Numerical study of the squareroot conformation tensor formulation for confined and freesurface viscoelastic fluid flows
 Irineu L. Palhares Junior^{1},
 Cassio M. Oishi^{1},
 Alexandre M. Afonso^{2},
 Manuel A. Alves^{2} and
 Fernando T. Pinho^{3}Email author
https://doi.org/10.1186/s4032301500544
© Palhares Junior et al. 2016
Received: 9 June 2015
Accepted: 1 December 2015
Published: 23 February 2016
Abstract
We present a numerical study of a stabilization method for computing confined and freesurface flows of highly elastic viscoelastic fluids. In this approach, the constitutive equation based on the conformation tensor, which is used to define the viscoelastic model, is modified introducing an evolution equation for the squareroot conformation tensor. Both confined and freesurface flows are considered, using two different numerical codes. A finite volume method is used for confined flows and a finite difference code developed in the context of the markerandcell method is used for confined and freesurface flows. The implementation of the squareroot formulation was performed in both numerical schemes and discussed in terms of its ability and efficiency to compute steady and transient viscoelastic fluid flows. The numerical results show that the squareroot formulation performs efficiently in the tested benchmark problems at highWeissenberg number flows, such as the liddriven cavity flow, the flow around a confined cylinder, the crossslot flow and the impacting drop free surface problem.
Keywords
Background
Many engineering applications deal with viscoelastic (or nonNewtonian) fluids, characterizing soft materials such as polymer solutions (fluids containing polymer molecules which typically have thousands to millions of atoms per macromolecule), colloidal suspensions, gels, emulsions, or surfactants. These viscoelastic fluids can be represented by appropriate constitutive equations, that describe the rheological behavior of the material as a relation between the stress (force per unit area) and strain (a measure of deformation history) or rate of strain. Such constitutive equations depend on the structure of the fluid, and can be represented in the form of algebraic, differential, integral, or integrodifferential equations [1, 2]. Two dimensionless numbers are frequently used to represent ratios of relevant forces or ratios of time scales in viscoelastic fluid flows, namely, the Weissenberg (Wi) and Deborah (De) numbers. The Deborah number, \(De=\lambda U/H\), is the ratio between the fluid relaxation time \(\lambda \) and a flow time scale [3]. The Weissenberg number [4] is a dimensionless parameter that measures the degree of anisotropy or orientation generated by the deformation, and is defined as the ratio between elastic and viscous forces [5].
When numerical methods are applied to flows of viscoelastic fluids, the momentum and mass conservation equations are inherently coupled to the constitutive equation for the extrastress, \(\varvec{\tau }\) (or conformation tensor, \(\mathbf {A}\)). The inclusion of the constitutive equation does not only increase the total number of degrees of freedom of the problem but also modifies the type of the resulting system of governing equations [6]. Moreover, the evolutionary character of the constitutive models and the hyperbolic nature of the equations require preserving the positive definiteness of the conformation tensor [6, 7], and numerical discretization errors could, eventually, lead to the loss of such positive definiteness, resulting in a loss of topological evolutionary that can trigger Hadamard instabilities [6]. This numerical breakdown, which occurs when Wi increases, is known as the High Weissenberg Number Problem (HWNP), and has been a great challenge for those working on numerical simulations of viscoelastic fluid flows. The HWNP was first identified by the breakdown of the numerical schemes for macroscopic continuum mechanics constitutive equations. This numerical failure at moderate/low Weissenberg/Deborah numbers, was accompanied by numerical inaccuracies and lack of meshconvergent solutions, particularly when geometrical corners or stagnation points are present, due to the exponential growth of the normal stresses at such locations characterized by large deformation rates and low velocities. Therefore, in order to perform numerical simulations at high elasticity, the constitutive equation needs to be solved using an appropriate, stable, convergent and positivity preserving numerical method.
Although a definite solution to the HWNP is still an open problem in Computational Rheology, several effective stabilization methods have been developed during the past years. For direct numerical simulations in turbulent flow, Sureshkumar and Beris [8] introduced an artificial stress diffusion term into the evolution equation of the conformation tensor, leading to successful results when used with spectral methods. Vaithianathan et al. [9] developed another method that also guarantees positive eigenvalues of the conformation tensor, while preventing overextension for dumbbellbased models, such as the OldroydB, FENEP and Giesekus models. Their finite difference method (FDM) was coupled with a pseudospectral scheme for homogeneous turbulent shear flow.
In the framework of laminar flow computational rheology, and specifically for the inertialess (\(Re \sim 0\)) HWNP case, Lozinski and Owens [10] presented theoretical nonlinear (energy) estimates for the stress and velocity components in a general setting, for an OldroydB fluid. The authors worked with the configuration tensor, and derived a method that guarantees a wellposed evolutionary Hadamard problem.
