k-Mode Turing Instability in Spatial Systems

• k-Mode Turing instability is a mode-resolved diffusion‐driven symmetry breaking phenomenon in which a homogeneous state loses stability to nonuniform eigenmodes.
• The methodology relies on linear stability analysis using Laplacian eigenfunctions and characteristic polynomial criteria to identify specific unstable modes.
• The framework applies broadly—from continuous reaction–diffusion systems to discrete networks, higher-order structures, and quantum models—highlighting its practical significance.

k-Mode Turing instability is the mode-resolved form of diffusion-driven symmetry breaking: a homogeneous state remains stable to spatially uniform perturbations but loses stability to one or more nonuniform eigenmodes of the Laplacian, of a network coupling operator, or of an effective averaged operator. In most of the literature, the label $k$ denotes a specific spatial Fourier or Laplacian mode whose growth rate becomes positive; in some discrete-network treatments, the same expression also denotes a regime with $k$ distinct unstable modes. The concept has been developed for continuous reaction–diffusion systems, nonlocal oscillator chains, Cartesian product and time-varying networks, heterogeneous domains, directed higher-order structures, and semiclassical limits of open quantum systems (Challenger et al., 2014, Asllani et al., 2014, Petit et al., 2017, Krause et al., 2019, Dorchain et al., 2024, Comparato et al., 8 Jul 2026).

1. Classical modal formulation

In the standard two-species reaction–diffusion setting,

$$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$

one starts from a homogeneous steady state $(\bar\phi,\bar\psi)$ satisfying $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$. The homogeneous kinetics is linearly stable when

$$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$

Expanding perturbations on Laplacian eigenfunctions $W_k$ with $-\nabla^2W_k=k^2W_k$ yields a decoupled $2\times2$ problem for each mode $k$, with linear operator

$k$0

The characteristic polynomial is

$k$1

with

$k$2

$k$3

Because the homogeneous fixed point is stable, $k$4, so a specific mode $k$5 is unstable exactly when $k$6. The classical closed Turing inequalities are

$k$7

which ensure the existence of a band of nonzero modes with positive dominant growth rate (Challenger et al., 2014).

In nonlocal discrete media the same logic persists, but the spectral parameter replacing $k$8 is set by the coupling kernel. For a one-dimensional chain with power-law coupling $k$9, the discrete Fourier modes $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$0 evolve through a mode-dependent factor

$$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$1

so that local coupling corresponds to $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$2, whereas global coupling makes all nonzero modes share essentially the same effective diffusion eigenvalue. The unstable band is therefore determined by which discrete $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$3-values place $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$4 between the roots of the mode determinant, and the sharpness of mode selection depends strongly on the coupling range (Viana et al., 2011).

2. Meanings of “k” and multimode usage

The dominant usage identifies a $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$5-mode Turing instability with instability of one particular spatial mode. In that sense, $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$6 is a spectral label: a Fourier wave number in continuum systems, a Laplacian eigenmode on a graph, or an analogous eigenvector of a coupling operator. The instability criterion is modewise, and pattern selection near onset is governed by the mode whose real growth rate is maximal.

A distinct but closely related usage appears on Cartesian product networks $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$7. There the Laplacian eigenmodes are tensor products

$$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$8

and the linearized $$\frac{\partial \phi}{\partial t}=f(\phi,\psi)+D_\phi \nabla^2\phi,\qquad \frac{\partial \psi}{\partial t}=g(\phi,\psi)+D_\psi \nabla^2\psi,$$9 block for mode $(\bar\phi,\bar\psi)$0 is controlled by

$(\bar\phi,\bar\psi)$1

The relevant characteristic polynomial has coefficients $(\bar\phi,\bar\psi)$2 and $(\bar\phi,\bar\psi)$3, and because $(\bar\phi,\bar\psi)$4, the mode $(\bar\phi,\bar\psi)$5 is unstable iff

$(\bar\phi,\bar\psi)$6

The paper then defines

$(\bar\phi,\bar\psi)$7

so a “$(\bar\phi,\bar\psi)$8-mode Turing instability” means that $(\bar\phi,\bar\psi)$9 distinct tensor-product modes are simultaneously unstable (Asllani et al., 2014).

