Transfer-Tensor Method in Open Quantum Systems

• Transfer-Tensor Method is a systematic, data-driven approach that transforms discrete dynamical maps into convolutional propagators capturing full memory effects in open quantum systems.
• The method extracts transfer tensors from short-time dynamics, enabling efficient long-time propagation and linking discrete maps to continuous memory-kernel formalisms.
• TTM has practical applications in quantum noise spectroscopy, non-Markovianity quantification, and localization diagnostics, offering scalable simulation for complex quantum environments.

The transfer-tensor method (TTM) is a systematic, data-driven approach for the efficient simulation and analysis of non-Markovian dynamics in open quantum systems. By transforming discrete sets of dynamical maps, TTM provides an exact, algebraic convolutional propagator capturing the full memory effects of a system's interaction with its environment. TTM plays a crucial role in quantum noise spectroscopy, non-Markovianity quantification, long-time propagation based on numerically exact short-time evolution, and the diagnosis of localization and transport in disordered systems.

1. Mathematical Formulation of the Transfer-Tensor Method

TTM starts by discretizing time with uniform steps $\Delta t$, generating a series of dynamical maps. For an open quantum system initialized in $\rho(0)$, the reduced state at time $t_n = n\Delta t$ is given by

$\rho(t_n) = \mathcal{E}_n \,\rho(0)$

where $\mathcal{E}_n$ is the completely positive trace preserving (CPTP) map describing the system's evolution up to $t_n$.

The central insight of TTM is to express $\{\mathcal{E}_n\}$ as a convolution over past states with a set of time-translation-invariant superoperators $\{T_k\}$ ("transfer tensors"):

$$\mathcal{E}_n = \sum_{k=1}^n T_k \mathcal{E}_{n-k},\quad n\geq1$$

with $T_1 \equiv \mathcal{E}_1$ and for $\rho(0)$0,

$\rho(0)$1

The propagation of arbitrary states is performed via the discrete recurrence

$\rho(0)$2

where $\rho(0)$3 is the memory cutoff set by the decay of $\rho(0)$4.

In the continuous-time limit ($\rho(0)$5), the TTM reproduces the Nakajima–Zwanzig memory-kernel formalism:

$\rho(0)$6

The discrete transfer tensors relate to the memory kernel as

$\rho(0)$7

and

$\rho(0)$8

for $\rho(0)$9. At finite $t_n = n\Delta t$0, TTM is exact as constructed from the discrete maps, with the explicit connection to the continuous memory kernel holding in the $t_n = n\Delta t$1 limit (Chen et al., 2019, Morillas-Rozas et al., 9 Mar 2026, Bose, 2024).

2. Extraction and Algorithmic Implementation

TTM implementation requires access to tomographically complete or numerically exact reduced dynamics for a short initial time window. The practical extraction of transfer tensors proceeds as follows:

1. Obtain short-time dynamical maps: Use either process tomography (experiment) or numerically exact solvers such as HEOM, QuAPI, stochastic path-integral, or tensor-train/inchworm algorithms to sample $t_n = n\Delta t$2 up to the bath's memory time (Buser et al., 2017, Wang et al., 14 Jun 2025, Bose, 2024).
2. Recursively form transfer tensors: Use the backward recurrence

$t_n = n\Delta t$3

to derive $t_n = n\Delta t$4.

1. Long-time propagation: Propagate to arbitrarily long times via

$t_n = n\Delta t$5

for $t_n = n\Delta t$6.

2. Convergence and accuracy: Monitor the decay of $t_n = n\Delta t$7 and, if present, inhomogeneous terms for initial system–environment correlations. Increase $t_n = n\Delta t$8 if necessary to ensure accuracy (Buser et al., 2017, Wang et al., 14 Jun 2025).

In the context of contemporary tensor network methods, extracting TTM from tensor-train bath influence functionals and coupling to inchworm or path-integral approaches allows deterministic, linearly-scaling simulation of non-Markovian dynamics far beyond the bath memory time (Wang et al., 14 Jun 2025).

3. Application to Noise Characterization and Spectroscopy

TTM enables protocolized quantum noise spectroscopy:

• Preparation: Choose a tomographically complete set of initial states.
• Process tomography: Measure the evolved states at a series of time points, reconstruct the discrete maps $t_n = n\Delta t$9.
• Memory kernel reconstruction: Extract $\rho(t_n) = \mathcal{E}_n \,\rho(0)$0 and relate to the discrete memory kernel as above.
• Spectral density extraction: From the (second-order) memory kernel,

$\rho(t_n) = \mathcal{E}_n \,\rho(0)$1

The quantum noise power spectrum is obtained via Fourier transform:

$\rho(t_n) = \mathcal{E}_n \,\rho(0)$2

(Chen et al., 2019, Chen et al., 2020).

The Spectral Transfer Tensor Maps (SpecTTM) protocol extends TTM, enabling SPAM-free, eigenvalue-based extraction of transfer tensors, particularly for Pauli channels and their twirled approximations, with efficient RHP non-Markovianity quantification and noise-spectrum reconstruction (Chen et al., 2020).

4. Non-Markovianity Quantification and Extension

TTM provides several diagnostics for non-Markovianity:

• Tensor norm signature: In Markovian processes, $\rho(t_n) = \mathcal{E}_n \,\rho(0)$3, so nonzero higher-order tensor norms directly quantify memory effects.
• Bloch-volume dynamics: For qubits, TTM-propagated affine maps yield a volume $\rho(t_n) = \mathcal{E}_n \,\rho(0)$4 whose transient increases mark non-Markovianity (see

$\rho(t_n) = \mathcal{E}_n \,\rho(0)$5

).