Fattal and Kuperfman [11, 12] proposed a reformulation of the constitutive laws which describe viscoelastic fluids, using a formulation based on a tensorial transformation of the conformation tensor, a method known as logconformation, which reformulates the constitutive equation using the natural logarithm of the positivedefinite conformation tensor, thus linearizing the exponential stress growth in regions near singularities. Lee and Xu [13] presented a class of positivity preserving discretization schemes applied for ratetype viscoelastic constitutive equations, using a semiLagrangian approach and a finite element method (FEM), based on the observation that the ratetype constitutive equations can be cast into the general form of the Riccati differential equations, and demonstrated that their method is secondorder accurate in both time and space. Cho [14] proposed a vector decomposition of the Maxwelltype evolution equations of the conformation tensor. In his transformation, the vectorized equations were considered as the sum of dyadics of the conformation tensor. This vector decomposition preserved the positive definiteness of the conformation tensor and avoided the solution of the eigenvalue problem at every calculation step, decreasing the computation cost. Nevertheless, in a generic 3D simulation, the vector decomposition requires the calculation of nine components instead of the six independent components as in the logconformation tensor approach, thus limiting its efficiency in 3D numerical calculations, when compared with the logconformation transformation approach. Balci et al. [15] proposed a method in which the square root of the conformation tensor is used for OldroydB and FENEP models. They derive an evolution equation for the square root of the conformation tensor, by taking advantage of the fact that the positive definite symmetric polymer conformation tensor possesses a unique symmetric squareroot tensor that satisfies a closedform evolution equation. Balci et al. [15] claimed that their method can be easily implemented in numerical simulations, because it does not require the determination of eigenvectors and eigenvalues of the conformation tensor at every time step, resulting in a significant reduction of the computational time. Afonso et al. [16] proposed a generic framework with applications to a wide range of matrix transformations of the conformation tensor evolution equation. In order to present a robust algorithm for solving steady solutions at high Weissenberg number flows, Saramito [17] used a nonsingular logconformation formulation based on the resolution by a Newton method. Another modification in the original logconformation formulation was proposed in [18] where a fullyimplicit method is used to solve a new constitutive equation. The application of all these stabilization methods invariably has showed good stability properties for solving challenging problems in viscoelastic flows [19–26].
In the present study, we employ the stabilization method proposed by Balci et al. [15] for confined and freesurface flows, using two different numerical codes and methods. A finite volume method [23, 27–29] (FVM) is used for confined flows and a finite difference code developed in the context of the MarkerAndCell (MAC) method [30, 31] is used for confined and freesurface flows. The squareroot formulation is implemented in both numerical schemes and the results obtained are discussed in terms of stability and efficiency to compute transient viscoelastic fluid flows.
The remainder of the paper is organized as follows. The governing equations are discussed in Sect. “Governing equations”. Section “Squareroot conformation tensor methodology” describes the mathematical formulation of the symmetric square root representation of the conformation tensor and its application to the constitutive models used in the present study. The numerical implementation of the algorithm is presented in Sect. “Numerical method”, for both the finite difference and the finite volume methods used. The validations of the numerical formulations are presented in Sect. “Validation: laminar liddriven cavity flow”. Results and discussion of the numerical simulations of flow problems at highWeissenberg number flows, such as the confined liddriven cavity flow, the flow around a confined cylinder, the crossslot flow and the impacting drop problem, are presented in Sect. “Applications”. Finally, the main conclusions from the study are summarized in Sect. “Conclusions”.
Governing equations

Inlets: The normal velocity component is specified while the tangential velocity component is set to zero. Moreover, the extrastress tensor \(\varvec{\tau }\) is computed assuming fullydeveloped flow conditions. Once the value of \(\varvec{\tau }\) is imposed, the conformation tensor can be obtained from Eq. (4).

Outlets: The homogeneous Neumann conditions are employed for the velocity field and the extrastress tensor.

Walls: The noslip condition is used (\(\mathbf {u}=\mathbf {u}_{wall}\)) for the velocity field.

Moving free surfaces: In the absence of surface tension effects, the normal and tangential components of the total stress must be continuous across any free surface, i.e.,$$\begin{aligned}&\mathbf n \cdot \varvec{\sigma }\cdot \mathbf n ^{\mathrm{T}}=0, \end{aligned}$$(5)where \(\varvec{\sigma }\) is the total stress tensor, given by$$\begin{aligned}&\mathbf m \cdot \varvec{\sigma }\cdot \mathbf n ^{\mathrm{T}}=0, \end{aligned}$$(6)and \(\mathbf D \) is the rate of deformation tensor defined by \(\mathbf D = \frac{1}{2}(\nabla \mathbf u + \nabla \mathbf u ^{T} )\), with \(\left( \nabla \mathbf u \right) _{i,j} = {\partial u_{i}} / {\partial x_{j}}\).$$\begin{aligned} \varvec{\sigma }=p\mathbf I +\beta \frac{2}{Re}{} \mathbf D +\varvec{\tau }, \end{aligned}$$(7)
In Eqs. (5) and (6), \(\mathbf n \) represents a unit vector normal to the free surface and pointing outwards, and \(\mathbf m \) is a unit vector tangent to the free surface. Equation (3) is enforced at free surface cells for computing the conformation tensor, and after this, \(\varvec{\tau }\) is directly obtained from Eq. (4).
