Variational Boundary Fluctuations as a First-Principles Origin of Langevin Noise

This paper proposes a first-principles derivation of Langevin noise by demonstrating that fluctuations in the boundary data of Hamilton's principle induce state-dependent, multiplicative stochastic forces through the gradient of the on-shell action, with homogeneous additive noise emerging only as a specific Markovian limit.

The Gist

The Big Idea: Where Does Randomness Come From?

Usually, when scientists talk about randomness (or "noise") in physics—like a pollen grain jittering in water—they assume it comes from the environment. Imagine a billiard ball being hit by invisible, tiny molecules. The standard way to explain this is to say, "We can't track every single molecule, so we just pretend there is a random force pushing the ball around."

This paper proposes a different origin. It suggests that randomness doesn't necessarily come from a chaotic environment pushing on the object. Instead, it can come from imperfect starting and ending points in the laws of motion themselves.

Think of it like this: If you try to draw a perfect line from point A to point B, but your hand shakes slightly at the very beginning or the very end, the whole line you draw will be slightly different. This paper argues that this "shaky hand" at the boundaries is enough to create the appearance of random noise in the middle of the journey, even if the journey itself follows strict, deterministic rules.

---

The Core Mechanism: The "Shaky Hand" Analogy

1. The Perfect vs. The Real

In classical physics (Hamilton's Principle), we usually imagine a particle traveling from a start point to an end point with perfectly fixed coordinates. It's like aiming a laser pointer at a specific dot on a wall. The path the laser takes is the most efficient, "perfect" path.

However, in the real world, we can never be 100% precise. Maybe the laser pointer wobbles slightly when you turn it on (the start), or your hand trembles when you stop it (the finish). The paper calls these "fluctuating endpoint data."

2. The Ripple Effect

The authors show that if you wiggle the start or end point just a tiny bit, it doesn't just change the start or end; it changes the entire path the particle takes.

• The Analogy: Imagine you are rolling a marble down a smooth, curved hill.
  • Scenario A (Fixed): You place the marble exactly at the top of the hill. It rolls down a specific, predictable line.
  • Scenario B (Fluctuating): You place the marble slightly to the left or right of the top, or you stop it slightly early or late. Because the hill is curved, that tiny shift at the start changes the speed and direction of the marble all the way down the hill.

The paper calculates exactly how that tiny "wobble" at the edge gets transported down the hill.

3. The "Ghost Force"

Here is the magic part: When you look at the marble's motion from the perspective of someone who doesn't know about the wobble at the start, it looks like the marble is being pushed by a mysterious, random force.

The paper proves that this "random force" (which physicists call Langevin noise) is actually just the gradient (the slope) of the change in the "action" (a measure of the path's efficiency) caused by the wobble.

• Simple Translation: The "random push" isn't a new thing being added to the system. It is the mathematical shadow of the uncertainty at the starting line.

---

Key Findings in Plain English

1. The Noise is "Multiplicative" (It Depends on Where You Are)

In many simple models, random noise is treated like rain falling evenly everywhere (additive noise). If you are at the top of the hill or the bottom, the rain is the same.

This paper says: No, the noise depends on where you are.

• The Analogy: Imagine the "wobble" at the start is like a ripple in a pond. If you are standing in deep water, the ripple moves slowly. If you are in shallow water, the ripple crashes and changes shape.
• The Result: The "random force" the particle feels changes depending on the particle's current position and speed. The paper calls this state-dependent noise. The shape of the "hill" (the physics of the system) filters the noise.

2. The "Filter" (The Hessian)

The paper introduces a mathematical tool called the Hessian. You can think of this as the curvature of the path.

• If the path is very curved (like a sharp turn), a tiny wobble at the start gets amplified into a big change in direction.
• If the path is flat, the wobble doesn't change much.
• Conclusion: The system acts like a filter. It takes the raw "wobble" at the boundary and shapes it into a specific type of noise based on the geometry of the path.

3. When Does It Look Like Standard Randomness?

The paper admits that sometimes, if you look at the motion over a long time and "blur" the details (a process called coarse-graining), this complex, position-dependent noise looks like the simple, uniform rain we usually assume.

• The Catch: This only happens if you ignore the fine details. If you look closely, the noise is never truly uniform; it is always tied to the shape of the path.

---

A Concrete Example: The Spring

The authors tested this idea using a simple spring (a harmonic oscillator).

• Standard View: A spring bouncing up and down with random jitters.
• This Paper's View: The jitters come from the fact that we didn't pull the spring back to exactly the same spot every time we started the experiment.
• The Result: Even for a simple spring, the "random force" isn't just a constant push. It has two parts:
  1. A part related to where the spring is (the position).
  2. A part related to how fast the "wobble" at the start was changing (the speed of the error).

Summary

This paper flips the script on how we think about randomness in physics.

• Old View: The environment is messy, so we add random forces to our equations.
• New View (from this paper): The laws of motion are perfect, but our boundaries (start and end points) are fuzzy. That fuzziness travels through the system, creating an effective random force that looks like noise but is actually a geometric consequence of imperfect boundaries.

