Decay Rates in Interleaved Benchmarking with Single-Qubit References

This paper establishes a rigorous theoretical foundation for using single-qubit reference sequences in cross-entropy benchmarking by deriving a refined analytical expression that corrects systematic overestimations, thereby enabling reliable and high-precision characterization of multi-qubit entangling gates on superconducting quantum processors.

The Gist

Here is an explanation of the paper using simple language and creative analogies.

The Big Picture: Testing a Quantum Computer's "Muscle"

Imagine you have a brand-new, high-tech gym (a quantum computer) and you want to know how strong its machines are. Specifically, you want to test the two-person machines (entanglement gates) where two people have to work together perfectly.

To test this, you usually have two options:

1. The "Hard" Way: You put two people on the machine at once and see how well they coordinate. This is accurate, but it's slow, expensive, and the people get tired easily (high error rates).
2. The "Easy" Way: You test each person individually on their own machine, then assume that if they are both strong, they will be strong together. This is fast and cheap, but scientists have always worried: "Is this assumption actually true? Or are we tricking ourselves?"

This paper says: "The 'Easy' Way works, but only if you do the math correctly."

---

The Problem: The "Additive" Trap

For years, researchers used the "Easy" way (called Interleaved Cross-Entropy Benchmarking or XEB). They tested single qubits (the "people") separately and added their error rates together to guess the error rate of the two-qubit machine.

Think of it like this:

• Person A drops a ball 1% of the time.
• Person B drops a ball 1% of the time.
• The Old Assumption: If they work together, they will drop the ball 2% of the time.

The authors of this paper discovered that this math is wrong. When two people work together in a quantum gym, their mistakes don't just add up like simple numbers. They interact in a complex way. If you use the old "add them up" math, you end up thinking the machine is better than it actually is. It's like thinking you are a perfect athlete because your two legs are strong individually, ignoring the fact that they might trip over each other when running.

The Solution: A New Formula

The team developed a new, more accurate formula to fix this.

Instead of just adding the errors, they looked at how the errors "decay" (fade out) when you run a long sequence of random tests.

• The Old View: They thought the decay was a straight line (like a car slowing down at a constant rate).
• The New View: They found the decay is actually curved and wiggly (like a car slowing down on a bumpy road).

They derived a specific equation (Equation 5 in the paper) that accounts for this "bumpy road." When you use this new formula, the results from the "Easy" way (testing individuals) match perfectly with the "Hard" way (testing the pair together).

The "Magic" Ingredient: The CZ Gate

You might wonder: "If I only test individuals, how do I know they can work together?"

The paper shows that if you insert a specific "teamwork" move (a CZ gate, which is like a handshake between the two qubits) in the middle of the individual tests, it forces the system to behave like a chaotic, complex team.

The Analogy:
Imagine you are testing a dance troupe.

• The Test: You make the dancers spin individually (random single-qubit gates).
• The Twist: In the middle, you make them do a specific, complex lift (the CZ gate).
• The Result: Even though they started by spinning alone, that one lift forces them to coordinate so perfectly that the whole group looks like a chaotic, unified mess (mathematically, this is called "Porter-Thomas statistics").

Because of this "chaos," the single-qubit tests become a valid proxy for the two-qubit test. The single-qubit references are sufficient to "twirl" the errors into a predictable pattern.

Why This Matters: The "High-Precision" Shortcut

The biggest win of this paper is efficiency.

• The Old Standard (IRB): To test a two-qubit gate, you had to build complex reference circuits using two-qubit logic. This is like building a full-scale replica of a car engine just to test a single spark plug. It's slow and introduces its own errors.
• The New Method (XEB with Single-Qubit Refs): You use simple, single-qubit circuits as your reference. This is like testing the spark plug on a workbench. It's fast, clean, and has fewer errors.

The Result:
The authors tested this on a real superconducting quantum processor (a type of quantum computer). They found that:

1. The new method gave the exact same accuracy as the old, heavy-duty method.
2. The new method was more precise because the reference circuits were simpler and made fewer mistakes.

The Takeaway

This paper removes the "theoretical doubt" surrounding a popular shortcut. It proves that:

1. You cannot simply add up single-qubit errors to get two-qubit errors (the old math was wrong).
2. If you use the new, corrected math, you can safely use simple single-qubit tests to benchmark complex multi-qubit gates.

In everyday terms: You don't need to build a massive, expensive test rig to check if a team works well together. If you test the individuals carefully and use the right formula to combine the results, you get a perfect picture of the team's performance, saving time and money while getting better data.

Technical Summary

Here is a detailed technical summary of the paper "Decay Rates in Interleaved Benchmarking with Single-Qubit References."

1. Problem Statement

Cross-entropy benchmarking (XEB) using single-qubit reference sequences has become a standard method for characterizing multi-qubit gates in large-scale quantum processors. This approach is favored because it reduces experimental overhead by leveraging simple, routine single-qubit calibration operations and supports non-Clifford target gates.

