betaFeedback

WorksPhysics and Astronomy

Polymeric diffusive instability leading to elastic turbulence in plane Couette flow

Abstract

Elastic turbulence is a chaotic flow state observed in dilute polymer solutions in the absence of inertia. It was discovered experimentally in circular geometries and has long been thought to require a finite amplitude perturbation in parallel flows. Here we demonstrate, within the commonly-used Oldroyd-B and FENE-P models, that a self-sustaining chaotic state can be initiated via a linear instability in a simple inertialess shear flow caused by the presence of small but non-zero diffusivity of the polymer stress. Numerical simulations show that the instability leads to a three-dimensional selfsustaining chaotic state, which we believe is the first reported in a wall-bounded, parallel, inertialess viscoelastic flow.

Dilute polymer solutions are ubiquitous in everyday life (e.g. foods, shampoo, paints, cosmetics) and understanding how they behave is important for many industrial processes (e.g. plastics, oil, pharmaceuticals and chemicals). The stretching and subsequent relaxation of the polymers introduces new viscoelastic stresses in the flow which depend on the flow’s deformation history. As a result, polymer flows can exhibit startlingly different behaviour from that of a Newtonian fluid like water ( e.g. rod-climbing, die swell and elastic recoil [1] ). Perhaps most strikingly, a chaotic flow state – so called “elastic turbulence” (ET) – can occur for vanishing inertia, in stark contrast to Newtonian fluid mechanics where inertia provides the only nonlinearity. ET has important applications in small scale flows where, for example, enhanced mixing for chemical reactions or heat transfer for cooling computer chips are highly desirable [2, 3]. Even though the origin of this behaviour is still not understood, the key ingredients are believed to be fluid elasticity provided by the polymers and streamline curvature which together give rise to a new elastic linear instability [4, 5, 6, 7]. Experiments in curved geometries confirm the linear instability leads to sustained ET [8]. In contrast, wall-bounded inertialess parallel polymer flow has been presumed linearly stable, and experimental work there has focused on triggering a finite amplitude instability instead. Obstacles in the flow have been used to provide the required streamline curvature believed necessary for this elastic instability and ultimately ET [9, 10, 11, 12]. However, the requirements for both initiating such a transition and for the existence of a self-sustaining chaotic state in a planar geometry are unknown.

Recently, by exploring the large relaxation-time limit of the polymers, a new elastic linear instability has been identified for a parallel flow of an Oldroyd-B fluid in a pipe or channel at finite inertia [13, 14] and at vanishing inertia in a channel only [15]. This instability is a ‘centre mode’ – concentrated around the centreline of the channel – and is strongly subcritical, giving rise to an arrowhead-shaped travelling wave solution [16] over a large region of the parameter space. This solution has been seen in simulations at both finite [17] and vanishing inertia [18, 19], and may play a role in two-dimensional elasto-inertial turbulence where inertia is important [20]. While no link has yet been found between this structure and ET in wall-bounded flows, similar arrowhead-like flow structures have been observed in doubly-periodic ‘Kolmogorov flow’ [21, 22] where a self-sustaining, two-dimensional chaotic state is maintained in the absence of inertia. The chaotic dynamics found here can be directly connected to a linear instability of the basic state [23, 24] – though whether this instability is driven by the same mechanism as the centre mode, or whether the nonlinear chaotic state is a manifestation of the three-dimensional ET found in wall bounded flows is an open problem. In contrast to these configurations, the simpler case of constant shear between two differentially-moving, parallel plates, known as plane Couette flow, has been considered linearly stable for all inertia and elasticity parameters [13].

In this Letter we report that viscoelastic plane Couette flow is linearly unstable if polymer stress diffusion is included in the model. This diffusion is generally so small that it is ignored as an important physical effect, but is reintroduced as a much larger ‘artificial’ diffusion to stabilise time-stepping schemes if their inherent numerical diffusion isn’t sufficient. This diffusion-induced linear instability is distinctly different from the centre-mode present in channels, being concentrated instead at the walls. Significantly, there is no smallest diffusion threshold below which the instability vanishes: the wavelength of the instability decreases with the size of the diffusion so the instability could be misunderstood as a numerical instability. The growth rate of the instability tends to a non-zero limit as the polymer diffusion goes to zero so the vanishing-diffusion limit is singular. The new diffusive instability exists over a very wide area of the parameter space and is robust to the choice of boundary conditions on the polymer conformation. Direct numerical simulations (DNS) show that the instability saturates onto a low-amplitude limit cycle in two-dimensions. Three-dimensional simulations show a transition to sustained spatiotemporal chaos, which we believe to be the first reported computation of such a state in a planar geometry.

We consider the inertialess flow of an incompressible, viscoelastic fluid between infinite plates at y=±hy=\pm h moving with velocity ±U0x^\pm U_0\hat{\mathbf{x}}. The governing equations are

∇p=βΔu+(1−β)∇⋅T(C),(1)\nabla p=\beta\Delta\mathbf{u}+(1-\beta)\nabla\cdot\mathbf{T}(\mathbf{C}), \tag*{(1)}
∇⋅u=0,(2)\nabla\cdot\mathbf{u}=0, \tag*{(2)}
∂tC+(u⋅∇)C+T(C)=C⋅∇u+(∇u)T⋅C+ϵΔC.(3)\partial_t\mathbf{C}+(\mathbf{u}\cdot\nabla)\mathbf{C}+\mathbf{T}(\mathbf{C})=\mathbf{C}\cdot\nabla\mathbf{u}+(\nabla\mathbf{u})^{T}\cdot\mathbf{C}+\epsilon\Delta\mathbf{C}. \tag*{(3)}

where the polymeric stress T\mathbf{T} is related to the conformation tensor C\mathbf{C} using the FENE-P model,

T(C):=1Wi[C1−(tr⁡C−3)/Lmax⁡2−I].\mathbf{T}(\mathbf{C}):=\frac{1}{Wi}\left[\frac{\mathbf{C}}{1-(\operatorname{tr}\mathbf{C}-3)/L_{\max}^{2}}-\mathbf{I}\right].

