betaFeedback

WorksPhysics and Astronomy

No need to stay positive: a practical approach to direct numerical simulations of elastic turbulence

Abstract

Successfully performing direct numerical simulations of polymeric flows remains a major challenge in computational fluid mechanics. In addition to the velocity field, such simulations must resolve polymeric degrees of freedom, often expressed via the conformation tensor, c\mathbf{c}, which captures the local stretch of polymer molecules. A key difficulty here lies in maintaining the physical requirement Tr c>3\mathrm{Tr}\, \mathbf{c}>3, which is not explicitly enforced by the governing equations. Consequently, simulations initiated from physical conditions may silently drift into unphysical states with Tr c<0\mathrm{Tr}\, \mathbf{c}<0, indicating a loss of positive-definiteness of the conformation tensor. Existing numerical methods to prevent this are costly, making direct numerical simulations of chaotic polymer flows, such as elastic turbulence, heavily reliant on high-performance computing. Here, we ask whether simulations that violate Tr c>3\mathrm{Tr}\, \mathbf{c}>3 can still yield meaningful physical insight into the underlying dynamics. We simulate a model dilute polymer solution driven through a plane channel at low Reynolds number and observe the transition to elastic turbulence. Our simulations exhibit two threshold resolutions: below the first, they become numerically unstable and exhibit a finite-time blow-up; above the second, they maintain positive-definiteness. In between, simulations remain stable and chaotic despite local violations of Tr c>3\mathrm{Tr}\, \mathbf{c}>3. Surprisingly, these violations do not affect mid-plane statistics of velocity, its gradients, or polymer stretch, which match results from fully positive-definite simulations. This suggests that resolving flow structures or key flow statistics may not require the extreme resolutions needed to preserve positive-definiteness, potentially lowering computational barriers for studying elastic turbulence.

Introduction

Understanding instabilities in flows of complex fluids is one of the outstanding problems of fluid mechanics [37] . An archetypal example of such a challenge is posed by elastic (ET) [29, 59] and elasto-inertial (EIT) [54, 20] turbulence – chaotic flows of polymer solutions at low and moderate levels of fluid inertia.

The dynamics of such flows are commonly characterised by two dimensionless parameters: The Weissenberg number, Wi\mathrm{Wi}, defined as the product of a characteristic shear gradient magnitude and the polymer relaxation time, quantifies the ability of the flow to stretch polymer molecules, while the Reynolds number, Re\mathrm{Re}, measures the relative importance of inertial and viscous effects. In this framework, ET corresponds to the regime of large Wi\mathrm{Wi} and small Re\mathrm{Re}, while EIT arises when strong elastic stresses interact with finite inertia.

Recently, significant progress was made in understanding the transition to EIT and ET ([59, 60, 19, 20, 47]), particularly in flows in straight channels, where it has been discovered that such flows are organised around coherent structures, similar to their Newtonian counterparts ([28]). In contrast to Newtonian turbulence, however, viscoelastic coherent structures are driven by the anisotropic distribution of elastic stresses that are localised either at the walls of the domain (the wall mode) or at its midplane (the centre mode). The distinction between wall and centre modes, rooted in linear stability analysis results ([27, 14, 15, 34, 33, 19, 13]), is also reflected in the nonlinear states that organise EIT and ET. In the EIT regime, a growing body of experimental, numerical, and theoretical work has connected the wall modes to coherent near-wall dynamics resembling self-sustained (viscoelastic) Tollmien–Schlichting waves ([54, 21, 58, 16, 43, 57, 55, 17, 56, 20]). At significantly lower values of Re\mathrm{Re}, however, polymeric flows have been shown to organise around travelling-wave solutions: arrowheads in the EIT and narwhals in the ET regimes, respectively. Such solutions, localised around the centreline/centreplane of the flow geometry, were reported in Kolmogorov flow ([7, 8, 50, 41]), pressure-driven channel flow ([52, 46, 12, 38, 4, 39, 42, 47]), flow through a cylinder array ([66, 35]), and curved pipe flow ([44]).

A unified pathway linking the coherent-structure dynamics of ET and EIT was proposed by [26] and [4].

The ET and EIT phenomena are driven predominantly by the anisotropic distribution of the polymer stress, while the velocity field is mostly responding to its changes. While the latter can routinely be measured in experiments, there are currently no reliable techniques to measure time- and space-resolved profiles of the former. Under such circumstances, the only reliable source of information about the polymeric stress field can be obtained from direct numerical simulations (DNS), which have been crucial in uncovering the physical scenario presented above. Such simulations, however, are notoriously difficult to perform owing to a broad range of numerical instabilities and artefacts arising at sufficiently high values of Wi\mathrm{Wi}, commonly referred to as the High-Weissenberg Number Problem ([51]). While the recent work by Kerswell and colleagues indicate that some of these instabilities might be caused by the presence of the polymer stress diffusion term in the constitutive equation ([6, 5, 18, 40]), the main origin of the High-Weissenberg Number Problem is rooted in the loss of positive-definiteness of the polymer conformation tensor C\mathbf{C}.

There are two ways to introduce the polymer conformation tensor and understand the significance of its positive-definiteness. The first stems from kinetic theory of dilute polymer solutions ([9, 37, 48]), which identifies C\mathbf{C} with the ensemble average dyadic product ⟨RR⟩\langle\mathbf{R}\mathbf{R}\rangle, where R\mathbf{R} is the end-to-end vector of a polymer molecule. The conformation tensor is normalised with the typical size of the polymer molecule in thermal equilibrium such that in the absence of flow C=I\mathbf{C}=\mathbf{I}, where I\mathbf{I} is the identity matrix. Depending on the microscopic force-extension law assumed, the polymer stress tensor τ\boldsymbol{\tau} can then be expressed as an algebraic function of c\mathbf{c}, that reduces to τ∼c−I=0\boldsymbol{\tau} \sim\mathbf{c} - \mathbf{I} = 0, in equilibrium [1]. Within this interpretation, under flow the eigenvalues of c\mathbf{c} can be viewed as the square of the polymer extension along its principle axes and should thus all be positive. Alternatively, one can introduce c\mathbf{c} without any appeal to kinetic theory. Within such an approach, the conformation tensor plays the role of an auxiliary variable related to the stress tensor through an algebraic relation. For a wide class of constitutive models, it is then possible to prove mathematically that the equation of motion for c\mathbf{c} is evolutionary: if at time t0t_0, the eigenvalues of c\mathbf{c} are non-negative everywhere in the domain, they remain non-negative at any later time t>t0t > t_0 [31, 32]. Since the rest state, c=I\mathbf{c} = \mathbf{I}, is clearly non-negative definite, the eigenvalues of c\mathbf{c} should always remain non-negative. For a wide class of models, this bound can be made stronger, identifying physical solutions with the requirement that Tr⁡c>d\operatorname{Tr}\mathbf{c} > d everywhere in the domain, where Tr⁡\operatorname{Tr} denotes the trace and dd is the number of spatial dimensions [30, 64]. In this work, we employ this stronger condition as an indicator of physical admissibility of a particular conformation tensor (and, hence, τ\boldsymbol{\tau}) distribution.

