betaFeedback

WorksPhysics and Astronomy

Cutting, Shuffling and Diffusion as a Coarsening Process

Abstract

Cutting and shuffling (CS) rearranges fluid without stretching, but homogenisation requires diffusion (D); this interaction is not well understood. We show that CSD is a coarsening process, with scale ℓˉ=L(ncPe/2)−1/3\bar\ell=L(n_c\mathrm{Pe}/2)^{-1/3} rather than the Batchelor scale, and that cutting acts as a fractional Laplacian in wavenumber, giving the exact mixing rate γ=21/3∣a1′∣nc2/3λPe−1/3γ=2^{1/3}|a_1'|n_c^{2/3}λ\mathrm{Pe}^{-1/3} for ncn_c cuts per event at rate λλ, with a1′=−1.0188…a_1'=-1.0188\dots the first zero of Ai′\mathrm{Ai}'. This holds in any dimension for random and deterministic protocols, extends to finite-time mixing, and places CSD in a known non-equilibrium universality class.

Shuffling a deck of cards rearranges without deforming any card. Fluids can be mixed the same way: cut into pieces and reassembled isometrically [1, 2, 3], forming piecewise isometries (PWIs). This cutting and shuffling (CS) arises naturally in discontinuously deforming media such as granular [4, 5, 6, 7, 8] and yield-stress [9, 10] materials, but is also remarkably prevalent in continuous fluids. Indeed, CS can arise in any flow with persistent bifurcations, including valved systems [11], porous [12] and fractured [13] media, microfluidic [14] and static [3] mixers.

Because CS is a measure-preserving isometry, it conserves scalar variance σ2\sigma^2 exactly—it stirs but cannot physically mix [15]. Note however that CS can mix ergodically; PWIs only exhibit weak ergodic mixing [16] under specific conditions [17]. Conversely, scalar homogenization involves diffusion (D) acting on the discontinuities that CS generates. Despite its prevalence, a quantitative theory of CSD mixing is lacking. Figure 1 shows these processes are intertwined: cutting manufactures the gradients diffusion consumes, while diffusion erases the interfaces cutting requires. What emerges from this competition—how fast scalar fields homogenise, and how these are distributed—are the questions we answer here.

Figure 1. Broadening, merger and coarsening of concentration fronts under CSD: (a) transient 1D mixing at Pe=106\mathrm{Pe}=10^6 with nc=6n_c=6 at t=4/λt=4/\lambda (black), 6/λ6/\lambda (blue), 8/λ8/\lambda (red) and (b) snapshot of 2D mixing at Pe=104\mathrm{Pe}=10^4 via the “shear” F1 protocol. Note fronts of various ages (widths).

To date these have resisted answer. The ergodic theory of the 1D archetype, the interval exchange transformation (IET), is mature [16, 17, 18, 19], and design rules for good finite-time shuffling are established [14, 20, 21]. CS in the presence of diffusion has revealed unexpectedly rich behaviour, including accelerated homogenisation [1, 22], optimal stirring [23], non-monotonic dependence of mixing rates on diffusivity [24], and apparent finite cut-off phenomena [25]. These studies characterise mixing via global norms, interface counts and decay rates, with scales estimated by the classical Batchelor scaling Pe−1/2\mathrm{Pe}^{-1/2} [25, 26].

Our central claim is that discontinuous mixing is a coarsening problem. Each cut creates interfaces; diffusion broadens them; neighbours overlap and merge. The scalar field is therefore a population of diffusing fronts with a distribution of ages, and everything—spectrum, length scale, mixing rate—exactly follows from the balance between front creation and front coalescence. Three

physical consequences are worth highlighting:

(i) The relevant small scale is not the Batchelor scale. Fronts are created at rate κ=ncλ/L\kappa=n_c\lambda/L per unit length, (with ncn_c cuts per CS event, λ\lambda the cutting rate, LL the domain size) and destroyed once their diffusive width reaches the mean spacing ℓˉ\bar{\ell}, after a time ∼ℓˉ2/Dm\sim\bar{\ell}^2/D_m. Balancing birth against death gives

ℓˉ=L(ncPe)−1/3,Pe≡λL2/Dm.(1)\bar{\ell}=L(n_c\mathrm{Pe})^{-1/3}, \qquad\mathrm{Pe}\equiv\lambda L^2/D_m. \tag*{(1)}

This is the balance Rácz solved for 1D diffusion-controlled annihilation with a steady source [27, 28]: thus discontinuous mixing belongs to a known non-equilibrium universality class. The Batchelor-type scale kb=λ/2Dmk_b=\sqrt{\lambda/2D_m} only appears as the upper cut-off of a distinct spectral band; they separate as kbℓˉ∼(Pe/nc2)1/6k_b\bar{\ell}\sim(\mathrm{Pe}/n_c^2)^{1/6}, so no single-scale description can survive at large Pe.

(ii) The spectrum is exact and counts cuts. With no stretching there is no Batchelor k−1k^{-1} range [29]; CS is scale-free. Instead, the spectrum is built from discontinuities and their diffusive relics, giving k−2k^{-2} and k−4k^{-4} bands respectively. The amplitude of the terminal k−2k^{-2} band is independent of DmD_m and equals 4nc/L24n_c/L^2: the high-kk floor of the spectrum counts the cuts in ncn_c.

