What is the foundational concept behind cross-entropy benchmarking (XEB) and how is it used to measure the fidelity of quantum circuits?

/ / Published in Artificial Intelligence, EITC/AI/TFQML TensorFlow Quantum Machine Learning, Quantum supremacy, Extracting coherence information from random circuits, Examination review

Cross-entropy benchmarking (XEB) is a pivotal technique employed in the realm of quantum computing to evaluate the fidelity of quantum circuits, particularly in the context of demonstrating quantum supremacy. The foundational concept behind XEB revolves around the comparison of experimentally obtained probability distributions with theoretically predicted distributions for a quantum circuit, typically a random circuit. This method quantifies how closely the output of a quantum device adheres to the expected quantum mechanical behavior, thereby providing a measure of the device's performance and coherence.

To understand XEB, it is essential to consider the principles of quantum circuits and the nature of randomness in quantum mechanics. Quantum circuits consist of a series of quantum gates applied to qubits, which are the fundamental units of quantum information. When these gates are applied in a specific sequence, they create a transformation of the qubits' state, leading to a final output state that can be measured. In the context of XEB, the circuits are often designed to be random, meaning that the sequence of gates is chosen according to some random distribution. This randomness is important as it ensures that the circuit's behavior is complex and difficult to simulate classically, which is a key aspect of demonstrating quantum supremacy.

The process of cross-entropy benchmarking begins with the construction of a random quantum circuit. Once the circuit is defined, it is executed on a quantum device multiple times to gather a set of measurement outcomes. Each execution, or shot, yields a bitstring representing the state of the qubits after the circuit has been applied. The frequency of these bitstrings forms an empirical probability distribution.

Parallel to the experimental execution, the theoretical probability distribution of the bitstrings is computed using classical simulation methods. This involves calculating the probabilities of each possible bitstring outcome based on the quantum mechanical description of the circuit. However, for sufficiently large and complex circuits, this classical computation becomes infeasible, which is why XEB is particularly relevant for circuits that are on the edge of classical simulability.

The core metric in XEB is the cross-entropy difference, which is derived from the cross-entropy between the empirical distribution obtained from the quantum device and the theoretical distribution. Cross-entropy is a measure from information theory that quantifies the difference between two probability distributions. Specifically, it is defined as:

    \[ H(P, Q) = -\sum_{x} P(x) \log Q(x) \]

where P is the true distribution (in this case, the empirical distribution from the quantum device), and Q is the predicted distribution (the theoretical distribution). The cross-entropy difference is then given by:

    \[ \Delta H = H(P, Q) - H(P, P) \]

Here, H(P, P) is the Shannon entropy of the empirical distribution, which serves as a baseline. The cross-entropy difference thus measures the deviation of the empirical distribution from the theoretical prediction. A smaller cross-entropy difference indicates higher fidelity, meaning the quantum device's output is closer to the expected quantum mechanical behavior.

XEB has been instrumental in landmark experiments aimed at demonstrating quantum supremacy. For instance, in Google's 2019 quantum supremacy experiment, a 53-qubit processor named Sycamore was used to execute random quantum circuits. The output bitstrings were collected and compared against theoretical predictions using XEB. The results showed that the quantum device could produce distributions that were infeasible to simulate on classical supercomputers, thereby providing evidence of quantum supremacy.

The practical implementation of XEB involves several steps and considerations. First, the design of random circuits must ensure that they are sufficiently complex to challenge classical simulation capabilities. Second, the empirical data collection requires a large number of shots to obtain a reliable distribution. Third, the classical computation of the theoretical distribution, while challenging, can be approximated using advanced simulation techniques for smaller instances of the circuit.

Moreover, XEB is sensitive to various sources of noise and errors in the quantum device, such as gate errors, decoherence, and readout errors. These factors can affect the fidelity of the output distribution, making it important to account for and mitigate these errors to obtain an accurate measure of performance. Advanced error correction and mitigation techniques are often employed to improve the reliability of XEB results.

In addition to its role in benchmarking quantum supremacy, XEB provides valuable insights into the coherence properties of quantum circuits. By analyzing the deviations between the empirical and theoretical distributions, researchers can infer the presence and impact of different types of noise and errors. This information is vital for the development of more robust and error-tolerant quantum devices.

To illustrate the application of XEB, consider a simple example with a small quantum circuit. Suppose we have a 3-qubit circuit with a sequence of randomly chosen gates. After executing the circuit on a quantum device for 1000 shots, we obtain an empirical distribution of bitstrings. Simultaneously, we compute the theoretical distribution using classical simulation. The empirical distribution might look something like this:

    \[ P_{\text{empirical}} = \{000: 0.25, 001: 0.15, 010: 0.20, 011: 0.10, 100: 0.10, 101: 0.05, 110: 0.10, 111: 0.05\} \]

The theoretical distribution might be:

    \[ P_{\text{theoretical}} = \{000: 0.24, 001: 0.16, 010: 0.21, 011: 0.09, 100: 0.11, 101: 0.06, 110: 0.09, 111: 0.04\} \]

Using these distributions, we can calculate the cross-entropy and the cross-entropy difference. The Shannon entropy of the empirical distribution H(P_{\text{empirical}}, P_{\text{empirical}}) is calculated as:

    \[ H(P_{\text{empirical}}, P_{\text{empirical}}) = -\sum_{x} P_{\text{empirical}}(x) \log P_{\text{empirical}}(x) \]

Similarly, the cross-entropy H(P_{\text{empirical}}, P_{\text{theoretical}}) is:

    \[ H(P_{\text{empirical}}, P_{\text{theoretical}}) = -\sum_{x} P_{\text{empirical}}(x) \log P_{\text{theoretical}}(x) \]

The cross-entropy difference \Delta H is then:

    \[ \Delta H = H(P_{\text{empirical}}, P_{\text{theoretical}}) - H(P_{\text{empirical}}, P_{\text{empirical}}) \]

By computing these values, we obtain a quantitative measure of how well the quantum device's output matches the theoretical prediction. This measure helps in assessing the device's fidelity and guiding improvements in quantum circuit design and error mitigation.

Cross-entropy benchmarking is a powerful and essential tool in the evaluation of quantum circuits, particularly in the context of quantum supremacy. By providing a quantitative measure of fidelity, XEB enables researchers to gauge the performance of quantum devices, understand the impact of noise and errors, and drive advancements in quantum computing technology.

Other recent questions and answers regarding Examination review:

More questions and answers:

Tagged under: , , , , ,