Unfortunately, this criterion is easily violated in DNS, even when staring from a positive-definite state, where accumulation of numerical errors leads to the loss of positive-definiteness and makes the results unphysical. There exist various numerical strategies of mitigating this problem, including changes of variables (e.g., the log-conformation [25], square-root [2], higher-root [1] or eigendecomposition [62] formulations), adaptive diffusivity [24, 23], or intrinsically dissipative space discretisation [45, 36, 22, 20]. Such strategies, however, are numerically costly and are often difficult to combine with highly-accurate spectral methods due to the infinite-order nonlinearities involved in such strategies, thus leading to an infinite-order dealiasing problem. The demand for spatial accuracy often outweighs the concerns of preserving the positive-definiteness of the conformation tensor, and spectral simulations are usually performed in primitive variables thus allowing for potentially unphysical values of c\mathbf{c}. While in some cases this results in a numerical instability and a finite-time blow-up, often such unphysical simulations exhibit superficially reasonable behaviour. Therefore, to ensure physically-meaningful results, each viscoelastic simulation has to be explicitly checked to preserve the positive-definite nature of the conformation tensor (or Tr⁡c>d\operatorname{Tr}\mathbf{c} > d) everywhere in the domain at all times.

To date, however, a significant fraction of publications reporting DNS results for ET or EIT do not explicitly comment on positive-definiteness of c\mathbf{c} and it is unclear whether the checks described above have been implemented as a routine part of the numerical protocol. With the currently growing number of DNS of chaotic flows of polymeric fluids, there is a possibility that some of these simulations do not satisfy the stringent criterion described above, and are thus formally unphysical. There is therefore an urgent need to access the importance of strictly preserving the positive-definite nature of the conformation tensor and to determine whether quantitatively correct results can still be obtained in formally unphysical simulations.

Here, to determine to what extent strict preservation of positive-definiteness is required for obtaining quantitatively reliable results, we perform DNS of a model polymeric fluid driven by an externally applied pressure gradient through a straight channel. Based on our previous work, we start from the recently discovered three-dimensional narwhal states that organise ET in this system [39], and construct a series of simulations at different resolutions designed to continuously switch from strictly positive-definite conformation tensor to the situations where Tr⁡c>d\operatorname{Tr}\mathbf{c} > d is strongly violated in scattered, highly localised regions of the domain. In particular, these narrow regions are found at the interface between near-laminar regions and regions of large polymeric stress. Surprisingly, we discover that beyond a certain numerical resolution required to ensure numerical stability, the presence of unphysical values of c\mathbf{c} does not alter the statistical properties of ET as quantified, for instance, by the velocity, velocity-gradient and polymer stretch distribution functions. On the basis of these results, we provide guidance for the cost-effective numerical simulation of elastic and elasto-inertial turbulence in wall-bounded shear flows.

Governing equations and numerical details

We consider a model polymer fluid confined within a straight, three-dimensional channel formed by the gap between two parallel, infinite plates. The geometry is conveniently described using a Cartesian coordinate system (x,y,z)(x,y,z), aligned with the streamwise, wall-normal, and spanwise directions, respectively. A constant external pressure gradient is imposed along the streamwise direction to drive the flow. The fluid motion is governed by the dimensionless simplified Phan-Thien Tanner (sPTT) constitutive model (Phan-Thien & Tanner 1977) given by

∂tc+u⋅∇c−(∇u)T⋅c−c⋅(∇u)=κ∇2c−c−IWi[1+ϵTr⁡(c−I)],(1)\partial_t \mathbf{c}+\mathbf{u}\cdot\nabla\mathbf{c}-(\nabla\mathbf{u})^{T}\cdot\mathbf{c}-\mathbf{c}\cdot(\nabla\mathbf{u})=\kappa\nabla^2\mathbf{c}-\frac{\mathbf{c}-\mathbf{I}}{\mathrm{Wi}}\left[1+\epsilon\operatorname{Tr}(\mathbf{c}-\mathbf{I})\right], \tag*{(1)}
∂tu+u⋅∇u=−∇p+βRe∇2u+1−βRe Wi∇⋅c+2Rex^,(2)\partial_t\mathbf{u}+\mathbf{u}\cdot\nabla\mathbf{u}=-\nabla p+\frac{\beta}{\mathrm{Re}}\nabla^2\mathbf{u}+\frac{1-\beta}{\mathrm{Re}\,\mathrm{Wi}}\nabla\cdot\mathbf{c}+\frac{2}{\mathrm{Re}}\hat{\mathbf{x}}, \tag*{(2)}
∇⋅u=0.(3)\nabla\cdot\mathbf{u}=0. \tag*{(3)}

Here, c\mathbf{c} is the polymer conformation tensor, u\mathbf{u} is the fluid velocity, pp is the pressure, and x^\hat{\mathbf{x}} is a unit vector in the streamwise direction. The superscript (⋅)T(\cdot)^T denotes the transpose and I\mathbf{I} is the identity matrix. The governing equations are rendered dimensionless by using the following characteristic scales: all lengths are scaled with the channel half-width dd, velocities with U0U_0, time with d/U0d/U_0, stress with ηpU0/d\eta_p U_0/d, and pressure with (ηs+ηp)U0/d(\eta_s+\eta_p)U_0/d. Here, ηs\eta_s and ηp\eta_p are the solvent and polymeric contributions to the viscosity, respectively, while the velocity scale U0U_0 corresponds to the laminar centreline velocity of a Newtonian fluid with the viscosity ηs+ηp\eta_s+\eta_p at the same value of the applied pressure gradient.