(iii) The mixing rate is exact and algebraic. CS acts as a fractional Laplacian in wavenumber, giving an Airy eigenvalue problem, with exact mixing rate γ=σ˙2/σ2\gamma= \dot{\sigma}^{2}/\sigma^{2}

γλ=21/3∣a1′∣nc2/3Pe⁡−1/3,(2)\frac{\gamma}{\lambda} = 2^{1/3}|a'_1|n_c^{2/3}\operatorname{Pe}^{-1/3}, \tag*{(2)}

where ∣a1′∣=1.0188…|a'_1| = 1.0188\ldots is the magnitude of the first zero of Ai⁡′\operatorname{Ai}'. CS is a genuine mixing mechanism that beats pure diffusion by a factor Pe⁡2/3\operatorname{Pe}^{2/3} but γ/λ\gamma/\lambda vanishes as Pe⁡→∞\operatorname{Pe} \to\infty; in the absence of stretching, mixing always stays diffusion-limited. This is in sharp contrast with chaotic advection, whose decay rate is set by the stretching rate and is asymptotically Pe-independent [30, 31].

Spectral Theory and DNS. Before extending to multiple dimensions, we first consider a scalar c(x,t)c(x,t) with diffusivity DmD_m on the 1D periodic domain x∈[0,L]x \in[0,L]; full derivations are in [32]. A CS event is the measure-preserving PWI ϕ\phi, c(x,t+)=c(ϕ(x),t−)c(x,t^+) = c(\phi(x),t^-), built from intervals with random or deterministic cut positions and creating ncn_c cuts, i.e. ncn_c is twice the number of interval exchanges per CS event. We first consider CS events forming a Poisson (P) process in time at rate λ\lambda, before extending to regular (R) protocols with fixed temporal intervals Δt=1/λ\Delta t = 1/\lambda. The theory requires only that successive cuts decorrelate; randomised cut positions achieve this, but so does deterministic and ergodic CS [32]. We therefore consider ϕ\phi with ergodic cut positions, the exception being fixed IETs, whose resonances dominate earlier studies [20, 24]. To generate DNS data we integrate CS with diffusion pseudo-spectrally [32] for λ=1\lambda= 1, L=1L = 1 and Dm=10−4D_m = 10^{-4}–10−810^{-8} under Poisson and regular protocols, recording energy spectra after each CS event.

Periodicity confines the spectrum to the wavenumber lattice k∈(2π/L)Zk \in(2\pi/L)\mathbb{Z} with spacing Δk=2π/L\Delta k = 2\pi/L; all sums run over this lattice. Writing c=∑k>0c^(k,t)eikxc = \sum_{k>0}\hat{c}(k,t)e^{ikx} and E(k,t)≡2⟨∣c^(k,t)∣2⟩E(k,t) \equiv2\langle|\hat{c}(k,t)|^{2}\rangle for k>0k > 0, so that σ2=∑k>0E(k,t)\sigma^{2} = \sum_{k>0}E(k,t), a CS event acts as c^(k,t+)=∑k′Akk′c^(k′,t−)\hat{c}(k,t^+) = \sum_{k'} A_{kk'}\hat{c}(k',t^-) with Akk′=L−1∫0Le−ikϕ−1(x)+ik′x dxA_{kk'} = L^{-1}\int_{0}^{L}e^{-ik\phi^{-1}(x)+ik'x}\,dx. Assuming phases decorrelate between events, the covariance Ckk′(t)≡⟨c^(k,t)c^(k′,t)∗⟩≃EP(k,t)δkk′C_{kk'}(t) \equiv\langle\hat{c}(k,t)\hat{c}(k',t)^*\rangle\simeq E_P(k,t)\delta_{kk'}, with EP(k,t)E_P(k,t) the Poisson energy spectrum. For a Poisson CS process of rate λ\lambda with diffusion, EP(k,t)E_P(k,t) evolves as

∂tEP=λ∑k′[T(k,k′)−δkk′]EP(k′,t)−2Dmk2EP,(3)\partial_t E_P = \lambda\sum_{k'}\left[T(k,k')-\delta_{kk'}\right]E_P(k',t)-2D_m k^2E_P, \tag*{(3)}

where T(k,k′)≡⟨∣Akk′∣2⟩T(k,k') \equiv\langle|A_{kk'}|^2\rangle represents CS and the last term represents diffusion. The reduction of Ckk′C_{kk'} to EP(k,t)E_P(k,t) is not obvious as IETs are never strongly mixing [16], but it is accurate: measuring TT directly from the map and comparing ∑k′T(k,k′)EP(k′,t−)\sum_{k'}T(k,k')E_P(k',t^-) with EP(k,t+)E_P(k,t^+) gives a sampling error less than 1% [32]. This closure is exact in the ensemble mean for uniform, independent cut positions, and holds to measurement precision whenever successive cuts decohere [32]. As expected, this closure fails when a cut positions do not decohere, and γ\gamma falls 2–8× below (2) but remains exponential [32].