However, the theoretical foundation of this method is weak. It relies on two unproven assumptions:

1. Additivity of Errors: The assumption that the joint fidelity decay of simultaneous single-qubit sequences is simply the product (or additive sum of error rates) of individual qubit decays.
2. Sufficiency of Randomization: The assumption that single-qubit gates alone can effectively "twirl" (randomize) errors in an interleaved circuit to mimic the behavior of multi-qubit Clifford groups.

If these assumptions are false, XEB with single-qubit references could lead to systematic overestimation of gate fidelities and obscure the identification of dominant error mechanisms.

2. Methodology

The authors developed a rigorous theoretical framework to analyze interleaved benchmarking with single-qubit references and validated it through experiments on a superconducting fluxonium-based quantum processor.

• Theoretical Derivation:
  • They modeled local noise as independent single-qubit depolarizing channels.
  • They derived an analytical expression for the joint decay of simultaneous single-qubit benchmarking ($F_{single}$) by explicitly summing over all possible measurement outcomes generated by single-qubit Clifford operations, rather than assuming a simple exponential decay.
  • They analyzed the output probability distributions (bitstring probabilities) to determine if single-qubit references induce sufficient randomization. They compared the distribution of outcomes from factorized single-qubit sequences against the Porter-Thomas (PT) distribution, which is characteristic of fully chaotic, multi-qubit random circuits.
• Experimental Validation:
  • Simulations: Simulated XEB experiments for 2-qubit and 3-qubit systems under local noise to compare standard multi-qubit Clifford benchmarking against simultaneous single-qubit benchmarking.
  • Hardware Experiments: Performed interleaved benchmarking on a superconducting processor.
    • Control: Standard Interleaved Randomized Benchmarking (IRB) using multi-qubit Clifford reference sequences.
    • Test: Interleaved XEB using simultaneous single-qubit reference sequences with a target CZ gate.
  • Post-Processing: Applied a refined fidelity estimator derived from their analytical model to correct the single-qubit reference data.

3. Key Contributions

A. Refutation of the Additive Approximation

The paper proves that the commonly used additive approximation ($F \approx (1 - \sum e_i)^m$) is invalid for simultaneous single-qubit benchmarking.

• They derived an exact analytical formula (Eq. 5) for the fidelity decay of $n$ simultaneous qubits:
  $$F_{single} = \frac{2^n \prod (2+p_i^m) + 3^n - \prod (3+p_i^m) - 4^n}{6^n + 3^n - 2 \cdot 4^n}$$
• This expression exhibits a non-exponential decay that deviates significantly from the simple additive model, particularly as the number of qubits increases.

B. Validation of Randomization (Twirling)

Despite the non-exponential decay of the reference sequences, the authors demonstrated that interleaving a multi-qubit entangling gate (like a CZ gate) with single-qubit references does generate effective global randomization.

• They showed that the output probability distribution of the interleaved circuit converges to the Porter-Thomas distribution (the same as multi-qubit Clifford sampling) once the circuit depth is sufficient (e.g., depth $m=4$ for CZ).
• This confirms that single-qubit references are sufficient to "twirl" the noise for the target gate, provided the statistical criterion is met.

C. Refined Fidelity Estimator

The authors introduced a corrected formula for extracting the target gate fidelity ($p_G$) from single-qubit reference data.

• Instead of the naive ratio, the fidelity must be calculated using the effective depolarizing parameter of the multi-qubit reference model ($p_{multi}$) derived from individual qubit errors ($e_i$):
  $$p_G = \frac{p_{int}}{p_{multi}} = \frac{p_{int}}{1 - \frac{4}{5}(e_1 + e_2)}$$
• Neglecting this correction leads to a systematic overestimation of the gate fidelity.

4. Results

• Theoretical Agreement: The derived analytical expression for $F_{single}$ perfectly matched numerical simulations of simultaneous single-qubit XEB under local noise.
• Experimental Consistency: Experiments on the superconducting processor showed that:
  • The fidelity of a CZ gate extracted via standard IRB (multi-qubit references) was 0.9830(25).
  • The fidelity extracted via interleaved XEB (single-qubit references), when using the refined estimator, was 0.9835(5).
  • The results agree within uncertainty, validating the theoretical framework.
• Precision Improvement: The single-qubit reference method achieved higher precision (smaller error bars) than standard IRB. This is attributed to the significantly lower error rates of single-qubit gates compared to multi-qubit Clifford sequences, which require deeper circuits to synthesize.
• Systematic Error Quantification: The study quantified the error of the naive additive approach, showing it would have overestimated the CZ gate fidelity to 0.9850, a significant deviation in high-fidelity regimes.

5. Significance

• Theoretical Foundation: This work establishes the first rigorous theoretical justification for using single-qubit references in interleaved benchmarking, removing a major conceptual barrier to its widespread adoption.
• Practical Utility: It provides a reliable, low-overhead method for characterizing multi-qubit entangling gates (crucial for scaling quantum computers) without the need for complex, error-prone multi-qubit Clifford reference sequences.
• Accuracy and Precision: By correcting the decay model, the method prevents systematic overestimation of gate fidelities while simultaneously offering higher statistical precision due to reduced reference-sequence noise.
• Future Directions: The paper outlines criteria for detecting non-local errors and suggests that for current-generation processors, the local noise model holds, making this approach robust for immediate application in quantum hardware characterization.