On the upper critical dimension of the KPZ universality class: KPZ and related equations on a fully connected graph

By numerically and analytically studying stochastic growth equations on a fully connected graph, this paper demonstrates that the Kardar-Parisi-Zhang (KPZ) nonlinearity becomes irrelevant in the infinite-dimensional limit, causing the system to converge to Edwards-Wilkinson behavior with asymptotically flat interfaces.

The Gist

Imagine you are watching a crowd of people trying to build a sandcastle together, but instead of standing in a line or a grid, everyone is standing in a giant circle where every single person can talk to every other person instantly. There are no corners, no edges, and no "middle" of the crowd; everyone is equally connected to everyone else.

This is the setup of the paper you're asking about. The scientists are studying how a "surface" (like a pile of sand or a growing crystal) behaves when it exists in this perfectly connected, infinite-dimensional world.

Here is the breakdown of their discovery using simple analogies:

The Three Rules of the Game

The researchers tested three different "rules" for how the sand grows:

1. The Smooth Rule (EW Equation): Imagine the sand wants to be flat. If one person piles up a grain too high, surface tension (like water smoothing out a puddle) immediately pulls it down to match their neighbors. This is the "Edwards-Wilkinson" (EW) model.
2. The Bumpy Rule (KPZ Equation): This is the real-world scenario. Sand doesn't just smooth out; it also piles up based on how steep the slope is. If you pour sand on a steep slope, it slides down and piles up at the bottom, making the hill steeper. This creates a feedback loop that makes the surface very rough and bumpy. This is the famous "Kardar-Parisi-Zhang" (KPZ) model.
3. The Wild Rule (TKPZ Equation): This is the Bumpy Rule but with no smoothing at all. It's like pouring sand on a cliff with zero friction. It's chaotic and hard to simulate.

The Big Question: Does the "Bumpiness" Matter?

In normal life (on a flat 2D table or a 3D room), the "Bumpy Rule" (KPZ) creates a very specific, complex type of roughness that is different from the "Smooth Rule." Scientists have been trying to figure out: At what point does the world get so "connected" or "high-dimensional" that the bumpiness stops mattering?

They call this the Upper Critical Dimension. Think of it like a volume knob. If you turn the "dimension" knob up high enough, does the complex, bumpy behavior disappear, and does the system just act like the simple, smooth one?

The Experiment: The "All-You-Can-Connect" Graph

To test this, the scientists didn't build a physical 100-dimensional room (which is impossible). Instead, they used a Complete Graph.

• The Analogy: Imagine a party where everyone shakes hands with everyone else. In a normal room, you only shake hands with the people next to you. In this "Complete Graph" party, you are connected to everyone.
• The Result: In physics, this setup is mathematically equivalent to having infinite dimensions.

What They Found

They ran massive computer simulations of these growing surfaces on this "perfectly connected" graph. Here is what happened:

1. The Smooth Rule (EW):
As expected, the surface stayed flat. As the number of people ($N$) in the graph got huge, the roughness vanished. It was like a calm lake. The math predicted this, and the computer confirmed it.

2. The Bumpy Rule (KPZ):
This was the surprise.

• Small Groups: When they simulated a small group (say, 100 people), the surface looked bumpy and chaotic, just like the "Bumpy Rule" predicts.
• Huge Groups: As they increased the group size to 10,000 or 100,000, something magical happened. The bumpiness disappeared. The complex, non-linear feedback loop that usually makes the surface rough simply stopped working.
• The Metaphor: Imagine a rumor spreading in a small town. It gets wild and distorted (bumpy). But if you try to spread that same rumor in a city where everyone talks to everyone else instantly, the noise averages out, and the rumor becomes boring and flat. The "non-linearity" (the wild part) became irrelevant.

3. The Wild Rule (TKPZ):
This one was tricky. Because there was no smoothing, the numbers in the computer simulation tried to explode (get infinitely large). The scientists had to use a "safety valve" (a control function) to stop the numbers from crashing the computer. They found that this safety valve accidentally turned the chaotic system into a simple "Random Deposition" system (just dropping sand randomly), which is a different kind of simplicity.