This multimode viewpoint is pushed further in a codimension-3 framework for two-component reaction–diffusion systems on $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$0 with Neumann boundary conditions. There one can choose parameters so that the spatially homogeneous $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$1-mode undergoes a Hopf instability while the $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$2- and $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$3-modes undergo Turing instabilities simultaneously. For the analytically treated $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$4 case, center-manifold reduction yields a cubic normal form for a complex Hopf amplitude $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$5 and real Turing amplitudes $f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$6,

$f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$7

$f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$8

$f(\bar\phi,\bar\psi)=g(\bar\phi,\bar\psi)=0$9

Here “multimode Turing instability” is literal: two spatial modes are critical at once, and their quadratic and cubic interactions organize mixed patterns, invariant tori, period-doubling, and chaotic behavior (Izuhara et al., 2023).

3. Networks, averaged operators, and directed higher-order structures

On a static network, the continuum $$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$0 is replaced by Laplacian eigenvalues $$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$1. For a homogeneous equilibrium $$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$2, the linearized averaged mode equation has the standard form

$$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$3

so each graph eigenmode is treated exactly as a discrete spatial mode. In the theory of time-varying networks, this mode decomposition is recovered by averaging the periodic Laplacian,

$$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$4

A “$$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$5-mode Turing instability” is then defined with respect to an eigenvalue $$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$6 of $$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$7, via

$$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$8

The theorem of averaging implies that, for sufficiently small $$\operatorname{tr}J=f_\phi+g_\psi<0,\qquad \det J=f_\phi g_\psi-f_\psi g_\phi>0.$$9, the fast time-varying system stays $W_k$0-close to the averaged one over $W_k$1, so the unstable modes of the original time-varying system are the unstable modes of the averaged Laplacian. For piecewise constant periodic networks, the same statement can be sharpened using the monodromy matrix

$W_k$2

and for smoothly varying periodic networks via the Floquet–Magnus expansion, whose leading term is precisely $W_k$3 (Petit et al., 2017).

Directed higher-order structures require a more elaborate spectral language. On $W_k$4-directed $W_k$5-hypergraphs, linearization produces two Laplacian-type matrices: a symmetric head–head Laplacian $W_k$6 and a generally asymmetric head–tail Laplacian $W_k$7. Under generalized natural coupling, they combine into an effective Laplacian

$W_k$8

whose eigenvalues $W_k$9 can be complex. Each spatial mode $-\nabla^2W_k=k^2W_k$0 then obeys

$-\nabla^2W_k=k^2W_k$1

and instability occurs when the dominant eigenvalue of that $-\nabla^2W_k=k^2W_k$2 block has positive real part. Writing $-\nabla^2W_k=k^2W_k$3, the unstable region in the complex plane is described by

$-\nabla^2W_k=k^2W_k$4

with $-\nabla^2W_k=k^2W_k$5 determined only by the reaction and coupling Jacobians. This permits stationary or wave-like Turing patterns and shows explicitly that directionality can create instabilities absent in the symmetric case (Dorchain et al., 2024).

4. Beyond homogeneous fixed points

The modewise Turing picture extends beyond fixed homogeneous equilibria. For a stable limit cycle $-\nabla^2W_k=k^2W_k$6, each spatial mode $-\nabla^2W_k=k^2W_k$7 satisfies a periodic linear system

$-\nabla^2W_k=k^2W_k$8

and the instability is controlled by Floquet multipliers of the first-return map. A tractable approximation replaces the monodromy by

$-\nabla^2W_k=k^2W_k$9

where

$2\times2$0

The resulting generalized Turing inequalities are

$2\times2$1

$2\times2$2

In this framework, oscillation death is reinterpreted as a Turing instability of the first-return map: a homogeneous synchronous oscillation is stable as an orbit, yet some nonuniform mode acquires positive Floquet growth and drives the system to a stationary heterogeneous state (Challenger et al., 2014).

In spatially heterogeneous media, global trigonometric modes are no longer appropriate. For

$2\times2$3

linearization about a heterogeneous state $2\times2$4 yields a variable-coefficient eigenvalue problem. WKBJ analysis produces localized modes of the form

$2\times2$5

where $2\times2$6 are eigenvalues of $2\times2$7. The local zero-mode stability conditions are

$2\times2$8

and the local Turing conditions are

$2\times2$9

Unstable modes are confined to Turing regions $k$0, and the support $k$1 shrinks as $k$2. The unstable “$k$3-modes” are therefore localized WKBJ eigenfunctions rather than global cosines (Krause et al., 2019).