This model successfully predicted the phenomenon of elasto-inertial turbulence (EIT) in 2010 [25] which was observed a year later in experiments [26, 27].

The equations are non-dimensionalised by the half the gap width, hh, and the plate speed U0U_0, which defines the Weissenberg number Wi:=λU0/hWi:=\lambda U_0/h (the ratio of the polymer relaxation time λ\lambda to a flow timescale). The parameter β:=μs/μT\beta:=\mu_s/\mu_T is the ratio of the solvent-to-total viscosities while ϵ:=D/U0h\epsilon:=D/U_0h is the nondimensionalisation of the polymer diffusivity DD [28]. In this configuration, the laminar basic state is simply U=yx^\mathbf{U}=y\hat{\mathbf{x}} with only Txx=2WiT_{xx}=2Wi and Txy=1T_{xy}=1 being non-zero stress components for an Oldroyd-B fluid (Lmax⁡→∞L_{\max}\to\infty). The polymer equation (1c) changes character from hyperbolic at ϵ=0\epsilon=0 to parabolic for ϵ≠0\epsilon\ne0 and extra boundary conditions are then needed. Three boundary conditions are considered: (i) application of the governing equations with ϵ=0\epsilon=0 at the walls [29], (ii) application of the governing equations with only the term ϵ∂y2Cij\epsilon\partial_y^2 C_{ij} removed [17] and (iii) Neumann, so ∂yCij=0\partial_y C_{ij}=0.

The linear stability of the basic state is examined by introducing small perturbations of the form ϕ′(x,t)=ϕ^(y)exp⁡(ikx(x−ct))+c.c.\phi'(x,t)=\hat{\phi}(y)\exp(ik_x(x-ct))+\mathrm{c.c.}, where kx∈Rk_x\in\mathbb{R} is the streamwise wavenumber and c=cr+icic=c_r+ic_i a complex wavespeed, with instability if ci>0c_i>0. The linear eigenvalue problem is solved by expanding each flow variable using the first NN Chebyshev polynomials (N=300N=300 is usually sufficient to ensure convergence). An example eigenvalue spectrum for an Oldroyd-B fluid with Wi=100Wi=100, ϵ=10−3\epsilon=10^{-3}, β=0.95\beta=0.95 and kx=3k_x=3 is reported in Figure 1 (the bottom right inset shows the equivalent spectrum with ϵ=0\epsilon=0). The continuous spectra in the absence of polymeric diffusion are regularised with the introduction of ϵ≠0\epsilon\ne0 [30], with a pair of linear instabilities emerging with wavespeeds cr∼±1c_r\sim\pm1.

Eigenvalue spectrum

Figure 1. Spectrum at Re=0Re=0, Wi=100Wi=100, β=0.95\beta=0.95, kx=3k_x=3 for two different resolutions Ny=300N_y=300 (orange circles) and Ny=400N_y=400 (purple dots) with ϵ=10−3\epsilon=10^{-3} within the domain and ϵ=0\epsilon=0 at the boundaries. Top inset: zoom-in to the unstable eigenvalues ci>0c_i>0. Bottom right inset: spectrum for the same parameters but ϵ=0\epsilon=0 everywhere in the domain and boundaries and two different resolutions Ny=300N_y=300 (red circles) and Ny=400N_y=400 (blue dots). The most unstable eigenvalues with finite ϵ\epsilon are c=±1.0052607137+4.513293285i×10−5c=\pm1.0052607137+4.513293285i\times10^{-5}. Note the apparent instability in the continuous spectrum seen in the ϵ=0\epsilon=0 results is associated with poor resolution of the (non-smooth) eigenfunctions in our numerics (e.g. [31]) and can be suppressed if the resolution is increased further.

We map out the unstable region for various parameters and boundary conditions in Figure 2(a). In the top left panel of Figure 2(a) we observe that the instability persists as ϵ→0\epsilon\to0 i.e. this is a singular limit for all three boundary conditions and occurs at a constant value of WiWi for diffusivities ϵ≤10−2\epsilon\le10^{-2}, requiring an increasingly

Neutral curves for Oldroyd-B and FENE-P models

Figure 2. (a) Neutral curves for Oldroyd-B in the ε\varepsilon-WiWi space for β=0.9\beta= 0.9 with different boundary conditions: (i) ε=0\varepsilon= 0 at the walls (blue), (ii) ε→0\varepsilon\to0 only for ∂yyC\partial_{yy}C (red) and (iii) ∂yC=0\partial_y C = 0 (purple). Dashed lines show the neutral curves for individual wavenumbers {1,2,3}\{1, 2, 3\} with b.c. (ii). Inset shows the wavenumber kxk_x along the neutral curves and its scaling with ε−1/2\varepsilon^{-1/2} for all boundary conditions. (b) Neutral curves in the ε\varepsilon-WiWi for different β\beta in Oldroyd-B. Inset: collapse of the curves in the ultra-dilute limit β→1\beta\to1, Wi→∞Wi \to\infty. (c) Effect of the adding finite extensibility to plane Couette flow (red solid lines) and plane Poiseuille flow (blue dashed lines) in the FENE-P model. Curves show different values of Lmax⁡={∞,600,200,100,60}L_{\max} = \{\infty, 600, 200, 100, 60\}, curves from right (lighter shade) to left (darker shade) for ε=10−3\varepsilon= 10^{-3}.

large streamwise wavenumber kx∝ϵ−1/2/8k_x\propto\epsilon^{-1/2}/8 ([30] did not consider these large wavenumbers and so failed to find instability). The imaginary part of the wavespeed scales with the square root of the diffusivity, ci∝ϵ1/2c_i\propto\epsilon^{1/2}, so that the growth rate of the instability, kxcik_xc_i, remains O(1)O(1) as ϵ→0\epsilon\to0. The unstable region appears unbounded as Wi→∞Wi\to\infty for boundary conditions (i) and (ii) but stability is restored in this limit for (iii). Henceforth boundary condition (i) is used.

