KPZ Universality Class
Overview
The Kardar-Parisi-Zhang (KPZ) universality class describes non-equilibrium stochastic growth of interfaces. It is fundamentally different from equilibrium critical phenomena: there is no partition function, no free energy, and the relevant exponents characterise dynamic scaling of a growing surface rather than static thermodynamic singularities.
The KPZ equation in $d$ spatial dimensions is
\[\frac{\partial h}{\partial t} = \nu_0 \nabla^2 h + \frac{\lambda}{2}(\nabla h)^2 + \eta(\mathbf{x}, t)\]
where $h(\mathbf{x}, t)$ is the interface height, $\nu_0$ is a smoothing coefficient, $\lambda$ is the non-linear growth coupling, and $\eta$ is Gaussian white noise with $\langle\eta(\mathbf{x},t)\eta(\mathbf{x}',t')\rangle = 2D\,\delta^d(\mathbf{x}-\mathbf{x}')\delta(t-t')$.
Systems in this class: ballistic deposition, Eden growth, polynuclear growth, directed polymers in random media, TASEP (totally asymmetric simple exclusion process).
$1+1$ Dimensions –- Exact Exponents
In $1+1$D (one spatial + one temporal dimension), the KPZ exponents are known exactly.
Growth Exponents
| Exponent | Value | Definition | Reference |
|---|---|---|---|
| $\beta$ | $1/3$ | Growth exponent: $W(t) \sim t^\beta$ at early times | KPZ (1986) |
| $\alpha$ | $1/2$ | Roughness exponent: $W_{\mathrm{sat}} \sim L^\alpha$ | KPZ (1986) |
| $z$ | $3/2$ | Dynamic exponent: $t_{\times} \sim L^z$ | KPZ (1986) |
Here $W(t) = \sqrt{\langle(h - \langle h\rangle)^2\rangle}$ is the interface width (roughness).
These are not the standard $\alpha, \beta$ of thermal phase transitions. The KPZ $\beta$ is the growth exponent (width vs time), and $\alpha$ is the roughness exponent (saturation width vs system size). Do not confuse with order-parameter or specific-heat exponents.
Galilean Invariance Constraint
The non-linear term $(\nabla h)^2$ endows the KPZ equation with Galilean invariance under tilted-frame transformations. This symmetry enforces the exact relation
\[\alpha + z = 2\]
which, combined with the scaling relation $z = \alpha / \beta$, fixes all three exponents from a single one:
\[\alpha = 1/2, \quad z = 3/2, \quad \beta = \alpha/z = 1/3\]
Exact Distribution
Beyond the exponents, the full probability distribution of the height fluctuations is known exactly in $1+1$D:
- Flat initial condition: $\chi \sim t^{1/3} \xi_{\mathrm{GOE}}$ (Tracy-Widom GOE distribution)
- Curved initial condition: $\chi \sim t^{1/3} \xi_{\mathrm{GUE}}$ (Tracy-Widom GUE distribution)
This was proven rigorously via the connection to the TASEP and random matrix theory (Sasamoto-Spohn 2010, Amir-Corwin-Quastel 2011).
Higher Dimensions –- Numerical Estimates
For $d \geq 2$ spatial dimensions, no exact solution is known. QAtlas returns the best-published numerical estimates with their quoted uncertainties.
$2+1$D –- Pagnani & Parisi (2015)
| Exponent | Value | Statistical $1\sigma$ | Source |
|---|---|---|---|
| $\beta$ | $0.2415$ | $\pm 0.0015$ | Pagnani–Parisi 2015 |
| $\alpha$ | $0.393$ | $\pm 0.005$ | Pagnani–Parisi 2015 |
| $z$ | $1.613$ | $\pm 0.009$ | Pagnani–Parisi 2015 |
Galilean invariance $\alpha + z = 2$ is satisfied within combined error bars: $0.393 + 1.613 = 2.006 \pm 0.014$.
$3+1$D –- Kelling & Ódor (2011)
| Exponent | Value | Estimated $1\sigma$ | Source |
|---|---|---|---|
| $\beta$ | $0.18$ | $\sim 0.01$ | Kelling–Ódor 2011 |
| $\alpha$ | $0.31$ | $\sim 0.01$ | Kelling–Ódor 2011 |
| $z$ | $1.51$ | $\sim 0.01$ | Kelling–Ódor 2011 |
The numerical estimates above give $\alpha + z \approx 1.82$, well below the symmetry-required value of $2.0$. This is a known open issue in the d≥3 KPZ literature, not a violation of Galilean invariance — it reflects the difficulty of extracting $\alpha$ and $z$ from finite-size simulations far from the d=1 fixed point. Treat the d=3 entry as a best-numerical pointer, not as a sharp reference value.
The upper critical dimension of KPZ (if it exists) remains an open problem; QAtlas does not expose values for $d \geq 4$.
QAtlas API
using QAtlas
# 1+1D KPZ: exact growth exponents (Rational{Int})
g1 = QAtlas.fetch(Universality(:KPZ), GrowthExponents(); d=1)
# (β_growth = 1//3, α_rough = 1//2, z = 3//2)
# 2+1D KPZ: numerical estimates with 1σ companions
g2 = QAtlas.fetch(Universality(:KPZ), GrowthExponents(); d=2)
# (β_growth = 0.2415, β_growth_err = 0.0015,
# α_rough = 0.393, α_rough_err = 0.005,
# z = 1.613, z_err = 0.009)
# 3+1D KPZ
g3 = QAtlas.fetch(Universality(:KPZ), GrowthExponents(); d=3)
# (β_growth = 0.18, α_rough = 0.31, z = 1.51, … _err = 0.01 each)
# d ≥ 4 raises an ErrorException.The _err fields are absent at $d=1$ (exact) and present at $d=2,3$ (numerical). Use haskey(g, :β_growth_err) to branch on exact-vs-numerical if needed.
Note the use of GrowthExponents() rather than CriticalExponents() to reflect the non-equilibrium nature of the KPZ class.
References
- M. Kardar, G. Parisi, Y.-C. Zhang, "Dynamic scaling of growing interfaces", Phys. Rev. Lett. 56, 889 (1986) –- original KPZ equation and exponent prediction.
- M. Prähofer, H. Spohn, "Universal distributions for growth processes in 1+1 dimensions and random matrices", Phys. Rev. Lett. 84, 4882 (2000) –- d=1 exact distribution.
- T. Sasamoto, H. Spohn, "One-dimensional Kardar-Parisi-Zhang equation: an exact solution and its universality", Phys. Rev. Lett. 104, 230602 (2010) –- exact height distribution.
- G. Amir, I. Corwin, J. Quastel, "Probability distribution of the free energy of the continuum directed random polymer in 1 + 1 dimensions", Comm. Pure Appl. Math. 64, 466 (2011).
- I. Corwin, "The Kardar-Parisi-Zhang equation and universality class", Random Matrices Theory Appl. 1, 1130001 (2012) –- review.
- A. Pagnani, G. Parisi, "Numerical estimate of the Kardar-Parisi-Zhang universality class in (2+1) dimensions", Phys. Rev. E 92, 010101(R) (2015) –- d=2 numerical estimates.
- J. Kelling, G. Ódor, "Extremely large-scale simulation of a Kardar-Parisi-Zhang model using graphics cards", Phys. Rev. E 84, 061150 (2011) –- d=3 numerical estimates.
- T. Halpin-Healy, "(2+1)-dimensional directed polymer in a random medium: scaling phenomena and universal distributions", Phys. Rev. Lett. 109, 170602 (2012) –- d=2 cross-method consistency.