A phenomenological but influential reinterpretation appears in wall-bounded shear flow. Manneville reformulated Waleffe’s four-variable self-sustaining-process model as a reaction–diffusion system, Wa97RD, with variables $k$4. In a reduced two-variable approximation, the homogeneous “featureless turbulent” state becomes unstable to a finite wavenumber $k$5 when the effective diffusivities are sufficiently different, with threshold conditions

$k$6

$k$7

and

$k$8

This gives the band pattern an intrinsic wavelength $k$9, although the model treats the diffusivities phenomenologically and does not determine the physical orientation or absolute scale of the modulation (Manneville, 2012).

5. Mode competition, wavelength selection, and nonlinear outcomes

Near onset, the fastest-growing mode often predicts the observed pattern, but the literature also shows systematic departures from that linear picture. In dryland vegetation models, the dominant Turing wavelength $k$00 governs the pattern selected from random initial conditions within the Turing-unstable regime. Beyond that regime, however, periodic patterns coexist with uniform states, localized states arise through homoclinic snaking, and wavelength selection is controlled by the interaction of the dominant Turing wavelength $k$01, the snaking wavelength $k$02, and the tail wavelength $k$03 extracted from spatial eigenvalues. Under repeated local disturbances, the effective wavelength

$k$04

can shift below or above $k$05, so mode selection becomes strongly history-dependent rather than purely dispersion-driven (Zelnik et al., 2017).

A microscopic mode-selection theorem is available for two interacting Ising lines with Kac interactions. After Fourier transform, each mode $k$06 evolves under a $k$07 matrix $k$08, and Turing instability means that $k$09 is linearly stable while some $k$10 has $k$11. The paper constructs a unimodular regime in which only $k$12 are unstable. On the fluctuation scale, if

$k$13

then the rescaled modes $k$14 converge to zero for all $k$15, whereas the $k$16 modes converge to nontrivial Gaussian limits. At the critical time

$k$17

the unstable $k$18 modes become order one and stay uniformly away from zero with high probability. Here k-mode instability is not merely a linear diagnostic: it controls the mesoscopic stochastic pattern that emerges (Capanna et al., 2017).

At codimension-3 Hopf–Turing–Turing points, the reduced amplitude system supports mixed-mode equilibria, limit cycles, invariant 2-tori, invariant 3-tori, heteroclinic cycles, period-doubling, and chaotic attractors. Numerically, this structure is reflected in full reaction–diffusion systems such as the Schnakenberg and Mimura–Murray models, where spatially heterogeneous periodic branches connect to homogeneous periodic branches and different Turing modes coexist or compete (Izuhara et al., 2023).

A quantum-open counterpart has now been formulated for a chain of bosonic modes governed by a GKSL master equation. In the semiclassical limit, the mean fields satisfy

$k$19

For the null state, the modewise growth rates are

$k$20

In the minimal $k$21 case, one can tune $k$22 so that first mode $k$23, and then modes $k$24, become unstable. The deterministic system then exhibits a stationary branch dominated by $k$25, an oscillatory mixed-mode branch when $k$26, and finally a stationary pattern dominated by $k$27. Reduced Wigner functions and collective quadrature observables track the same modal competition on the quantum side (Comparato et al., 8 Jul 2026).

6. Limits of the concept and major caveats

A k-mode Turing instability does not, by itself, guarantee the existence of a stable stationary Turing pattern. The sharpest counterexample is the reaction–diffusion–ODE class

$k$28

with one non-diffusing and one diffusing component. For a constant state $k$29, diffusion-driven instability occurs under the autocatalysis condition

$k$30

The same mechanism, however, destabilizes every continuous heterogeneous stationary solution. For a regular stationary pattern $k$31, the linearized operator has continuous spectrum containing

$k$32

and if autocatalysis holds somewhere then $k$33, so the unstable half-plane is reached through continuous spectrum. The consequence is that the model admits diffusion-driven instability of homogeneous states but no stable continuous Turing patterns (Marciniak-Czochra et al., 2013).

A broader implication is that “k-mode Turing instability” is best understood as a statement about the linear or weakly nonlinear fate of specific spatial modes, not as a synonym for stationary pattern formation. Depending on the substrate and nonlinear saturation, the endpoint may be a stationary pattern, a mixed-mode pattern, oscillation death, a wave-like state, a localized structure, a torus, or a chaotic attractor. That interpretation is consistent with the network, heterogeneous-media, turbulence, multimode, and quantum formulations surveyed above.