We present our simulation results for the benchmark problem of the ow past a Rudimentary Landing Gear (RLG) using a General Galerkin (G2) nite element method, also referred to as Adaptive DNS/LES. In G2 no explicit subgrid model is used, instead the compuational mesh is adaptively re ned with respect to an a posteriori error es-timate of a quantity of interest in the computation, in this case the drag force on the RLG. Turbulent boundary layers are modeled using a simple wall layer model with the shear stress at walls proportional to the skin friction, which here is assumed to be small and, therefore, can be approximated by zero skin friction. We compare our results with experimental data and other state of the art computations, where we nd good agreement in sound pressure levels, surface velocities and ow separation. We also compare with detailed surface pressure experimental data where we nd largely good agreement, apart from some local dierences for which we discuss possible explanations.
In this work we propose a new stabilized approach for solving the incompressible Navier-Stokes equations on fixed overlapping grids. This new approach is based on the partition of unity finite element method, which defines the solution fields as weighted sums of local fields, supported by the different grids. Here, the discrete weak formulation of the problem is re-set in cG(1)cG(1) stabilized form, which has the dual benefit of lowering grid resolution requirements for convection dominated flows and allowing for the use of velocity and pressure discretizations which do not satisfy the inf-sup condition. Additionally, we provide an outline of our implementation within an existing distributed parallel application and identify four key options to improve the code efficiency namely: the use of cache to store mapped quadrature points and basis function gradients, the intersection volume splitting algorithm, the use of lower order quadrature schemes, and tuning the partition weight associated with the interface elements. The new method is shown to have comparable accuracy to the single mesh boundary-fitted version of the same stabilized solver based on three transient flow tests including both 2D and 3D settings, as well as low and moderate Reynolds number flow conditions. Moreover, we demonstrate how the four implementation options have a synergistic effect lowering the residual assembly time by an order of magnitude compared to a naive implementation, and showing good load balancing properties.
For problems involving large deformations of thin structures, simulating fluid-structure interaction (FSI) remains a computationally expensive endeavour which continues to drive interest in the development of novel approaches. Overlapping domain techniques have been introduced as a way to combine the fluid-solid mesh conformity, seen in moving-mesh methods, without the need for mesh smoothing or re-meshing, which is a core characteristic of fixed mesh approaches. In this work, we introduce a novel overlapping domain method based on a partition of unity approach. Unified function spaces are defined as a weighted sum of fields given on two overlapping meshes. The method is shown to achieve optimal convergence rates and to be stable for steady-state Stokes, Navier-Stokes, and ALE Navier-Stokes problems. Finally, we present results for FSI in the case of 2D flow past an elastic beam simulation. These initial results point to the potential applicability of the method to a wide range of FSI applications, enabling boundary layer refinement and large deformations without the need for re-meshing or user-defined stabilization.
In this paper. we identify and propose solutions for several issues encountered when designing a mesh adaptation package, such as mesh-to-mesh projections and mesh database design, and we describe an algorithm to integrate a mesh adaptation procedure in a physics solver. The open-source MAdLib package is presented as an example of such a mesh adaptation library. A new technique combining global node repositioning and mesh optimization in order to perform arbitrarily large deformations is also proposed. We then present several test cases to evaluate the performances of the proposed techniques and to show their applicability to fluid-structure interaction problems with arbitrarily large deformations. Copyright (C) 2009 John Wiley & Sons, Ltd.
A unified approach for the numerical simulation of vowels is presented, which accounts for the self-oscillations of the vocal folds including contact, the generation of acoustic waves and their propagation through the vocal tract, and the sound emission outwards the mouth. A monolithic incompressible fluid-structure interaction model is used to simulate the interaction between the glottal jet and the vocal folds, whereas the contact model is addressed by means of a level set application of the Eikonal equation. The coupling with acoustics is done through an acoustic analogy stemming from a simplification of the acoustic perturbation equations. This coupling is one-way in the sense that there is no feedback from the acoustics to the flow and mechanical fields. All the involved equations are solved together at each time step and in a single computational run, using the finite element method (FEM). As an application, the production of vowel [i] has been addressed. Despite the complexity of all physical phenomena to be simulated simultaneously, which requires resorting to massively parallel computing, the formant locations of vowel [i] have been well recovered.