In Figure 2(b)-(Figure 2c) we also examine the effect of the viscosity ratio and finite extensibility on the diffusive instability. The instability is realised for decreasing values of WiWi at fixed ϵ\epsilon as β\beta is reduced (i.e. increasing polymer concentration), and the marginal stability curves collapse when plotted against Wi(1−β)/βWi(1-\beta)/\beta in the ultradilute limit β→1\beta\to1, which is magnitude of the perturbation stresses τxx′\tau'_{xx} and τxy′\tau'_{xy} relative to the diffusion terms in the momentum equation. Furthermore, the instability survives for realistic values of the polymer extensibility of Lmax⁡=O(100)L_{\max}=O(100), in the FENE-P model, though it is pushed to increasingly low values of β\beta and suppressed beyond a critical WiWi. Figure 2(c) shows that this instability is also present in plane Poiseuille flow driven by a non-dimensional pressure gradient ∂xP=−2\partial_xP=-2, the neutral curves nearly overlapping when WiWi is rescaled by the shear rate at the wall, Uwall′U'_{\mathrm{wall}}. The quantitative differences observed between the two flow configurations for the specific case considered in 2(c) arise due to the relatively large value of ε=10−3\varepsilon= 10^{-3} (the boundary layer has thickness ε1/2\varepsilon^{1/2}), so that the instability is affected by the non-monotonic channel profile. A neutral eigenfunction is shown in Figure 3(a), where we visualise contours of the perturbation trace of the polymer conformation and streamlines for an example set of parameters Wi=47.84Wi = 47.84, β=0.9\beta= 0.9, ε=10−3\varepsilon= 10^{-3}. The spanwise vorticity is shown below in Figure 3(b) for various values of ε\varepsilon, β\beta and WiWi.

Figure 3. (a) Pseudocolor corresponding to tr⁡(C)\operatorname{tr}(C) of the upper wall eigenfunctions on the neutral curves for the orange point indicated in Figure 2(b). Contour lines indicate the streamfunction (flow left to right). (b) Spanwise vorticity for the corresponding coloured dots in Figure 2(b). The eigenfunctions are normalised so that ∥vmax⁡∥=1\lVert v_{\max}\rVert= 1.

The mechanism of instability is reminiscent of the destabilizing effect of viscosity in Newtonian channel flow which subtly adjusts the relative phases of key dynamical [32]. In the polymeric diffusive instability perturbations to the polymer field are highly stretched in a boundary layer of size O(ε1/2)O(\varepsilon^{1/2}) at one of the walls where the perturbation vorticity is concentrated. Without stress diffusion, the phase of the ensuing polymer stress is poorly matched to that of the velocity field so there is not enough positive feedback in the momentum equation. A diffusive phase lag in the polymer stretch response, however, allows the polymer stress to sufficiently reinforce the flow which created it, thereby producing an instability (see Supplementary Material for a more detailed description).

Direct numerical simulations of inertialess viscoelastic plane Couette flow

Figure 4. Direct numerical simulations of inertialess viscoelastic plane Couette flow for Wi=100Wi = 100, β=0.9\beta= 0.9, Lmax=600L_{max} = 600, ϵ=10−3\epsilon= 10^{-3}. (a) Instanteneous visualization of ⟨tr⁡(C)⟩\langle\operatorname{tr}(C) \rangle of the two-dimensional periodic orbit identified. (c) Instanteneous visualization of the chaotic three-dimensional state identified. The horizontal slice is a pseudocoulour of tr⁡(C)\operatorname{tr}(C) at y=−0.99y = -0.99, the vertical slice is a pseudocoulour of vertical velocity, vv. The contour indicate positive and negative values of the spanwise velocity, ww. Flow from left to right. (b) Time series of ⟨v2⟩\langle v^2 \rangle for the two-dimensional nonlinearly saturated state corresponding to a low-amplitude high-frequency periodic orbit. (d) time series of the volume average tr⁡(C)\operatorname{tr}(C) for the chaotic three-dimensional flow with the parameters above (red), the two-dimensional nonlinearly saturated state (blue), and the laminar state (orange). (e)-(g) and (h) time average mean quantities – denoted (⋅)‾\overline{(\cdot)} – for the two-dimensional limit cycle and the three-dimensional chaotic state; (e) deviation of the mean velocity from U=yU = y (2D blue, 3D red). (f) urmsu_{\mathrm{rms}} (purple: 2D light, 3D dark) and vrmsv_{\mathrm{rms}} (orange: 2D light, 3D dark) (g) tr⁡(C)/Lmax\operatorname{tr}(C)/L_{max} (2D blue, 3D red). (h): CxyC_{xy} (2D blue, 3D red).

The instability described here is realised for commonly used parameters in numerical simulations of dilute polymer solutions (e.g. Wi∼50Wi \sim50 at β=0.9\beta= 0.9) over a wide range of realistic diffusivities. For example, in a microfluidic context, ϵ\epsilon can range from O(10−6)O(10^{-6}) for long polymer molecules formed by random chains with contour length much larger than the persistence length to O(10−3)O(10^{-3}) for short polymers. The instability exists in a boundary layer of size ≈10ϵ\approx10\sqrt{\epsilon} (see Figure 3(b)) and has an optimal wavenumber kx≈0.125/ϵk_x \approx0.125/\sqrt{\epsilon} consistent with this (see Figure 1(a) in Supplementary Material).