Squareroot conformation tensor methodology
Numerical method
Overview of the finite difference code
The MACtype scheme used in the present paper was firstly introduced in [30] (see also [31]). In this section, we describe the corresponding modifications to introduce the squareroot conformation tensor in the context of the finite difference methodology.
Initially, a pressuresegregation method is applied in order to uncouple the velocity and pressure fields. This projection method is widely used for solving the NavierStokes Eqs. (1) and (2) [32].
 1.
Given \(\mathbf{Q }^{(n)}\), \(\mathbf{u }^{(n)}\) and \(\mathbf{G }^{(n)}\), solve the evolution Eq. (27) to obtain the intermediate value \({\overline{\mathbf{Q }}}^{(n+1)}\);
 2.
From \({\overline{\mathbf{Q }}}^{(n+1)}\), compute the intermediate value of the conformation tensor \({\overline{\mathbf{A }}}^{(n+1)}\) using Eq. (8). Then, Eq. (4) is directly applied to determine the extrastress tensor \({\overline{\varvec{\tau }}}^{(n+1)}\) which is used to obtain an intermediate velocity from the momentum equation (substep of the projection method).
 3.
Apply the projection method to obtain the final velocity \(\mathbf{u }^{(n+1)}\) and pressure \(p^{(n+1)}\) fields (more details can be found in [30]).
 4.
Construct the matrix \({\overline{\mathbf{G }}}^{(n+1)}\) (see Eq. 28).
 5.
Calculate the final value of the squareroot matrix \(\mathbf{Q }^{(n+1)}\) using Eq. (26). At this stage, the final conformation and extrastress tensors are computed using Eqs. (8) and (4), respectively.
Remark 1
To start the algorithm, homogeneous isotropic initial data are imposed, i.e., \(\mathbf{Q }^{2}_{t=t_0} = \mathbf I \).
Remark 2
Overview of the finite volume code
The squareroot formulation was also implemented in a Finite Volume Method (see [23, 27–29], for more details). This FVM uses collocated nonorthogonal meshes, central differences for the discretization of diffusive terms, a first or second order backward implicit time discretization, and the SIMPLEC [33] algorithm to ensure simultaneously the momentum balance and mass conservation.
These transformations are a necessary step towards using a general FVM based on the collocated mesh arrangement, as described in Oliveira et al. [27] and Alves et al. [28, 29]. In the discretization of Eq. (31), the \(\beta _{lk}\) coefficients are replaced by area components of the surface whose normal vector points towards direction l, the Jacobian J is replaced by the cell volume V, and the derivatives \(\partial \)/\(\partial \zeta _{l}\) become differences between values along direction l [27].
 1.
First, the squareroot tensor components \(Q_{ij}\) are initialized;
 2.
The components of the antisymmetric tensor \(G_{ij}\) are calculated with the information of the known velocity gradient using Eq. (19);
 3.
The components of the symmetric tensor \(K_{ij}\) are calculated using Eq. (17) with the information of the velocity gradient \(G_{ij}\) and \(Q_{ij}\);
 4.
The evolution equation for \(\mathbf {Q}\) is solved implicitly (Eq. 32);
 5.
The conformation tensor matrix \(\mathbf {A}\) is recovered using Eq. (8);
 6.
The new extrastress components \(\tau _{ij}\) are now calculated from the conformation tensor using Eq. (4);
 7.
The momentum equations are solved for each velocity component, \(u_{i}\) to determine the new velocity field;
 8.
As generally the velocity components do not satisfy the continuity equation, this step of the algorithm involves a correction to \(u_{i}\) and to the pressure field p, so that the updated velocity field \(u_{i}\) and the corrected pressure field p satisfy simultaneously the continuity and the momentum equations. This part of the algorithm remains unchanged and is described in detail in Oliveira et al. [27];
 9.
Steps 2–8 are repeated until convergence is reached (steadystate calculations), or until the desired final time is reached (unsteady calculations)
Remark 3
The CUBISTA highresolution scheme [29] is used to discretize the advection terms of the governing equations in the FDM and FVM.
Validation: laminar liddriven cavity flow
In this section, we present the validation of both numerical implementations, using the liddriven cavity benchmark flow problem [34]. This flow is generated by the motion of one or more walls of a closed cavity. There are several experimental and numerical studies involving liddriven cavity flows, mainly with Newtonian fluids [35], whereas for viscoelastic fluids, the interest is fairly recent, and mainly used to assess numerical methods for highly elastic flows [12, 36–40] which requires a regularization on the lid motion due to the singular behavior near the corners.
The standard problem relies on the following regularized parabolic profile for the top lid, \(u(x,t)=8[1+\tanh (8t4)]x^{2}(1x)^{2}\). The remaining cavity walls are stationary and the noslip boundary condition is imposed in the four walls. We have fixed the Reynolds number, \(Re=0.01\), and the solvent viscosity ratio, \(\beta =0.5\). To assess the mesh convergence of both the finite difference and finite volume methods, the cavity flow was simulated using two uniform meshes: M1 (\(h=\min (\Delta x,\Delta y)=\frac{1}{128},\,128\times 128\) cells) and M2 (\(h=\min (\Delta x,\Delta y)=\frac{1}{256},\,256\times 256\) cells).