The flow is controlled by several dimensionless quantities: the strength of fluid elasticity is described by the Weissenberg number, Wi=λU0/d\mathrm{Wi}=\lambda U_0/d, the relevance of inertia - by the Reynolds number, Re=ρU0d/(ηs+ηp)\mathrm{Re}=\rho U_0d/(\eta_s+\eta_p), the relative contribution of the polymers to the fluid viscosity - by the ratio β=ηs/(ηs+ηp)\beta=\eta_s/(\eta_s+\eta_p), while the degree of shear-thinning is set by ϵ\epsilon. Here, ρ\rho is the density of the fluid, λ\lambda is its Maxwell relaxation time, and DD is the diffusion coefficient of a polymer molecule. The above equations include a (global) polymer diffusion term where the stress diffusivity κ=D/dU0\kappa=D/dU_0 compares the typical distance diffused by a polymer molecule in one time unit with the channel half-width. In what follows, time intervals are reported in terms of t/Wit/\mathrm{Wi}, which in our dimensionless units corresponds to measuring physical time in terms of the Maxwell relaxation time λ\lambda.

To simulate an unbounded flow in the streamwise and spanwise directions, these directions are set to be periodic, with the domain lengths Lx=Lz=10L_x=L_z=10. At the walls, we impose no-slip boundary conditions for the velocity, i.e. u(x,y=±1,z,t)=0\mathbf{u}(x,y=\pm1,z,t)=0, while the boundary conditions for the conformation tensor are obtained by requiring that c\mathbf{c} at the walls is equal to the corresponding value obtained by solving the constitutive model (1) with κ=0\kappa=0 (Thomas et al. 2006).

The equations of motion, Eqs. (2.1)-(2.3), are solved numerically using an MPI-parallel fully-dealiased pseudo-spectral code developed within the Dedalus framework (Burns et al. 2020). Time-discretisation is performed using a 4th-order semi-implicit BDF scheme (Wang & [63] with the timestep 2.5×10−32.5 \times10^{-3}. The velocity, conformation tensor, and pressure fields are represented through a spectral decomposition based on Fourier-Chebyshev-Fourier modes in the streamwise, wall-normal, and spanwise directions, respectively. The spectral resolution is set by specifying the number of modes, (Nx,Ny,Nz)(N_x, N_y, N_z), used in each direction.

In this work, we focus on a single set of fluid parameters given by Wi=150\mathrm{Wi} = 150, Re=10−2\mathrm{Re} = 10^{-2}, β=0.9\beta= 0.9, ϵ=5×10−5\epsilon= 5 \times10^{-5}, and κ=5×10−5\kappa= 5 \times10^{-5}. Compared to the parameters used in [39], the current fluid is significantly more elastic and less shear thinning, intentionally making it more prone to developing numerical instabilities and making it more difficult to ensure positive-definiteness of the conformation tensor. We assess the influence of the spatial resolution on the positive-definiteness of the conformation tensor by performing a set of numerical simulations with systematically increased NxN_x, NyN_y, and NzN_z. This process was terminated when we reached the resolution that strictly preserved the PD nature of c\boldsymbol{c} at all times. The details of all simulations are given in Table 1. We note that all simulations run at resolutions below that of A1 exhibited a finite-time blow-up and were therefore excluded from the analysis.

(Nx,Ny,Nz)(N_x, N_y, N_z)Mdof/Mdof(A1)M_{dof}/M_{dof}^{(\mathrm{A1})}χ/10−4\chi/10^{-4}P(Trc<3)/10−5\mathcal{P}(\mathrm{Tr}\boldsymbol{c} < 3)/10^{-5}T/WiT/\mathrm{Wi}
A1(128,512,128)(128, 512, 128)17.26.668.0
A2(512,128,128)(512, 128, 128)13.12.915.0
A3(256,256,256)(256, 256, 256)21.21.115.0
A4(512,256,128)(512, 256, 128)22.92.815.0
A5(512,128,256)(512, 128, 256)21.1×10−31.1 \times10^{-3}1.0×10−31.0 \times10^{-3}15.0
A6(512,128,512)(512, 128, 512)40068.0

Table 1. Summary of the simulations used in this work. Here, (Nx,Ny,Nz)(N_x, N_y, N_z) describes the spatial resolution, χ\chi is the fraction of midplane points where Trc<3\mathrm{Tr}\boldsymbol{c} < 3, P(Trc<3)\mathcal{P}(\mathrm{Tr}\boldsymbol{c} < 3) is the probability of such events (see (4)-(5)), while T/WiT/\mathrm{Wi} is the total run time in polymer relaxation time units. For all datasets, snapshots have been sampled each 0.670.67 T/WiT/\mathrm{Wi} time units. The total number of degrees of freedom in a simulation, Mdof=10NxNyNzM_{dof} = 10N_xN_yN_z, corresponding to the spectral decomposition of 6 independent components of c\boldsymbol{c}, three components of u\boldsymbol{u}, and one component of pp, is compared to that of the A1 run, Mdof(A1)∼8.4×107M_{dof}^{(\mathrm{A1})} \sim8.4 \times10^7.

To ensure meaningful comparison, all simulations presented below were started from the same initial condition. To this end, we performed a long simulation (not shown) for the current set of parameters with Nx=Nz=256N_x = N_z = 256 and Ny=1024N_y = 1024, until it reached statistically-steady elastic turbulence, following the protocol from [39]. We then selected a single, positive-definite configuration from the statistically steady-state of that run and re-interpolated it on the A1-A6 spatial grids (see Table 1). As confirmed below in Figure 2, all initial conditions thus obtained are positive-definite, even at the lowest resolution considered.

Instantaneous value of the minimum polymer stretch

Figure 2. Instantaneous value of min⁡(Tr⁡c−3)\min(\operatorname{Tr}\mathbf{c}-3) as a function of time. The dashed line indicates the admissibility limit, min⁡(Tr⁡c)=3\min(\operatorname{Tr}\mathbf{c})=3.

Results

Here, we present the results of the A1-A6 runs started from the same initial condition, as discussed above. Since we are dealing with a chaotic system, small numerical differences between the initial conditions at different resolutions will amplify in the course of the simulations, and we expect the associated chaotic trajectories to lose resemblance at sufficiently long times. To quantify this behaviour, we introduce two observables, the kinetic

energy E(t)=⟨∥u(r,t)∥2/2⟩E(t) = \langle\lVert\mathbf{u}(\mathbf{r},t)\rVert^2/2\rangle, and a measure of the total polymer stretch ⟨Tr⁡c(r,t)⟩\langle\operatorname{Tr}\mathbf{c}(\mathbf{r},t)\rangle, where ⟨⋯ ⟩\langle\cdots\rangle denotes the volume average and r\mathbf{r} is the spatial position in the simulation domain. In Figure 1, we monitor the time-evolution of these quantities and confirm that the trajectories follow each other closely for t/Wi≲8t/\mathrm{Wi}\lesssim8. This timescale is in line with the typical stress correlation times observed previously by [39] in elastic turbulence at somewhat different fluid parameters.

