⚛️ quantum physics

The Hidden Nature of Non-Markovianity

This paper demonstrates that under mild assumptions, the trajectories of non-Markovian evolutions can be fully reproduced by time-dependent Markovian Lindbladians, rendering non-Markovianity undetectable when analyzing single trajectories alone.

Original authors: Jihong Cai, Advith Govindarajan, Marius Junge

Published 2026-02-20

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Jihong Cai, Advith Govindarajan, Marius Junge

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: The "Magic Trick" of Quantum Memory

Imagine you are watching a movie of a quantum system (like a tiny atom or a qubit) moving through time. You see it start at point A, wiggle around, and end up at point B.

In the world of quantum physics, there are two ways a system can move:

Markovian (The "Forgetful" Walker): The system has no memory. Its next step depends only on where it is right now. It's like a drunk person stumbling; they don't remember where they were five seconds ago, only where their foot is right now.
Non-Markovian (The "Remembering" Walker): The system has memory. It can "backflow" information from the environment. It's like a person walking who remembers a path they took earlier and uses that memory to correct their steps or even retrace their path.

The Paper's Shocking Discovery:
The authors (Cai, Govindarajan, and Junge) discovered a mind-bending secret: If you only watch the movie of the walker's path (the "trajectory"), you cannot tell if they are forgetful or remembering.

You can watch a "remembering" walker take a specific path, and then watch a "forgetful" walker take the exact same path in the exact same way. To an outside observer, the two look identical. The "memory" is hidden.

The Analogy: The GPS vs. The Driver

To understand this better, let's use a GPS navigation analogy.

The Trajectory: This is the line drawn on the map showing where the car went.
The Markovian Driver: This driver follows the GPS perfectly. They only look at the current street and turn where the GPS says. They have no memory of the traffic they saw 10 minutes ago.
The Non-Markovian Driver: This driver is a genius. They remember a traffic jam from 10 minutes ago, so they take a weird detour to avoid it, or they speed up because they know a light is about to turn green.

The Paper's Finding:
The authors prove that for any specific route a "genius" driver takes (even a weird, memory-based detour), you can always invent a "dumb" GPS driver who follows the exact same route.

If you only look at the line on the map (the trajectory), you cannot know if the driver was using their memory or just following a complex set of instructions. The "memory" is invisible if you only look at the path.

Why is this a big deal?

For a long time, scientists thought: "If we see a quantum system doing something weird, like an entanglement suddenly coming back to life (a 'revival'), that proves it has memory (Non-Markovianity)."

This paper says: Not necessarily.

That "weird revival" could just be a very complex, standard (Markovian) process happening in a specific way. Unless you know the entire rulebook of how the system interacts with its environment (the "Dynamical Map"), you can't tell the difference just by watching the system move.

The "Magic" Construction (How they did it)

The authors didn't just guess this; they built a mathematical "magic trick."

Imagine you have a specific path a particle took.

The Problem: Usually, to describe a path, you need a "generator" (a rule that says how the particle moves). For "forgetful" systems, this rule must always be positive (it can't push the particle in a forbidden direction).
The Solution: The authors found a way to create a "Lindbladian Lift." Think of this as a custom-made engine for that specific path.
- If the path looks like it requires "memory" (like a negative decay rate), they construct a special, time-changing engine that mimics that behavior using only standard, "forgetful" rules.
- It's like taking a car that drives backward (which seems impossible for a standard car) and building a custom transmission that makes a standard car drive backward perfectly.

The "Exponential" Surprise

The paper goes even further. They showed that even if you watch many different paths at once (like watching 100 different cars), you still might not be able to tell if they have memory.

They proved that you could have a massive family of "remembering" paths, and a single "forgetful" engine could generate all of them simultaneously. It's as if one simple, forgetful robot could perfectly mimic the complex, memory-based behavior of a whole army of geniuses.

The "Boundary" Problem (Why it's tricky)

There is one catch. The paper mentions that this trick works best when the path is "smooth" and doesn't hit the "walls" of the quantum world (where the system becomes impossible, like having negative probability).

The Wall: Imagine a ball rolling on a table. If it hits the edge, it stops.
The Trick: If the "remembering" ball hits the edge and bounces back in a way that a "forgetful" ball physically couldn't do, then you can tell them apart.
The Reality: But for most smooth, realistic paths that don't hit these hard walls, the "forgetful" engine can mimic the "remembering" one perfectly.

The Takeaway for Everyday Life

1. You can't judge a book by its cover (or a path by its line).
Just because a system looks like it's remembering the past, it doesn't mean it actually is. It might just be following a very complicated set of "now-only" rules.

2. Memory is a global property, not a local one.
"Non-Markovianity" (memory) isn't a feature of a single path; it's a feature of the entire system and how it interacts with the universe. If you only look at a single slice of the movie, you miss the whole story.

3. Why do we care?
Even though we can't see the memory in the path, memory is still useful! It's like having a secret weapon. A "remembering" system might be able to do things (like preserve quantum information) more efficiently than a "forgetful" one, even if they end up in the same place. The "forgetful" engine might just need a much bigger, more complex machine to do the same job.

In short: The paper reveals that the "ghost" of memory in quantum systems is invisible if you only look at the footprints. To see the ghost, you have to look at the whole house, not just the path on the floor.

1. Problem Statement

The paper addresses a fundamental question in the theory of open quantum systems: Can Non-Markovianity be detected solely by observing the trajectory of a quantum state?

Context: Open quantum systems are typically modeled using the Gorini-Kossakowski-Sudarshan-Lindblad (GKSL) master equation, $\dot{\rho}_t = \mathcal{L}_t \rho_t$ , where $\mathcal{L}_t$ is a time-local Lindbladian generator. Evolutions generated by such operators are Markovian (specifically, CP-divisible) and exhibit properties like data processing inequality and monotonic decoherence.
The Challenge: Many physical systems exhibit Non-Markovian behavior (memory effects, information backflow, entanglement revival). Various definitions of Non-Markovianity exist (e.g., based on CP-divisibility, information backflow, or process tensors), but they are generally defined at the level of the dynamical map (the global evolution operator) or multi-time processes, not individual state trajectories.
The Core Question: If an experimentalist observes a single continuous path of density matrices $\rho_t$ (or even a finite ensemble of such paths), can they definitively determine whether the underlying evolution was Markovian or Non-Markovian?

2. Methodology

The authors employ a geometric and constructive approach to the Lindbladian Lifting Problem.

The Lifting Problem: Given a trajectory $\rho_t$ in the space of density matrices $\mathcal{D}(\mathcal{H})$ , does there exist a family of time-dependent Lindbladians $\{\mathcal{L}_t\}$ such that $\dot{\rho}_t = \mathcal{L}_t(\rho_t)$ ?
Geometric Analysis: The authors analyze the tangent cone of the state space. They utilize the characterization that the set of kinematically admissible velocities at a state $\rho$ is exactly the set of vectors generated by Lindbladians: $T^+_\rho \mathcal{D}(\mathcal{H}) = \{ \mathcal{L}(\rho) \mid \mathcal{L} \text{ is a Lindbladian} \}$ .
Regularity Assumptions: To ensure the existence of smooth lifts, the paper imposes mild regularity conditions on the trajectory $\rho_t$ $ρ_{t}$ :
1. The eigenspaces of $\rho_t$ depend continuously on time $t$ .
2. The set of times where the rank of $\rho_t$ changes has no accumulation points.
Constructive Proofs: The authors explicitly construct families of Lindbladians (specifically "replacer Lindbladians") that reproduce specific trajectories known to be generated by Non-Markovian dynamics. They also utilize reparametrization techniques (geometric lifts) to handle cases where strict parametric lifts might encounter singularities.

3. Key Contributions and Results

A. Theorem 1: Lindbladian Lifting (Main Result)

The paper proves that any sufficiently regular ( $C^2$ ) trajectory of density matrices $\rho_t$ can be realized as the solution to a time-local GKSL master equation with a continuous Lindbladian $\mathcal{L}_t$ .

Implication: Under mild assumptions, Non-Markovianity is invisible when observing a single trajectory. A trajectory generated by a Non-Markovian process (e.g., one with information backflow) can be perfectly reproduced by a Markovian process with a suitably engineered, time-dependent Hamiltonian and jump operators.

B. Indistinguishability of Single and Multiple Trajectories

Single Path: The authors demonstrate that a specific Non-Markovian pure dephasing trajectory (where the decay rate $\gamma(t)$ changes sign and diverges) can be exactly reproduced by a Markovian evolution using a "replacer Lindbladian" construction:
$\mathcal{L}_t = \frac{1}{\epsilon_t} (\mathcal{R}_{\rho_t + \epsilon_t \dot{\rho}_t} - \mathbb{I})$
where $\mathcal{R}_\sigma$ is the replacer channel mapping any state to $\sigma$ .
Exponential Indistinguishability: The paper extends this to $n$ -qubit systems. They construct a scenario where $2^n$ distinct trajectories (generated by a specific Non-Markovian evolution) are simultaneously reproduced by a single Markovian dynamics (a tensor product of local Lindbladians). This proves that even checking an exponentially large number of initial states is insufficient to distinguish the nature of the evolution without global information.
Discrete Sampling: Even if one only observes discrete time samples of a trajectory, the authors show that a piecewise affine interpolation of these points can always be lifted to a Markovian evolution. Thus, discrete data cannot detect Non-Markovianity.

C. Regional Indistinguishability

The authors show that a single Markovian generator can lift an entire connected region of trajectories arising from the same Non-Markovian evolution. By varying a parameter $z$ in the initial state, one obtains a family of paths, all of which can be generated by a single time-local Lindbladian.

4. Significance and Discussion

Fundamental Non-Identifiability: The paper establishes that Markovianity is not an invariant property of individual quantum trajectories. It is a property of the global dynamical map (the multi-time process), not the path itself.
Limitations of Trajectory-Based Detection: Many experimental protocols rely on monitoring single trajectories or ensembles to detect memory effects (e.g., via information backflow). This work suggests that such detection requires a priori knowledge of the evolution or optimization over initial conditions to find "witness" pairs of states. Without global constraints, the trajectory alone is ambiguous.
Geometric Origin: The indistinguishability arises from the geometry of the quantum state space. The state space has a tangent cone structure rather than a vector space structure. At the boundary, certain directions are forbidden for Markovian evolutions (to preserve positivity), but the authors show that for almost all interior paths, the "memory" can be encoded into the time-dependence of the generator rather than the path geometry.
Resource Perspective: While Non-Markovianity is "hidden" in the trajectory, the authors note it remains a valuable resource. Reproducing a Non-Markovian trajectory via a Markovian model may require significantly larger resources (e.g., larger environment registers or more complex control Hamiltonians).

5. Conclusion

The paper concludes that Non-Markovianity is fundamentally unidentifiable from trajectory data alone. Whether a system is Markovian or Non-Markovian cannot be determined by observing the path $\rho_t$ or even a finite collection of such paths. Non-Markovianity is a global property of the dynamical map, and its detection requires information beyond the instantaneous state evolution, such as the structure of the multi-time process or constraints on the system-environment interaction.