For all figures in this section, the profiles of uvelocity and of the nonNewtonian \(\tau _{xx}\) component are plotted along the vertical line \(x=0.5\) while the vvelocity component is reported at the horizontal line \(y=0.75\). The origin of the Cartesian coordinate system is placed at the lower left corner of the square cavity, and the moving wall is located at \(y=1\), from \(x=0\) to \(x=1\).
In order to assess the implementation of the codes, we compare our results of the liddriven cavity flow with those of Fattal and Kupferman [12] and Pan et al. [36]. The results are presented in Figs. 1 and 2 for \(Wi=1\) and \(Wi=2\), respectively, and confirm that the results are independent of the numerical method when the meshes are sufficiently fine, as expected.
In Fig. 1, we plot the velocity components u and v for \(Wi=1\) at the dimensionless time \(t=40\), showing that the squareroot formulation converges to the literature results. For \(Wi=2\), we have simulated until \(t=80\), and according to Fig. 2 the results are again in good agreement with the literature.
In order to provide additional data for this benchmark problem, we have plotted in Fig. 3 the \(\tau _{xx}\) component profile along the vertical line \(x=0.5\), illustrating the convergence with mesh refinement of the squareroot formulation for \(Wi=1\) and \(Wi=2\). Note that in Fig. 3, the extrastress is normalized as \(\tau _{xx}' / (\frac{\eta _{0}U}{L})\), where U is the maximum velocity of the lid.
Applications
Confined flow: 2D flow around a cylinder
The twodimensional (2D) flow around a confined cylinder in a channel is an important benchmark test in computational rheology [41]. It is representative of fundamental flow dynamics of viscoelastic fluids around submerged solid bodies and it can be encountered in many engineering processes, namely in the food industry, in composite and textile coating operations and flows through porous media. From a numerical point of view, this flow is considered a smooth flow, due to the absence of geometrical singularities, such as a reentrant or salient corners found in entry flows. However, it also introduces important challenges associated with the development of thin stress layers on the cylinder wall and large normal stresses along the centerline in the cylinder rear wake, imposing a limiting Weissenberg number for which steadystate solutions can be obtained. For all these reasons this flow was selected as a benchmark problem in computational rheology in the VIIIth international workshop on numerical methods for nonNewtonian flows [41].
In this problem, the ratio of channel halfwidth h to cylinder radius R is set equal to 2, which corresponds to the benchmark 50 % blockage ratio case [41]. The Weissenberg number based on the fluid relaxation time, \(\lambda \), the average inlet velocity, U, and the cylinder radius, R, is here defined as \(Wi=\lambda U/R\). The computational domain is 200R long, with 99R upstream and 99R downstream of the forward and rear stagnation points of the cylinder, respectively. The downstream length is large enough for the flow to become fullydeveloped and to avoid any effect of the Neumann outflow boundary condition upon the flow in the vicinity of the cylinder. Zero axial gradients are applied to all variables, including the pressure gradient, at the outlet plane. Noslip boundary conditions are imposed at both the cylinder surface (\(r=R\): \(u=0\), \(v=0\)) and the channel walls (\(y'=\pm h\): \(u=0\), \(v=0\)).
In this confined flow problem, the simulations were restricted to the squareroot formulation using the FVM. Simulations were performed in two meshes MC1 and MC2, mapping the complete flow domain, i.e., no symmetry boundary condition was imposed along the centerline. Mesh MC1 is composed of 30 cells placed radially and nonuniformly from the cylinder to the channel wall, leading to a minimum cell spacing normalized with the cylinder radius along the radial (\(\Delta r\)) and the azimuthal (\(\Delta s = \Delta \theta \)) directions, of 0.008 and 0.0012, respectively. In mesh MC2, those dimensions are reduced to 0.004 and 0.0006, respectively. Mesh MC2 is the same as \(M60_{WR}\) used in previous works [16, 23, 42] and corresponds to highly refined meshes along the wake.
All the calculations were carried out at a low Reynolds number, \(Re=\rho UR/\eta =0.01\). The viscosity ratio was fixed at \(\beta =0.59\), a value that characterizes the MIT Boger fluid used in previous experiments [41]. All simulations started from a quiescent state velocity field (\(u=v=0\)), meaning that \(\mathbf {A}\left( \mathbf {x},t=0 \right) =\mathbf {I}\) and for the squareroot tensor, \(\mathbf {Q}\left( \mathbf {x},t=0 \right) =\mathbf {I}\).
In order to assess the stability of the squareroot method for high Wi, we carried out additional simulations at a higher value of elasticity, \(Wi=1\). For this Wi number, the nonstabilized version of the conformation tensor formulation diverges, while the use of the squareroot formulation allowed the simulation of unsteady flows.