Dimensionally, this is a lengthscale of 10ϵh=10R/WikBTλ/6πμsR310\sqrt{\epsilon}h = 10R/\sqrt{Wi} \sqrt{k_BT\lambda/6\pi\mu_sR^3} using the Stokes-Einstein relation (kBTk_BT is the thermal energy and RR is the polymer gyration radius) and with typical values (e.g. see [33]) reduces to ≈10R/Wi\approx10R/\sqrt{Wi} consistent with the simple scaling relationship D∼R2/λD \sim R^2/\lambda. The optimal instability lengthscale therefore approaches the polymer lengthscale for Wi≳100Wi \gtrsim100 although larger-scale instability modes, which offer a better scale separation, also exist away from the neutral curve (see figure 1(b) in the Supplementary Material).

We now show using the FENE-P model that this instability forms a pathway to chaos in planar Couette flow. The nonlinear evolution of the diffusive instability is examined by first conducting two-dimensional DNS of the governing equations (1) using the open-source codebase Dedalus [34]. We perform calculations at parameter settings Wi=100Wi = 100, β=0.9\beta= 0.9, Lmax=600L_{max} = 600 and ϵ=10−3\epsilon= 10^{-3} in a box of length of length Lx=2πL_x = 2\pi with resolution [Nx,Ny]=[128,256][N_x, N_y] = [128, 256]. The simulations are initialised with low amplitude white noise, which excites the most unstable mode kx=3k_x = 3. The unstable mode amplifies exponentially before saturating onto a limit cycle with period T∼h/U0T \sim h/U_0. We report a snapshot of the instantaneous polymer stretch field in figure 4(a) alongside the volume-averaged – denoted ⟨⋅⟩\langle\cdot\rangle – vertical velocity squared in figure 4(b). While the polymer is significantly stretched relative to the laminar value, the velocity fluctuations are relatively low amplitude, v=O(10−2U0)v = O(10^{-2}U_0).

In flows with inertia, EIT can be realised in purely two-dimensional configurations [35], which is not something we observe in our inertialess calculations. The stability of the limit cycle is examined in a three-dimensional configuration with spanwise width Lz=2πL_z = 2\pi by adding small-amplitude white noise in the velocity component ww. Three-dimensional simulations were performed with resolutions ranging from [Nx,Ny,Nz]=[64,128,32][N_x, N_y, N_z] = [64, 128, 32] to [Nx,Ny,Nz]=[128,256,64][N_x, N_y, N_z] = [128, 256, 64] to check robustness. A time series of the volume-averaged polymer trace from this computation is reported in figure 4(d) which shows a departure from the simple time-periodic solution to a chaotic trajectory which is maintained for over a 104h/U010^4h/U_0 time period. We believe this to be a numerical realisation of elastic turbulence in parallel flow.

An instantaneous snapshot of the chaotic flow can be seen in figure 4(c). The horizontal pseudocolor plane located at y=−0.99hy = -0.99h (in the boundary layer at the lower wall for ϵ=10−3\epsilon= 10^{-3}) shows how tr(C)tr(C) is modulated in the both the streamwise and spanwise direction forming high and low high stretch regions. The vertical back plane contours the vertical velocity vv, which is amplified in small scale near-wall patches reminiscent of suction and ejection events in high Reynolds number wall bounded turbulence. The three-dimensional contours (in the box) represent ww, illustrating the spanwise motion of the flow.

The statistics of both the limit cycle and the three-dimensional chaotic attractor are examined further in panels (e)–(h) of figure 4, which indicates (i) that the limit cycle is a significant departure from the laminar base state, (ii) the chaos departs from the limit cycle while retaining certain features (iii) the polymer is substantially stretched in the centre of the domain for both the limit cycle and in the chaotic flow. There are also sharp variations of urmsu_{rms} in a boundary layer δ∝ϵ1/2\delta\propto\epsilon^{1/2} at the walls (verified in computations at various ϵ\epsilon but not shown). The values of urmsu_{rms} in ET are significantly different from the limit cycle and reach their largest magnitude at the centreline.

The presence of polymeric stress diffusivity is commonly disregarded in the linear stability analyses of viscoelastic flows [11, 13, 14, 36, 37, 38]. However, we have found that its presence, even at vanishingly small values in the FENE-P model, fundamentally changes the stability of viscoelastic plane Couette flow in the absence of inertia (i.e. vanishing polymer stress diffusion is a singular limit). This polymeric diffusive instability (PDI) is a ‘wall’ mode, travelling with roughly the wall speed, and has a streamwise wavelength comparable to the boundary layer thickness. This last feature could easily have led to the instability being dismissed in the past as a numerical instability. The onset of instability is found at Wi≈8Wi \approx8 independently of LmaxL_{max} and ϵ\epsilon for a broad range of β\beta. DNS of the instability leads to a stable periodic orbit in 2D and to spatiotemporal chaos in 3D, the first numerical realisation of such a flow state in a wall-bounded parallel flow configuration. This chaotic state is reproducible in other viscoelastic shear flows with and without inertia [39].

The fact that the PDI operates on a scale approaching the polymer gyration radius at higher WiWi (≳100\gtrsim100) brings into question whether (a) FENE-P remains a good viscoelastic model there, and (b) whether this instability could be experimentally observed. Certainly at lower Wi=O(10)Wi = O(10) and for less optimal (larger scale) PDI modes, there should be sufficient scale separation for the continuum approximation. The bigger issue is whether FENE-P with polymer stress diffusion incorporates enough physics to be realistic when the PDI drives the dynamics. Preliminary calculations in the two-fluid model of [40] (simplified to a solution of Hookean dumbbells [41, 42], see Supplementary Material) suggest that the added presence of polymer concentration diffusion stabilises the PDI. This would suggest that the PDI will be not be seen in the laboratory. If that is the case, our results then call into question the continuing use of the FENE-P model which can only be numerically integrated forward in time if polymer stress diffusion is present (either artificial or numerical).