Conservation of variance makes AA unitary, so ∑kT(k,k′)=1\sum_k T(k,k') = 1 and the source amplitude in (3) is fixed with no kk-independent background.

with no kk-independent background. Conservation of the mean gives A0k′=0A_{0k'} = 0, hence T(k,k′)∝k2T(k,k') \propto k^2 as k→0k \to0: transfer to large scales is dipolar and no variance reaches the mean. On a periodic domain this constraint acts only below the fundamental mode, and the observable spectrum saturates at a plateau for k≪kbk \ll k_b [32]. Writing the asymptotic spectrum as a decaying stationary strange eigenmode [22] ES,P(k)E_{S,P}(k), i.e. EP(k,t)→tσ2(t)ES,P(k)E_P(k,t) \xrightarrow{t} \sigma^2(t)E_{S,P}(k) with σ˙2=−γσ2\dot{\sigma}^2 = -\gamma\sigma^2 and mixing rate γ\gamma, Eq. (3) shows ES,P(k)E_{S,P}(k) transitions around kb=Λ0/2Dmk_b = \sqrt{\Lambda_0/2D_m}, where

ES,P(k)=Q(k)Λ0+2Dmk2,(4)E_{S,P}(k) = \frac{Q(k)}{\Lambda_0 + 2D_m k^2}, \tag*{(4)}

with Q(k)≡λ∑k′T(k,k′)ES,P(k′)Q(k) \equiv\lambda\sum_{k'}T(k,k')E_{S,P}(k') and Λ0=λ−γ\Lambda_0 = \lambda- \gamma. Hence Λ0→λ\Lambda_0 \to\lambda at large Pe. From (3), cutting is a difference kernel [32] on the lattice, i.e. CS is a jump process:

T(k,k′)=2nc[L(k−k′)]2,k≠k′.(5)T(k,k') = \frac{2n_c}{[L(k-k')]^2}, \qquad k \ne k'. \tag*{(5)}

If the lattice spacing is negligible against the width of the spectrum, i.e. Δk ℓ≪1\Delta k\,\ell\ll1, the sum over k′k' may be replaced by an integral and EPE_P is governed by a fractional Laplacian (−∂k2)1/2(-\partial_k^2)^{1/2} with a harmonic potential,

∂tEP=−κ(−∂k2)1/2EP−2Dmk2EP.(6)\partial_t E_P = -\kappa(-\partial_k^2)^{1/2}E_P - 2D_m k^2E_P. \tag*{(6)}

Hence CS is a non-local, heavy-tailed random walk that spreads energy to large ∣k∣|k| while diffusion is a confining potential that removes energy at large ∣k∣|k|. Thus (6) is a continuous-kk approximation whose error is quantified below. Because kk lives on a lattice of spacing 2π/L2\pi/L, the variable conjugate to kk is the physical lag rr on the circle [0,L)[0,L), and the transform of EPE_P is the autocorrelation C(r,t)=∑k>0EP(k,t)cos⁡(kr)C(r,t)=\sum_{k>0}E_P(k,t)\cos(kr). In this variable the lattice multiplier of the cutting term is elementary [32],

∂tC=2Dm∂r2C−κ∣r∣(1−∣r∣L)C(7)\partial_t C=2D_m\partial_r^2 C-\kappa|r|\left(1-\frac{|r|}{L}\right)C \tag*{(7)}

with Neumann condition ∂rC(0,t)=0\partial_r C(0,t)=0. In (7), κ∣r∣(1−∣r∣/L)\kappa|r|(1-|r|/L) is the rate at which a cut falls between two points separated by rr, while diffusion acts on the separation with diffusivity 2Dm2D_m. The continuous-kk approximation ψ(r,t)\psi(r,t) of CC is precisely the neglect of the curvature term ∣r∣2/L|r|^2/L, leaving the linear potential in the Airy equation

∂tψ=L(ψ)≡2Dm∂r2ψ−κ∣r∣ψ(8)\partial_t\psi=\mathcal{L}(\psi)\equiv2D_m\partial_r^2\psi-\kappa|r|\psi \tag*{(8)}

which is accurate because the eigenmodes are supported on r∼ℓˉ≪Lr\sim\bar{\ell}\ll L. The dominant eigenmode is ψ1(r)=Ai⁡((∣r∣−γ/κ)/ℓ)\psi_1(r)=\operatorname{Ai}((|r|-\gamma/\kappa)/\ell), and the corresponding eigenvalue is the mixing rate γ\gamma of (2); the full ladder γj=(∣aj′∣/∣a1′∣)γ\gamma_j=(|a'_j|/|a'_1|)\gamma follows from the zeros aj′a'_j of Ai⁡′\operatorname{Ai}'. First-order perturbation theory in the neglected term gives γ/γAiry−1≃(M2/∣a1′∣)ℓˉ/L\gamma/\gamma_{\mathrm{Airy}}-1\simeq(M_2/|a'_1|)\bar{\ell}/L with M2=∫0∞s2Ai⁡2/∫0∞Ai⁡2M_2=\int_0^\infty s^2\operatorname{Ai}^2/\int_0^\infty\operatorname{Ai}^2 evaluated at s+a1′s+a'_1 [32], so the continuous-kk error is O(ℓˉ/L)=O(Pe−1/3)O(\bar{\ell}/L)=O(\mathrm{Pe}^{-1/3}) and vanishes in the mixing limit. A jump in ψ′′′(r)\psi'''(r) at r=0r=0 produces a k−4k^{-4} band in Es,PE_{s,P} as k→∞k\to\infty [32], and the full stationary spectrum is

