Non-equilibrium generalized Langevin equation for multi-dimensional observables

This paper derives a non-equilibrium generalized Langevin equation for multi-dimensional observables using the Mori-Zwanzig formalism, revealing a unique instantaneous friction contribution that vanishes only for uncorrelated components, and demonstrates its application to modeling coupled protein folding kinetics in human islet amyloid polypeptide fibril formation.

The Gist

Imagine you are trying to predict the path of a single dancer in a crowded, chaotic ballroom. If you try to track every single person in the room (the "microscopic" view), the math becomes impossible. So, instead, you decide to track just the dancer's average position and speed. This is what scientists call a "coarse-grained" view.

However, there's a catch: the dancer isn't moving in a vacuum. They are bumping into others, getting pushed by the music, and reacting to the crowd's mood. To predict their next move, you can't just look at where they are now; you have to remember where they were a second ago, and how the crowd reacted to them back then.

This paper is about creating a new, super-precise mathematical rulebook (a Generalized Langevin Equation, or GLE) to describe exactly how that dancer moves when the ballroom itself is changing (a non-equilibrium system) and when you are tracking multiple dancers at once (a multi-dimensional system).

Here is the breakdown of their discovery using simple analogies:

1. The "Memory" of the Crowd (Non-Markovian Friction)

In simple physics, if you push a ball through honey, the honey slows it down immediately. But in complex systems (like proteins folding), the "honey" has a memory. If you push the dancer today, the crowd might still be reacting to a push from five seconds ago.

• The Analogy: Imagine the dancer is walking through a room full of people holding elastic bands attached to their waists. If the dancer moves, the bands stretch. But the bands don't snap back instantly; they wiggle and pull for a while. This "wiggling pull" is the memory kernel. The paper provides the exact formula to calculate how strong that pull is based on the history of the dancer's movement.

2. The "Instantaneous" Drag (Markovian Friction)

Usually, scientists thought that if you look at a system at a specific moment, the drag (friction) only depends on the current speed. But this paper found something surprising: If your dancers are linked together, they create an instant drag on each other.

• The Analogy: Imagine two dancers holding hands. If one spins, the other feels an immediate tug. The paper proves that if you are tracking two linked variables (like the shape of a protein and its distance to another protein), they create a "friction force" that acts right now, not just in the future.
• The Big Surprise: The authors found that this "instant drag" only exists if the variables are connected. If the two dancers are completely unrelated (uncorrelated), this instant drag disappears. It's like realizing that the only reason you feel a sudden tug is because you are holding hands with someone else.

3. The "External DJ" (Time-Dependent Hamiltonian)

Most old rulebooks assumed the ballroom music (the energy of the system) never changed. But in real life, the DJ changes the beat, or the temperature changes, or a chemical reaction starts.

• The Analogy: The paper accounts for a DJ who changes the music while the dance is happening. This changes how the dancers move and how the crowd reacts. The new equation includes a special term to handle these sudden changes in the "rules of the game."

4. The Real-World Test: The Amyloid Fibril

To prove their math works, the authors looked at IAPP, a protein involved in Type 2 diabetes. When this protein misfolds, it clumps together into sticky fibers (fibrils) that damage cells.

• The Scenario: They tracked two things at once:
  1. How much the protein is folding inside itself (Intra-layer).
  2. How far apart the layers of the fiber are (Inter-layer).
• The Result: They found that these two processes are linked, but in a specific way. The "instant drag" between them was zero because, statistically, their movements were uncorrelated in the long run. However, the "memory" (the elastic bands from the crowd) was still there. This allowed them to model the protein's behavior perfectly using their new, simplified equation.

Why Does This Matter?

Before this paper, scientists had to guess how to simplify complex systems. They often made assumptions that weren't true.

• The Takeaway: This paper gives us a "universal translator" for complex systems. It tells us exactly when we can ignore the "instant drag" and when we must include it. It helps us understand how biological machines (like proteins) work, how chemicals react, and how energy moves through complex materials, all by accounting for the fact that everything is connected to everything else, and the past always influences the present.

In short: If you want to predict the future of a complex system, you can't just look at the present. You have to know who your friends are (correlations), what the DJ is doing (time-dependence), and how the crowd remembers your past moves (memory). This paper gives us the math to do all three at once.

Technical Summary

Here is a detailed technical summary of the paper "Non-equilibrium generalized Langevin equation for multi-dimensional observables" by Héry et al.

1. Problem Statement

The Mori-Zwanzig formalism is a rigorous theoretical framework used to derive Generalized Langevin Equations (GLEs) for coarse-grained observables in complex systems. While scalar (one-dimensional) GLEs are well-established for both equilibrium and non-equilibrium systems, multi-dimensional GLEs remain an under-explored area, particularly for non-equilibrium systems.

Many biological and chemical processes involve coupled reaction coordinates (e.g., protein folding and aggregation) that exhibit coupled kinetics. Existing frameworks often fail to systematically account for:

1. Multi-dimensionality: The coupling between multiple observables.
2. Non-equilibrium conditions: Time-dependent Hamiltonians (external driving forces).
3. Instantaneous Friction: The specific contribution of Markovian friction arising from correlations between different components of a multi-dimensional observable.

The authors aim to derive a rigorous, non-approximate, multi-dimensional non-equilibrium GLE (md-neq Mori GLE) and analyze its structure under various limiting conditions to understand how friction and memory kernels emerge from microscopic dynamics.

2. Methodology

The authors employ the Mori-Zwanzig projection operator formalism combined with time-dependent evolution operators.

• System Definition: They consider a system of $N$ classical particles with a time-dependent, non-autonomous Hamiltonian:
  $$H(t, \vec{w}) = H_0(\vec{w}) - H_1(t, \vec{r})$$
  where $H_0$ is the time-independent many-body Hamiltonian and $H_1$ represents a time-dependent external perturbation.
• Liouville Dynamics: The system evolves according to the Liouville equation with a time-dependent Liouville operator $L(t) = L_0 - L_1(t)$. The probability density is propagated using a time-ordered Schrödinger-type operator.
• Observable Definition: A $d$-dimensional observable of interest $\vec{A}(t_0, t, \vec{w})$ is defined via the Heisenberg-type propagator acting on a time-independent Schrödinger observable $\vec{A}_S(\vec{r})$.
• Projection Operator: A multi-dimensional Mori projection operator ($P_M$) is constructed. This operator projects the full phase space dynamics onto the subspace spanned by the observable $\vec{A}$, its time derivative $\dot{\vec{A}}$, and the identity. The projection is defined using Gram matrices ($\hat{G}_A$ and $\hat{G}_{\dot{A}}$) to handle correlations between components.
• Derivation: Using the Dyson operator identity and the decomposition $I = P + Q$ (where $Q$ is the orthogonal complement), the authors derive an exact equation of motion for the acceleration $\ddot{\vec{A}}$.

3. Key Contributions

The paper provides the first rigorous derivation of a multi-dimensional non-equilibrium Mori GLE without ad-hoc assumptions. Key theoretical contributions include:

1. General Equation Structure: The derived equation (Eq. 28) for the $i$-th component of the observable includes:
   • An effective non-equilibrium force $D_i(t)$.
   • A linear force proportional to a time-dependent stiffness matrix $\hat{K}(t)$.
   • A Markovian friction force proportional to a time-dependent friction matrix $\hat{\gamma}(t)$.
   • Memory kernels (integrals over time) coupling to position and velocity.
   • An orthogonal (random) force $F_M$.
2. Identification of Instantaneous Friction: A critical finding is the presence of a Markovian friction term ($\hat{\gamma}(t) \cdot \dot{\vec{A}}$) that is linear in the velocity. The authors demonstrate that this term arises specifically from the correlations between different components of the multi-dimensional observable.
3. Limiting Case Analysis: The authors rigorously analyze three limiting cases to simplify the general equation:
   • Uncorrelated Limit: Components of $\vec{A}$ are uncorrelated.
   • Equilibrium Limit: The system is at equilibrium ($H_1 \to 0$).
   • Uncorrelated Equilibrium Limit: Both conditions are met.

4. Results

A. General Non-Equilibrium Case

The general md-neq GLE contains a complex interplay of forces. Notably, the Markovian friction matrix $\hat{\gamma}(t)$ is non-zero only if the components of the observable are correlated. Specifically, the off-diagonal terms of the friction matrix depend on the cross-correlations between the velocity of one component and the acceleration of another.

B. The Uncorrelated Limit

When the components of $\vec{A}$ are uncorrelated (the covariance matrix is diagonal):

• The Markovian friction matrix $\hat{\gamma}(t)$ vanishes entirely (becomes zero).
• The stiffness matrix becomes diagonal.
• Result: For uncorrelated reaction coordinates, there are no instantaneous Markovian friction forces acting on the coordinates. Friction only appears via non-Markovian memory kernels.

C. The Equilibrium Limit

When the system is at equilibrium ($H_1 \to 0$):

• The non-equilibrium force $D(t)$ and constant memory kernel $\Gamma_1$ vanish.
• The stiffness and friction matrices become time-independent.
• Crucial Finding: Even at equilibrium, if the components of $\vec{A}$ are coupled (correlated), a Markovian friction force persists. This contradicts the intuition that friction is purely non-Markovian in equilibrium; it shows that coupling induces instantaneous dissipation.

D. The Uncorrelated Equilibrium Limit

When the system is both at equilibrium and the components are uncorrelated:

• The Markovian friction matrix vanishes.
• The equation simplifies to a form with only a diagonal stiffness term and a non-Markovian memory kernel.
• Result: In this specific case, there are no Markovian friction forces at all.

E. Application to IAPP Fibril Formation

The authors apply their theory to the formation of human islet amyloid polypeptide (IAPP) fibrils, a process linked to Type 2 diabetes.

• Observables: Two coupled coordinates were chosen: $A_1$ (intra-peptide folding, H-bonds) and $A_2$ (inter-peptide distance).
• Analysis: Molecular dynamics simulations showed that the free energy landscape factorizes and the cross-correlation between $A_1$ and $A_2$ decays rapidly.
• Conclusion: The IAPP system behaves as an uncorrelated equilibrium system. Therefore, the dynamics can be modeled using the simplified GLE (Eq. 45), where dynamical coupling occurs only at the non-Markovian level (via off-diagonal terms in the memory kernel), and no instantaneous friction exists.

5. Significance

• Theoretical Clarity: The paper resolves a long-standing ambiguity regarding the origin of Markovian friction in multi-dimensional coarse-grained models. It proves that instantaneous friction is a direct consequence of correlations between reaction coordinates. If coordinates are uncorrelated, friction is purely non-Markovian (memory-dependent).
• Modeling Guidance: It provides a systematic protocol for modeling complex biological systems. By analyzing the covariance matrix of the observables, researchers can determine a priori whether their system requires a full multi-dimensional GLE with instantaneous friction or a simplified version.
• Biological Relevance: The application to IAPP demonstrates the practical utility of the formalism in distinguishing between different kinetic regimes (coupled vs. uncorrelated) in protein aggregation, offering a more accurate framework for simulating drug targets related to amyloid diseases.
• Non-Equilibrium Extension: By extending the formalism to time-dependent Hamiltonians, the work opens the door to modeling driven biological processes (e.g., active matter, force-induced unfolding) with rigorous statistical mechanics.

In summary, this work establishes that friction in multi-dimensional systems is not an intrinsic property of the coordinates themselves but emerges from their statistical coupling, fundamentally changing how coarse-grained kinetic models should be constructed for complex, non-equilibrium systems.