Confined flow: 2D flow in a crossslot
The crossslot flow, illustrated in Fig. 7a), and other strong extensional flows, such as the four roll mill and opposed jet apparatus, have been extensively investigated because of the need to develop methods for measuring the extensional viscosity of polymer solutions [46].
It is now well established that viscoelastic flows often generate purelyelastic flow instabilities, which lead to unsteady flows, even under creeping flow conditions. Of particular relevance to the study of elastic instabilities is the experimental observation of flow instabilities in a crossslot microchannel by Arratia et al. [47], which motivated the numerical work of Poole et al. [48] on the twodimensional crossslot flow of an upperconvected Maxwell (UCM) fluid under low Reynolds number flow conditions. Poole et al. [48] were able to predict the first type of flow instability (steady and asymmetric flow) in a twodimensional crossslot channel flow of an UCM fluid, which led to a reduction of pressure loss, and reported also the stabilizing effect of inertia. This flow was proposed recently as a benchmark problem by Cruz et al. [49], due to its conceptually simple geometry and well defined steady asymmetric flow instability. The benchmark results were presented for a wide range of differential constitutive equations, namely the UCM, OldroydB and PhanThien and Tanner models, showing that, in the limit of negligible inertia, i.e. when Re approaches zero, the flow exhibits two types of purelyelastic instabilities for fluids with high extensional viscosity. Above a first critical value of the Deborah number, \(De_{crit}\), the steady flow becomes spatially asymmetric, even though the geometry is perfectly symmetric; at higher De a second instability occurs and the flow becomes timedependent. Recently, Cruz and Pinho [50] presented a general analytical solution for the twodimensional steady planar extensional flow with wallfree stagnation point for the UCM model.
OldroydB model data (DQ and Wi), with \(\beta =1/9\), for a crossslot with sharp corners
De \(=\) 0.1  De \(=\) 0.2  De \(=\) 0.3  De \(=\) 0.35  De \(=\) 0.4  De \(=\) 0.42  

FVM  
\(DQ_{log}\)  0.000  0.000  0.000  0.000  0.555  0.652 
\(DQ_{sqroot}\)  0.000  0.000  0.000  0.000  0.554  0.653 
FDM  
\(DQ_{log}\)  0.000  0.000  0.000  0.000  0.563  0.660 
\(DQ_{sqroot}\)  0.000  0.000  0.000  0.000  0.563  0.661 
DQ Ref. [49]  
FVM  0.000  0.000  0.000  0.000  0.562  0.666 
\(Wi_{log}\)  0.317  0.517  0.589  0.601  0.506  0.493 
\(Wi_{sqroot}\)  0.317  0.517  0.589  0.600  0.505  0.493 
FDM  
\(Wi_{log}\)  0.313  0.516  0.583  0.576  0.500  0.486 
\(Wi_{sqroot}\)  0.319  0.516  0.579  0.587  0.502  0.489 
Wi Ref. [49]  0.322  0.522  0.591  0.602  0.497  0.487 
Free surface flow: the impacting drop problem
In this problem, we compute the time evolution of the shape of a 2D drop that falls under the action of gravity from a distance H above a rigid plate. Figure 9 illustrates the drop shape and uvelocity field for different phases of the transient motions for \(Wi= \lambda U/D = 1\), where U is the initial velocity of the drop at height H, and D is the initial diameter of the drop.
Typical difficulties in this simulation are the numerical instability for highly elastic flows in following the large deformation of the interface between liquid and air making the timedependent impacting drop problem a popular benchmark for free surface flows of viscoelastic fluids [51–54]. However, the literature is scarce for highWeissenberg number flows due to the inherent numerical difficulties. Therefore, in order to obtain more insight into the physics of this free surface flow, we have used the squareroot formulation in the context of FDM for solving this problem at high Wi flows.
The parameters used in this study were: \(D = 0.02\,\mathrm {m}\), \(U = 1\,\mathrm {m/s}\) and \(g =9.81\,\mathrm {m/s^2}\). We consider a drop with an initial velocity of \(v_0=1\,\mathrm {m/s}\) at a distance \(H = 0.04\,\mathrm {m}\) measured from the center of the drop to the impacting rigid plate.
To assess the influence of fluid viscoelasticity on the impacting drop problem, Fig. 11 presents, for mesh M2, the time evolution of the drop width for different values of Wi. In this figure, we have also included the results for the Newtonian fluid (\(Wi=0\)) considering \(Re=5\) and \(Re=50\). We can observe that increasing Wi leads to an increase of the drop width with a more efficient spreading, as also observed when comparing in Fig. 9. In order to compare the squareroot stabilization results with the logconformation method in this free surface benchmark, we have also plotted in Fig. 11 the drop width obtained by the logconformation formulation at \(Wi=1\) and 200. From this figure, we can conclude that the squareroot formulation produces similar results as the logconformation method.