Short-time evolution of the total kinetic energy and average polymer stretch

Figure 1. Short-time evolution of the total kinetic energy E(t)E(t) (top panel) and the average polymer stretch ⟨Tr⁡c⟩\langle\operatorname{Tr}\mathbf{c}\rangle for all datasets.

First, we focus on the short-time behaviour of the A1-A6 simulations. Although Figure 1 suggests that the global scalar quantities, such as the kinetic energy and the polymer stretch, are virtually indistinguishable between different runs, the PD properties of the corresponding conformation tensors vary significantly. This observation is quantified in Figure 2, where we plot the global minimum of Tr⁡c\operatorname{Tr}\mathbf{c} within the simulation domain as a function of time. Since a physical conformation tensor requires Tr⁡c>3\operatorname{Tr}\mathbf{c}>3 at each point in space ([30]; [49]), observing a negative value of Tr⁡c−3\operatorname{Tr}\mathbf{c}-3 implies the presence of numerical artefacts and unphysical values of c\boldsymbol{c}. As can be seen from Figure 2, the loss of positive-definiteness occurs for most of the resolutions, with only the A6 simulation remaining strictly positive-definite at all times. While the minimum value of Tr⁡c\operatorname{Tr}\boldsymbol{c} seems to be rather insensitive to the resolution in the wall-normal direction, NyN_y, sufficient numbers of Fourier modes NxN_x and NzN_z are required to ensure Tr⁡c≥3\operatorname{Tr}\boldsymbol{c} \ge3.

In addition to the degree to which the PD properties of c\boldsymbol{c} are violated, we quantify how often and where they are violated for a given resolution. As was noted in the Introduction, elastic turbulence in channel flows is characterised by spot-like coherent structures strongly localised around the channel midplane (Lellep et al. 2024). Since the largest deviations of the elastic stress from its laminar value take place around the midplane, below we focus on the analysis of the polymer stretch at the midplane, Tr⁡c(x,y=0,z,t)\operatorname{Tr}\boldsymbol{c}(x,y=0,z,t). We confirmed that all violations of positive-definiteness in our simulations happen in a narrow vicinity of the midplane plane. In Figure 3 we present the comparison of the polymer stretch for the A1 and A6 runs corresponding to the lowest and highest resolutions studied, respectively. We can see that unphysical stresses emerge in the vicinity of the stagnation regions of the coherent structures where the polymer stress is largest. To analyse the statistics of the midplane polymer stretch, in Figure 4 we present the PDF of Tr⁡c(x,y=0,z,t)\operatorname{Tr}\boldsymbol{c}(x,y=0,z,t) constructed from the values of the midplane polymer stretch for all configurations recorded up to t/Wi≲8t/\mathrm{Wi}\lesssim8. It allows us to make two important observations. Firstly, we note that the lower is the resolution, the larger is the number of spatial points where c\boldsymbol{c} loses its positive-definiteness. We quantify this by introducing the probability of observing a local non-physical value of c\boldsymbol{c},

Midplane profiles of $\operatorname{Tr}\boldsymbol{c}$ at $t/\mathrm{Wi}=2.7$ from the A1 and A6 runs

Figure 3. Midplane profiles of Tr⁡c\operatorname{Tr}\boldsymbol{c} at t/Wi=2.7t/\mathrm{Wi}=2.7 from the A1 (left panels) and A6 (right panels) runs, respectively. The bottom-row panels are magnifications of the top-row images for 3.5≤x≤53.5 \le x \le5 and 0≤z≤30 \le z \le3. Physically inadmissible points with Tr⁡c<3\operatorname{Tr}\boldsymbol{c}<3 can be seen close to the bottom-center coherent structure in the left panels.

PDF of $\operatorname{Tr}\mathbf{c}$ at the midplane

Figure 4. PDF of Tr⁡c\operatorname{Tr}\mathbf{c} at the midplane. The sampling time is that of the data shown in Figure 1 and Figure 2. The inset is a magnification of the xx-axis region [−2000,2000][-2000, 2000]. The vertical dark grey line indicates Tr⁡c=3\operatorname{Tr}\mathbf{c} = 3.

P(Tr⁡c<3)=∫−∞3dX Pc(X),(4)\mathcal{P}(\operatorname{Tr}\boldsymbol{c}<3)=\int_{-\infty}^{3}\mathrm{d}X\,P_{\boldsymbol{c}}(X), \tag*{(4)}

where PcP_c denotes the probability density function (PDF) of Tr⁡c(x,y=0,z,t)\operatorname{Tr}\mathbf{c}(x, y = 0, z, t), and the fraction of midplane points where such events occur,

χ=Nevents P(Tr⁡c<3)NxNz.(5)\chi= \frac{N_{\mathrm{events}}\,\mathcal{P}\left(\operatorname{Tr}\mathbf{c} < 3\right)}{N_xN_z}. \tag*{(5)}

The denominator corresponds to the total number of degrees of freedom in the channel midplane, and NeventsN_{\mathrm{events}} is the total number of local measurements used to construct the PDF. The short-time estimates for these quantities are reported in Table 1 for each dataset. As we can see, both P(Tr⁡c<3)\mathcal{P}(\operatorname{Tr}\mathbf{c} < 3) and χ\chi systematically decrease with an increased resolution and can be made arbitrarily small by adjusting NxN_x, NyN_y, and NzN_z. The second key observation to be made from the data shown in Figure 4 is that the PDFs of the physical values, Tr⁡c(x,y=0,z,t)>3\operatorname{Tr}\mathbf{c}(x, y = 0, z, t) > 3, are virtually indistinguishable and do not change with the resolution. This observation implies that the presence of unphysical events does not alter the global statistics of the polymer stress.