Es,P(k)=4LAi⁡(−∣a1′∣)∫0∞cos⁡(kr)Ai⁡[rℓ−∣a1′∣] dr(9)E_{s,P}(k)=\frac{4}{L\operatorname{Ai}(-|a'_1|)}\int_0^\infty\cos(kr)\operatorname{Ai}\left[\frac{r}{\ell}-|a'_1|\right]\,\mathrm{d}r \tag*{(9)}

which is normalised, ∑k>0Es,P(k)=1\sum_{k>0}E_{s,P}(k)=1, with ℓˉ\bar{\ell} from (1). As shown by Fig. 3a, its high-kk floor (16) is set by ncn_c [32],

Es(k)→C∞k2,C∞=4ncL2.(10)E_s(k)\to\frac{C_\infty}{k^2},\qquad C_\infty=\frac{4n_c}{L^2}. \tag*{(10)}

Figure 3a. The compensated floor k2Esk^2E_s is independent of DmD_m and counts cuts: C∞=4nc/L2C_\infty=4n_c/L^2.

The transient spectrum EP(k,t)E_P(k,t) is also given by the complete eigenmodes of L\mathcal{L}. Note that closed form approximations for (9) have also been developed [32]. Figure 2 shows that (9) captures the DNS data almost exactly except for the k−2k^{-2} tail, which can be explained by consideration of regular CS events as follows.

Stationary energy spectra for Poisson CS

Figure 2. Stationary spectra ES,P(k)E_{S,P}(k) for Poisson CS from DNS (symbols) and theory Eq. (9) (dashed lines) and Eq. (13) (solid lines). Regular spectra ES,R+E_{S,R}^{+} (not shown) are almost identical. Guides show k0k^{0}, k−2k^{-2} and k−4k^{-4} scaling ranges of Eq. (16).

Regular CS. The Poisson solution (9) can be extended to regular CS events with fixed spacing Δt=1/λ\Delta t=1/\lambda. Denoting ER+(k,t)E_R^+(k,t), ER−(k,t)E_R^-(k,t) respectively as the regular spectrum immediately after and before a CS event, then

ER−(k,t+Δt)=e−zER+(k,t),z≡2Dmk2λ(11)E_R^-(k,t+\Delta t)=e^{-z}E_R^+(k,t),\qquad z\equiv\frac{2D_mk^2}{\lambda} \tag*{(11)}

and the CS transfer operator K\mathcal{K} updates ER−E_R^- as ER+(k,t+Δt)=KER−(k,t)E_R^+(k,t+\Delta t)=\mathcal{K}E_R^-(k,t). This finite-step can be connected to (9) by consideration of a new discontinuity which produces a large-kk spectrum P(k)∼k−2P(k)\sim k^{-2}, which then ages with time via diffusive attenuation as exp⁡(−zλt)\exp(-z\lambda t). For

Figure 3b. The Poisson γP\gamma_P (black dots) and regular γR\gamma_R (red dots, obscured by black dots) DNS mixing rate matches the exact theory (2) (line) and the relative difference between γR\gamma_R and γP\gamma_P scales as Pe−1/3\mathrm{Pe}^{-1/3}. The branch at Pe<104\mathrm{Pe}<10^4 (not shown) scales diffusively, i.e. γ/λ∝Pe−1\gamma/\lambda\propto\mathrm{Pe}^{-1}.

continuous and discrete cutting events, the dimensionless residence factors are respectively

λ∫0∞e−zλt dt=1z,∑m=0∞e−mz=11−e−z,(12)\lambda\int_0^\infty e^{-z\lambda t}\,\mathrm{d}t=\frac{1}{z},\qquad\sum_{m=0}^\infty e^{-mz}=\frac{1}{1-e^{-z}}, \tag*{(12)}

hence the finite-step spectrum is given as

ER+(k)≃EP(k)z1−e−z,(13)E_R^+(k)\simeq E_P(k)\frac{z}{1-e^{-z}}, \tag*{(13)}

and the mixing rate under the regular CS protocol is also slightly larger, in [32] we show

γR=γP[1+O(Pe−1/3)],(14)\gamma_R=\gamma_P\left[1+O(\mathrm{Pe}^{-1/3})\right], \tag*{(14)}

as confirmed in Fig. 3b, hence γR→γP\gamma_R\to\gamma_P as Dm→0D_m\to0. Note that the results are sampled after the diffusion step, then from (11), the above spectrum is multiplied by e−ze^{-z}. As z/(1−e−z)=1+z/2+O(z2)z/(1-e^{-z})=1+z/2+O(z^2), the continuous spectrum is unchanged for z≪1z\ll1 and for z≫1z\gg1, ER+∼k−2E_R^+\sim k^{-2}, i.e. the finite-step spectrum exhibits a k−2k^{-2} tail. As dln⁡ER+/dln⁡k≃−2−z/(ez−1)\mathrm{d}\ln E_R^+/\mathrm{d}\ln k\simeq-2-z/(e^z-1), the midpoint of the transition (slope =−3=-3) is given by zb=1.256z_b=1.256. Hence the transition kbk_b and coarsening kck_c crossovers occur at

kb=(zbλ2Dm)1/2,kc=1ℓ=(λ2Dm)1/3(15)k_b=\left(\frac{z_b\lambda}{2D_m}\right)^{1/2},\qquad k_c=\frac{1}{\ell}=\left(\frac{\lambda}{2D_m}\right)^{1/3} \tag*{(15)}

which separate as kb/kc=(zb3Pe/2nc2)1/6k_b/k_c=(z_b^3\mathrm{Pe}/2n_c^2)^{1/6}. In [32] we show that these results are consistent with a physically-motivated model based on diffusive fronts generated by CS events that coarsen with age. This model shows that the fronts generated by regular CS form two distinct populations; fresh discontinuities (k−2k^{-2}) and aged fronts (k−4k^{-4}), leading to the overall spectrum