The authors gratefully acknowledge the support of EP-SRC through grant EP/V027247/1. The authors declare no competition of interest.

Supplementary material

Linear stability analysis

Adding a small perturbation, ϕ∗\phi^*, to the base state Φ\Phi so that the total field is ϕ=Φ+ϕ∗\phi= \Phi+\phi^*, the linearised equations derived from (1a)-(1c) in the main manuscript are

∂xp∗=β(∂xxu∗+∂yyu∗)+(1−β)(∂xτxx∗+∂yτxy∗),(4)\partial_xp^*=\beta\left(\partial_{xx}u^*+\partial_{yy}u^*\right)+(1-\beta)\left(\partial_x\tau^*_{xx}+\partial_y\tau^*_{xy}\right), \tag*{(4)}
∂yp∗=β(∂xxv∗+∂yyv∗)+(1−β)(∂xτxy∗+∂yτyy∗),(5)\partial_yp^*=\beta\left(\partial_{xx}v^*+\partial_{yy}v^*\right)+(1-\beta)\left(\partial_x\tau^*_{xy}+\partial_y\tau^*_{yy}\right), \tag*{(5)}
∂tcxx∗+U∂xcxx∗+Cxx′v∗+τxx∗=2(Cxx∂xu∗+U′cxy∗+Cxy∂yu∗)+ε(∂xxcxx∗+∂yycxx∗),(6)\partial_tc^*_{xx}+U\partial_xc^*_{xx}+C'_{xx}v^*+\tau^*_{xx}=2\left(C_{xx}\partial_xu^*+U'c^*_{xy}+C_{xy}\partial_yu^*\right)+\varepsilon\left(\partial_{xx}c^*_{xx}+\partial_{yy}c^*_{xx}\right), \tag*{(6)}
∂tcyy∗+U∂xcyy∗+Cyy′v∗+τyy∗=2(Cxy∂xv∗+Cyy∂yv∗)+ε(∂xxcyy∗+∂yycyy∗),(7)\partial_tc^*_{yy}+U\partial_xc^*_{yy}+C'_{yy}v^*+\tau^*_{yy}=2\left(C_{xy}\partial_xv^*+C_{yy}\partial_yv^*\right)+\varepsilon\left(\partial_{xx}c^*_{yy}+\partial_{yy}c^*_{yy}\right), \tag*{(7)}
∂tczz∗+U∂xczz∗+Czz′v∗+τzz∗=ε(∂xxczz∗+∂yyczz∗),(8)\partial_tc^*_{zz}+U\partial_xc^*_{zz}+C'_{zz}v^*+\tau^*_{zz}=\varepsilon\left(\partial_{xx}c^*_{zz}+\partial_{yy}c^*_{zz}\right), \tag*{(8)}
∂tcxy∗+U∂xcxy∗+Cxy′v∗+τxy∗=(Cxx∂xv∗+U′cyy∗+Cyy∂yu∗)+ε(∂xxcxy∗+∂yycxy∗),(9)\partial_tc^*_{xy}+U\partial_xc^*_{xy}+C'_{xy}v^*+\tau^*_{xy}=\left(C_{xx}\partial_xv^*+U'c^*_{yy}+C_{yy}\partial_yu^*\right)+\varepsilon\left(\partial_{xx}c^*_{xy}+\partial_{yy}c^*_{xy}\right), \tag*{(9)}
∂xu∗+∂yv∗=0,(10)\partial_xu^*+\partial_yv^*=0, \tag*{(10)}

where

τij∗=1Wi(tr⁡(c∗)f2(tr⁡C)Lmax⁡2Cij+f(tr⁡C)cij∗),(11)\tau^*_{ij}=\frac{1}{Wi}\left(\frac{\operatorname{tr}(c^*)f^2(\operatorname{tr}C)}{L_{\max}^2}C_{ij}+f(\operatorname{tr}C)c^*_{ij}\right), \tag*{(11)}
∂xτij∗=1Wi(∂xtr⁡c∗f2(tr⁡C)Lmax⁡2Cij+f(tr⁡C)∂xcij∗),(12)\partial_x\tau^*_{ij}=\frac{1}{Wi}\left(\partial_x\operatorname{tr}c^*\frac{f^2(\operatorname{tr}C)}{L_{\max}^2}C_{ij}+f(\operatorname{tr}C)\partial_xc^*_{ij}\right), \tag*{(12)}
∂yτij∗=1Wi(∂ytr⁡c∗f2(tr⁡C)Lmax⁡2Cij+tr⁡C∂y[f2(tr⁡C)Lmax⁡2]Cij+cij∗∂yf(tr⁡C)+f(tr⁡C)∂ycij∗).(13)\partial_y\tau^*_{ij}=\frac{1}{Wi}\left(\partial_y\operatorname{tr}c^*\frac{f^2(\operatorname{tr}C)}{L_{\max}^2}C_{ij}+\operatorname{tr}C\partial_y\left[\frac{f^2(\operatorname{tr}C)}{L_{\max}^2}\right]C_{ij}+c^*_{ij}\partial_yf(\operatorname{tr}C)+f(\operatorname{tr}C)\partial_yc^*_{ij}\right). \tag*{(13)}

(primed quantities denote base flow derivatives with respect to the wall-normal direction and Cxx′=Cxy′=Cyy′=Czz′=0C'_{xx}=C'_{xy}=C'_{yy}=C'_{zz}=0 for plane Couette flow). We consider the exponential growth of perturbations and apply a Fourier transform in the homogeneous xx direction by introducing the ansatz ϕ∗=ϕ^(y)eikx(x−ct)\phi^*=\hat{\phi}(y)e^{ik_x(x-ct)}, where kxk_x is the streamwise wavenumber and c=cr+icic=c_r+ic_i the complex phase speed of the perturbations, with ci>0c_i>0 for linear instability.