A key point to assess is the extent to which negative definite stresses alter flow structures and statistics at large and small scale. For a quantitative assessment, we therefore consider midplane velocity-field and velocity-gradient statistics. Figure 5 presents the standardised PDFs of the streamwise and spanwise velocity fluctuations, u′u' and w′w', for all datasets. Most importantly, no significant difference can be observed in the velocity-components statistics across the datasets. The streamwise velocity fluctuations are approximately Gaussian irrespective of the resolution and the presence of negative definite stresses in datasets A1-A5. In comparison, extreme fluctuations of spanwise velocity are more likely, as can be seen from the super-Gaussian, exponential tails of PDFs, which again coincide across all datasets, except for a small amount of noise in the tails. Figure 6 shows the PDFs of the longitudinal and transverse gradients of the streamwise and spanwise velocity components. Similar to the velocity statistics, apart from the expected noise in the tails, we do not observe significant differences between the datasets. That is, both velocity and velocity-gradient statistics are unaffected by the negative definite stresses present in datasets A1-A5. In summary, coarse resolution that leads to negative-definite stresses does not alter the flow and stress statistics.

Standard PDFs of the streamwise and spanwise velocity component fluctuations

Figure 5. Standard PDFs of the streamwise (left) and spanwise (right) velocity component fluctuations for all the datasets, sampled at the midplane for t/Wi∈[0.0,9.2]t/\mathrm{Wi} \in[0.0, 9.2], with the standard deviations σu\sigma_u and σw\sigma_w, respectively. The grey dashed line corresponds to a Gaussian with a zero mean and unit variance.

Standard PDFs of longitudinal and transverse gradients of the streamwise and spanwise velocity components

Figure 6. Standard PDFs of longitudinal and transverse gradients of the streamwise (left) and spanwise (right) velocity components for all the datasets, sampled at the midplane for t/Wi∈[0.0,9.2]t/\mathrm{Wi} \in[0.0, 9.2], with the standard deviations σu,x\sigma_{u,x}, σu,z\sigma_{u,z}, σw,x\sigma_{w,x} and σw,z\sigma_{w,z}, respectively. The grey dashed line corresponds to a Gaussian with a zero mean and unit variance.

The conclusion reached so far is based on short-time simulations that, by construction, visually appeared indistinguishable. We now demonstrate that the same conclusion holds in general, even for chaotic trajectories that do not resemble each other. To this end, we focus on the lowest, A1, and highest, A6, resolutions studied here, and continue the corresponding simulations to significantly larger values of t/Wit/\mathrm{Wi}. In Figure 7 we present the time-evolution of the kinetic energy and the polymer stretch, where the A1 and A6 trajectories strongly deviate from each other and thus are likely to have decorrelated for t/Wi≳8t/\mathrm{Wi} \gtrsim8. As previously reported by [39], elastic turbulence in channel flows exhibits spatio-temporal intermittency at sufficiently high Wi, with the number of localised turbulent ‘spots’ fluctuating strongly in time. This is indeed what we observe here as well. While both trajectories start from a configuration where a large portion of the midplane is occupied by turbulent spots,

Long time series of mean kinetic energy, polymer extension, and midplane visualisations for datasets A6 and A1

Figure 7. (a,b): Long time series of (a) mean kinetic energy and (b) polymer extension for datasets A6 and A1. The symbols indicate instances in time corresponding to the midplane visualisations of Tr⁡c\operatorname{Tr}\mathbf{c} for A6 and A1 shown in (c) and (d), respectively. The colour scheme and extent are the same as in Figure 3. Note the barely visible regions with Tr⁡c<3\operatorname{Tr}\mathbf{c} < 3 in (d).

as can be seen from Figure 3, at later times the number of spots decreases in both simulations, albeit at different t/Wit/\mathrm{Wi}, see Figure 7.

In the following, we focus our analysis on the time interval t/Wi≥46t/\mathrm{Wi} \ge46. For these times, both trajectories exhibit the same dynamical state with a single localised structure in the midplane, as can be seen in Figure 7, yet their spatial realisations are very different between the two runs. We repeat the analysis performed above for the early-time trajectories and demonstrate that the A1 and A6 runs are sampling two very similar statistical states despite the latter simulation having four times the number of the degrees of freedom of the former. This difference in the resolution becomes even more substantial along the xx-zz midplane where A6 is 16 times more resolved than A1.

In Figure 8 we present the PDF of the polymer stretch, similar to Figure 4, and observe that while the A1 run occasionally exhibits non-physical values at high-stress regions of the channel midplane (see Figure 7), the A6 run remains strictly positive-definite at all times. For the physical values of Tr⁡c\operatorname{Tr}\mathbf{c}, both PDF are virtually the same, sharing a peak located at Tr⁡c≈1500\operatorname{Tr}\mathbf{c} \approx1500 which is absent from the short-time PDF, see Figure 4. Since the latter is associated with a large number of the midplane coherent structures, we conclude that the presence of such a peak is a signature of a single turbulent spot.

PDFs of $\mathrm{Tr}\,\mathbf{c}-3$ at the midplane for A1 and A6 in the single-structure state

Figure 8. PDFs of Tr c−3\mathrm{Tr}\,\mathbf{c}-3 at the midplane for A1 and A6 in the single-structure state, sampled for 46≤t/Wi≤6046 \le t/Wi \le60. The vertical dark grey line indicates Tr c=3\mathrm{Tr}\,\mathbf{c}=3. The inset shows the PDF for values in an interval around zero.

Figure 9 presents standard PDFs of streamwise and spanwise velocity fluctuations. In both cases, the PDFs are qualitatively similar, albeit slightly less so than for the early-time results shown in Figure 5. A striking difference can be observed between the u′u' PDFs in the space-filling and the single-structure state by comparison of the left panels of Figure 5 and Figure 9. While u′u' is Gaussian distributed in the space-filling case, in the single-structure case the PDF is clearly bimodal and has non-zero skewness. In contrast, the spanwise velocity component PDF in the single-structure case shows similar slightly super-Gaussian tails as in the space-filling case, as can be seen by comparison of the right panels of Figure 5 and Figure 9. The physical origin of the deviation from Gaussian statistics for the streamwise velocity fluctuations in the single-structure state, as opposed to the space-filling state, will be investigated in future work.

Standard PDFs of the streamwise and spanwise velocity component fluctuations

Figure 9. Standard PDFs of the streamwise (left) and spanwise (right) velocity component fluctuations for A1 and A6 in the single-structure state, with the standard deviations σu\sigma_u and σw\sigma_w, respectively. The sampling time corresponds to that of Figure 8. The grey dashed line corresponds to a Gaussian with a zero mean and unit variance.