Es(k)≃{6.042 ℓˉ/L,k≪kc,2ncλk−4/DmL2,kc≪k≪kb,C∞k−2,k≫kb.(16)E_s(k)\simeq \begin{cases} 6.042\,\bar{\ell}/L, & k\ll k_c,\\ 2n_c\lambda k^{-4}/D_mL^2, & k_c\ll k\ll k_b,\\ C_\infty k^{-2}, & k\gg k_b. \end{cases} \tag*{(16)}

The k−4k^{-4} band corresponds to the fraction of fronts still sharp at scale 1/k1/k, which falls as (kc/k)2(k_c/k)^2 and multiplies the single-front k−2k^{-2}. The terminal k−2k^{-2} band is carried by the discontinuities of the most recent cut. Matching the two gives k2=λ/2Dmk^2=\lambda/2D_m, recovering exactly the kbk_b of the independent balance (4).

Figure 2 shows that theory (13) closely matches the Poisson spectra (<1%<1\%), and the regular spectra (not shown) are almost identical. This is a sampling-protocol artefact; the Poisson spectra are measured immediately following a CS event, and so the k−2k^{-2} tail of the fresh front dominates, whereas the specific timing of older ages fronts plays a minor role. This is confirmed by Fig. 3a which shows that the Poisson spectra and theory (13) yield an invariant k−2k^{-2} tail that is independent of DmD_m. Similarly, Fig. 3b shows the mixing rate γ\gamma follows (2) closely and both the Poisson γP\gamma_P and regular γR\gamma_R rates are very similar, with the relative difference scaling as Pe−1/3\mathrm{Pe}^{-1/3}.

The theory predicts three different scalings for the cut number ncn_c: C∞∝ncC_\infty\propto n_c, ℓˉ∝nc−1/3\bar{\ell} \propto n_c^{-1/3} and γ∝nc2/3\gamma\propto n_c^{2/3}. Figure 4 tests these scalings across two CS protocol families (sub-arc reversals and mm-interval IETs) and over three values of Pe\mathrm{Pe}. We find the DNS results agree with these predicted exponents to within 1% rms error.

Figure 4. Scaling of various quantities with ncn_c. (a) L2C∞/4ncL^2C_\infty/4n_c, Eq. (10), (thin line), for sub-arc reversals and IETs. (b) ℓˉ∝nc−1/3\bar{\ell} \propto n_c^{-1/3}, Eq. (1). (c) γ∝nc2/3\gamma\propto n_c^{2/3}, Eq. (2).

Finite-time mixing. In general, the finite-time spectrum EP(k,t)E_P(k,t) is given by all eigenmodes of L\mathcal{L} [32]

EP(k,t)=∑j=1∞bjFjkA(kkA)exp⁡[−21/3nc2/3∣aj′∣λPe−1/3t],(17)E_P(k,t)=\sum_{j=1}^{\infty}\frac{b_jF_j}{k_A}\left(\frac{k}{k_A}\right)\exp\left[-2^{1/3}n_c^{2/3}|a'_j|\lambda\mathrm{Pe}^{-1/3}t\right], \tag*{(17)}

with kA=1/ℓˉk_A=1/\bar{\ell}, Ai⁡′(−∣aj′∣)=0\operatorname{Ai}'(-|a'_j|)=0 and bjb_j known [32] for the initial condition c(x,0)=10<x<1/2c(x,0)=\mathbb{1}_{0<x<1/2}. Hence each Airy mode decays as γj=21/3nc2/3∣aj′∣λPe−1/3\gamma_j=2^{1/3}n_c^{2/3}|a'_j|\lambda\mathrm{Pe}^{-1/3}, and the transient mixing rate evolves as

γ(t)=21/3nc2/3λPe−1/3∑j∣aj′∣bjAi⁡(−∣aj′∣)e−∣aj′∣τ∑jbjAi⁡(−∣aj′∣)e−∣aj′∣τ→t→∞γP,(18)\gamma(t)=2^{1/3}n_c^{2/3}\lambda\mathrm{Pe}^{-1/3}\frac{\sum_j |a'_j|b_j\operatorname{Ai}(-|a'_j|)e^{-|a'_j|\tau}}{\sum_j b_j\operatorname{Ai}(-|a'_j|)e^{-|a'_j|\tau}}\xrightarrow{t\to\infty}\gamma_P, \tag*{(18)}

with τ=t/tc\tau=t/t_c, which relaxes to the dominant mode as exp⁡(−Δaτ)\exp(-\Delta_a\tau), Δa=∣a2′∣−∣a1′∣=2.229\Delta_a=|a'_2|-|a'_1|=2.229, where the coarsening time is

tc=ℓˉ22Dm=(Lncλ)2/3(2Dm)−1/3=∣a1′∣γP.(19)t_c=\frac{\bar{\ell}^2}{2D_m}=\left(\frac{L}{n_c\lambda}\right)^{2/3}(2D_m)^{-1/3}=\frac{|a'_1|}{\gamma_P}. \tag*{(19)}

Extension to regular CS events requires the discrete propagator ER+(k,t=N/λ)=(Ke−z)NER+(k,0)E_R^+(k,t=N/\lambda)=(\mathcal{K}e^{-z})^N E_R^+(k,0) rather than the asymptotic correction (13).