Figure 5 shows that the optimum wavenumber for the PDI scales as kx≈0.125/εk_x\approx0.125/\sqrt{\varepsilon} and that 1−cr=O(ε)1-c_r=O(\sqrt{\varepsilon}) for the PDI localised at the boundary moving at 1 on the neutral curve. This is consistent with the scaling of the growth rate kxci=O(1)k_xc_i=O(1) as ε→0\varepsilon\to0 away from the neutral curve.

Figure 5. (a) Wavenumbers kxk_x (in blue) corresponding to PDI for Oldroyd-B along the neutral curves in Figure 2(a) for the same values of β\beta. Dashed red line shows that kx≈1/8ε−1/2k_x \approx1/8\varepsilon^{-1/2} for the critical wavenumber of the instability. Solid red shaded curves show kx∣cr−U∣k_x|c_r-U|, of the PDI along the same neutral curves. (b) Unstable wavenumbers for PDI for β=0.9\beta=0.9, ε=10−3\varepsilon=10^{-3}. The shaded region shows how the range of unstable kxk_x increases beyond criticality.

Instability mechanism

For an Oldroyd-B fluid, the instability persists in the ultra-dilute limit of Wi→∞Wi\to\infty and β→1\beta\to1 with Λ=Wi(1−β)/β\Lambda=Wi(1-\beta)/\beta staying finite: see Figure 2 in the main text. The leading order equations for the perturbation in this limit are

ikxp∗−(∂y2−kx2)u∗=ikxτxx+Dτxy,(14)ik_xp^*-\left(\partial_y^2-k_x^2\right)u^*=ik_x\tau_{xx}+D\tau_{xy}, \tag*{(14)}
∂yp∗−(∂y2−kx2)v∗=ikxτxy,(15)\partial_yp^*-\left(\partial_y^2-k_x^2\right)v^*=ik_x\tau_{xy}, \tag*{(15)}
[ikx(U−c)−ε(∂y2−kx2)]τxx=4ikxΛu∗+2U′τxy,(16)\left[ik_x(U-c)-\varepsilon\left(\partial_y^2-k_x^2\right)\right]\tau_{xx}=4ik_x\Lambda u^*+2U'\tau_{xy}, \tag*{(16)}
[ikx(U−c)−ε(∂y2−kx2)]τxy=2ikxΛv∗,(17)\left[ik_x(U-c)-\varepsilon\left(\partial_y^2-k_x^2\right)\right]\tau_{xy}=2ik_x\Lambda v^*, \tag*{(17)}
ikxu∗+∂yv∗=0.(18)ik_xu^*+\partial_yv^*=0. \tag*{(18)}

where perturbation variables are taken proportional to exp⁡(ikx(x−ct))\exp(ik_x(x-ct)) and τij:=(1−β)τij∗\tau_{ij}:=(1-\beta)\tau^*_{ij} is a rescaled stress. Note the appearance of Λ\Lambda in the polymer conformation evolution equations (4c and 4d) – this is the rescaled base normal stress, TxxT_{xx}. We have performed numerical experiments that indicate that the terms in equations (4a–4e) highlighted with underbraces can be dropped without suppressing the instability. We can therefore describe the instability mechanism as follows: In equation (4d) the highly-tensioned base state polymers are tilted by ∂xv∗\partial_xv^*, to generate a perturbation shear stress τxy\tau_{xy}. This perturbation shear stress is then deformed by the base shear to drive a streamwise stress perturbation, τxx\tau_{xx}, in equation (4c) which can then enhance the streamwise velocity in the momentum equation (4a), provided there is a phase adjustment through polymer stress diffusion. The phase adjustment through the diffusion terms in (4c) and (4d) is crucial for the positive feedback as the reduced system is stable without it. For example, at Λ=3\Lambda=3, kx=1.4k_x=1.4 and ε=0.001\varepsilon=0.001, using Neumann boundary conditions on the stress for simplicity, we find PDI with ci=0.0024c_i=0.0024, which vanishes on setting ε=0\varepsilon=0.

Numerical method

The direct numerical simulations (DNS) of eqs. (1a)-(1c) in the main text were performed using the numerical codebase Dedalus [34] (the Dedalus codebase has been extensively used in flow simulation [17, 19, 34]). We consider a computational domain of size [Lx,Ly,Lz]=[2π,2,2π]h[L_x,L_y,L_z]=[2\pi,2,2\pi]h. The quantities C\mathbf{C} and u\mathbf{u} are expanded in NxN_x and NzN_z Fourier modes in the streamwise (xx) and spanwise (zz) directions respectively. An expansion in NyN_y Chebyshev modes is employed in the wall-normal (yy) direction. Time integration is performed with a 4th-order exponential Runge-Kutta (RK443) scheme with fixed time step Δt=5×10−3\Delta t=5\times10^{-3}. The two-dimensional simulations were performed with a resolution [Nx,Ny]=[128,256][N_x,N_y]=[128,256] and the three-dimensional ones used [Nx,Ny,Nz]=[64,128,32][N_x,N_y,N_z]=[64,128,32]. The consistency of the results was checked with double the resolution in both cases, obtaining the same dynamical behaviour. The two-dimensional simulations were initialised by adding random noise to the undisturbed laminar state. We tested the three-dimensional stability of the two-dimensional state by adding small amplitude random perturbation in the spanwise component of the velocity w\mathbf{w}.