• Trace distance backflow: Non-monotonic dynamics of the trace distance $\rho(t_n) = \mathcal{E}_n \,\rho(0)$6 indicates information backflow, signifying non-Markovianity.
• RHP measure: Using SpecTTM, the RHP non-Markovianity measure is computed as the area of negative decoherence rates,

$\rho(t_n) = \mathcal{E}_n \,\rho(0)$7

(Chen et al., 2019, Chen et al., 2020).

For multi-qubit and spatially correlated environments, TTM quantifies correlated noise and cross-talk via the structure of multi-qubit transfer tensors $\rho(t_n) = \mathcal{E}_n \,\rho(0)$8 (Chen et al., 2019).

5. Analytical Case Studies and Model Systems

TTM has been analytically dissected in exactly solvable scenarios. In the Jaynes–Cummings model of a two-level atom in a lossy cavity, the coherence and population sectors admit closed-form dynamical maps and transfer tensors. Critical insight is provided into "stroboscopic Markovianity," i.e., at special time steps in the underdamped regime, higher-order tensors vanish and the sampled dynamics appears CP-divisible. In contrast, in overdamped regimes, nonzero $\rho(t_n) = \mathcal{E}_n \,\rho(0)$9 persist for all finite $\mathcal{E}_n$0 (Morillas-Rozas et al., 9 Mar 2026). This demonstrates subtlety: non-Markovianity depends both on underlying dynamics and time-discretization.

In disordered many-body systems, e.g., Anderson and Aubry–André–Harper models, ensemble-averaged dynamics necessitate memory terms in the transfer tensors to avoid spurious site-sampling and distinguish static from dynamic disorder. "Eternal memory" (non-decaying $\mathcal{E}_n$1) is necessary, but not sufficient, for localization, as shown by the outgoing-pseudoflux metric extracted algebraically from TTM (Anderson et al., 22 Sep 2025).

6. Extensions, Advantages, and Limitations

Advantages

• Generality: TTM is nonperturbative and data-driven—applicable wherever the memory kernel can be sampled or reconstructed.
• Efficiency: After the initial short-time window, propagation to arbitrarily long times is inexpensive, scaling as $\mathcal{E}_n$2.
• Compatibility: TTM can be flexibly coupled atop any exact short-time solver (HEOM, QuAPI, sPI, inchworm, tensor networks).
• Lindblad inclusivity: Once memory is extracted, empirical Lindblad processes are incorporated at no extra cost (Bose, 2024, Wang et al., 14 Jun 2025).
• Spectral and thermometric diagnostics: Direct access to correlation functions, emission/absorption spectra, and even reverse estimation of inverse temperature from spectra (Buser et al., 2017).

Limitations

• Short-time cost: The expensive part is generating the short-time kernel; large systems or long bath memory can remain numerically intensive (Bose, 2024).
• Discretization sensitivity: Highly structured or oscillatory kernels may require small $\mathcal{E}_n$3 and/or large $\mathcal{E}_n$4 for convergence.
• SPAM in tomography: Full process tomography in experiment is SPAM-sensitive; SpecTTM protocol mitigates this for Pauli and related channels but is not scalable to large system Hilbert spaces (Chen et al., 2020).
• Initial correlations: For initially correlated system–environment states, extra inhomogeneous corrections are needed, but these too decay on the bath memory timescale (Buser et al., 2017).

Advances in low-rank factorizations of the transfer tensors, tensor-train compression, and adaptive memory cutoff have further improved large-system applicability (Wang et al., 14 Jun 2025, Bose, 2024).

7. Selected Applications and Benchmarks

• Quantum hardware noise benchmarking: IBM Quantum Experience devices have been analyzed with TTM, revealing mild non-Markovian dissipation and spatial correlations, with quantitative predictions of coherence decay and collective decoherence validated in both single- and two-qubit settings (Chen et al., 2019).
• Disordered system transport/localization: Identification of localization regimes via outgoing-pseudoflux and memory measures in Anderson and AAH models (Anderson et al., 22 Sep 2025).
• Molecular spectra and thermometry: TTM applied to multichromophoric systems yields efficient, accurate emission/absorption spectra and allows extraction of thermodynamic parameters from spectral data (Buser et al., 2017).
• Hybrid path-integral/Lindblad evolution: Combined TTM–PI–Lindblad approaches enable simulation of mixed Markovian/non-Markovian dissipation with minimal additional complexity, as demonstrated in photosynthetic complexes (Bose, 2024).
• Tensor-network/inchworm acceleration: Coupling TTM to tensor-train-based inchworm solvers enables deterministic, linearly scaling simulation of general open quantum systems across extended timescales (Wang et al., 14 Jun 2025).

---

By transforming numerically or experimentally accessible short-time information into a compact discrete-memory propagator, the transfer-tensor method has unified and advanced the analysis of open-system dynamics, non-Markovianity quantification, noise spectroscopy, and long-time quantum simulation across physics, chemistry, and quantum information science (Chen et al., 2019, Morillas-Rozas et al., 9 Mar 2026, Anderson et al., 22 Sep 2025, Chen et al., 2020, Buser et al., 2017, Wang et al., 14 Jun 2025, Bose, 2024).