It might seem counterintuitive, but for smaller Wi the normal stress growth after the impact is faster (due to the smaller relaxation time compared with the deformation time) and this leads to some recoil effect observed due to the viscoelasticity. Increasing Wi leads to larger normal stresses, but occurring only for larger times and this reduces the resistance due to normal stress in the initial times after the impact. As a consequence, for \(Wi \ge 50\) the drop diameter does not suffer a significant overshoot as observed for lower Wi. In particular, for the largest Wi the polymer stresses do not have time to grow during the fast deformation process after the drop impact and in practice the fluid response is identical to the Newtonian solvent with viscosity \(\beta \eta _0\), thus the curve for \(Wi=500\) approaches the Newtonian case (\(Wi=0\)) with an effective Reynolds number of \(Re= 5/ \beta = 50\), as shown in Fig. 11.
Conclusions
This work presented the application of the squareroot conformation tensor stabilization method for computation of confined and freesurface flows of viscoelastic fluids. The results showed the ability of this methodology for efficient and stable computations of highWeissenberg number flows using the OldroydB model. The numerical studies include the liddriven cavity, the flow around a confined cylinder, the flow in a crossslot and the impacting drop problem.
Both the FDM and FVM implementations of the squareroot formulation presented stable and efficient results for all the 2D confined and free surface benchmark flows studied in this work. When compared with the nonstabilized conformation tensor version, the squareroot formulation allowed calculations at higher Weissenberg numbers. When both methods converge to a steady solution, the use of the squareroot formulation provides similar results as those of the nonstabilized conformation tensor version. When compared to other stabilized methodologies, the accuracy and stability of the squareroot formulation is found to be similar to the logconformation approach or with the kernelconformation with the equivalent transformation functions. From the numerical performance point of view, and by design, the squareroot formulation can introduce some reduction of the computational time since it does not require the computation of eigenvectors and eigenvalues of the conformation tensor at every time step needed in both the logconformation or kernelconformation approaches, thus offering an alternative methodology for addressing the HighWeissenberg Number Problem. Nevertheless, this CPU time reduction can be compensated in the logconformation and kernelconformation methodologies using efficient techniques to compute the conformation tensor diagonalization.
Declarations
Author’s contributions
ILPJ and CMO developed the finite difference code and performed the corresponding simulations. AMA, MMA and FTP worked on the development of the finite volume code and the corresponding simulations were carried out by AMA. FTP and MMA defined the cases for investigation and coordinated the two teams’ work. All authors participated in the writing, review and revision of the manuscript. All authors read and approved the final manuscript.
Acknowledgements
I. L. Palhares Junior and C. M. Oishi acknowledge the financial support of FAPESP (Fundação de Amparo à Pesquisa do Estado de São Paulo), Grants 2014/173482 and 2013/073750, and CNPq (Conselho Nacional de Desenvolvimento Cientifico e Tecnologico), Grants 473589/20133 and 309514/20134. A. M. Afonso acknowledges Fundação para a Ciência e a Tecnologia (FCT) for financial support through the scholarship SFRH/BPD/75436/2010. M. A. Alves acknowledges funding from the European Research Council (ERC) under the European “Ideas” specific programme of the 7th framework programme (Grant agreement 307499). F. T. Pinho acknowledges funding from project MEC/MCTI/CAPES/CNPq/FAPs No 61/2011 within the framework of the Brazilian “Programa Ciência sem Fronteiras—Bolsas no País”.
Competing interests
The authors declare that they have no competing interests.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 Bird RB, Armstrong RC, Hassager O. Dynamics of polymeric liquids. Fluid mechanics, vol. 1. 2nd ed. New York: Wiley; 1987.Google Scholar
 Bird RB, Curtiss CF, Armstrong RC, Hassager O. Dynamics of polymeric liquids. Kinetic theory, vol. 2. 2nd ed. New York: Wiley; 1987.Google Scholar
 Reiner M. The Deborah number. Phys Today. 1964;17(1):62.View ArticleGoogle Scholar
 White JL. Dynamics of viscoelastic fluids, melt fracture, and the rheology of fiber spinning. J Appl Polym Sci. 1964;8:2339–57.View ArticleGoogle Scholar