Polymer concentration diffusion

The effect of introducing polymer concentration diffusion was checked by considering the two-fluid model of [40] simplified to a solution of Hookean dumbells [41, 42]. Using our non-dimensionalisation, this model is

∇p=βΔu+(1−β)∇⋅τ,\nabla p=\beta\Delta\mathbf{u}+(1-\beta)\nabla\cdot\boldsymbol{\tau},
∇⋅u=0,\nabla\cdot\mathbf{u}=0,
∂tC+u⋅∇C+τ=C⋅∇up+(∇up)T⋅C+ε∇2C,\partial_t\mathbf{C}+\mathbf{u}\cdot\nabla\mathbf{C}+\boldsymbol{\tau}=\mathbf{C}\cdot\nabla\mathbf{u}_p+(\nabla\mathbf{u}_p)^{T}\cdot\mathbf{C}+\varepsilon\nabla^2\mathbf{C},
∂tn+u⋅∇n=ε∇⋅(∇n−Wi∇⋅τ)\partial_t n+\mathbf{u}\cdot\nabla n=\varepsilon\nabla\cdot(\nabla n-Wi\nabla\cdot\boldsymbol{\tau})

with stress

τ:=1Wi(C−nI),\boldsymbol{\tau}:=\frac{1}{Wi}(\mathbf{C}-n\mathbf{I}),

where nn is the polymer concentration and the definition of the mass-average velocity is

up:=u+γεWi(−∇n+Wi∇⋅τ).(19)\mathbf{u}_p:=\mathbf{u}+\gamma\varepsilon Wi(-\nabla n+Wi\nabla\cdot\boldsymbol{\tau}). \tag*{(19)}

We solve the linear stability problem for infinitesimal disturbances superimposed on the base flow

(ub,τxx,τxy,τyy,n,p)=(yx^,2Wi,1,0,1,1)(20)(\mathbf{u}_b,\tau_{xx},\tau_{xy},\tau_{yy},n,p)=(y\hat{\mathbf{x}},2Wi,1,0,1,1) \tag*{(20)}

with no-slip boundary conditions on the velocity field and no flux conditions on the stress and the concentration field (the latter given by n^⋅up=0\hat{\mathbf{n}}\cdot\mathbf{u}_p=0). PDI is present within this model for γ<2⋅10−3\gamma<2\cdot10^{-3}, however for γ≥2⋅10−3\gamma\geq2\cdot10^{-3} (and γ=1\gamma=1 in [40]) no linear instability was found over β∈[0.001,0.9998]\beta\in[0.001,0.9998] and Wi∈[4,1400]Wi\in[4,1400].

References

References

  1. [1]R. Bird, R. Curtiss, R. Armstrong, and O. Hassager, Dynamics of Polymeric Liquids: Fluid Mechanics, vol 1, 2nd Ed (1987).
  2. [2]A. Groisman and V. Steinberg, Efficient mixing at low Reynolds numbers using polymer additives, Nature 410, 905 (2001).DOI
  3. [3]T. M. Squires and S. R. Quake, Microfluidics: Fluid physics at the nanoliter scale, Reviews of Modern Physics 77, 977 (2005).DOI
  4. [4]E. S. G. Shaqfeh, Purely elastic instabilities in viscometric flows, Annual Review of Fluid Mechanics 28, 129 (1996).
  5. [5]A. Groisman and V. Steinberg, Elastic turbulence in a polymer solution flow, Nature 405, 53 (2000).pubmed.ncbi.nlm.nih.gov/10811214
  6. [6]R. G. Larson, Turbulence without inertia, Nature 405, 27 (2000).pubmed.ncbi.nlm.nih.gov/10811202
  7. [7]A. Groisman and V. Steinberg, Elastic turbulence in curvilinear flows of polymer solutions, New Journal of Physics 6, 29 (2004).DOI
  8. [8]V. Steinberg, Elastic turbulence: an experimental view on interminable random flow, Annual Review of Fluid Mechanics 53, 27 (2021).DOI
  9. [9]P. Pakdel and G. H. McKinley, Elastic instability and curved streamlines, Physical Review Letters 77, 2459 (1996).pubmed.ncbi.nlm.nih.gov/10061959
  10. [10]L. Pan, A. Morozov, C. Wagner, and P. E. Arratia, Nonlinear elastic instability in channel flows at low Reynolds numbers, Physical Review Letters 110, 174502 (2013).pubmed.ncbi.nlm.nih.gov/23679736
  11. [11]R. van Buel and H. Stark, Characterizing elastic turbulence in the three-dimensional von Kármán swirling flow using the Oldroyd-B model, Physics of Fluids 34, 043112 (2022).DOI
  12. [12]S. Datta, A. M. Ardekani, P. E. Arratia, A. N. Beris, I. Bischofberger, G. H. McKinley, J. G. Eggers, J. E. López-Aguilar, S. M. Fielding, A. Frishman, et al., Perspectives on viscoelastic flow instabilities and elastic turbulence, Physical Review Fluids 7, 080701 (2022).arxiv.org/abs/2108.09841
  13. [13]P. Garg, I. Chaudhary, M. Khalid, V. Shankar, and G. Subramanian, Viscoelastic pipe flow is linearly unstable, Physical Review Letters 121, 024502 (2018).DOI
  14. [14]M. Khalid, I. Chaudhary, P. Garg, V. Shankar, and G. Subramanian, The centre-mode instability of viscoelastic plane Poiseuille flow, Journal of Fluid Mechanics 915 (2021).DOI
  15. [15]M. Khalid, V. Shankar, and G. Subramanian, Continuous pathway between the elasto-inertial and elastic turbulent states in viscoelastic channel flow, Physical Review Letters 127, 134502 (2021).DOI
  16. [16]J. Page, Y. Dubief, and R. R. Kerswell, Exact traveling wave solutions in viscoelastic channel flow, Physical Review Letters 125, 154501 (2020).DOI
  17. [17]Y. Dubief, J. Page, R. R. Kerswell, V. E. Terrapon, and V. Steinberg, First coherent structure in elasto-inertial turbulence, Physical Review Fluids 7, 073301 (2022).DOI
  18. [18]A. Morozov, Coherent structures in plane channel flow of dilute polymer solutions with vanishing inertia, Physical Review Letters 129, 017801 (2022).DOI
  19. [19]G. Buza, M. Beneitez, J. Page, and R. R. Kerswell, Finite-amplitude elastic waves in viscoelastic channel flow from large to zero Reynolds number, Journal of Fluid Mechanics 951, A3 (2022).DOI
  20. [20]G. H. Choueiri, J. M. Lopez, A. Varshney, S. Sankar, and B. Hof, Experimental observation of the origin and structure of elastoinertial turbulence, Proceedings of the National Academy of Sciences 118, e210250118 (2021).DOI
  21. [21]S. Berti, A. Bistagnino, G. Boffetta, A. Celani, and S. Musacchio, Two-dimensional elastic turbulence, Physical Review E 77, 055306 (2008).pubmed.ncbi.nlm.nih.gov/18643127
  22. [22]S. Berti and G. Boffetta, Elastic waves and transition to elastic turbulence in a two-dimensional viscoelastic kolmogorov flow, Physical Review E 82, 036314 (2010).DOI
  23. [23]G. Boffetta, A. Celani, A. Mazzino, A. Puliafito, and M. Vergassola, The viscoelastic kolmogorov flow: eddy viscosity and linear stability, Journal of Fluid Mechanics 523, 161 (2005).DOI
  24. [24]A. Bistagnino, G. Boffetta, A. Celani, A. Mazzino, A. Puliafito, and M. Vergassola, Nonlinear dynamics of the viscoelastic Kolmogorov flow, Journal of Fluid Mechanics 590, 61 (2007).
  25. [25]Y. Dubief, C. M. White, E. S. G. Shaqfeh, and V. E. Terrapon, Polymer maximum drag reduction: a unique transitional state, Annual Research Briefs , Stanford Centre for Turbulence Research, 395 (2010).arxiv.org/abs/1106.4482
  26. [26]B. Hof, D. Samanta, and C. Wagner, The maximum drag reduction asymptote, APS Division of Fluid Dynamics, Baltimore, Abstract M8-005 (2011).
  27. [27]D. Samanta, Y. Dubief, M. Holzner, C. Schafer, A. N. Morozov, C. Wagner, and B. Hof, Elasto-inertial turbulence, Proceedings of the National Academy of Sciences 110, 10557 (2011).DOI
  28. [28]A. W. El-Kareh and L. G. Leal, Existence of solutions for all Deborah numbers for a non-Newtonian model modified to include diffusion, Journal of Non-Newtonian Fluid Mechanics 33, 257 (1989).
  29. [29]R. Sureshkumar, A. N. Beris, and R. A. Handler, Direct numerical simulation of the turbulent channel flow of a polymer solution, Physics of Fluids 9, 743 (1997).DOI
  30. [30]P. Chokshi and V. Kumaran, Stability of the plane shear flow of dilute polymeric solutions, Physics of Fluids 21, 014109 (2009).DOI
  31. [31]H. J. Wilson, M. Renardy, and Y. Renardy, Structure of the spectrum in zero reynolds number shear flow of the ucm and oldroyd-b liquids, Journal of non-newtonian fluid mechanics 80, 251 (1999).
  32. [32]C.-C. Lin, The Theory of Hydrodynamic Stability (Cambridge University Press, 1955).
  33. [33]N. Burshtein, K. Zografos, A. Q. Shen, R. J. Poole, and S. J. Haward, Inertioelastic flow instability at a stagnation point, Phys. Review X 7, 041039 (2017).
  34. [34]K. J. Burns, G. M. Vasil, J. S. Oishi, D. Leacoanet, and B. P. Brown, Dedalus: A flexible framework for numerical simulations with spectral methods, Physical Review Research 2, 023068 (2020).DOI
  35. [35]S. Sid, V. Terrapon, and Y. Dubief, Two-dimensional dynamics of elasto-inertial turbulence and its role in polymer drag reduction, Physical Review Fluids 3, 011301 (2018).DOI
  36. [36]R. G. Larson, E. S. Shaqfeh, and S. J. Muller, A purely elastic instability in Taylor–Couette flow, Journal of Fluid Mechanics 218, 573 (1990).DOI
  37. [37]L. Pan, A. Morozov, C. Wagner, and P. Arratia, Nonlinear elastic instability in channel flows at low Reynolds numbers, Physical Review Letters 110, 174502 (2013).
  38. [38]H. A. C. Sanchez, M. R. Jovanovic, S. Kumar, A. Morozov, V. Shankar, G. Subramanian, and H. J. Wilson, Understanding viscoelastic flow instabilities: Oldroyd-B and beyond, Journal of Non-Newtonian Fluid Mechanics 302, 104742 (2022).DOI
  39. [39]Manuscript in preparation.
  40. [40]V. G. Mavrantas and A. N. Beris, Modelling of the rheology and the flow-induced concentration changes in polymer solutions, Phys. Re. Lett. 69, 273 (1992).
  41. [41]A. Beris and V. G. Mavrantzas, On the compatibility between various macroscopic formalisms for the concentration and flow of dilute polymer solutions, J. Rheology 38, 1235 (1994).DOI
  42. [42]W. B. Black and M. D. Graham, Slip, concentration fluctuations, and flow instability in sheared polymer solutions, Macromolecules 34, 5731 (2001).

Paper details

vlnc:W_qx4mdqvr
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