Longitudinal and transverse velocity-gradient PDFs obtained from A1 and A6 for the single-structure state are presented in Figure 10. Apart from the PDFs for the A1-data having consistently noisier tails than the higher-resolved A6-data with positive definite polymer stresses, there is no significant difference between the two cases. Compared to the space-filling state shown in Figure 6, both longitudinal and transverse w′−w^{\prime-} gradients fluctuate similarly in the single-structure state. This also applies to the longitudinal gradient of the streamwise velocity component. A different between the space-filling and the single-structure state can only be discerned for the transverse u′−u^{\prime-}-gradient, whose PDF has an approximately Gaussian core in the space-filling state that is not present in the single-structure case.

Standard PDFs of the longitudinal and transverse gradients of the streamwise and spanwise velocity components for A1 and A6

Figure 10. Standard PDFs of the longitudinal and transverse gradients of the streamwise (left) and spanwise (right) velocity components for A1 and A6 in the single-structure state, with the standard deviations σu,x\sigma_{u,x}, σu,z\sigma_{u,z}, σw,x\sigma_{w,x} and σw,z\sigma_{w,z}, respectively. The sampling time corresponds to that of Figure 8. The grey dashed line corresponds to a Gaussian with a zero mean and unit variance.

Conclusions

The results presented above indicate that the behaviour of direct numerical simulations of elastic turbulence in channel flows is governed by two distinct spatial resolution thresholds. The first threshold corresponds to the minimum resolution required for stability thus yielding simulations that do not exhibit numerical, finite-time divergences. For the fluid parameters used in this study, this threshold is represented by the A1 run (see Table 1). The second threshold corresponds to the minimum resolution necessary to ensure physically admissible values of the conformation tensor at all times throughout the domain. In our case, this condition is satisfied only by the highest-resolution simulation, A6.

Simulations performed at intermediate resolutions are numerically stable yet formally unphysical, as they fail to satisfy the condition Tr⁡c>3\operatorname{Tr}\mathbf{c} > 3 pointwise at all times. Remarkably, here we showed that such simulations, runs A1 through A5, still provide an accurate description of elastic turbulence. Key observables, including the statistics of velocity and velocity-gradient fluctuations, quantities previously used to characterise elastic turbulence in both experiments ([29], [10]) and direct numerical simulations ([21]), result in PDFs that are virtually indistinguishable from those obtained in the high-resolution A6 run, which strictly preserved the physical nature of the conformation tensor.

These findings have significant implications for direct numerical simulations of chaotic flows in polymeric liquids, particularly in wall-bounded shear flows. To date, there exists only a limited number of successful simulations of such flows (e.g. [65]; [21]; [39]; [26]; [47]) largely due to the extremely high spatial resolutions required to ensure the physical nature of the conformation tensor. For instance, the A6 run presented here took 1.6⋅1061.6 \cdot10^{6} core hours to complete, requiring access to a Tier 1 High-Performance Computing facility. In contrast, the A1 run completed using only 2⋅1052 \cdot10^{5} core hours - a computational cost attainable on most modern computing clusters. Given that the A1 run yields flow structures and turbulent statistics that are virtually indistinguishable from the physically accurate A6 case, we propose that such low-resolution, formally unphysical simulations can serve as an efficient and practical tool for studying chaotic flows in polymeric liquids.

Alternatively, our findings support a hybrid-resolution strategy, motivated by the observation that the A1 and A6 runs sampled similar statistically-steady flow states. This approach would entail conducting a long, low-resolution run followed by multiple short high-resolution simulations initialised from snapshots along this trajectory. Our results suggest that only a few timesteps are needed in the high-resolution runs to eliminate unphysical regions in the conformation tensor, thereby producing a sequence of fully physical configurations suitable for further analysis.

The existence of a typical resolution required to preserve the physical nature of the polymeric stress implies the existence of an emergent microscopic lengthscale that needs to be resolved for a direct numerical simulation to be admissible, similar to the Kolmogorov scale in Newtonian turbulence. While the origin of this lengthscale in elastic turbulence is currently unknown, here we speculate that it is related to the width of the stress filament that outlines the ‘body’ and the ‘tusk’ of the narwhal state, as discussed in [47]. We defer the study of this lengthscale to our future work.

In conclusion, we hope that our results will encourage a broader range of studies on elastic turbulence through direct numerical simulations, particularly those using (pseudo-)spectral methods. The computational strategies discussed above should help to lower the entry barrier to this field and pave the way for a wider exploration of elastic turbulence and its mechanism.

Acknowledgement.

The authors would like to thank EPSRC and UKRI for the computational time made available on the UK supercomputing facility ARCHER2 via the UK Turbulence Consortium (EP/X035484/1) and the ‘Access to High Performance Computing Facilities’ call (APP95424). Support from the ARCHER2 team (https://www.archer2.ac.uk) are gratefully acknowledged.

For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission.

Declaration of interests.

The authors report no conflicts of interest.