Adaptive DNS/LES (direct numerical simulation/large-eddy simulation) is used to compute the drag coefficient CD for the flow past a sphere at Reynolds number Re = 10(4). Using less than 10(5) mesh points, CD is computed to an accuracy of a few percent, corresponding to experimental precision, which is at least an order of magnitude cheaper than standard non-adaptive LES computations in the literature. Adaptive DNS/LES is a General Galerkin G2 method for turbulent flow, where a stabilized Galerkin finite element method is used to compute approximate solutions to the Navier-Stokes equations, with the mesh being adaptively refined until a stopping criterion is reached with respect to the error in a chosen output of interest, in this paper CD. Both the stopping criterion and the mesh refinement strategy are based on a posteriori error estimates, in the form of a space-time integral of residuals multiplied by derivatives of the solution of an associated dual problem, linearized at the approximate solution, and with data coupling to the output of interest. There is no filtering of the equations, and thus no Reynolds stresses are introduced that need modelling. The stabilization in the numerical method is acting as a simple turbulence model.
We review our work oil adaptivity and error control for turbulent, flow, and we present, recent. development's oil turbulent, boundary layer flow. The computational method G2 is not based oil filtering of the Navier-Stokes (NS) equations, and thus no Reynolds (subgrid) stresses are introduced. Instead the mathematical basis is E-weak solutions to the NS equations and weak uniqueness of such epsilon-weak solutions. Based oil this mathematical framework we construct adaptive finite element methods for the computation of (mean value) Output ill turbulent HOW. Where the mesh is refined with respect to it posteriori estimates of the error in the out put of interest. The a posteriori error estimates are based oil stability information front the numerical solution of an associated dual (adjoint) problem with data given by the output, of interest. To model turbulent, boundary layer separation we use, it skill friction boundary layer model, and we also consider the case of zero skin friction corresponding to solving the inviscid Euler equations with slip boundary conditions, which we refer to as an EG2 method. The results of EG2 a new resolution to the d'Alembert paradox, and it new scenario for turbulent boundary layer separation.
We compute the time average of the drag in two benchmark bluff body problems: a surface mounted cube at Reynolds number 40000, and a square cylinder at Reynolds number 22000, using adaptive DNS/LES. In adaptive DNS/LES the Galerkin least-squares finite element method is used, with adaptive mesh refinement until a given stopping criterion is satisfied. Both the mesh refinement criterion and the stopping criterion are based on a posteriori error estimates of a given output of interest, in the form of a space-time integral of a computable residual multiplied by a dual weight, where the dual weight is obtained from solving an associated dual problem computationally, with the data of the dual problem coupling to the output of interest. No filtering is used, and in particular no Reynolds stresses are introduced. We thus circumvent the problem of closure, and instead we estimate the error contribution from subgrid modeling a posteriori, which we find to be small. We are able to predict the mean drag with an estimated tolerance of a few percent using about 105 mesh points in space, with the computational power of a PC.
In this thesis we consider the following aspects of computational modeling of complex flows: (i) subgrid modeling, (ii) stability, (iii) a posteriori error estimation, and (iv) computational platform. <p />We develop a framework for adaptivity and error control for multiscale problems, in particular for turbulent flow, based on a posteriori error estimates. The a posteriori error estimates take the form of a space-time integral of residuals times dual weights, where discretization residuals relate to numerical errors from discretization, modeling residuals relate to modeling errors from subgrid modeling, and the dual weights govern the propagation of errors in space-time, and is given by solutions of the dual linearized Navier-Stokes equations. We compute approximate such dual solutions for both laminar and turbulent flows. <p />A framework for subgrid modeling based on scale similarity in a Haar Multiresolution analysis is developed. These models are shown to be effective in the case of convection-diffusion-reaction equations with fractal data, simulating turbulent data, and we use the models to estimate modeling residuals in turbulent flow. <p />A computational study of different mechanisms for transition to turbulence in shear flow is presented. <p />We develop the computational platform DOLFIN, an object-oriented library in C++ for finite element computation, where the methods in this thesis are implemented.
