Divergence-free drifts decrease concentration

This paper demonstrates that bounded divergence-free drifts in the advection-diffusion equation on $\mathbb{R}^d$ reduce solution concentration—manifested as increased variance, entropy, and decreased $L^p$ norms compared to the heat equation—though this effect does not hold on the torus $\mathbb{T}^d$.

The Gist

Here is an explanation of the paper "Divergence-free drifts decrease concentration," translated into everyday language with analogies.

The Big Picture: The "Stirring" Paradox

Imagine you have a cup of hot coffee (the Heat Equation). If you leave it alone, the heat spreads out evenly. The coffee gets cooler, and the "hot spot" in the middle becomes less intense. This is pure diffusion: things naturally spread out to fill the space.

Now, imagine you have a second cup of coffee, but this time you have a magical spoon that stirs the liquid perfectly. This stirring represents the Advection-Diffusion Equation. The "stirring" is the drift (the vector field $u$).

The paper asks a simple question: Does stirring make the coffee stay hotter in one spot (more concentrated), or does it help it spread out faster (less concentrated)?

Most people might guess that if you stir a drop of dye in water, you can squeeze it into a tighter knot, making it more concentrated. Or perhaps you think stirring just mixes it up.

The authors' surprising discovery: If your stirring motion is "perfect" (mathematically speaking, divergence-free, meaning it doesn't compress or expand the fluid, just moves it around), it actually makes the substance spread out faster and become less concentrated than if you just let it sit and diffuse on its own.

In short: Perfect stirring helps diffusion win.

---

Key Concepts Explained

1. The "Perfect Stirrer" (Divergence-Free Drift)

In fluid dynamics, a "divergence-free" flow is like a crowd of people moving in a room where no one is entering or leaving, and no one is being squished into a corner or stretched out. The total volume of the crowd stays the same; they just swap places.

• Analogy: Imagine a deck of cards. If you shuffle them perfectly (divergence-free), you aren't creating new cards or destroying old ones. You are just rearranging them.
• The Paper's Finding: Even though you are just rearranging the cards, this specific type of rearrangement actually helps the "heat" (or dye) spread out more than if you did nothing.

2. "Concentration" (The Modulus of Absolute Continuity)

The authors use a fancy mathematical term called the "modulus of absolute continuity" to measure how "clumped up" a substance is.

• Simple Definition: Imagine you take a snapshot of your coffee. How much of the total "heat" is contained in the smallest possible circle you can draw?
  • High Concentration: All the heat is in a tiny, burning dot.
  • Low Concentration: The heat is spread out so thinly that no small circle contains much of it.
• The Result: The paper proves that if you start with a "symmetric" drop (like a perfect sphere of dye), the version with the perfect stirrer will always be less clumpy than the version that just sits there.

3. Variance and Entropy (Spreading Out)

The paper translates this "clumpiness" into two concepts you might know:

• Variance: How far the particles have traveled from the center.
  • Analogy: If you drop a marble in a pond, variance is how wide the ripples get. The paper says the stirred marble ripples get wider faster than the unstirred one.
• Entropy: A measure of disorder or randomness.
  • Analogy: A neat stack of papers has low entropy. A messy pile has high entropy. The paper says the stirred coffee becomes more messy (higher entropy) faster than the coffee left alone.

The "But Wait..." Twist: The Torus (The Donut World)

The paper has a fascinating second act. It says, "This rule works perfectly on an infinite flat plane (like an endless ocean), but it breaks on a Torus."

• What is a Torus? Imagine a video game screen where if you walk off the right edge, you appear on the left. It's a donut shape.
• The Twist: On this donut-shaped world, you can find a stirring pattern that actually traps the substance, making it stay more concentrated than if you just let it diffuse.
• Why? On an infinite plane, the substance can run away forever. On a donut, it eventually runs into itself. The authors show that by timing the "stirring" just right, you can use the geometry of the donut to bunch the substance up, defeating the natural tendency to spread out.

Why Does This Matter?

This isn't just about coffee or dye. This math applies to:

1. Weather and Climate: How pollutants or heat move through the atmosphere.
2. Ocean Currents: How oil spills or nutrients spread in the ocean.
3. Fluid Dynamics: Understanding turbulence.

The paper gives us a fundamental rule: In an open space, if you move a fluid without compressing it, you are actually helping it mix and spread out faster than nature intended. You can't cheat the system to keep things clumped together using just "perfect" movement.

Summary in One Sentence

If you stir a fluid without squeezing it (divergence-free), you actually make it spread out faster and become less concentrated than if you let it sit still, unless you are trapped in a loop (like a donut), where you might be able to trick the system.

Technical Summary

Here is a detailed technical summary of the paper "Divergence-free drifts decrease concentration" by Hess-Childs, Raqu´epas, and Rowan.

1. Problem Statement

The paper investigates the concentration behavior of solutions to the advection-diffusion equation with a bounded, divergence-free vector field drift ($u$) compared to the solution of the pure heat equation (diffusion only).

The core question is whether the presence of a divergence-free drift can "concentrate" a solution more than pure diffusion. Specifically, the authors compare the solution $\theta$ to:
$$\partial_t \theta - \Delta \theta + u \cdot \nabla \theta = 0$$
against the solution $\phi$ to the heat equation:
$$\partial_t \phi - \Delta \phi = 0$$
Both equations share the same initial data (or comparable initial data). The authors aim to determine if the drift $u$ (where $\nabla \cdot u = 0$) causes the solution to become more concentrated (higher peaks, smaller variance) or less concentrated than the pure diffusion case.

2. Methodology

The authors employ a combination of symmetric decreasing rearrangements, operator splitting, and fluctuation-dissipation relations.

• Modulus of Absolute Continuity: Instead of directly analyzing $L^p$ norms or variance, the authors define a central object $C_\mu(\alpha)$, the modulus of absolute continuity of a measure $\mu$ with respect to the Lebesgue measure. It is defined as the supremum of the measure of any set $E$ with volume $\alpha$:
  $$C_\mu(\alpha) := \sup \{ \mu(E) : E \subseteq \mathbb{R}^d \text{ Borel}, |E| = \alpha \}$$
  They introduce a partial order $\mu \preceq \nu$ ("$\mu$ is less concentrated than $\nu$") if $C_\mu(\alpha) \leq C_\nu(\alpha)$ for all $\alpha$.

• Operator Splitting (Trotter Product Formula): The proof strategy relies on approximating the advection-diffusion operator $P^u_t$ by alternating steps of pure advection ($e^{-tu \cdot \nabla}$) and pure diffusion ($e^{t\Delta}$).

  • Advection Step: Since the drift $u$ is divergence-free, the flow map is volume-preserving. Consequently, the advection step preserves the modulus of absolute continuity ($C_{e^{-tu \cdot \nabla}f} = C_f$).
  • Diffusion Step: The authors prove that the heat semigroup is order-preserving with respect to the concentration relation $\preceq$, provided the reference measure is symmetric decreasing. This relies heavily on the Riesz rearrangement inequality.

• Approximation: The general case (bounded, measurable, time-dependent $u$) is reduced to the case of smooth, time-independent, Lipschitz vector fields and $L^2$ initial data via mollification and density arguments.

3. Key Contributions and Results

A. Main Theorem (Theorem 1.7)

The central result states that for any bounded, measurable, divergence-free vector field $u$, if the initial measure $\mu$ is "less concentrated" than a symmetric decreasing measure $\nu$ (i.e., $\mu \preceq \nu$), then the solution to the advection-diffusion equation remains less concentrated than the solution to the heat equation with initial data $\nu$:
$$P^u_t \mu \preceq e^{t\Delta} \nu \quad \forall t \geq 0$$
This implies that divergence-free drifts always decrease concentration (or at best, do not increase it) compared to pure diffusion on $\mathbb{R}^d$.

B. Corollaries on Physical Quantities

From the main theorem, the authors derive concrete inequalities for standard physical quantities (Corollary 1.10):

1. $L^p$ Norms: For all $p \in [1, \infty]$, the $L^p$ norm of the advection-diffusion solution is bounded by that of the heat equation solution:
   $$\|P^u_t \mu\|_{L^p} \leq \|e^{t\Delta} \nu\|_{L^p}$$
2. Variance: The variance of the solution is larger (more spread out) than the pure diffusion case:
   $$\text{Var}(P^u_t \mu) \geq \text{Var}(e^{t\Delta} \nu)$$
   This resolves a specific question about whether divergence-free flows can suppress variance growth; the answer is no, they enhance it relative to pure diffusion.
3. Entropy: Under mild moment conditions, the entropy of the advection-diffusion solution is greater than that of the heat equation solution, indicating a more "spread out" distribution.

C. Application to Navier-Stokes

The results are applied to the vorticity form of the 2D Navier-Stokes equation. If the initial vorticity is symmetric decreasing and of fixed sign, the $L^p$ norms of the vorticity are bounded by the heat kernel solution, and the dissipation (gradient norm) is enhanced.

D. Counterexample on the Torus (Theorem 1.14)

The authors demonstrate that the result fails on the torus $\mathbb{T}^d$. They construct a specific smooth, divergence-free vector field $u$ with compact time support such that, for large times, the $L^2$ norm of the advection-diffusion solution is strictly greater than that of the heat equation solution.

• Mechanism: On the torus, the heat kernel's level sets become non-circular at long times. The constructed flow exploits this by rotating these non-circular level sets to align with the geometry of the torus in a way that minimizes the surface area of super-level sets (via the isoperimetric inequality), thereby reducing the effectiveness of diffusion.

4. Significance and Implications

• Passive Scalar Turbulence: The paper provides a rigorous mathematical foundation for the intuition that divergence-free flows (common in fluid dynamics) act to "mix" or spread out passive scalars more effectively than diffusion alone. It confirms that such flows enhance dissipation and variance growth.
• Rigorous Comparison Principles: It establishes a sharp comparison principle for advection-diffusion equations against the heat equation, extending previous work on rearrangement inequalities to include drift terms.
• Domain Dependence: The distinction between $\mathbb{R}^d$ and $\mathbb{T}^d$ highlights the critical role of geometry and boundary conditions in transport phenomena. While drifts always spread mass on unbounded domains, bounded domains allow for specific flow configurations that can temporarily "trap" or concentrate mass relative to diffusion.
• Fluctuation-Dissipation: The results connect the variance of stochastic processes (Itô diffusions) to the dissipation of the associated PDE, reinforcing the link between enhanced dissipation and enhanced variance growth in divergence-free flows.

In summary, the paper proves that on Euclidean space, divergence-free drifts cannot concentrate a solution more than pure diffusion; they invariably lead to a "flatter," more spread-out distribution with higher variance and entropy. This behavior is unique to the unbounded domain, as the topology of the torus allows for exceptions.