References

  1. [1]Alves, MA, Oliveira, PJ & Pinho, FT 2021 Numerical methods for viscoelastic fluid flows. Annu. Rev. Fluid Mech. 53, 509–541.
  2. [2]Balci, Nusret, Thomases, Becca, Renardy, Michael & Doering, Charles R. 2011 Symmetric factorization of the conformation tensor in viscoelastic fluid models. J. Non-Newtonian Fluid Mech. 166, 546–553.
  3. [4]Beneitez, Miguel, Page, Jacob, Dubief, Yves & Kerswell, Rich R. 2024a Multistability of elasto-inertial two-dimensional channel flow. J. Fluid Mech. 981, A30.
  4. [5]Beneitez, M., Page, J., Dubief, Y. & Kerswell, R. R. 2024b Transition route to elastic and elasto-inertial turbulence in polymer channel flows. Phys. Rev. Fluids 9, 123202.arxiv.org/abs/2408.11508
  5. [6]Beneitez, Miguel, Page, Jacob & Kerswell, Rich R. 2023 Polymer diffusive instability leading to elastic turbulence in plane Couette flow. Phys. Rev. Fluids 8, L101901.
  6. [7]Berti, S., Bistagnino, A., Boffetta, G., Celani, A. & Musacchio, S. 2008 Two-dimensional elastic turbulence. Phys. Rev. E 77, 055306.
  7. [8]Berti, S. & Boffetta, G. 2010 Elastic waves and transition to elastic turbulence in a two-dimensional viscoelastic Kolmogorov flow. Phys. Rev. E 82, 036314.
  8. [9]Bird, R. B., Curtiss, C. F., Armstrong, R. C. & Hassager, O. 1987 Dynamics of polymeric liquids, 2nd edn., vol. 2. Kinetic theory. New York: Wiley.
  9. [10]Burghelea, Teodor, Segre, Enrico & Steinberg, Victor 2007 Elastic turbulence in von Karman swirling flow between two disks. Phys. Fluids 19, 053104.
  10. [12]Buza, Gergely, Beneitez, Miguel, Page, Jacob & Kerswell, Rich R. 2022 Finite-amplitude elastic waves in viscoelastic channel flow from large to zero Reynolds number. J. Fluid Mech. 951, A3.
  11. [13]Castillo Sánchez, Hugo A., Jovanović, Mihailo R., Kumar, Satish, Morozov, Alexander, Shankar, V., Subramanian, Ganesh & Wilson, Helen J. 2022 Understanding viscoelastic flow instabilities: Oldroyd-B and beyond. J. Non-Newtonian Fluid Mech. 302, 104742.
  12. [14]Chaudhary, Indresh, Garg, Piyush, Shankar, V & Subramanian, Ganesh 2019 Elasto-inertial wall mode instabilities in viscoelastic plane Poiseuille flow. J. Fluid Mech. 881, 119–163.
  13. [15]Chaudhary, Indresh, Garg, Piyush, Subramanian, Ganesh & Shankar, V. 2021 Linear instability of viscoelastic pipe flow. J. of Fluid Mech. 908, A11.
  14. [16]Choueiri, George H, Lopez, Jose M & Hof, Björn 2018 Exceeding the asymptotic limit of polymer drag reduction. Phys. Rev. Lett. 120, 124501.
  15. [17]Choueiri, George H., Lopez, Jose M., Varshney, Atul, Sankar, Sarath & Hof, Björn 2021 Experimental observation of the origin and structure of elasto-inertial turbulence. Proc. Natl. Acad. Sci. U.S.A. 118 (45).
  16. [18]Couchman, Miles M.P., Beneitez, Miguel, Page, Jacob & Kerswell, Rich R. 2024 Inertial enhancement of the polymer diffusive instability. J. Fluid Mech. 981, A2.
  17. [19]Datta, Sujit S., Ardekani, Arezoo M., Arratia, Paulo E., Beris, Antony N., Bischofberger, Irmgard, McKinley, Gareth H., Eggers, Jens G., López-Aguilar, J. Esteban, Fielding, Suzanne M., Frishman, Anna, Graham, Michael D., Guasto, Jeffrey S., Haward, Simon J., Shen, Amy Q., Hormozi, Sarah, Morozov, Alexander, Poole, Robert J., Shankar, V., Shaqfeh, Eric S. G., Stark, Holger, Steinberg, Victor, Subramanian, Ganesh & Stone, Howard A. 2022 Perspectives on viscoelastic flow instabilities and elastic turbulence. Phys. Rev. Fluids 7, 080701.
  18. [20]Dubief, Yves, Terrapon, Vincent E & Hof, Björn 2023 Elasto-inertial turbulence. Annu. Rev. Fluid Mech. 55, 675–705.
  19. [21]Dubief, Yves, Terrapon, Vincent E. & Soria, Julio 2013 On the mechanism of elasto-inertial turbulence. Phys. Fluids 25 (11), 110817.
  20. [22]Dubief, Yves, Terrapon, Vincent E., White, Christopher M, Shaqfeh, Eric SG, Moin, Parviz & Lele, Sanjiva K. 2005 New answers on the interaction between polymers and vortices in turbulent flows. Flow Turbul. Combust. 74, 311–329.
  21. [23]Džanic, V., From, C.S. & Sauret, E. 2022a The effect of periodicity in the elastic turbulence regime. J. of Fluid Mech. 937, A31.
  22. [24]Dzanic, V., From, C.S. & Sauret, E. 2022b A hybrid lattice Boltzmann model for simulating viscoelastic instabilities. Comput. Fluids 235, 105280.DOI
  23. [25]Fattal, Raanan & Kupferman, Raz 2004 Constitutive laws for the matrix-logarithm of the conformation tensor. J. Non-Newtonian Fluid Mech. 123, 281–285.DOI
  24. [26]Foggi Rota, Giulio, Amor, Christian, Le Clainche, Soledad & Rosti, Marco Edoardo 2024 Unified view of elastic and elasto-inertial turbulence in channel flows at low and moderate Reynolds numbers. Phys. Rev. Fluids 9, L122602.DOI
  25. [27]Garg, Piyush, Chaudhary, Indresh, Khalid, Mohammad, Shankar, V. & Subramanian, Ganesh 2018 Viscoelastic pipe flow is linearly unstable. Phys. Rev. Lett. 121, 024502.DOI
  26. [28]Graham, Michael D. & Floryan, Daniel 2021 Exact coherent states and the nonlinear dynamics of wall-bounded turbulent flows. Annu. Rev. Fluid Mech. 53 (1), 227–253.DOI
  27. [29]Groisman, Alexander & Steinberg, Victor 2000 Elastic turbulence in a polymer solution flow. Nature 405 (6782), 53–55.
  28. [30]Hu, Dan & Lelièvre, Tony 2007 New entropy estimates for Oldroyd-B and related models. Commun. Math. Sci. 5, 909.
  29. [31]Hulsen, Martien A. 1988 Some properties and analytical expressions for plane flow of Leonov and Giesekus models. J. Non-Newtonian Fluid Mech. 30, 85–92.DOI
  30. [32]Hulsen, Martien A. 1990 A sufficient condition for a positive definite configuration tensor in differential models. J. Non-Newtonian Fluid Mech. 38, 93–100.DOI
  31. [33]Khalid, Mohammad, Chaudhary, Indresh, Garg, Piyush, Shankar, V. & Subramanian, Ganesh 2021a The centre-mode instability of viscoelastic plane Poiseuille flow. J. Fluid Mech. 915, A43.DOI