In this paper we propose and study a subgrid model for linear convection-diffusion-reaction problems with fractal rough coefficients. The subgrid model is based on extrapolation model of a modeling residual from coarser scales using a computed solution without subgrid model on a finest scale as reference. We present a priori and a posteriori error estimates, and we show in experiments that a solution with subgrid model on a scale h corresponds to a solution without subgrid model on a scale less than h/4
A dynamic scale similarity model is proposed. The subgrid model is tested for model problems related to time dependent non-linear convection-diffusion-reaction systems with fractal solutions. The error of an approximate solution with subgrid model on a scale h is typically smaller than that of a solution without subgrid model on the scale h/2. We also consider the problem of a posteriori error estimation for fractal solutions, splitting the total computational error into a numerical error, related to the discretization of the continuous equations, and a modelling error, taking into account the quality of the subgrid model.
General Galerkin (G2) is a new computational method for turbulent flow. where a stabilized Galerkin finite element method is used to compute approximate weak solutions to the Navier-Stokes equations directly, without any filtering of the equations as in a standard approach to turbulence simulation. such as large eddy simulation, and thus no Reynolds stresses are introduced, which need modelling. In this paper, G2 is used to compute the drag coefficient c(D) for the flow Past a circular cylinder at Reynolds number Re=3900, for which the flow is turbulent. It is found that it is possible to approximate c(D) to an accuracy of a few percent, corresponding to the accuracy in experimental results for this problem, using less than 10(5) mesh points, which makes the simulations possible using a standard PC. The mesh adaptively refined until a stopping criterion is reached with respect to the error in a chosen output of interest, which in this paper is c(D). Both the stopping criterion and the mesh-refinement strategy are based on a posteriori error estimates, in the form of a space-time integral of residuals times derivatives of the solution of it dual problem, linearized at the approximate solution, and with data coupling to the output of interest.
In the context of flow visualization, a triple decomposition of the velocity gradient into irrotational straining flow, shear flow, and rigid body rotational flow was proposed by Kolar in 2007 [V. Kolar, "Vortex identification: New requirements and limitations," Int. J. Heat Fluid Flow, 28, 638-652 (2007)], which has recently received renewed interest. The triple decomposition opens for a refined energy stability analysis of the Navier-Stokes equations, with implications for the mathematical analysis of the structure, computability, and regularity of turbulent flow. We here perform an energy stability analysis of turbulent incompressible flow, which suggests a scenario where at macroscopic scales, any exponentially unstable irrotational straining flow structures rapidly evolve toward linearly unstable shear flow and stable rigid body rotational flow. This scenario does not rule out irrotational straining flow close to the Kolmogorov microscales, since there viscous dissipation stabilizes the unstable flow structures. In contrast to worst case energy stability estimates, this refined stability analysis reflects the existence of stable flow structures in turbulence over extended time.
We derive a posteriori error estimates for the filtered velocity field in a large eddy simulation in various norms and linear functionals. The a posteriori error estimates take the form of an integral in space-time of a discretization residual and a modeling residual times a dual weight. The discretization residual is directly computable, and the modeling residual is estimated by a scale similarity model. We approximate the dual weight by solving an associated linearized dual problem numerically. Computational examples from transition to turbulence in Couette flow are presented.
In this paper we use a General Galerkin (G2) method to simulate drag crisis for a sphere, where the unresolved turbulent boundary layer is modeled as a decreasing skin friction.