At early times the first set of fronts retains a strong imprint on the system, leading to EP∼k−2E_P\sim k^{-2}, but for 0<τ<10<\tau<1, the spectrum develops an Airy core with a surviving k−2k^{-2} range, followed by a transition to the dominant mode, EP→Es,PE_P\to E_{s,P}. Before coalescence sets in, every front created since t=0t=0 survives, ages are uniform on [0,t][0,t], and each front of age τ\tau dissipates at rate Dm/2πτ\sqrt{D_m/2\pi\tau}, yielding

γ(t)=4ncλLDmt2π(t≪tc),(20)\gamma(t)=\frac{4n_c\lambda}{L}\sqrt{\frac{D_mt}{2\pi}}\qquad(t\ll t_c), \tag*{(20)}

Equation (20) holds once the cuts outnumber those of the initial condition, i.e. for t>ρ0L/ncλt>\rho_0L/n_c\lambda with ρ0\rho_0 the initial interface density [32]. It reproduces the measured rate to 1–4%, and γ(t)/γP\gamma(t)/\gamma_P collapses on t/tct/t_c over four decades in Pe\mathrm{Pe} [Figure 5(a)]. Thus finite-time decay is not a single exponential but a t3/2t^{3/2} super-exponential rolling over into a single exponential at tct_c. This explains why fitting a single stretched exponential exp⁡[−(t/τ)β]\exp[-(t/\tau)^\beta] across both regimes [25] returns an effective β\beta between 1 and 3/2 that drifts towards 3/2 as Pe increases.

Transient and 2D CSD

Figure 5. Transient and 2D CSD. (a) Transient decay rate γ(t)/γP\gamma(t)/\gamma_P collapsed on t/tct/t_c with tct_c from Eq. (19); the dashed line is the parameter-free Eq. (20). (b) Shell-integrated 2D spectra; the dotted guide is the k−2k^{-2} floor of Eq. (16). (c) Compensated spectra (Lk)2E(Lk)^2E, showing the DmD_m-independent discontinuity floor. (d) Mixing rate γ∝Dm1/3\gamma\propto D_m^{1/3} for PWI families F1 (shear) and F2 (block).

Multiple Dimensions

The CSD construction and picture of diffusing and coalescing fronts extends naturally to multiple dimensions. A PWI ϕ\phi in any dimension has codimension-one discontinuities—points in 1D, lines in 2D, surfaces in 3D—so each front broadens transversely with age τ\tau as 4Dmτ\sqrt{4D_m\tau} and coalesces with its neighbours after τc∼s2/Dm\tau_c \sim s^2/D_m. For a general PWI the geometry of the discontinuity set determines how Fourier modes are redistributed, and there is no requirement that the resulting spectral operator be isotropic. In particular, a generic dd-dimensional fractional-Laplacian does not imply a generic Pe−1/3\mathrm{Pe}^{-1/3} scaling.

A relevant class is PWIs that generate codimension-one scalar fronts; these can be described in terms of coordinates (s,r)(s,r) tangent and normal to the front, such that at scales smaller than the radius of curvature of the front, diffusion acts in the normal direction ∂tc≈Dm∂r2c\partial_t c \approx D_m\partial_r^2 c. Thus the small-scale front dynamics are locally equivalent to the 1D problem. For a fixed front orientation, with normal kn=k⋅nk_n=\mathbf{k}\cdot\mathbf{n} and tangential k⊥\mathbf{k}_\perp wavevectors, the corresponding transfer operator TT has the structure

T(kn,p),(kn′,q)(d)=T(1)(kn,kn′)δpq,(21)T^{(d)}_{(k_n,p),(k'_n,q)}=T^{(1)}(k_n,k'_n)\delta_{pq}, \tag*{(21)}

while diffusion contributes exp⁡[−Dm(kn2+∣k⊥∣2)Δt]\exp[-D_m(k_n^2+|\mathbf{k}_\perp|^2)\Delta t]. Random CS orientations and reconnections couple these sectors but do not introduce a new diffusive length scale. Hence the dominant normal balance therefore retains the 1D coarsening scale: ℓˉ∼Dm1/3\bar{\ell}\sim D_m^{1/3}, kb∼Dm−1/3k_b\sim D_m^{-1/3}, γ∼Dm1/3\gamma\sim D_m^{1/3}.