  32. [34]Khalid, Mohammad, Shankar, V. & Subramanian, Ganesh 2021b Continuous pathway between the elasto-inertial and elastic turbulent states in viscoelastic channel flow. Phys. Rev. Lett. 127, 134502.DOI
  33. [35]King, Jack R. C., Broadley, Henry M. & Benitez, Miguel 2026 Elasto-inertial transitions in viscoelastic flows through cylinder arrays, arXiv: 2604.05892.arxiv.org/abs/2604.05892
  34. [36]Kurganov, Alexander & Tadmor, Eitan 2000 New high-resolution central schemes for nonlinear conservation laws and convection–diffusion equations. J. Comput. Phys. 160, 241–282.
  35. [37]Larson, R. G. 1999 The structure and rheology of complex fluids. Oxford University Press.
  36. [38]Lellep, Martin, Linkmann, Moritz & Morozov, Alexander 2023 Linear stability analysis of purely elastic travelling-wave solutions in pressure-driven channel flows. J. Fluid Mech. 959, R1.
  37. [39]Lellep, Martin, Linkmann, Moritz & Morozov, Alexander 2024 Purely elastic turbulence in pressure-driven channel flows. Proc. Natl. Acad. Sci. U.S.A. 121 (9), e2318551121.
  38. [40]Lewy, Theo & Kerswell, Rich 2024 The polymer diffusive instability in highly concentrated polymeric fluids. J. Non-Newtonian Fluid Mech. 326, 105212.
  39. [41]Lewy, Theo & Kerswell, Rich R. 2025 Revisiting two-dimensional viscoelastic Kolmogorov flow: a centre-mode-driven transition. J. Fluid Mech. 1007, A55.
  40. [42]Lin, Fenghui, Liao, Zi-Mo, Zhao, Zhiye, Liu, Nansheng, Lu, Xi-Yun & Khomami, Bamin 2025 GPU acceleration of a hi-fidelity algorithm for direct numerical simulation of polymer-induced/modified turbulence. J. Non-Newtonian Fluid Mech. 342, 105437.DOI
  41. [43]Lopez, Jose M., Choueiri, George H. & Hof, Björn 2019 Dynamics of viscoelastic pipe flow at low Reynolds numbers in the maximum drag reduction limit. J. Fluid Mech. 874, 699–719.
  42. [44]Lu, Ziyin & Hof, Björn 2026 Multiple states of turbulence at vanishing inertia, arXiv: 2602.23498.arxiv.org/abs/2602.23498
  43. [45]Min, Taegee, Yoo, Jung Yul & Choi, Haecheon 2001 Effect of spatial discretization schemes on numerical solutions of viscoelastic fluid flows. J. Non-Newtonian Fluid Mech. 100, 27–47.
  44. [46]Morozov, Alexander 2022 Coherent structures in plane channel flow of dilute polymer solutions with vanishing inertia. Phys. Rev. Lett. 129, 017801.DOI
  45. [47]Morozov, Alexander, Lellep, Martin, Capocci, Damiano & Linkmann, Moritz 2025 Narwhals and their blessings: exact coherent structures of elastic turbulence in channel flows, arXiv: 2509.03175.arxiv.org/abs/2509.03175
  46. [48]Morozov, Alexander & Spagnolie, Saverio E 2015 Introduction to complex fluids. In Complex fluids in biological systems, pp. 3–52. Springer.DOI
  47. [49]Musacchio, Stefano 2003 Effects of friction and polymers on 2D turbulence. Phd thesis, Università degli Studi di Torino, Torino, Italy.
  48. [50]Nichols, Jeffrey, Guy, Robert D. & Thomases, Becca 2025 Period-doubling route to chaos in viscoelastic Kolmogorov flow. Phys. Rev. Fluids 10, L041301.
  49. [51]Owens, Robert G & Phillips, Timothy N 2002 Computational rheology. World Scientific.
  50. [52]Page, Jacob, Dubief, Yves & Kerswell, Rich R. 2020 Exact traveling wave solutions in viscoelastic channel flow. Phys. Rev. Lett. 125, 154501.DOI
  51. [54]Samanta, Debarjan, Dubief, Yves, Holzner, Markus, Schäfer, Christof, Morozov, Alexander N., Wagner, Christian & Hof, Björn 2013 Elasto-inertial turbulence. Proc. Natl. Acad. Sci. U.S.A. 110 (26), 10557–10562.
  52. [55]Shekar, Ashwin, McMullen, Ryan M., McKeon, Beverley J. & Graham, Michael D. 2020 Self-sustained elasto-inertial Tollmien–Schlichting waves. J. Fluid Mech. 897, A3.
  53. [56]Shekar, Ashwin, McMullen, Ryan M., McKeon, Beverley J. & Graham, Michael D. 2021 Tollmien-Schlichting route to elastoinertial turbulence in channel flow. Phys. Rev. Fluids 6, 093301.DOI
  54. [57]Shekar, Ashwin, McMullen, Ryan M., Wang, Sung-Ning, McKeon, Beverley J. & Graham, Michael D. 2019 Critical-layer structures and mechanisms in elastoinertial turbulence. Phys. Rev. Lett. 122, 124503.DOI
  55. [58]Sid, S., Terrapon, V. E. & Dubief, Y. 2018 Two-dimensional dynamics of elasto-inertial turbulence and its role in polymer drag reduction. Phys. Rev. Fluids 3, 011301.
  56. [59]Steinberg, Victor 2021 Elastic turbulence: an experimental view on inertialess random flow. Annu. Rev. Fluid Mech. 53, 27–58.
  57. [60]Steinberg, Victor 2022 New direction and perspectives in elastic instability and turbulence in various viscoelastic flow geometries without inertia. Low Temp. Phys. 48, 492–507.
  58. [62]Vaithianathan, T. & Collins, Lance R. 2003 Numerical approach to simulating turbulent flow of a viscoelastic polymer solution. J. Comput. Phys. 187, 1–21.
  59. [63]Wang, Dong & Ruuth, Steven J 2008 Variable step-size implicit-explicit linear multistep methods for time-dependent partial differential equations. J. Comput. Math. 26, 838–855.
  60. [64]Yerasi, Sumithra R, Picardo, Jason R, Gupta, Anupam & Vincenzi, Dario 2024 Preserving large-scale features in simulations of elastic turbulence. J. Fluid Mech. 1000, A37.
  61. [65]Zhang, Hong-Na, Li, Feng-Chen, Cao, Yang, Tomoaki, Kunugi & Yu, Bo 2013 Direct numerical simulation of elastic turbulence and its mixing-enhancement effect in a straight channel flow. Chin. Phys. B 22, 024703.
  62. [66]Zhu, Lu & Kerswell, Rich R. 2024 Early turbulence in viscoelastic flow past a periodic cylinder array, arXiv: 2410.12033.arxiv.org/abs/2410.12033

Paper details

vlnc:W_bk4bqzb6
Version
v1 · latest
Published
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