In recent years adaptive stabilized finite element methods, here referred to as General Galerkin (G2) methods, have been developed as a general methodology for the computation of mean value output in turbulent flow. In earlier work, in the setting of bluff body flow, the use of no slip boundary conditions has been shown to accurately capture the separation from a laminar boundary layer, in a number of benchmark problems. In this paper we extend the G2 method to problems with turbulent boundary layers, by including a simple wall-model in the form of a friction boundary condition, to account for the skin friction of the unresolved turbulent boundary layer. In particular, we use G2 to simulate drag crisis for a circular cylinder, by adjusting the friction parameter to match experimental results. By letting the Reynolds number go to infinity and the skin friction go to zero, we get a G2 method for the Euler equations with slip boundary conditions, which we here refer to as an EG2 method. The only parameter in the EG2 method is the discretization parameter, and we present computational results indicating that EG2 may be used to model very high Reynolds numbers flow, such as geophysical flow.
In this paper we study a subgrid model based on extrapolation of a modeling residual, in the case of a linear convection-diffusion-reaction problem Lu=f in two dimensions. The solution u to the exact problem satisfies an equation Lhu=[f]h+Fh(u), where Lh is the operator used in the computation on the finest computational scale h, [f]h is the approximation of f on the scale h, and Fh(u) is a modeling residual, which needs to be modeled. The subgrid modeling problem is to compute approximations of Fh(u) without using finer scales than h. In this study we model Fh(u) by extrapolation from coarser scales than h, where Fh(u) is directly computed with the finest scale h as reference. We show in experiments that a solution with subgrid model on a scale h in most cases corresponds to a solution without subgrid model on a mesh of size less than h/4.
We consider the problem of computational simulation of turbulence, where we study turbulent solutions to the incompressible Navier- Stokes equations. We construct approximate weak solutions using a stabilized Galerkin finite element method, here referred to as General Galerkin G2, for which we investigate uniqueness in output (or weak uniqueness ) by solving an associated dual problem computationally, with data coupling to the particular output we are interested in. For simulation of turbulent flow we refer to the adaptive version of G2 as Adaptive DNS/LES, with part of the flow being resolved in a Direct Numerical Simulation DNS, and part of the flow being left unresolved in a Large Eddy Simulation LES , with the stabilization in G2 acting as a dissipative subgrid model. We present computational results using Adaptive DNS/LES, where we find that for the problem of simulating the turbulent flow past various bluff bodies we are able to compute mean value output, such as drag, using 10-100 times less degrees of freedom than in typical LES computations using ad hoc mesh refinement. We further use Adaptive DNS/LES to simulate the turbulent flow past a cylinder rolling along ground.
The goal of the FEniCS project is to develop open-source software for the automation of Computational Mathematical Modeling (CMM). FEniCS is a joint project between the Toyota Technological Institute at Chicago, the University of Chicago, and Chalmers University of Technology. The vision of FEniCS is to set a new standard in CMM, which can be described as the Automation of CMM, towards the goals of generality, efficiency, and simplicity, concerning mathematical methodology, implementation, and application.
In this paper we first review our recent work on a new framework for adaptive turbulence simulation: we model turbulence by weak solutions to the Navier-Stokes equations that are wellposed with respect to mean value output in the form of functionals, and we use an adaptive finite element method to compute approximations with a posteriori error control based on the error in the functional output. We then derive a local energy estimate for a particular finite element method, which we connect to related work on dissipative weak Euler solutions with kinetic energy dissipation due to lack of local smoothness of the weak solutions. The ideas are illustrated by numerical results, where we observe a law of finite dissipation with respect to a decreasing mesh size.
We present a framework for adaptive finite element computation of turbulent flow and fluid structure interaction, with focus on general algorithms that allow for complex geometry and deforming domains. We give basic models and finite element discretization methods, adaptive algorithms and strategies for efficient parallel implementation. To illustrate the capabilities of the computational framework, we show a number of application examples from aerodynamics, aero-acoustics, biomedicine and geophysics. The computational tools are free to download open source as Unicorn, and as a high performance branch of the finite element problem solving environment DOLFIN, both part of the FEniCS project.