This prediction was tested against two 2D PWI families [32]: F1 consisting of randomly positioned strip shear-reconnection events, while F2 randomly partitions the torus into square blocks and applies a block permutation (Figure 1). These protocols generate front geometries in fundamentally different ways. The shell-integrated spectrum E(k)≡∑∣k′∣∈k±Δk/22∣c^k′∣2E(k)\equiv\sum_{|\mathbf{k}'|\in k\pm\Delta k/2}2|\hat{c}_{\mathbf{k}'}|^2, with Δk=2π/L\Delta k=2\pi/L the lattice spacing, for Dm=10−4−10−6D_m=10^{-4}-10^{-6} under F1 is shown in Figure 5b (F2 is also similar [32]), which exhibits the same k−4k^{-4} and k−2k^{-2} branches as the 1D case, as well as a DmD_m-independent discontinuity floor (Figure 5c). Hence the high wavenumber spectrum retains the 1D spectrum. Note that the k0k^0 plateau at small kk is not observed due to lack of averaging. Figure 5d also shows that the 2D mixing rates scale as γ∝Dm1/3\gamma\propto D_m^{1/3}, with different prefactors due to the different PWI protocols.

One important consideration is irreducibility. Similar to 1D CSD [24], decoherence is also required in multiple dimensions for the closure (3) to Ep(k,t)E_p(k,t) to hold. For block PWIs, a fixed block lattice does not decohere; random offsets are required to restore the Dm1/3D_m^{1/3} cascade: this scaling is universal for PWIs which are irreducible on relevant scales.

Discussion

As 1D CS is scale-free, random CSD is governed by the coarsening of a population of diffusing fronts. CS creates codimension-one fronts at a rate set by the PWI, while diffusion broadens and annihilates them. Their birth–death balance gives a non-Batchelor coarsening scale ℓˉ=L(Pe/2)−1/3\bar{\ell}=L(\mathrm{Pe}/2)^{-1/3}, and algebraic mixing rate γ/λ∼n2/3Pe−1/3\gamma/\lambda\sim n^{2/3}\mathrm{Pe}^{-1/3}. Hence while mixing is diffusive (γ∼Pe−1\gamma\sim\mathrm{Pe}^{-1}) for Pe≲103\mathrm{Pe}\lesssim10^3, at larger Pe, CS accelerates diffusion by Pe2/3\mathrm{Pe}^{2/3} while remaining asymptotically diffusion-limited; γ/λ→0\gamma/\lambda\to0 as Pe→∞\mathrm{Pe}\to\infty.

The same mechanism extends to transient and 2D mixing. Transient simulations using two geometrically distinct, measure-preserving PWIs recover γ∝Dm1/3\gamma\propto D_m^{1/3} and the characteristic k−4→k−2k^{-4}\to k^{-2} spectral structure. This suggests that the 1/3 exponent is a geometric consequence of irreducible renewal of codimension-one fronts.

These results place discontinuous mixing in the non-equilibrium universality class of diffusion-controlled coarsening [27, 28], so that the extensive exact results available for that class [33, 34] may be brought to bear on higher-order mixing statistics and on reactive and multicomponent transport. This framework applies wherever CS organise transport, from granular [4, 5, 6, 7, 8], porous [12] and fractured [13] media to biological tissues, valved systems [11] and microfluidic devices [3, 14].

Data availability statement. The data that support the findings of this article are publicly available [35].

