiently. This certification process examines the circuit under a pessimistic loss model, where qubits can be lost before or after any operation and reloaded between any two operations. The team tested the certification process on 73 circuits, including distance-three exports of LightStim protocols and two-round qLDPC memory circuits generated with qLDPC.

The framework accounts for qubit loss by effectively canceling gates that would have acted on the lost qubit, a technique visualized in the paper’s diagrams. The identity placeholders maintain the physical time steps, ensuring the loss is accurately modeled even between operations. This gate cancellation, combined with the common reference, allows Ellis to batch updates for multiple shots, significantly reducing computational overhead. The researchers note that the same certification and sampling techniques apply to leakage, treating it identically to qubit loss within their model.

The team’s approach diverges from other simulators like PPVM and Clifft in how they handle loss. PPVM maintains separate tableaus and loss flags for each shot, while Clifft updates Clifford coordinates online to account for dropped gates. Ellis, however, uses the certified common reference to perform masked Pauli-frame updates, streamlining the simulation process. Timing experiments, excluding circuit parsing and sampler setup, demonstrate the substantial performance gains achieved by this method.

ELLIS Achieves 952,000 Shots Per Second with Loss

The simulator’s efficiency stems from a novel approach to handling circuits where qubits are lost during computation, a common source of error in physical quantum systems. Unlike previous simulators, Ellis uses a “common reference” technique for a specific class of circuits termed “frame-CSS circuits,” enabling exact simulation even with both qubit loss and Pauli noise. The team’s work identifies that membership in the frame-CSS circuit class is sufficient for a common reference under any loss model, though not strictly necessary. The core innovation lies in how Ellis handles dropped operations resulting from qubit loss.

This streamlined approach avoids the need for individual tracking, allowing for parallel processing and a dramatic increase in speed. The ability to accurately and efficiently simulate circuits under loss is important for evaluating quantum error correction schemes. The framework’s design allows it to compute a single noiseless measurement reference using tableau simulation, then represent each shot as a difference from that reference.

This method, combined with the efficient handling of qubit loss, positions Ellis as a valuable resource for the quantum computing community. This unified approach simplifies the simulation process and broadens the applicability of Ellis to a wider range of error models.

Defining Frame-CSS Circuits Relative to Loss Models

Researchers formalized these circuits, defining them relative to a loss model that specifies where qubits can be lost and subsequently reloaded, an important distinction from previously studied “secretly CSS” circuits. Nine circuits, Y-state injection, six S-gate teleportation circuits, the S round trip and the five-qubit code, lacked a common reference, with contradiction searches taking up to 91 seconds to confirm this absence. Lattice-surgery distillation also proved not to be frame-CSS, timing out after two minutes, but the structural checks completed quickly.

The researchers note that the invariant underlying frame-CSS circuits may also benefit decoding, as the detector error model is similarly dependent on the loss pattern, a direction they intend to explore in future work. “We gave a procedure that decides, under stated assumptions, whether a circuit satisfies these conditions, and we released the open-source simulator Ellis,” the team writes.

The work builds upon and generalizes conditions for efficient simulation previously established by other groups, formally defining their “secretly CSS” circuits as frame-CSS circuits relative to a loss model. This formalization provides a more comprehensive understanding of the conditions under which accelerated simulation is possible, and the developed certification procedure fills a gap left by that earlier work.

Beyond simulation, the researchers suggest that the principles behind frame-CSS circuits could extend to improvements in quantum decoding, potentially streamlining the error correction process itself. During the preparation of the manuscript, the team became aware of concurrent work simulating CSS circuits at a cost of O(1) per gate, a method closely related to Pauli-frame simulation, further highlighting the active research in this area.

👉 More information
🗞 Ellis: Extremely Fast Quantum Circuit Simulation under Qubit Loss and Leakage
✍️ Yuewen Hou and Gokul Subramanian Ravi (University of Michigan, Ann Arbor, USA)
🧠 ArXiv: https://arxiv.org/abs/2611.26173v1

Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of Muhammad Rohail T.

Latest Posts by Muhammad Rohail T.: