arXiv:2610.01814
Collision models: Markovian and Non-Markovian impurity models
Jacob Werner
Abstract
Collision models provide a flexible framework for studying open quantum systems and introducing non-Markovianity. We present a novel class of collision models, which we term Markovian impurity models, motivated by physical scenarios in which information backflow is periodically blocked. Using the Breuer-Laine-Piilo (BLP) measure of non-Markovianity, we show that if a $p$-impurity model is non-Markovian, then the corresponding $(p+1)$-impurity model is also non-Markovian, implying the existence of a critical $p$ at which the process transitions from Markovian to non-Markovian. Numerical simulations are used to investigate how this critical parameter depends on the system-ancilla and ancilla-ancilla interaction strengths. We find that, altough exceptions exist, increasing either interaction strength generally lowers the critical $p$, while sufficiently weak ancilla-ancilla interactions lead to a regime in which the dynamics remain Markovian even in the absence of impurities. Finally, we briefly introduce the complementary non-Markovian impurity model.
AI Summary
The paper studies how periodically interrupting environmental memory affects information backflow in collision models. It introduces a $p$-impurity model in which a system qubit sequentially interacts with thermal ancilla qubits through partial swaps, while nearest-neighbour ancilla–ancilla interactions are omitted every $p$ collisions. Thus $p=1$ gives a memoryless collision model, and $p\to\infty$ removes the interruptions. Using the discrete-time Breuer–Laine–Piilo (BLP) criterion, it proves that non-Markovianity for a given $p$ implies non-Markovianity for $p+1$. The proof combines the reset of system–incoming-ancilla correlations between blocks with the fact that an ancilla-only unitary cannot immediately change the reduced system state. Consequently, whenever the uninterrupted model exhibits backflow, there is a finite smallest impurity period $p'$ that permits it. Sampling only at block boundaries instead gives repeated CPTP maps and no trace-distance increases. Numerical calculations explore $p'$ versus system–ancilla and ancilla–ancilla interaction strengths, mainly for ancilla population parameter $a=0.3$. Stronger interactions generally lower $p'$, sometimes to $2$, but nonmonotonic exceptions occur. The results suggest a weak ancilla–ancilla coupling regime without backflow even without interruptions. The numerical BLP calculation uses only the $\sigma_x$ eigenstate pair rather than optimizing over initial states. Searches generally stop at $p=40$, with selected checks to $100$ steps, so finite large thresholds are not conclusively distinguished from absence of backflow. A complementary model with isolated memory-enabling interactions is introduced but not systematically analyzed.
Error found: The reported numerical critical spacings use only the sigma_x eigenstate pair, so without an optimality argument they do not establish the actual BLP critical spacings or certify Markovianity where that witness vanishes. The claimed impurity-free Markovian regime is supported by finite-time simulations rather than an argument excluding trace-distance revivals at arbitrarily late times.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.