It suggests that what we call "noise" might just be the universe's way of telling us that we can never pin down the exact start and finish of a process.

Technical Summary

Technical Summary: Variational Boundary Fluctuations as a First-Principles Origin of Langevin Noise

Problem Statement
Stochastic dynamics, particularly Langevin noise, is traditionally introduced either by postulating fluctuating forces directly in equations of motion or by eliminating environmental degrees of freedom (e.g., via projection-operator methods, influence functionals, or Brownian bath models). In these standard frameworks, randomness enters the bulk dynamics, the reservoir, or the underlying evolution law. This paper addresses a gap in the theoretical foundation: whether stochastic forcing can arise from a purely variational origin within a deterministic bulk, without assuming primitive noise or hidden reservoirs. The authors investigate how fluctuating endpoint data in Hamilton's principle—representing environmental uncertainty in preparation and termination constraints—propagates into the system's dynamics.

Methodology
The authors develop a deterministic framework where the bulk Hamiltonian remains fixed, but the boundary conditions of the variational problem are subject to fluctuations. The derivation proceeds through the following logical chain:

1. Variational Boundary Fluctuations: Instead of fixing endpoints $q(t_i)$ and $q(t_f)$, the authors consider a family of extremals with displaced endpoints $\eta_i$ and $\eta_f$. On-shell, this induces a perturbation in Hamilton's principal function $S$, denoted as $\delta S_\partial = p_f \eta_f - p_i \eta_i$.
2. Hamilton–Jacobi Propagation: This boundary perturbation is treated as a field imprint transported by the Hamilton–Jacobi flow. The authors decompose the principal function into an unperturbed part $S_0$ and a transported boundary imprint $S_\partial$.
3. Emergent Source Term: By substituting this decomposition into the Hamilton–Jacobi equation, the authors identify a residual term $\zeta = -D^{(0)}_t S_\partial$, where $D^{(0)}_t$ is the material derivative along the reference flow. This $\zeta$ represents the local failure of the boundary imprint to be freely advected.
4. Force Derivation: The effective Langevin force $\xi$ is defined as the gradient of this residual, $\xi = \nabla \zeta$. This establishes that the stochastic force is not an external postulate but the gradient of a transported action perturbation.
5. Geometric Filtering: For mechanical Hamiltonians, the relationship between the boundary displacement and the resulting force is mediated by the Hessian of the principal function, $\nabla^2 S$. This acts as a geometric filter, converting boundary uncertainty into momentum uncertainty.

Key Contributions and Results

• Derivation of State-Dependent Noise: The paper derives an explicit expression for the emergent force (Eq. 16):
  $$\xi_i = -S_{ik} \dot{\eta}_k - (D_t S_{ik})\eta_k - (\partial_i v_j)S_{jk}\eta_k$$
  where $S_{ik} = \partial_i \partial_k S$. This result demonstrates that the inherited noise is inherently multiplicative, anisotropic, and state-dependent, filtered by the geometry of the action field (the Hessian).
• Non-Equivalence to Additive Noise: The authors prove that this variational origin is not equivalent to standard homogeneous additive Langevin noise. While a phenomenological covariance can be matched locally, the global structure is constrained by the Hessian modulation. Homogeneous additive forcing is recovered only as a Markovian coarse-grained limit where the boundary fluctuations are white noise and the Hessian is effectively uniform.
• Harmonic Oscillator Benchmark: Applying the theory to a harmonic oscillator ($V = kq^2/2$), the authors show that even in this simple case, the force is not generic additive noise. It separates into a positional sector ($k\eta$) and a phase-inertial sector ($-A\dot{\eta}$), where $A(t)$ is the local curvature of the principal function. Consequently, even with short-correlated boundary fluctuations, the resulting force is generally colored, becoming white only after Markovian coarse graining.
• Overdamped and Field Extensions: The mechanism persists in the overdamped limit ($m\ddot{q} \ll \gamma\dot{q}$), retaining the Hessian filtering. The authors also outline a natural extension to field theories, where fluctuating boundary data on spatial boundaries or Cauchy hypersurfaces induce stochastic force densities via functional differentiation of the transported action source.

Significance and Claims
The paper claims to provide a variational route to stochastic dynamics that operates beyond phenomenological noise models. Its primary significance lies in identifying a first-principles origin for Langevin forcing that does not require postulating random forces in the bulk or integrating out environmental degrees of freedom. Instead, stochasticity emerges strictly from the uncertainty in the variational boundary data, propagated through the deterministic Hamilton–Jacobi flow.

The authors emphasize that the resulting force is a "causal object" derived from the geometry of Hamilton's principal function. This perspective reframes stochastic forcing not as a primitive element of the equations of motion, but as a consequence of how boundary variations imprint upon and propagate through the action field. The work suggests that standard Langevin equations with additive, state-independent noise are a specific, coarse-grained approximation of a more general, geometrically filtered variational process.