This chapter provides a description of the technology of Unicorn focusing on simple, efficient and general algorithms and software for the Unified Continuum (UC) concept and the adaptive General Galerkin (G2) discretization as a unified approach to continuum mechanics. We describe how Unicorn fits into the FEniCS framework, how it interfaces to other FEniCS components, what interfaces and functionality Unicorn provides itself and how the implementation is designed. We also present some examples in fluid–structure interaction and adaptivity computed with Unicorn.
We present a framework for coupled multiphysics in computational fluid dynamics, targeting massively parallel systems. Our strategy is based on general problem formulations in the form of partial differential equations and the finite element method, which open for automation, and optimization of a set of fundamental algorithms. We describe these algorithms, including finite element matrix assembly, adaptive mesh refinement and mesh smoothing; and multiphysics coupling methodologies such as unified continuum fluid-structure interaction (FSI), and aeroacoustics by coupled acoustic analogies. The framework is implemented as FEniCS open source software components, optimized for massively parallel computing. Examples of applications are presented, including simulation of aeroacoustic noise generated by an airplane landing gear, simulation of the blood flow in the human heart, and simulation of the human voice organ.
Developing multiphysics nite element methods (FEM) andscalable HPC implementations can be very challenging in terms of soft-ware complexity and performance, even more so with the addition ofgoal-oriented adaptive mesh renement. To manage the complexity we inthis work presentgeneraladaptive stabilized methods withautomatedimplementation in the FEniCS-HPCautomatedopen source softwareframework. This allows taking the weak form of a partial dierentialequation (PDE) as input in near-mathematical notation and automati-cally generating the low-level implementation source code and auxiliaryequations and quantities necessary for the adaptivity. We demonstratenew optimal strong scaling results for the whole adaptive frameworkapplied to turbulent ow on massively parallel architectures down to25000 vertices per core with ca. 5000 cores with the MPI-based PETScbackend and for assembly down to 500 vertices per core with ca. 20000cores with the PGAS-based JANPACK backend. As a demonstration ofthe high impact of the combination of the scalability together with theadaptive methodology allowing prediction of gross quantities in turbulent ow we present an application in aerodynamics of a full DLR-F11 aircraftin connection with the HiLift-PW2 benchmarking workshop with goodmatch to experiments.
We present work towards a parameter-free method for turbulent flow simulation based on adaptive finite element approximation of the Navier-Stokes equations at high Reynolds numbers. In this model, viscous dissipation is assumed to be dominated by turbulent dissipation proportional to the residual of the equations, and skin friction at solid walls is assumed to be negligible compared to inertial effects. The result is a computational model without empirical data, where the only parameter is the local size of the finite element mesh. Under adaptive refinement of the mesh based on a posteriori error estimation, output quantities of interest in the form of functionals of the finite element solution converge to become independent of the mesh resolution, and thus the resulting method has no adjustable parameters. No ad hoc design of the mesh is needed, instead the mesh is optimised based on solution features, in particular no bounder layer mesh is needed. We connect the computational method to the mathematical concept of a dissipative weak solution of the Euler equations, as a model of high Reynolds number turbulent flow, and we highlight a number of benchmark problems for which the method is validated.
The FEniCS Project aims towards the goals of generality, efficiency, and simplicity, concerning mathematical methodology, implementation and application, and the Unicorn project is an implementation aimed at FSI and high Re turbulent flow guided by these principles. Unicorn is based on the DOLFIN/FFC/FIAT suite and the linear algebra package PETSc. We here present some key elements of Unicorn, and a set of computational results from applications. The details of the Unicorn implementation are described in Chapter 18.