The Conclusion: The "Flat" Universe

The main takeaway is this: On a fully connected graph (which represents an infinite-dimensional world), the complex, bumpy behavior of the KPZ equation vanishes.

No matter how strong the "bumpiness" force is, if the system is connected enough, it behaves exactly like the simple, smooth rule. The interface becomes perfectly flat.

Why does this matter?
It suggests that the "Upper Critical Dimension" for the KPZ universality class might be infinite. In other words, you might need to go to a dimension higher than our universe can even imagine before the "bumpy" physics stops being special and just becomes "smooth" physics.

Summary in One Sentence

The scientists found that if you connect every point in a system to every other point, the complex, chaotic growth of a surface smooths itself out, proving that in an infinitely connected world, the "bumps" don't matter anymore.

Technical Summary

1. Problem Statement and Motivation

The Kardar–Parisi–Zhang (KPZ) equation is a fundamental model for non-equilibrium interface growth, characterized by a nonlinear gradient term $(\nabla h)^2$. A central unresolved issue in the field is the existence and precise value of the upper critical dimension ($d_u$) for the KPZ universality class.

• Current Knowledge: For the Edwards–Wilkinson (EW) equation (the linear limit of KPZ), $d_u = 2$. For KPZ, exact results exist only for $d=1$. Numerical simulations up to $d=15$ have not observed a crossover to EW behavior, suggesting $d_u$ might be very large or infinite. However, theoretical predictions vary widely ($d_u < 4$, $d_u > 4$, or $d_u = \infty$).
• The Gap: Previous attempts to study the infinite-dimensional limit using Bethe lattices or Cayley trees were flawed because these structures are dominated by boundary effects and lack topological homogeneity.
• Objective: This paper investigates the infinite-dimensional limit by simulating the EW, KPZ, and tensionless KPZ (TKPZ) equations on a fully connected graph (complete graph). This topology eliminates boundary effects and serves as an effective mean-field representation of infinite dimensions ($d \to \infty$).

2. Methodology

2.1 Model and Topology

• Substrate: A complete graph with $N$ nodes where every node is connected to every other node. All nodes are topologically equivalent, ensuring no distinction between surface and bulk.
• Equations:
  • EW: $\partial_t h = \nu \nabla^2 h + \eta$
  • KPZ: $\partial_t h = \nu \nabla^2 h + \frac{\lambda}{2}(\nabla h)^2 + \eta$
  • TKPZ: $\partial_t h = \frac{\lambda}{2}(\nabla h)^2 + \eta$ (where surface tension $\nu = 0$)
• Discretization: The Laplacian and gradient squared operators are adapted for the complete graph topology:
  $$\nabla^2 h_i = \sum_{j \neq i} (h_j - h_i), \quad (\nabla h_i)^2 = \sum_{j \neq i} (h_j - h_i)^2$$
  This leads to a discretized evolution equation where the nonlinear term scales with $N$.

2.2 Numerical Integration Schemes

The authors employed two integration methods to address the numerical instability inherent in the nonlinear KPZ term:

1. Standard (ST) Scheme: An explicit Euler method. It is formally unstable for any time step $\Delta t$ due to the quadratic nonlinearity but can remain stable for sufficiently small $\Delta t$ and large $N$.
2. Controlled Instability (CI) Scheme: A regularization method where the nonlinear term $(\nabla h)^2$ is replaced by a control function $f(x) = (1 - e^{-cx})/c$. This caps the gradient to prevent numerical overflow. The authors carefully analyzed that this modification should not alter hydrodynamic scaling if the control is weak, but found it can distort results if the nonlinearity is strong.

2.3 Observables

The study analyzed four key observables to characterize the universality class:

1. Global Roughness ($w(t)$): The second moment of height fluctuations.
2. Height Fluctuation Statistics: Probability density functions (PDF), skewness ($S$), and kurtosis ($K$) of rescaled fluctuations $\chi = u/w$.
3. Time Power Spectrum ($S(\omega)$): To extract dynamic exponents.
4. Two-Time Autocorrelation ($C_t(t, t_0)$): To investigate aging behavior and memory effects.

3. Key Results

3.1 Edwards–Wilkinson (EW) Equation

• Analytical Solution: Due to linearity and graph symmetry, the authors derived an exact expression for the roughness:
  $$w^2(t) = \frac{D(N-1)}{\nu N^2} (1 - e^{-2\nu N t})$$
• Scaling: As $N \to \infty$, $w^2(t) \to 0$. The interface becomes asymptotically flat, consistent with $d > d_u^{EW} = 2$.
• Statistics: Rescaled height fluctuations follow a Gaussian distribution.
• Aging: The system exhibits trivial aging. The two-time correlation decays exponentially with a time scale $\tau \sim 1/(\nu N)$, and there is no scale-free dependence on $t/t_0$ typical of low-dimensional critical systems.

3.2 Kardar–Parisi–Zhang (KPZ) Equation

• Numerical Stability: Strong nonlinearity ($\lambda$) and moderate $N$ cause numerical instabilities. The CI scheme is necessary to prevent overflow but can distort scaling if the control function is active too frequently.
• Convergence to EW:
  • Roughness: For large $N$, the roughness evolution matches the analytical EW prediction exactly.
  • Fluctuations: As $N$ increases, the distribution of rescaled fluctuations $\chi$ converges from a slightly non-Gaussian shape (reminiscent of Tracy-Widom distributions) to a Gaussian distribution. Skewness and kurtosis approach Gaussian values ($S \to 0, K \to 3$).
  • Power Spectrum: The spectrum scales as $S(\omega) \sim \omega^{-2}$, identical to the EW case (implying a hyperscaling relation $2\alpha + d = z$ rather than the KPZ Galilean relation $\alpha + z = 2$).
  • Aging: The system exhibits trivial aging identical to the EW case, with rapid decorrelation and no scale-free aging.
• Conclusion: In the thermodynamic limit ($N \to \infty$) on a complete graph, the KPZ nonlinearity becomes irrelevant. The system flows to the EW fixed point, resulting in a flat interface.

3.3 Tensionless KPZ (TKPZ) Equation

• Behavior: Without surface tension ($\nu=0$), gradients grow rapidly. The CI control function caps these gradients, effectively turning the nonlinear term into a constant drift $F = \lambda/(2c)$.
• Result: The system reduces to Random Deposition (RD) dynamics plus a constant drift. The roughness grows as $w^2(t) \sim t$, and fluctuations are Gaussian. This behavior is an artifact of the numerical regularization required for $\nu=0$ on this topology, not a distinct physical universality class in this limit.

4. Significance and Implications

• Upper Critical Dimension: The results strongly suggest that the upper critical dimension of the KPZ universality class is infinite ($d_u = \infty$). On a fully connected graph (representing $d \to \infty$), the nonlinear term is irrelevant, and the system behaves as a linear Gaussian process (EW).
• Mean-Field Interpretation: The complete graph acts as a perfect mean-field substrate. The absence of a KPZ phase in this limit implies that the fixed points associated with the KPZ universality class (strong coupling) and the roughening transition may flow to infinity as dimension increases, leaving only the Gaussian (EW) fixed point as the attractor.
• Methodological Insight: The paper highlights the critical importance of distinguishing between physical scaling and numerical artifacts. The CI scheme, while useful for stability, can alter the effective dynamics if not used carefully. The authors demonstrate that true asymptotic behavior is only recovered when numerical instabilities are suppressed by large system sizes and small time steps, allowing the ST scheme to yield reliable results.
• Resolution of Conflicts: The findings reconcile the lack of KPZ scaling in high-dimensional simulations (up to $d=15$) with theoretical expectations, suggesting that the KPZ regime may simply not exist or be unreachable in the infinite-dimensional limit.

In summary, the paper provides robust numerical and analytical evidence that on a fully connected graph, the KPZ equation loses its non-equilibrium roughening properties and converges to the linear Edwards–Wilkinson universality class, implying an infinite upper critical dimension for the KPZ class.