 Poole RJ. The Deborah and Weissenberg numbers. The British Society of Rheology. Rheol Bull. 2012;53:32–9.MathSciNetGoogle Scholar
 Joseph DD. Fluid dynamics of viscoelastic liquids. New York: Springer; 1990.View ArticleMATHGoogle Scholar
 Dupret F, Marchal JM. Loss of evolution in the flow of viscoelastic fluid. J Non Newton Fluid Mech. 1986;20:143–71.View ArticleMATHGoogle Scholar
 Sureshkumar R, Beris AN. Effect of artificial stress diffusivity on the stability of numerical calculations and the flow dynamics of timedependent viscoelastic flows. J Non Newton Fluid Mech. 1995;60(1):53–80.View ArticleGoogle Scholar
 Vaithianathan T, Robert A, Brasseur JG, Collins LR. An improved algorithm for simulating threedimensional, viscoelastic turbulence. J Non Newton Fluid Mech. 2006;140(1):3–22.View ArticleMATHGoogle Scholar
 Lozinski A, Owens RG. An energy estimate for the OldroydB model: theory and applications. J Non Newtonian Fluid Mech. 2003;112(2–3):161–76.View ArticleMATHGoogle Scholar
 Fattal R, Kupferman R. Constitutive laws for the matrixlogarithm of the conformation tensor. J Non Newton Fluid Mech. 2004;123(2–3):281–5.View ArticleMATHGoogle Scholar
 Fattal R, Kupferman R. Timedependent simulation of viscoelastic flows at high Weissenberg number using the logconformation representation. J Non Newton Fluid Mech. 2005;126:23–37.View ArticleMATHGoogle Scholar
 Lee YJ, Xu J. New formulations, positivity preserving discretizations and stability analysis for nonNewtonian flow models. Comput Methods Appl Mech Eng. 2006;195(9–12):1180–206.MathSciNetView ArticleMATHGoogle Scholar
 Cho KS. Vector decomposition of the evolution equations of the conformation tensor of Maxwellian fluids. Korea Aust Rheol J. 2009;21(2):143–6.Google Scholar
 Balci N, Thomases B, Renardy M, Doering CR. Symmetric factorization of the conformation tensor in viscoelastic fluid models. J Non Newton Fluid Mech. 2011;166:546–53.View ArticleMATHGoogle Scholar
 Afonso AM, Pinho FT, Alves MA. The kernelconformation constitutive laws. J Non Newton Fluid Mech. 2012;167–168:30–7.MATHGoogle Scholar
 Saramito P. On a modified nonsingular logconformation formulation for JohnsonSegalman viscoelastic fluids. J Non Newton Fluid Mech. 2014;211:16–30.View ArticleGoogle Scholar
 Knechtges P, Behr M, Elgeti S. Fullyimplicit logconformation formulation of constitutive laws. J Non Newton Fluid Mech. 2014;214:78–87.View ArticleGoogle Scholar
 Martins FP, Oishi CM, Afonso AM, Alves MA. A numerical study of the kernelconformation transformation for transient viscoelastic fluid flows. J Comput Phys. 2015;302:653–73.MathSciNetView ArticleGoogle Scholar
 Kwon Y. Finite element analysis of planar 4:1 contraction flow with the tensorlogarithmic formulation of differential constitutive equations. Korea Aust Rheol J. 2004;16:183–91.Google Scholar
 Coronado OM, Arora D, Behr M, Pasquali M. A simple method for simulating general viscoelastic fluid flows with an alternate logconformation formulation. J Non Newton Fluid Mech. 2007;147:189–99.View ArticleMATHGoogle Scholar
 Guénette R, Fortin A, Kane A, Hétu JF. An adaptive remeshing strategy for viscoelastic fluid flow simulations. J Non Newton Fluid Mech. 2008;153:34–45.View ArticleMATHGoogle Scholar
 Afonso A, Oliveira PJ, Pinho FT, Alves MA. The logconformation tensor approach in the finitevolume method framework. J Non Newton Fluid Mech. 2009;157:55–65.View ArticleMATHGoogle Scholar
 Jafari A, Fiétier N, Deville MO. A new extended matrix logarithm formulation for the simulation of viscoelastic fluids by spectral elements. Comput Fluids. 2010;39:1425–38.MathSciNetView ArticleMATHGoogle Scholar
 Damanik H, Hron J, Ouazzi A, Turek S. A monolithic FEM approach for the logconformation reformulation (LCR) of viscoelastic flow problems. J Non Newton Fluid Mech. 2010;165:1105–13.View ArticleMATHGoogle Scholar
 Chen X, Marschall H, Schafer M, Bothe D. A comparison of stabilization approaches for finitevolume simulation of viscoelastic fluid flow. Int J Comput Fluid Dyn. 2013;27:229–50.MathSciNetView ArticleGoogle Scholar
 Oliveira PJ, Pinho FT, Pinto GA. Numerical simulation of nonlinear elastic flows with a general collocated finitevolume method. J Non Newton Fluid Mech. 1998;79:1–43.View ArticleMATHGoogle Scholar
 Alves MA, Pinho FT, Oliveira PJ. Effect of a highresolution differencing scheme on finitevolume predictions of viscoelastic flows. J Non Newton Fluid Mech. 2000;93:287–314.View ArticleMATHGoogle Scholar
 Alves MA, Oliveira PJ, Pinho FT. A convergent and universally bounded interpolation scheme for the treatment of advection. Int J Numer Meth Fluids. 2003;41:47–75.View ArticleMATHGoogle Scholar