We present a time-resolved, adaptive finite element method for aerodynamics, together with the results from the HiLiftPW-2 workshop, where this method is used to compute the flow past a DLR-F11 aircraft model at realistic Reynolds number. The mesh is automatically constructed by the method as part of the computation, and no explicit turbulence model is needed. The effect of unresolved turbulent boundary layers is modeled by a simple parametrization of the wall shear stress in terms of the skin friction. In the extreme case of very high Reynolds numbers we approximate the small skin friction by zero skin friction, corresponding to a free slip boundary condition, which results in a computational model without any model parameter that needs tuning. Thus, the simulation methodology by- passes the main challenges posed by high Reynolds number CFD: the design of an optimal computational mesh, turbulence (or subgrid) modeling, and the cost of boundary layer res- olution. The results from HiLiftPW-2 presented in this report show good agreement with experimental data for a range of different angles of attack, while using orders of magnitude fewer degrees of freedom than what is needed in state of the art methods such as RANS.
This is a summary of preliminary results from simulations with the 30P30N high-lift device. We used the General Galerkin finite element method (G2), where no explicit subgrid model is used, and where the computational mesh is adaptively refined with respect to a posteriori error estimates for a quantity of interest. The mesh is fully unstructured and the solutions are time-resolved, which are key ingredients for solving challenging industrial applications in the field of aeroacoustics. We present preliminary results containing time-averaged quantities and snapshots of unsteady quantities, all reasonably agreeing with previous computational efforts. One important finding is that the use of adaptively generated meshes seems to be a more effcient way of computing aeroacoustic sources than by using "handmade" meshes.
We present a new mathematical theory explaining the fluid mechanics of sub-sonic flight, which is fundamentally different from the existing boundary layer-circulation theory by Prandtl-Kutta-Zhukovsky formed 100 year ago. The new the-ory is based on our new resolution of d’Alembert’s paradox showing that slightlyviscous bluff body flow can be viewed as zero-drag/lift potential flow modified by3d rotational slip separation arising from a specific separation instability of po-tential flow, into turbulent flow with nonzero drag/lift. For a wing this separationmechanism maintains the large lift of potential flow generated at the leading edgeat the price of small drag, resulting in a lift to drag quotient of size15
In this paper we present a General Galerkin (G2) method for the compressible Euler equations, including turbulent ow. The G2 method presented in this paper is a nite element method with linear approximation in space and time, with componentwise stabilization in the form of streamline diusion and shock-capturing modi cations. The method conserves mass, momentum and energy, and we prove an a posteriori version of the 2nd Law of thermodynamics for the method. We illustrate the method for a laminar shock tube problem for which there exists an exact analytical solution, and also for a turbulent flow problem
In this paper, we describe an incompressible Unified Continuum(UC) model in Euler (laboratory) coordinates with a moving mesh for tracking the fluid-structure interface as part of the discretization, allowing simple and general formulation and efficient computation. The model consists of conservation equations for mass and momentum, a phase convection equation and a Cauchy stress and phase variable theta as data for defining material properties and constitutive laws. We target realistic 3D turbulent fluid-structure interaction (FSI) applications, where we show simulation results of a flexible flag mounted in the turbulent wake behind a cube as a qualitative test of the method, and quantitative results for 2D benchmarks, leaving adaptive error control for future work. We compute piecewise linear continuous discrete solutions in space and time by a general Galerkin (G2) finite element method (FEM). We introduce and compensate for mesh motion by defining a local arbitrary Euler-Lagrange (ALE) map on each space-time slab as part of the discretization, allowing a sharp phase interface given by theta on cell facets. The Unicorn implementation is published as part of the FEniCS Free Software system for automation of computational mathematical modeling. Simulation results are given for a 2D stationary convergence test, indicating quadratic convergence of the displacement, a simple 2D Poiseuille test for verifying correct treatment of the fluid-structure interface, showing quadratic convergence to the exact drag value, an established 2D dynamic flag benchmark test, showing a good match to published reference solutions and a 3D turbulent flag test as indicated above.
