Practical Tomography of Multi-Time Processes

1. Problem Statement

The reliable operation of quantum devices is hindered by temporally correlated (non-Markovian) noise, where the system retains memory of past interactions with the environment. Standard error correction and control protocols often fail under these conditions because they assume independent noise channels.

To diagnose and mitigate such noise, a complete operational description is required via multi-time process tomography. This involves reconstructing the process matrix ($W$), which generalizes quantum states and channels to sequences of interventions at different times.

• The Challenge: Complete tomography requires an informationally complete set of probes.
• The Obstacle: Informational completeness necessitates non-deterministic intermediate operations (e.g., measurement-and-repreparation maps). On current quantum hardware, these are typically implemented using mid-circuit measurements, feed-forward, and reset.
• Limitations of Current Methods: Mid-circuit operations are slow, introduce significant noise, require ancilla resets, and are technologically restrictive. Consequently, most experiments are limited to restricted process matrices using only deterministic (unitary) operations, which are insufficient for full reconstruction.

2. Methodology

The authors propose a resource-efficient framework to achieve informational completeness without mid-circuit measurements or resets.

Core Concept: Sequential Ancilla Interaction

Instead of using fresh ancillas or mid-circuit resets for every time step, the protocol utilizes a single coherent qubit ancilla that interacts sequentially with the system across all intermediate labs (time steps).

• Setup: The system interacts with a single qubit ancilla via joint unitary operations ($U_{SA}$) at each time step.
• Measurement: The ancilla is measured only once at the very end of the sequence.
• Correlation: This creates a correlated multi-time probe (a superinstrument element) where correlations between time steps are mediated solely by the coherent evolution of the single ancilla.

Mathematical Framework

• Process Matrix Formalism: The process is described by a positive operator $W$. The probability of outcomes is given by the generalized Born rule: $p = \text{Tr}[W T]$, where $T$ is the probe operator.
• Informational Completeness: To reconstruct $W$, the set of probes $\{T_a\}$ must span the full operator space $L(\bigotimes_i H_i^I \otimes H_i^O)$.
• The Construction: The authors prove that by varying the joint unitaries and applying specific phase filters (linear combinations of different measurement outcomes), one can isolate specific product terms in the Choi representation of the probes.

3. Key Contributions and Theorems

Theorem 1: Single-Lab Completeness with One Qubit

The authors first establish that for a single intermediate lab acting on a $d$-dimensional system, a single qubit ancilla is sufficient to generate an informationally complete set of operations.

• Mechanism: By preparing the ancilla in $|0\rangle$, applying a joint unitary, and measuring in the computational basis, one can induce Kraus operators of the form $K = |\psi\rangle\langle a^*|$.
• Result: The linear span of these "measure-and-prepare" operations covers the full operator space ($d^4$ dimensions), proving that a $d$-dimensional ancilla is not required; a single qubit suffices.

Theorem 2: Multi-Lab Completeness with One Qubit (Main Result)

The central contribution is the proof that a single coherent qubit ancilla is sufficient to implement an informationally complete set of probes for arbitrary numbers of intermediate labs and arbitrary system dimensions.

• Constraint: The probes are constrained to have a Matrix Product Operator (MPO) structure with a fixed bond dimension (determined by the ancilla dimension).
• Proof Strategy:
  1. The correlated probe $T$ is expressed as a sum of product operators involving blocks of the joint unitaries ($K_{mn}$).
  2. By inserting phase gates ($R_\theta$) between interactions and taking specific linear combinations (filtering), the authors isolate specific terms where the ancilla state flips (e.g., $0 \to 1$ then $1 \to 0$).
  3. They demonstrate that by choosing appropriate unitaries, these isolated terms can generate any basis element of the local operator space (specifically, tensor products of generalized Pauli operators).
  4. Consequently, the linear span of the generated probes covers the entire multi-time operator space, ensuring informational completeness.

4. Results

• Resource Efficiency: The protocol reduces the ancillary requirement from potentially $N$ independent ancillas (or $d$-dimensional ancillas per step) to one single qubit for the entire process, regardless of the number of time steps ($N$) or system dimension ($d$).
• No Mid-Circuit Overhead: The method eliminates the need for mid-circuit measurements, feed-forward, and ancilla resets, which are major sources of noise and latency in current hardware.
• Full Reconstruction: The generated probes allow for the unique reconstruction of the full multi-time process matrix $W$, enabling the characterization of arbitrary temporal correlations.

5. Significance and Implications

• Feasible Multi-Time Tomography: This work provides a practical, experimentally viable route to complete multi-time process tomography on near-term quantum devices (e.g., superconducting qubits, trapped ions) that lack robust mid-circuit measurement capabilities.
• Beyond Tomography: Since the probes span the full operator space, the protocol allows for the evaluation of any linear functional of the process. This includes:
  • Memory Witnesses: Detecting non-Markovianity.
  • Control Objectives: Optimizing control strategies without full state reconstruction.
  • Verification: Verifying complex multi-time dynamics.
• Analogy to Circuit Cutting: The approach is described as a temporal analogue of "circuit cutting and stitching," where complex multi-time observables are inferred from simpler, experimentally friendly probe statistics via classical post-processing.
• Future Directions: While informational completeness is achieved, the authors note that the statistical performance (conditioning and noise robustness) of this specific probe family in finite-shot experiments requires further study. They also highlight that while unitary-only probes are insufficient for full tomography, they may still efficiently estimate specific physical quantities within their subspace.

In summary, the paper demonstrates that one qubit of coherent memory is the minimal and sufficient resource to fully characterize arbitrary multi-time quantum dynamics, bypassing the technological bottlenecks of mid-circuit measurements.