 Oishi CM, Martins FP, Tomé MF, Cuminato JA, McKee S. Numerical solution of the extended pompom model for viscoelastic free surface flows. J Non Newton Fluid Mech. 2011;166:165–79.View ArticleMATHGoogle Scholar
 Cuminato JA, Oishi CM, Figueiredo RA. Implicit methods for simulating low Reynolds number free surface flows: Improvements on the MACtype methods. In: Wakayama M, Anderssen RS, Cheng J, Fukumoto Y, McKibbin R, Polthier K, Takagi T, Toh KC, editors. The impact of applications on mathematics. Japan: Springer; 2014. p. 123–39.Google Scholar
 Chorin A, Marsden J. A mathematical introduction to fluid mechanics. New York: Springer; 2000.Google Scholar
 Van Doormaal JP, Raithby GD. Enhancements of the SIMPLE method for predicting incompressible fluid flows. Numer Heat Transf. 1984;7(2):147–63.MATHGoogle Scholar
 Pan F, Acrivos A. Steady flows in rectangular cavities. J Fluid Mech. 1967;28(4):643–55.View ArticleGoogle Scholar
 Shankar P, Deshpande M. Fluid mechanics in the driven cavity. Annu Rev Fluid Mech. 2000;32(1):93–136.MathSciNetView ArticleMATHGoogle Scholar
 Pan TW, Hao J, Glowinski R. On the simulation of a timedependent cavity flow of an OldroydB fluid. Int J Numer Meth Fluids. 2009;60:791–808.MathSciNetView ArticleMATHGoogle Scholar
 Yapici K, Karasozen B, Uludag Y. Finite volume simulation of viscoelastic laminar flow in a liddriven cavity. J Non Newton Fluid Mech. 2009;164(1):51–65.View ArticleMATHGoogle Scholar
 Poole R, Afonso A, Pinho FT, Oliveira P, Alves MA. Scaling of purelyelastic instabilities in viscoelastic liddriven cavity flow. In: XVIth International Workshop for numerical methods in nonNewtonian flows. 2010.Google Scholar
 Su J, Ouyang J, Wang X, Yang B, Zhou W. Lattice Boltzmann method for the simulation of viscoelastic fluid flows over a large range of Weissenberg numbers. J Non Newton Fluid Mech. 2013;194:42–59.View ArticleGoogle Scholar
 Habla F, Tan MW, Hablberger J, Hinrichsen O. Numerical simulation of the viscoelastic flow in a threedimensional liddriven cavity using the logconformation reformulation in OpenFOAM. J Non Newton Fluid Mech. 2014;212:47–62.View ArticleGoogle Scholar
 Brown RA, McKinley GH. Report on the VIIIth international workshop on numerical methods in viscoelastic flows. J Non Newton Fluid Mech. 1994;52(3):407–13.View ArticleGoogle Scholar
 Alves MA, Pinho FT, Oliveira PJ. The flow of viscoelastic fluids past a cylinder: finitevolume highresolution methods. J Non Newton Fluid Mech. 2001;97:207–32.View ArticleMATHGoogle Scholar
 Fan Y, Tanner RI, PhanThien N. Galerkin/leastsquare finiteelement methods for steady viscoelastic flows. J Non Newton Fluid Mech. 1999;84:233–56.View ArticleMATHGoogle Scholar
 Kim JM, Kim C, Ahn KH, Lee SJ. An efficient iterative solver and high precision solutions of the OldroydB fluid flow past a confined cylinder. J Non Newton Fluid Mech. 2004;123:161–73.View ArticleMATHGoogle Scholar
 Hulsen MA. Some properties and analytical expressions for plane flow of Leonov and Giesekus models. J Non Newton Fluid Mech. 1988;30:85–92.View ArticleGoogle Scholar
 Fuller GG, Cathey CA, Hubbard B, Zebrowski BE. Extensional viscosity measurements for low viscosity fluids. J Rheol. 1987;31:235–49.View ArticleGoogle Scholar
 Arratia PE, Thomas CC, Diorio JD, Gollub JP. Elastic instabilities of polymer solutions in crosschannel flow. Phys Rev Lett. 2006:144502.Google Scholar
 Poole RJ, Alves MA, Oliveira PJ. Purely elastic flow asymmetries. Phys Rev Lett. 2007;99(16):164503.View ArticleGoogle Scholar
 Cruz FA, Poole RJ, Afonso AM, Pinho FT, Oliveira PJ, Alves MA. A new viscoelastic benchmark flow: stationary bifurcation in a cross slot. J Non Newton Fluid Mech. 2014;214:57–68.View ArticleGoogle Scholar
 Cruz DOA, Pinho FT. Analytical solution of steady 2D wallfree extensional flows of UCM fluids. J Non Newton Fluid Mech. 2015;223:157–64.Google Scholar
 Rafiee A, Manzari MT, Hosseini M. An incompressible SPH method for simulation of unsteady viscoelastic freesurface flows. Int J Non Linear Mech. 2007;42:1210–23.View ArticleMATHGoogle Scholar
 Jiang T, Ouyang J, Yang B, Ren J. The SPH method for simulating a viscoelastic drop impact and spreading on an inclined plate. Comput Mech. 2010;45:573–83.MathSciNetView ArticleMATHGoogle Scholar
 Fang J, Owens RG, Tacher L, Parriaux A. A numerical study of the SPH method for simulating transient viscoelastic free surface flows. J Non Newton Fluid Mech. 2006;139:68–84.View ArticleMATHGoogle Scholar
 Oishi C, Martins FP, Tomé MF, Alves MA. Numerical simulation of drop impact and jet buckling problems using the extended pompom model. J Non Newton Fluid Mech. 2012;169–170:91–103.View ArticleGoogle Scholar