We present a framework for adaptive finite element computation of turbulent flow and fluid-structure interaction, with focus on general algorithms that allow for complex geometry and deforming domains. We give basic models and finite element discretization methods, adaptive algorithms and strategies for e cient parallel implementation. To illustrate the capabilities of the computational framework, we show a number of application examples from aerodynamics, aero-acoustics, biomedicine and geophysics. The computational tools are free to download open source as Unicorn, and as a high performance branch of the finite element problem solving environment DOLFIN, both part of the FEniCS project
In this paper we present a computational study of turbulent flow separation for a circular cylinder at high Reynolds numbers. We use a stabilized finite element method together with skin friction boundary conditions, where we study flow separation with respect to the decrease of a friction parameter. In particular, we consider the case of zero friction corresponding to pure slip boundary conditions, for which we observe an inviscid separation mechanism of large scale streamwise vortices, identified in our earlier work. We compare our computational results to experiments for very high Reynolds numbers. In particular, we connect the pattern of streamwise vorticity in our computations to experimental findings of spanwise 3d cell structures reported in the literature.
We present recent work on the following issues of CFD:(i) discretization of the non-stationary incompressible Navier-Stokes equations, (ii) solution of the discrete system at each time step, (iii) hydrodynamic stability, (iv) adaptive error control and a posteriori error estimates, (v) transition of turbulence, (vi) turbulence modeling
We discuss two topics of adaptive computational methods for differential equations: (i) individual time-stepping (ii) subgrid modeling, and we present some applications including the computability and predictability of the Solar System and aspects of subgrid modeling in convection-diffusion-reaction systems.
We present a new approach to computational fluid dynamics (CFD) using adaptive stabilized Galerkin finite element methods with duality based a posteriori error control for chosen output quantities of interest. We address the basic question of computability in CFD: For a given flow, what quantity is computable to what tolerance to what cost? We focus on incompressible Newtonian flow with medium to large Reynolds numbers involving both laminar and turbulent flow features. We estimate a posteriori the output of the computed solution with the output based on the exact solution to the Navier–Stokes equations, thus circumventing introducing and modeling Reynolds stresses in averaged Navier–Stokes equations. Our basic tool is a representation formula for the error in the quantity of interest in terms of a space–time integral of the residual of a computed solution multiplied by weights related to derivatives of the solution of an associated dual problem with data connected to the output. We use the error representation formula to derive an a posteriori error estimate combining residuals with computed dual weights, which is used for mesh adaptivity in space–time with the objective of satisfying a given error tolerance with minimal computational effort. We show in a concrete example that outputs such as a mean value in time of drag of a turbulent flow around a bluff body are computable on a PC with a tolerance of a few percent using a few hundred thousand mesh points in space. We refer to our methodology as adaptive DNS/LES, where automatically by adaptivity certain features of the flow are resolved in a direct numerical simulation (DNS), while certain other small scale turbulent features are left unresolved in a large eddy simulation (LES). The stabilization of the Galerkin method giving a weighted least square control of the residual acts as the subgrid model in the LES. The a posteriori error estimate takes into account both the error from discretization and the error from the subgrid model. We pay particular attention to the stability of the dual solution from (i) perturbations replacing the exact convection velocity by a computed velocity, and (ii) computational solution of the dual problem, which are the crucial aspects entering by avoiding using averaged Navier–Stokes equations including Reynolds stresses. A crucial observation is that the contribution from subgrid modeling in the a posteriori error estimation is small, making it possible to simulate aspects of turbulent flow without accurate modeling of Reynolds stresses.
We show that using adaptive finite element methods it is possible to accurately simulate turbulent flow with the computational power of a PC. We argue that this possibility should set a new agenda in CFD. The key to this break-through is (i) application of the general approach to adaptitive error control in Galerkin methods based on duality, coupled with (ii) crucial properties of turbulent flow allowing accurate computation of mean value quantities such as drag and lift without full resolution of all scales.
We present recent results using adaptive finite element methods, based on a posteriori error estimates, to compute various output functionals for incompressible flow problems in 3d, for both laminar and turbulent flows. The a posteriori error estimates are based on the solution of an associated dual problem with data connected to the output functional we want to compute.