References

References

  1. [1]R. Sturman, The role of discontinuities in mixing (Elsevier, 2012) pp. 51 – 90.DOI
  2. [2]I. C. Christov, R. M. Leuptow, and J. M. Ottino, Stretching and folding versus cutting and shuffling: An illustrated perspective on mixing and deformations of continua, American Journal of Physics 79, 359 (2011).DOI
  3. [3]L. D. Smith, P. B. Umbanhowar, R. M. Leuptow, and J. M. Ottino, The geometry of cutting and shuffling: An outline of possibilities for piecewise isometries, Physics Reports 802, 1 (2019), the geometry of cutting and shuffling: An outline of possibilities for piecewise isometries.
  4. [4]G. Juarez, R. M. Leuptow, J. M. Ottino, R. Sturman, and S. Wiggins, Mixing by cutting and shuffling, EPL (Europhysics Letters) 91, 20003 (2010).DOI
  5. [5]G. Juarez, I. C. Christov, J. M. Ottino, and R. M. Lueptow, Mixing by cutting and shuffling 3D granular flow in spherical tumblers, Chem. Eng. Sci. 73, 195 (2012).DOI
  6. [6]I. C. Christov, J. M. Ottino, and R. M. Lueptow, Streamline jumping: A mixing mechanism, Phys. Rev. E 81, 046307 (2010).pubmed.ncbi.nlm.nih.gov/20481827
  7. [7]Z. Zaman, M. Yu, P. P. Park, J. M. Ottino, R. M. Lueptow, and P. B. Umbanhowar, Persistent structures in a three-dimensional dynamical system with flowing and non-flowing regions, Nature Communications 9, 3122 (2018).DOI
  8. [8]P. P. Park, P. B. Umbanhowar, J. M. Ottino, and R. M. Lueptow, Mixing with piecewise isometries on a hemispherical shell, Chaos 26, 073115 (2016).DOI
  9. [9]L. D. Smith, M. Rudman, D. R. Lester, and G. Metcalfe, Impact of discontinuous deformation upon the rate of chaotic mixing, Phys. Rev. E 95, 022213 (2017).DOI
  10. [10]D. R. Lester and A. Chryss, Topological mixing of yield stress materials, Phys. Rev. Fluids 4, 064502 (2019).DOI
  11. [11]L. D. Smith, M. Rudman, D. R. Lester, and G. Metcalfe, Mixing of discontinuous deforming media, Chaos: An Interdisciplinary Journal of Nonlinear Science 26, 023113 (2016), https://doi.org/10.1063/1.4941851.
  12. [12]D. R. Lester, J. Heyman, Y. Méheust, and T. Le Borgne, A unified theory of pore-scale chaotic advection, Journal of Fluid Mechanics 1017, A13 (2025).DOI
  13. [13]S. Ascione, D. R. Lester, B. Pinier, T. Le Borgne, and J. Heyman, Cutting and shuffling control mixing in three-dimensional discrete fracture networks, Journal of Fluid Mechanics 1036, A64 (2026).DOI
  14. [14]L. D. Smith, P. B. Umbanhowar, J. M. Ottino, and R. M. Lueptow, Optimized mixing by cutting-and-shuffling, SIAM J. Appl. Dyn. Syst. 17, 2544 (2018).arxiv.org/abs/1803.08102
  15. [15]P. V. Danckwerts, The definition and measurement of some characteristics of mixtures, Appl. Sci. Res. A 3, 279 (1952).
  16. [16]A. Katok, Interval exchange transformations and some special flows are not mixing, Isr. J. Math. 35, 301 (1980).DOI
  17. [17]A. Avila and G. Forni, Weak mixing for interval exchange transformations and translation flows, Annals of Mathematics 165, 637 (2007).
  18. [18]M. Keane, Interval exchange transformations, Mathematische Zeitschrift 141, 25 (1975).DOI
  19. [19]C. F. Novak, Of Discontinuity-growth of interval-exchange maps, J. Mod. Dyn. 3, 379 (2009).DOI
  20. [20]M. K. Krotter, I. C. Christov, J. M. Ottino, and R. M. Lueptow, Cutting and shuffling a line segment: Mixing by interval exchange transformations, Int. J. Bifurcation Chaos 22, 1230041 (2012).DOI
  21. [21]M. Yu, P. B. Umbanhowar, J. M. Ottino, and R. M. Lueptow, Cutting and shuffling of a line segment: Effect of variation in cut location, Int. J. Bifurcation Chaos 26, 1630038 (2016).DOI
  22. [22]P. Ashwin, M. Nicol, and N. Kirkby, Acceleration of one-dimensional mixing by discontinuous mappings, Physica A 310, 347 (2002).DOI
  23. [23]G. Froyland, C. González-Tokman, and T. M. Watson, Optimal mixing enhancement by local perturbation, SIAM Rev. 58, 494 (2016).DOI
  24. [24]H. Kreczak, R. Sturman, and M. C. T. Wilson, Deceleration of one-dimensional mixing by discontinuous mappings, Phys. Rev. E 96, 053112 (2017).DOI
  25. [25]M. Wang and I. C. Christov, Cutting and shuffling with diffusion: Evidence for cut-offs in interval exchange maps, Phys. Rev. E 98, 022221 (2018).DOI
  26. [26]C. P. Schlick, I. C. Christov, P. B. Umbanhowar, J. M. Ottino, and R. M. Lueptow, A mapping method for distributive mixing with diffusion: Interplay between chaos and diffusion in time-periodic sine flow, Phys. Fluids 25, 052102 (2013).DOI
  27. [27]Z. Rácz, Diffusion-controlled annihilation in the presence of particle sources: Exact results in one dimension, Phys. Rev. Lett. 55, 1707 (1985).pubmed.ncbi.nlm.nih.gov/10031902
  28. [28]P. L. Kripavsky, S. Redner, and E. Ben-Naim, A Kinetic View of Statistical Physics (Cambridge University Press, 2010).
  29. [29]G. K. Batchelor, Small-scale variation of convected quantities like temperature in turbulent fluid part 1. general discussion and the case of small conductivity, J. Fluid Mech. 5, 113 (1959).DOI
  30. [30]R. Pierrehumbert, Tracer microstructure in the large-eddy dominated regime, Chaos, Solitons & Fractals 4, 1091 (1994), special issue: Chaos Applied to Fluid Mixing.DOI
  31. [31]J.-L. Thiffeault, Scalar decay in chaotic mixing, Lect. Notes Phys. 744, 3 (2008).DOI
  32. [32]See Supplemental Material for the full derivations, numerical methods and code, , which includes Refs. [16, 17, 20, 24–28, 36].
  33. [33]D. D. C. Torney and H. M. McConnell, Diffusion-limited reactions in one dimension, Proc. R. Soc. Lond. A 387, 147 (1983).DOI
  34. [34]D. ben Avraham, M. A. Burschka, and C. R. Doering, Statics and dynamics of a diffusion-limited reaction: Anomalous kinetics, nonequilibrium self-ordering, and a dynamic transition, J. Stat. Phys. 60, 695 (1990).
  35. [35]D. Lester, T. L. Borgne, and J. Heyman, Datasets for ”cutting, shuffling and diffusion as a coarsening process”, Figshare, doi.org/10.25439/rmt.34783593.

Paper details

vlnc:W_zdd7fs7c
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents