Farhat Research Group Seminars
At these meetings, a Farhat Research Group
team member will give a 20–45 minute talk on their current research. The
goal of these seminars is to promote feedback from the whole lab and
improve collaboration.
Email me at rtezaur@stanford.edu if you would like to present.
Upcoming Talks
Date & time 
Room  Speaker 
1/x/2015 12pm 
Durand 026 
Maciej Balajewicz 
Upcoming Talk Abstracts
Past Talks
12.9.14  Kyle Washabaugh
On weakly enforced boundary conditions for discrete nonlinear aerodynamic reducedorder models
Difficulties associated with weakly enforced boundary conditions for parameterdependent reducedorder models (ROMs) are discussed. A preliminary method for addressing these difficulties is proposed in the context of minimumresidual ROMs for nonlinear Computational Fluid Dynamics (CFD) applications. In the proposed method, the minimumresidual leastsquares problem is iteratively weighted to ensure that all weakly enforced boundary conditions are respected. Difficulties with weakly enforced boundary conditions are demonstrated for reducedorder simulations of airfoils with variable shape and variable freestream angle of attack. For these applications, the proposed method is shown to deliver improved results at predictive operating points.
12.2.14  Raunak Borker
A highorder discontinuous Galerkin method with Lagrange multipliers for advectiondiffusion Problems
A highorder Discontinuous Galerkin method with Lagrange Multipliers
(DGLM) is presented for the solution of steady and unsteady
advectiondiffusion problems in the high Peclet number regime. Like
the Discontinuous Enrichment Method (DEM), it chooses the basis
functions among the freespace solutions of the homogeneous form of
the governing differential equation, and relies on Lagrange
multipliers for enforcing a weak continuity of the approximated
solution across the element interface boundaries. This higherorder DG
framework delivers both the expected better accuracy for fixed spatial
and temporal resolutions, and a significantly better computational
efficiency for a fixed level of accuracy. This is because it operates
directly on the secondorder form of the governing equation, does not
require stabilization, and approximates the solution locally using
freespace solution of the homogeneous form of the governing equations
rather than polynomials which reduces the constant of its
convergence rate. This talk will discuss the theoretical and
computational aspects of the method, and results comparing DGLM to a
continuous polynomial FEM method will be shown.
11.19.14  Eli Turkel
Time reversal for identification of source and obstacles
Time reversal is based on advancing the solution of the relevant wave
problem "backward in time". One important
use of numerical timereversal is that of refocusing, where a reverse run
is performed to recover the location
of a source applied at an initial time based on measurements at a later
time. Usually, only partial, noisy,
information is available, at certain measurement locations, on the field
values that serve as data for the
reverse run. The question concerning the amount and characterization of the
available data
needed for a successful refocusing is discused for various wavelike
equations. A simple procedure
is proposed which exploits multiple measurement times, and is shown to be
very beneficial for refocusing.
A tradeoff between availability of spatial and temporal information is
discussed. The effect of
measurement noise is studied, and the technique is shown to be quite
robust, sometimes even in the
presence of very high noise levels. The use of the technique as a basis for
scatterer identification is also
discussed. A numerical study of these effects is presented,
11.4.14  Matthew Zahr
Fast algorithms for consistent reducedorder bases via proper orthogonal decomposition and lowrank SVD updates: application to local model reduction
An investigation into the offset used in the affine subspace approximation in Proper Orthogonal Decomposition (POD)based model reduction is performed. A notion of consistency is defined among the affine offset, ReducedOrder Basis (ROB), and set of snapshots from which the ROB is constructed. Four numerical examples are employed to demonstrate the importance of consistency, ranging from illustrative contrived examples to largescale CFD problems. A bottleneck in the offlineonline decomposition of model reduction occurs when the initial condition of a simulation is an online quantity, particularly if consistent reducedorder bases are desired.
Two algorithms for efficiently enforcing consistent reducedorder bases are introduced. The first constructs static affine subspaces offline and leverages precomputations to efficiently perform all required projections. The second approach uses fast, lowrank SVD modifications to efficiently use the online initial condition as the affine offset and enforce consistency. The potential of the two algorithms to efficiently enforce consistency, and generate more accurate reducedorder models for a given basis size than inconsistent counterparts, will be demonstrated on the four numerical examples.
10.28.14  Todd Chapman
ECSW with accelerated and scalable generation of reduced meshes
Hyperreduction of nonlinear structural ROMs is an active research area with exciting applications in industrial scale problems. Currently, there are two primary approaches to nonlinear hyperreduction. The first approach is to approximate the nonlinear quantity and project onto the appropriate reduced order subspace. The alternative approach is to project the nonlinear quantity into the reduced order subspace and then introduce the approximation. Energy conserving sampling and weighting (ECSW) takes the latter approach, constructing low dimensional nonlinear functions using a small subset of the original finite element mesh at a fraction of the computational cost of the unapproximated system.
Determining the optimal subset of elements required for an efficient and accurate hyperreduced model is challenging. It ultimately results in a L0 minimization problem that is known to be NPhard. Fortunately, the compressed sensing community has developed a wide variety of algorithms which achieve the same result as L0 minimization by instead minimizing the L1 norm. A quantitative comparison has been made between several well know L1minimization algorithms to determine their suitability for computing reduced meshes for the ECSW method.
Furthermore, in the compressed sensing community, little effort is expended on developing algorithms which are easily parallelizable for distributed memory architectures. This is due to the fact that the systems of equations that result in signal/image processing have fast implicit representations and do not explicit storage. A method of solving the least squares problem, that results at each iteration of nearly every L1 minimization algorithm, via conjugate gradients is given and demonstrated to be an effective way of parallelizing the computation of a reduced mesh.
10.21.14  Arthur PaulDuboisTaine
Parameterization framework for aeroelastic design optimization of wing structures
Recent advances in manufacturing technologies such as 3D printing or Electron Beam Melting have given airplane makers much greater design freedom, paving the way to the manufacturing of more complex and more efficient internal aircraft structures. Such technologies are also being used to manufacture insectlike wings for flapping Unmanned Air Vehicles, a field that has gained a lot of interest in the past few years.
This project aims at developing a general optimization framework for complex wing structures, both for classical fixed wing aircraft and flapping UAVs. The framework relies on an automated structural model generation tool associated with a simple but powerful wing structure parameterization. The Aero suite of code is then used to compute aeroelastic solutions affecting the desired objective function. This talk will present details about the design and optimization framework and present preliminary results for fixed wing aircraft optimization.
10.14.14  David Amsallem
Design optimization using hyperreducedorder models
Solving largescale PDEconstrained optimization problems presents computational challenges due to the large dimensional set of underlying equations that have to be handled by the optimizer.
Recently, projectionbased nonlinear reducedorder models have been proposed to be used in place of highdimensional models in a design optimization procedure. The dimensionality of the solution space is reduced using a reducedorder basis constructed by Proper Orthogonal Decomposition.
In the case of nonlinear equations, however, this is not sufficient to ensure that the cost associated with the optimization procedure does not scale with the high dimension. To achieve that goal, an additional reduction step, hyperreduction is applied. Then, solving the resulting reduced set of equations only requires a reduced dimensional domain and large speedups can be achieved.
In the case of design optimization, it is shown in this paper that an additional approximation of the objective function is required. This is achieved by the construction of a surrogate objective using radial basis functions. The proposed method is illustrated with two applications: the shape optimization of a simplified nozzle inlet model and the design optimization of a chemical reaction.
10.7.14  Alex Main
Implicit and higher order discretization methods for compressible multiphase and fluidstructure interaction problems
This thesis examines the design and implementation of embedded boundary methods for problems involving multiphysics, specifically those involving fluidfluid and fluidstructure interactions, with an emphasis on those problems involving complex, highly nonlinear phenomena. This thesis makes a variety of contributions to embedded boundary methods. First, an implicit time stepping scheme for embedded boundary methods is designed and implemented, showing speedups of 40x over the comparable explicit counterpart for problems. This framework is validated on problems involving hydrostatic implosions. Then, a second order accurate embedded boundary method is introduced for fluidfluid and fluidstructure interaction. Finally the accuracy and stability of these embedded boundary methods are examined, by introducing tools to quantify the stability of embedded boundary methods, and techniques are introduced to stabilize embedded boundary methods as necessary.
9.19.14  Daniel Neumann
An iterative approach for CFDbased aeroelastic computations in the frequency domain and its application to design optimization
The objective of the project is the efficient analysis of flutter properties for structures under certain (flight) conditions. The solution of the underlying Lagrange and Euler equations in the frequency domain allows us to directly compute and output the aeroelastic frequencies and damping ratios for the structural modes considered. The outcomes are compared to computational results in the time domain and experimental results based on wind tunnel tests. An application for the implemented approach is shown by applying it to the design optimization procedure of the ARW2 wing.
3.12.14  Guglielmo Scovazzi
Piecewise linear finite elements; arbitrary LagrangianEulerian methods; variational multi scale methods; fluid/structure interaction
A new tetrahedral finite element for transient dynamic computations of fluids and solids is presented. It utilizes the simplest possible finite element interpolations: Piecewise linear continuous functions are used for displacements and pressures (P1/P1), while the deviatoric part of the stress tensor (if present, as in the case of solids) is evaluated with simple singlepoint quadrature formulas. The variational multiscale stabilization eliminates the pressure checkerboard instabilities affecting the numerical solution in the case of the Darcytype operator related to compressible fluids computations, or the Stokestype operator related to solid dynamics computations. The formulation is extended to strong shock computations in fluids and to elasticplastic flow in solids. Extensive tests of shock flows in fluids, and of linear elasticity and nonlinear (finite) elastoplasticity (compressible as well as nearly incompressible) will be presented. Because of its simplicity, the proposed FEM method could favorably impact complex geometry, fluid/structure interaction, and embedded discontinuity computations. Time permitting, a number of preliminary results on fluidstructure interaction problems, and transient computations of viscoelastic solids will also be presented.
2.26.14  Mirko Myllykoski
Cyclic Reduction Type Poisson and Helmholtz Solvers on a GPU
The topic of the presentation are the block cyclic reduction type
linear system solvers and how they can be applied to the Poisson and Helmholtz
equations on a Graphics Processing Unit. The presented implementations are
based on a method called radixq PSCR. A total of three implementations are
presented: a simplified radix2 method, a simplified radix4 method and a
generalized radix4 method. The presentation will include a short overview to
GPU computing using OpenCL. The main matter of the presentation are the GPU
implementations and the obtained numerical results.
2.5.14  Patrick Lea
XFEM Primer to New Hybrid Method
This talk covers the extended finite element method (XFEM) starting with the basic building blocks. XFEM is a method of of modelling discontinuities within a finite element mesh. It has been used for both strong (fracture and dislocations) and weak discontinuities (material interfaces, grain boundaries, etc). Two new approaches will be discussed. The first approach is in the determination of the direction of crack propagation. The second approach uses a new concept of looking at propagation and combines it with a method to model branches and junctions. Results from academic problems as well as more real world situations will be presented.
12.11.13  Matthew Zahr
Rapid nonlinear topology optimization using precomputed reducedorder models
Topology optimization of structural systems is an important tool in engineering design to find the most efficient design of a structure for its intended use. The highdimensional nature of practical computational mechanics (CM) simulations make structural topology optimization a very largescale nonlinear program with nonconvex constraints defined by the discretized, nonlinear continuum equations. It is wellknown that such largescale, nonconvex nonlinear programs are difficult and computationally prohibitive to solve.
Despite the highdimensional nature of many CM models, the solution trajectory is typically confined to a lowdimensional affine subspace. This is the inherent assumption in Model Order Reduction (MOR), whereby a Reduced Order Model (ROM) is constructed with many fewer degrees of freedom than the original HighDimensional Model (HDM), without significant loss of fidelity.
The contribution of this work is a framework for using a reducedorder model in a topology optimiza tion problem, with an emphasis on nonlinear mechanical behavior. To speedup the topology optimization problem, all quantities scaling with the HDM must be reduced, namely, the state vector (generalized dis placements), governing equations, and optimization variable (usually defined over the elements of the mesh). In the proposed approach, Proper Orthogonal Decomposition (POD) and the method of snapshots are used to construct a reduced basis, which defines the reduction of the state variable. We assume that all nonlinear terms (internal force vector and tangent stiffness matrices) are polynomial in the state vector (generalized displacements). This polynomial assumption enables all terms scaling with the HDM to be precomputed and defines the reduction of the governing equations. Finally, the optimization variables are reduced using POD with snapshots defined apriori using manufacturability constraints to drive the snapshot selection. The reduction of these three quantities collectively define the reduced topology optimization problem.
This presentation will give a brief introduction to reducedorder models and topology optimization, cover the formulation of the proposed method as well as the underlying theory, and demonstrate the effectiveness of the method in rapidly designing a 3D wing box using loads that induce large deformations (over 500Ã— speedup with error on the order of 1%).
11.19.13  Dalei Wang
Fully implicit and semiimplicit hybrid discontinuous spacetime Galerkin methods for acoustic wave propagation
Medium frequency regime and multiscale wave propagation problems have been
a subject of active research in computational acoustics recently.
New techniques have attempted to overcome the limitations of existing
discretization methods that tend to suffer from dispersion.
One such technique, the discontinuous enrichment method (DEM), was developed for frequency domain problems such as the Helmholtz equation. DEM incorporates features of the governing partial differential equation
in the approximation, in particular, the solutions of the homogeneous
form of the equation. In this dissertation, based on this concept and on the conventional spacetime finite element method, a hybrid discontinuous Galerkin method (DGM) for the numerical solution of
transient problems governed by the wave equation in two and three
spatial dimensions is described. This hybrid DGM extends concepts of the existing DEM into time domain for problems that are better suited for analysis in time domain. The discontinuous formulation in both space and time enables the use of solutions to the homogeneous wave equation in the approximation. In this dissertation, within
each finite element, the solutions in the form of polynomial waves
are employed. The continuity of these polynomial waves is weakly
enforced through suitably chosen Lagrange multipliers. Preliminary results
for two and three dimensional model problems, in both low and mid frequency
regimes, show that the proposed DGM outperforms the conventional spacetime
finite element method and Newmark family semidiscrete schemes.
Additionally an alternative semiimplicit formulation is proposed where global
level linear systems stemming from the implicit formulation is traded in favor
of smaller and independent local systems. Preliminary results for two dimensional
model problems, in both low and mid frequency regimes, show that for a fixed
mesh resolution, the semiimplicit DGM requires far less memory than its fully
implicit counterpart. The semiimplicit scheme also parallelizes and scales very
well with the number of available CPUs. This makes the semiimplicit method an
attractive scheme both where memory resources are scarce and where processor
resources are abundant.
11.8.13  Alex Main
A secondorder FIVER method for highly nonlinear problems
The accurate and efficient computation of solutions to problems involving multimaterial flow and nonlinear fluidstructure interaction is important in many areas of research and industry. However, problems involving multiple fluids and those with structures undergoing large deformations remain particularly challenging to simulate. The advent of the FIVER method represented a significant advance in this area. In particular, the FVMERS method presents a uniform approach for handling both fluidfluid and fluidstructure interfaces within an Eulerian framework. However, like nearly all purely Eulerian methods for multifluid and fluidstructure interaction problems, the FVMERS is, as originally presented, first order accurate at the interface. Unfortunately, this results in a degradation of the global order of accuracy. Numerical experiments show that the typical global order of convergence for the original FVMERS is about 1.5, even if the chosen CFD scheme is high order accur
ate away from the interface. We will present work describing a simple extension to the original FVMERS that achieves both second order accuracy locally at the interface and exhibits global second order convergence on smooth problems. In addition to describing how to achieve second order accuracy on smooth problems, we also expound on the stability properties of both ghostfluid type methods and the FVMERS. We provide a framework for studying the stability of these methods, and give both theoretical and empirical results. Using these ideas we construct limiters that can be used in the FVMERS and its higher order extension for problems involving strong shocks and rarefactions in the neighborhood of material interfaces.
10.30.13  Mateo Ripepi
Model reduction methods for aeroelastic systems
The seminar will give an overview of the PhD research done at Politecnico di Milano, concerning the development of reduced order models for aeroelastic systems.
The model reduction will be applied to the aerodynamic subsystem, exploiting both system identification and projection methods.
In particular, an improved matrix fraction approximation of the aerodynamic transfer matrix is developed and compared with a PetrovGalerkin projection of the linearized aerodynamic subsystem onto Schur subspaces.
In this framework, a modified approach to gust modeling is also exploited, where the gust profile traveling towards the aircraft is approximated using a series of disturbance velocity shape functions fixed in the space, named gust modes.
10.25.13  Maciej Balajewicz
Nonlinear reduction of embedded boundary computational fluid dynamics
Embedded boundary methods for CFD and fluidstructure interaction problems alleviate computational challenges associated with meshing and large wall boundary motions, deformations, and even topological changes. Developing model order reduction methods for computational frameworks based on the embedded boundary method seems however to be challenging. Indeed, most popular model reduction techniques are projectionbased and rely on the computation of fluid basis functions based on simulation snapshots. In a traditional bodyfitted computational framework, this computation is straightforward because the fluid always occupies the same computational domain. In the embedded computational framework however, deriving global fluid basis functions is problematic {\it a priori} because the Eulerian fluid mesh is shadowed by the moving Lagrangian structural mesh. Snapshots collected from embeddedboundary simulations are, therefore, incomplete because flow states in shadowed regions are not defined. This prevents application of traditional snapshot matrix factorization techniques such as singular value decomposition because these techniques assume complete state observability. In this talk, we formulate the basis seeking problem as a lowrank approximation problem with missing data and summarize a computational optimization algorithm to solve it. To this effect, a model embeddedboundary CFD scheme that is representative of a large class of embedded boundary methods is developed. Then, we review an effective nonlinear model order reduction method based on a PetrovGalerkin projection and GaussNewton minimization. Finally, we report on successful model reduction results for twodimensional, vortexdominated, viscous flows.
10.18.13  Isaac Harari
Nitsche methods for Kirchhoff plates
An efficient procedure for embedding kinematic boundary conditions in thin plate bending, is based on a stabilized variational formulation, obtained by Nitsche's approach for enforcing boundary constraints. The absence of kinematic admissibility constraints allows the use of nonconforming meshes with noninterpolatory approximations, thereby providing added flexibility in addressing the higher continuity requirements typical of these problems. Workconjugate pairs weakly enforce kinematic boundary conditions. The enforcement of tangential derivatives of deflections obviates the need for pointwise enforcement of corner values in the presence of corners. A single stabilization parameter is determined from a local generalized eigenvalue problem, guaranteeing coercivity of the discrete bilinear form. The accuracy of the approach is verified by representative computations with bicubic Bsplines, exhibiting optimal rates of convergence and robust performance with respect to values of the stabilization parameter.
10.16.13  K.C. Park
Recent developments in external acousticstructure interaction models
The present talk surveys classical and computational approaches to model acousticflexible structure interaction problems, and offers challenges that would be needed for improved modeling of not only acousticstructure interaction but of sound source identification. Specifically, the talk begins with classical contributions made by H. Lamb, G. F. Carrier, M. C. Junger, D. Feit, H. Huang, as well as the developments of computationally attractive approximate models offered by T. L. Geers [1], C. A. Felippa[2], M. S. Lee et al [3], among others. The talk will then present recent efforts for improving interaction models via a combination of retarded potential and advanced potential representation [4].
Numerical performance of several recent improved acousticstructure interaction models will be presented in [5].
References
[1] T. L. Geers, Doubly asymptotic approximation for transient motion of submerged structures, J. Acoust. Soc. Am. 64 (1978) 15001508.
[2] C. A. Felippa, A family of early time approximations for fluid structure interaction, ASME J. Appl. Mech. 47 (4) (1980) 703708.
[3] M. S. Lee , YS Park, Y.G. Park, K.C. Park, New approximations of external acousticstructural interactions: Derivation and evaluation, Comput. Methods Appl. Mech. Engrg. 198 (2009) 13681388
[4] M. Fink, Time reversal acoustics, Phys. Today 50 (1997) 3440.
[5] H. K. Woo, Y. S. Shin, K. C. Park, Improvements of Doubly Asymptotic Approximations for External AcousticStructure Interaction Problems, Proc. Cupled2013, Ibiza, Spain.
4.4.13  Jakub Sistek
Parallel implementation of AdaptiveMultilevel BDDC method
In the first part of the talk, basic concepts of iterative
substructuring will be recalled along with very brief survey of domain
decomposition preconditioners. The components of the Balancing Domain
Decomposition based on Constraints (BDDC) method will be explained in
detail. The method solves large sparse systems of linear equations
resulting from discretization of elliptic PDEs.
In the second part of the talk, the algorithm of AdaptiveMultilevel
BDDC method will be described. It is based on a combination of two
extensions of the standard BDDC method: The Adaptive BDDC, which aims at
problems with severe numerical difficulties, generates constraints based
on prior analysis of the problem and eliminates the worst modes from the
space of admissible functions. The Multilevel BDDC is a recursive
application of the BDDC method to successive coarse problems aiming at
very large problems solved at thousands of compute cores.
Finally, the parallel implementation of these concepts within our
opensource BDDCML package will be presented. Performance of the solver
is analysed on several benchmark and engineering problems, and main
issues with an efficient implementation of both Adaptive and Multilevel
extensions are discussed. This is a joint work with Jan Mandel and
Bedrich Sousedik.
3.4.13  Thierry Coupez
Multiphase flow computation with anisotropic adaptive meshes
Combining stabilized finite element solvers and automatic tools to build
adapted unstructured anisotropic meshes with highly stretched elements gives
rise to a new way for performing accurate numerical simulations. The purpose of
this presentation is to explain our approach and method that targets multiphase
flow simulations. The meshing technology is based on a local topological
modification kernel using a minimum volume principle to create a conform
mesh. The element shape and size optimization process is extended to
account for anisotropy using a metric field. This metric is built directly at the
node using a new a posteriori estimate based on the length
distribution tensor approach and the associated edge based error analysis.
Recently, the proposed error analysis has been extended to time
adaptation using the same technique and providing automatically time steps or
time slabs of adapted meshes.
Anisotropic adaptive meshing has clear advantages for solving multiphase flow
problems. We refer to the use of a heterogeneous NavierStokes solver coupled
with a transport equation for the phase descriptor; here the local convected level
set function is used. It is possible to deal with the discontinuity appearing at the
interface by using regularization. The regularization parameter can be seen as the
thickness or the resolution of the interface. It will be shown that by using
adaptivity, one can choose a priori the resolution of the thickness to be very small
leading to very sharp interfaces, which makes it favorable for simulating turbulent
multiphase flow. Several numerical examples and applications will be given that
feature for fluidstructure interaction, bubble and droplet formation with surface
tension, and finally a free surface flow in hydraulic problem.
2.13.13  Kyle Washabaugh(kyle.washabaugh@stanford.edu)
A family of approaches for the reduction of discrete steady nonlinear aerodynamic models
A family of efficient snapshotgathering methods are proposed for constructing predic tive reducedorder steadystate CFD models. Along with converged steadystate solutions and the associated parametric statesensitivities, it is proposed to include the NewtonRaphson iterates and the Krylov search directions as snapshot information. Application to a transonic airfoil with a variable angle of attack shows that the proposed snapshot collection methods are capable of producing accurate ROMs using only a small number of training simulations.
11.28.12  Matthew Zahr(mzahr@stanford.edu)
Construction of parametricallyrobust reducedorder models via nonlinear programming
A greedy, physicsbased method for constructing parametricallyrobust reducedorder models is presented. A reducedorder model is progressively built by training in regions of the parameter space where it lacks accuracy. Regions of high error in the reducedorder model are determined by the solution of a small nonlinear program. The method is specialized to the case where the reducedorder model is to be used as a surrogate in PDEconstrained optimization. Application to shape optimization of a 1D nonlinear CFD simulation shows that the proposed method is effective in generating parametricallyrobust reducedorder models.
11.15.12  Milan Hanus
Numerical modeling of neutron transport
Accurate description of radiation transport phenomena is an important
task in many research and engineering areas, such as nuclear
engineering, oil industry, medicine or astrophysics. The general
mathematical model of radiation transport is represented by a
firstorder integrodifferential PDE posed in a sixdimensional phase
space. Highdimensionality and complicated structure of the equation
neccessitates the use of advanced computational methods for solving
realworld problems.
In this talk, we will introduce most popular approximations of the
transport equation together with appropriate numerical methods which are
being developed by our group. The coarsest of these approximations 
the diffusion approximation  has proven to be very effective for
largescale nuclear reactor calculations like the core reload
optimization. Moreover, combining the diffusion solver with a more
rigorous transport solution leads to a highfidelity, rapidly converging
iterative method suitable for solving problems with both diffusive and
nondiffusive regions. Fast diffusion solvers therefore play a
significant role in modern radiation transport codes and are the primary
focus of our group.
11.14.12  Alex Main(alexmain@stanford.edu)
Multigrid techniques for compressible flow on unstructured meshes
In this talk we discuss the theory and implementation of multigrid techniques for simulating steady, viscous, compressible flow using finite volume methods on unstructured grids. Multigrid methods fall into the category of "convergence acceleration techniques", which are designed to improve the rate of convergence to steady state of fluid dynamics problems. In particular, we will focus on the agglomerated multigrid method, a nonlinear geometric multigrid, and its implementation in AEROF. The multigrid scheme implemented in AEROF not only reduces the number of time steps required to reach steady state, but significantly reduces the required CPU time as well. In particular, on viscous, turbulent problems, we achieve speedups of 2.3 over the standard single grid scheme.
10.31.12  Youngsoo Choi(yc344@stanford.edu)
Simultaneous analysis and design in PDEconstrained optimization
We present new methods for solving certain types of PDE constrained optimization problems. The approach taken is to augment state of the art PDE methods. The PDE variables and the optimization variables are solved for simultaneously. This impacts the PDE method by changing the core from solving a system of nonlinear equations to that of finding a nonlinear saddle point. This in turn alters the character of the linear equations that are solved at each iteration.
The type of problem we address is one in which the objective function has to match a given target state. Both volume and boundary controls are considered in order to match the target. Regularization is added to the objective function to aid stability and to facilitate computing the solution of the linear systems that arise within the algorithm. Solving such linear systems has been
the focus of much research with many methods being proposed. How to do this simultaneously and efficiently for the specific system that arise for the PDE constrained optimization problems of interest is the main focus of the work we present. In the simultaneous approach, it is important to use effective preconditioner for a saddle point system. In this talk, we will go over some efficient preconditioners that exist and introduce novel preconditioners.
The new methods have been implemented by modifying the cuttingedge software AEROS. Numerical results are presented for a variety of problems including a flapping wing, a robotic control problem, and a thermal control problem.
10.24.12  Elie Hachem(ehachem@stanford.edu)
Threefield stabilized finite element method for fluidstructure interactions
We present some advancement towards a monolithic solution procedure for the numerical solution of fluidstructure interactions.
A new threefield stabilized finite element formulation is presented for modelling the interactions between the fluid (laminar or turbulent) and the structure (rigid or elastic).
We combine this method with anisotropic mesh adaptation to ensure an accurate capturing of the discontinuities at the interface.
The accuracy of the formulation is demonstrated by means of 2D and 3D numerical examples.
9.26.12  Adam Larat(adam.larat@ecp.fr)
A stable, robust and high order accurate numerical method for Eulerian
simulation of sprays and particles transport on unstructured meshes
The topic of this presentation is a new robust approach
for the numerical simulation of sprays dynamics. The talk will first
introduce the
Eulerian modelisation of the carried phase by the method of moments and
compare its features to the more classical Lagrangian method.
The Eulerian modelisation of the disperse phase leads to weakly to
strongly hyperbolic systems
of conservation laws which solutions are known to become singular in
finite time (strong discontinuity,
nonmeasurable solutions). A particular attention has now to be given to
the numerical approximation of
such systems. A new robust convex constraints preserving arbitrarily
high order discontinuous
Garlerkin formalism, \cite{ZhangXiaShu12}, is then presented.
It allows to keep
the realizability of the set of moments everywhere (the associated NDF can
always be reconstructed) while maintaining a high formal accuracy and
this on unstructured meshes
in order to cope with complex realistic geometries.
The convex constraint preserving property moreover ensures the numerical
procedure stays stable.
In order to illustrate the efficiency of the method, a set of
representative, numerically hard test cases
will eventually be given.
8.13.12  Evan Gawlik (egawlik@stanford.edu)
Highorder finite element methods for moving boundary problems
We develop a framework for the design of finite element methods for
moving boundary problems with arbitrarily high order of accuracy, both
in space and in time. At the core of our approach is the use of a
universal background mesh containing the domain of interest, which is
adjusted in such a way that the moving boundary coincides with a union
of curved element interfaces at any instant in time. The motion of
these element interfaces over a given time step introduces convective
terms to the governing equations that can be understood from an
Arbitrary Lagrangian Eulerian (ALE) standpoint. The governing
equations can be spatially discretized with a finite element method of
one's choosing, and the resulting semidiscrete equations can then be
integrated with any desired numerical integration scheme. The
framework so described stands independently of the numerical
representation of the boundary, be it with splines, level sets, or the
like.
We illustrate our approach through an application to two problems: the
Stefan problem, a movingboundary problem for which the aim is to
predict the evolution of a solidliquid interface during a melting
process, and incompressible NavierStokes flow around a moving
obstacle. We present numerical evidence for the order of accuracy of
our schemes with prescribed boundary evolution.
6.15.12  Mihailo Jovanovic (mihailo@umn.edu)
Dynamics and control of distributed systems
In the first part of the talk, techniques from control theory,
optimization, and compressive sensing are used to design feedback
gains that strike a balance between the performance of distributed
systems and the sparsity of the controller. Our approach consists
of two steps. First, we identify sparsity patterns of the feedback
gains by incorporating sparsitypromoting penalty functions into
the optimal control problem, where the added terms penalize the
number of communication links in the distributed controller.
Second, we optimize the feedback gains subject to the structural
constraints determined by the identified sparsity patterns. Several
examples of large dynamic networks are provided to illustrate the
effectiveness of the developed approach.
In the second part of the talk, we examine the efficacy of
streamwise traveling waves generated by surface blowing and suction
for controlling the onset of turbulence in a channel flow. For
small amplitude actuation, we utilize weaklynonlinear analysis to
determine base flow modifications and to assess the resulting net
power balance. Sensitivity analysis of the velocity fluctuations
around this base flow is then employed to design the traveling
waves. Our simulationfree approach reveals that, relative to the
flow with no control, the downstream traveling waves with properly
designed speed and frequency can significantly reduce sensitivity
which makes them wellsuited for controlling the onset of
turbulence. In contrast, the velocity fluctuations around the
upstream traveling waves exhibit larger sensitivity to
disturbances. Our theoretical predictions, obtained by perturbation
analysis (in the wave amplitude) of the linearized NavierStokes
equations, are verified using simulations of the nonlinear flow
dynamics. These show that a positive net efficiency as large as 25%
relative to the uncontrolled turbulent flow can be achieved with
downstream waves. We conclude that the theory developed for the
linearized flow equations with uncertainty has considerable ability
to predict fullscale phenomena.
5.29.12  Charbel Farhat(cfarhat@stanford.edu)
FIVER: A Finite Volume Method with Exact Riemann Problems for MultiMaterial Compressible Flow and FlowStructure Interaction
This lecture will present FIVER, a robust finite volume method for the solution of highspeed compressible flows in highly nonlinear multimaterial domains involving arbitrary equations of state and large density jumps. The global domain of interest can include a moving or deformable solid subdomain that furthermore may undergo topological changes due to, for example, crack propagation. The key components of FIVER, include: (a) the definition of a discrete surrogate material interface, (b) the computation first of a reliable inviscid approximation of the fluid state vector on each side of a discrete material interface via the construction and solution of a local, exact, twophase Riemann problem, (c) the algebraic solution of this auxiliary problem when the equation of state allows it, (d) the solution of this twophase Riemann problem using sparse grid tabulations otherwise, (e) a ghost fluid scheme for approximating next the diffusive and source terms, (f) a systematic procedure for populating the ghost or inactive fluid grid points that guarantees under specified conditions the desired order of spatial accuracy, and (g) an energy conserving algorithm for enforcing the equilibrium transmission condition at a fluidstructure interface.
5.23.12  Todd Chapman(todd_chapman@utexas.edu)
Hyper reduction of finite element simulations
I will present a new method for hyper reduction of finite element
simulations. This technique differs from other hyper reduction
techniques such as GNAT and SAGE in that it is an elementbased
reduction method as opposed to nodebased. Force matching over the
entire domain allows the nonlinear reduction algorithm to eliminate
unimportant elements while weighting remaining elements appropriately
to maintain a suitable model approximation. Offline time is also
reduced since no data collection is required from an intermediate
reduced order model.
5.2.12  David Amsallem (amsallem@stanford.edu)
Nonlinear model reduction based on local reducedorder bases
A new approach for the dimensional reduction via projection of nonlinear computational models based on the concept of local reducedorder bases is presented.
It is particularly suited for problems characterized by different physical regimes, parameter variations, or moving features such as discontinuities and fronts. Instead
of approximating the solution of interest in a fixed lowerdimensional subspace of global basis vectors, the proposed model order reduction method approximates
this solution in a lowerdimensional subspace generated by most appropriate local basis vectors. To this effect, the solution space is partitioned
into subregions and a local reducedorder basis is constructed and assigned to each subregion offline. During the incremental solution online of the
reduced problem, a local basis is chosen according to the subregion of the solution space where the current highdimensional solution lies.
This is achievable in realtime because the computational complexity of the selection algorithm scales with the dimension of the lowerdimensional solution space.
Because it is also applicable to the process of hyper reduction, the proposed method for nonlinear model order reduction is computationally efficient.
Its potential for achieving large speedups while maintaining good accuracy is demonstrated for two nonlinear computational fluid and fluidstructure interaction problems.
4.26.12  Xianyi Zeng (rhodusz@stanford.edu)
HighOrder Embedded Boundary Methods for FluidStructure Interactions
Embedded boundary methods are gaining popularity for solving fluidstructure interaction (FSI) problems because they simplify a number of computational issues. These range from gridding the fluid domain, to designing and implementing Eulerianbased algorithms for challenging fluidstructure applications characterized by large structural motions and deformations or topological changes. However, because they typically operate on non bodyfitted grids, embedded boundary methods also complicate other issues such as the treatment of wall boundary conditions in general, and fluidstructure transmission conditions in particular. These methods also tend to be at best firstorder spaceaccurate at the embedded boundaries. In some cases, they are also provably inconsistent at these locations.
To address this issue, a systematic approach is first presented for constructing higherorder embedded boundary methods for fluidstructure interaction problems. It is developed for inviscid compressible flows and nonpenetrating fluidstructure transmission conditions. The proposed approach leads to a departure from the current practice of populating ghost fluid values independently from the chosen spatial discretization scheme. Instead, it accounts for the pattern and properties of a preferred higherorder discretization scheme, and attributes ghost values as to preserve the formal order of spatial accuracy of this scheme. It is illustrated by its application to various finite difference (FD) and finite volume (FV) methods. Its impact is demonstrated by numerical experiments that confirm its theoretically proven ability to preserve higherorder spatial accuracy, including in the vicinity of the immersed interfaces.
The second component of the presentation focuses on the slope limiters in the secondorder FV methods. This is intended for fixing a computational issue associated with applying conventional slope limiter functions with nonuniform grids in the secondorder MUSCL methods, so that the work done previously for enhancing the spatial accuracy at the fluidstructure interface is not lost. In particular, it is observed that conventional limiter functions lead to loss of spatial accuracy when the fluid meshes are nonuniform.
The ReconstructionEvolveProject (REP) procedure is adopted for analysis and it is enhanced to account for the nonuniform grids.
Sufficient conditions are derived for slope limiters to achieve totalvariationaldiminishing (TVD) stability, secondorder spatial accuracy and symmetrypreserving property in the presence of nonuniform grids. A number of conventional slope limiter functions are enhanced to satisfy these conditions. The improvements in the accuracy by applying the enhanced limiters are demonstrated by various numerical experiments.
3.21.12  Frederic Alauzet (Frederic.Alauzet@inria.fr)
Extension of metricbased anisotropic mesh adaptation to timedependent problems involving moving geometries
Threedimensional reallife simulations are generally unsteady and involve moving geometries. Industries are currently still very
far from performing such simulations on a daily basis, mainly due to their extensive computational cost and the weakness of the numerical methods.
Anisotropic metricbased mesh adaptation, which has already proved its usefulness for steady CFD simulations, can certainly solve part of this issue
by enhancing both the accuracy and efficiency of CFD simulations. However, before extending these techniques to moving mesh simulations,
the fixedtopology constraint imposed by the classical ArbitraryLagrangianEulerian formulation (ALE) framework has to be released.
In a first part, this presentation deals with anisotropic mesh adaptation applied to unsteady simulations, especially those involving moving geometries.
We detail the extension of the multiscale anisotropic mesh adaptation, i.e. a control of the interpolation error in Lp norm,
to unsteady flows thanks to an enhanced version of the fixedpoint algorithm. Then, we are concerned with the development of this new
algorithm to moving mesh problems. We explain how the mesh displacement is taken into account in the mesh adaptation strategy.
In a second part, we propose a new way to improve moving mesh method for bodyfitted simulations.
First, we demonstrate numerically that moving threedimensional complex geometries with large displacements is feasible using
only vertex displacements and connectivity changes. This is presents several advantages over usual techniques.
Second, we describe a new methodology extending wellknown ALE schemes to changingconnectivity meshes.
Several CFD examples will be presented to illustrate the efficiency of the described methodologies.
3.7.12  Clement SaintJalm (saintjal@stanford.edu)
An embedded boundary nethod for compressible viscous flows and fluidstructure interaction problems
In this talk, I will present an embedded method for the solution of fluidstructure interaction problems characterized by compressible viscous flows, and large structural motions and deformations. The treatment of the interface combines a flux approach for semidiscretizing the convective terms based on the exact solution of local, onedimensional, fluidstructure Riemann problems, and a ghost fluid approach for semidiscretizing the diffusive terms. The proposed method is verified by comparison with reference solutions obtained using the LagrangianEulerian computational framework on simple test cases. The potential of this method is also highlighted by the results of its application to the simulation of flexible flapping wings.
2.22.12  Sebastien Brogniez(brogniez@stanford.edu )
A weaklydiscontinuous Galerkin method for the advectiondiffusion equation
The Discontinuous Enrichment Method (DEM) was developed as an alternative to standard Finite Element Methods and leads typically to solving smaller and sparser systems for a same level of accuracy.
To capture some features of the solutions, solutions of the homogeneous PDE are used in place of the standard polynomial field. Lagrange multipliers are the introduced to enforce a weak continuity between elements.
Even though DEM was proven to be computationally efficient, no theoretical framework has been derived to study this method for solving the advectiondiffusion equation yet. The objective of this presentation is to fill this gap for the case of a uniform advection field with no source term. A priori and a posteriori error estimates will be presented. These theoretical results will be confirmed by numerical simulations. Finally, a novel approach to deal with source terms without impacting on the size of the global system will be introduced.
2.9.12  Zhu Wang (wangzhu@vt.edu)
Reducedorder models of complex flows: modeling, analysis and computations
Reducedorder models are frequently used in the simulation of complex flows to overcome the high computational cost of direct numerical simulations, especially for threedimensional nonlinear problems. Proper orthogonal decomposition is one of the most commonly used methods to generate reducedorder models for turbulent flows dominated by coherent structures. To balance the low computational cost required by a reducedorder model and the complexity of the targeted turbulent flows, appropriate closure modeling strategies need to be employed. In this talk, we will present several new nonlinear closure methods for proper orthogonal decomposition reducedorder models. We will also present numerical results for the new models used in realistic applications such as uncertainty quantification in nuclear engineering, energy efficient building design and control, and climate modeling.
2.8.12  Laurent Monasse (monasse@stanford.edu)
In this talk, I will present the coupling method I have developed during
my PhD between a Discrete Element method and a compressible fluid solver
for fluidstructure interaction. I will first focus on the
energyconservative Discrete Element method used to simulate the solid.
I will then describe the coupling algorithm and its conservation
properties. We have centered our efforts on a form of Galilean
invariance and the conservation of mass, momentum and energy in the
system. I will conclude with numerical results showing the effectiveness
of the approach.
1.25.12  Alex Main (alexmain@stanford.edu)
Highorder schemes for multifluid problems
Like nearly all numerical schemes for multiphysics problems detailed in the literature, our methods as originally designed for multifluid problems are only formally first order accurate at the interface. This typically degrades the global order of convergence to only 1.41.5, even if the scheme is second order everywhere else. It is possible to further improve the performance by constructing schemes for these problems that are formally high order accurate even at the interface. I have worked on extending Xianyi's work to construct high order schemes for multifluid problems. Although our current results are for a second order accurate treatment of the interface, the interpolation/extrapolation method is very general and can achieve an arbitrary order of accuracy at the interface, including sixth order.
12.7.11  Xianyi Zeng (rhodusz@stanford.edu)
A second order embedded boundary method for inviscid compressible flow in fluidstructure interactions
A systematic approach is developed to enhance the spatial order accuracy of embedded boundary method for fluidstructure interactions. The method involves the ideas of replacing the physical fluidstructure interface by a surrogate interface, and using "operator matching" to populate ghost values in order to preserve the formal order of the spatial accuracy of a chosen CFD scheme. In previous work, the methodology is applied to solving advection equations by various finite difference schemes. Up to third order spatial accuracy is achieved for these problems.
In this presentation, the proposed approach is extended to finite volume methods, with application to a chosen CFD method, which combines the secondorder MUSCL scheme with Roe flux in space and secondorder TVD RungeKutta in time.
Its performance is evaluated by various numerical examples where the flow is modeled by 1D or 2D Euler equations.
Second order convergence is achieved for 1D and 2D problems with prescibed structural motions; and the convergence rates achieved for a 1D twoway coupled problem is roughly 1.5.
11.30.11  Dalei Wang(daleiw@stanford.edu)
Space time DGM for the wave equation
This presentation is divided into two parts. The first part discusses the application of the DGM (Discontinuous Galerkin Method) to wave equation in 2D. The performances of DGM with constant, bilinear, biquadratic Lagrange multipliers are assessed for planar propagating wave and circular waves. Accuracy and convergence rates are measured against SpaceTime FEM. The second part of the presentation discusses the use of a semiexplicit DGM scheme to wave equations. Semiexplicit schemes solve a number of local problems rather than one large global problems. These local problems can be posed independently of each other, thus are suitable for parallel computing. The presentation will discuss a parallel implementation in CUDA.
11.16.11  Kevin Wang (kevin.wgy@gmail.com)
Fluidstructure coupled simulation of crack propagation problems using an embedded boundary method
A numerical method for the simulation of dynamic crack propagation problems driven by
multiphase fluidstructure interaction will be presented. This method is based on: (1) coupling
XFEM with an embedded boundary method for multiphase CFD computations; (2) an accurate
and robust algorithm for tracking the wet surface of a structure undergoing large displacements,
deformations, and topological changes; (3) a level set technique for capturing fluidfluid material
interfaces arising from crack openings; and (4) a robust and secondorder coupled fluidstructure
timeintegrator for advancing in time the solution of such complex problems. The proposed
numerical method will be illustrated with applications in underwater implosion and pipeline explosion
11.2.11  Alex Sabbatini(alexsabb@stanford.edu)
A secondorder accurate time integration scheme for linearized aeroelastic systems
I develop a secondorder accurate time integration scheme for linearized aeroelastic systems. The first step of this scheme requires special attention since the 3point Backward Difference Scheme to be used at all other timesteps does not apply. I assess the order of accuracy of different explicit schemes for this first step including two variants on Forward Euler as well as a RungeKutta 2 technique. This is a jointeffort between Matt Zahr and myself.
10.31.11  Maciej Balajewicz (mjb47@duke.edu)
A new approach to model order reduction of the NavierStokes equations
It has often been remarked that turbulence is the last unsolved problem in Newtonian physics. The aggregate cost to our society resulting from our incomplete understanding of turbulence is quite staggering. Consider, for example, the environmental and economic costs associated with suboptimal performance of virtually every fluidthermal system such as the internal combustion engine and heat exchangers. The turbulence problem might appear puzzling to the outsider. After all, the equations governing turbulent flow, the NavierStokes equations, have been known for over a century and are deterministic and relatively compact. Turbulence remains a problem today because the only truly reliable tool for its approximation  direct numerical simulation  is computationally prohibitive. In this talk, I will summarize a new projectionbased approach to model order reduction of the NavierStokes equations. The key enabler in this proposed approach is a new nonempirical stabilization methodology.
10.24.11  Youngsoo Choi (yc344@stanford.edu)
PDEconstrained optimization and beyond
In this talk, I will start it by introducing a basic application of PDEbased optimal control. PDEbased optimal control is one of categories, which can be formulated as PDEconstrained optimization. What optimal control can do is so powerful that its impact on engineering field is large and keep increasing.
This introduction naturally leads to the significance of efficiently and accurately solving PDEbased optimal control problem. By going over three slightly different methods of solving Linear PDEconstrained problem, we will understand why research on developing efficient numerical method for PDEconstrained optimization. Then different formulations are introduced, which might be useful in different contexts: sparser control, parallel implementation, and DAEconstrained optimization. For each formulation, an appropriate numerical method will be described.
8.10.11  Eli Turkel (turkel@post.tau.ac.il), Tel Aviv University
Acceleration methods for NavierStokes equations and direct and inverse scattering based on wave and Helmholtz equations
5.31.11  Jari Toivanen (toivanen@stanford.edu)
A Hybrid discontinuous Galerkin method for nonlinear Schrödinger equation
The Schrödinger equation and its approximations are discussed for
fermions and bosons. For bosons, it leads to the GrossPitaevskii
equation (GPE) which is a nonlinear eigenvalue problem. GPE is also
called as the the nonlinear Schrödinger equation. A numerical method
for GPE is proposed. For the discretization, the threedimensional
computational domain is divided into hexahedral elements and in
each element the solution is approximated by a sum of basis
functions. Here polynomial and plane wave bases are considered.
The interelement continuity is enforced weakly using Lagrange
multipliers. The ground state is obtained by an iteration
minimizing the energy. Numerical experiments demonstrate that
the proposed method requires less computational work to reach
a given accuracy level that the standard finite element method.
5.19.11  Georg Hammerl (ghammerl@stanford.edu)
Adaptive time stepping for nonlinear hyperbolic problems
In implosion or explosion problems, the time step size is usually limited by a
small portion of the simulation due to stability and accuracy reasons. Hence, a
constant time step size throughout the simulation leads to unnecessary high
computational times and could be chosen larger in smooth portions without
compromising overall accuracy. Several adaptive time stepping strategies based
on estimation of the local truncation error are investigated. Error estimation
captures the degree of transitions in the fluid by comparing the solution of
two numerical schemes and adapts the time step size appropriately. Finally,
applications show the performance of the different approaches.
5.09.11  David Amsallem (amsallem@stanford.edu)
Model reduction with stability guarantee
Few model reduction techniques can guarantee stability of the resulting
reducedorder model (ROM). As a result, a novel and general procedure is
developed for generating linear reducedorder models with stability guarantee.
The method is not code intrusive and as a result can be used even with
blackbox codes. The MatLabbased procedure takes as an input a generated
projectionbased ROM (and associated bases if available) and returns a stable
ROM based on another projection. The method relies on the definition of a
convex optimization problem that can be efficiently implemented and solved
using existing semidefinite programming packages. It will be shown that the
method presents optimality properties, in the sense that it returns the
original ROM if it is stable. This optimality property is not shared by
procedures existing in the literature.
Applications to fluid and fluid/structure systems will be offered.
5.04.11  Rabia Djellouili (rabia.djellouli@csun.edu)
Can electrical current disrupt fibrous capsule tissue around biomaterial implants?
Implantation of biomaterials creates the socalled "foreignbody" responses
which includes the formation of a fibrous capsule to shield the body from the
foreign object by creating a fibrous wall of tissue of varying thickness in
between the two. This fibrous capsule surrounds or encapsulate the implant. The
key point here is that the formation of fibrous tissue around any implant is
normal. It is an integral part of the "woundhealing" process or the
body's response to the implant material. The problems begin to surface when
the fibrous capsule tightens and "squeezes" the implant to the point that it
hampers the performance of the implant and ultimately leads to its failure.
Surgery is then often required to remove the capsule tissue and/or the implant,
an invasive procedure that is not only traumatic, but costly as well.
The objective is to investigate the feasibility and the limitation of using
"moderate" electrical current for disrupting capsule formation tissue
around implants. We propose a mathematical model that describes the growth of
fibrous tissue around rigid, diskshaped implants. A solution methodology based
on an efficient regularized iterative method is presented to calibrate the
model from some measurements of the capsule tissue concentration. Numerical
results obtained with synthetic data are presented to demonstrate the ability
of the proposed solution methodology to determine the model parameters
corresponding to a given implant. In addition, numerical results obtained with
experimental data are presented to illustrate the validity of the proposed
model.
SHORT BIOGRAPHY:
Rabia Djellouli is Professor of Mathematics and Director of the
Interdisciplinary Research Institute for the Sciences (IRIS) at California
State University Northridge. He received his Ph.D. from University of ParisXI
and Ecole Polytechnique, FRANCE in 1988. His current research interest includes
the mathematical and numerical analysis of wave propagation phenomena, inverse
Problems, and some biomedical engineering applications.
5.03.11  Youngsoo Choi (yc344@stanford.edu)
An optimization algorithm for time dependent PDEconstrained optimization problems
One of the popular ways of solving PDEconstrained optimization problem is to
run two independent solvers (i.e. PDE solver and optimization solver). PDE
solver needs to provide optimization solver with objective function value or
sensitivity information. If the sensitivity is not provided, optimization
solver computes gradients via finite difference. In this case, there is a
drawback. Many objective function evaluation is required. Considering that each
objective function evaluation requires one call for PDE solver, it can be quite
expensive computationally. Even in the case that sensitivity is provided by PDE
solver, it is not always clear that it can be used or efficient, especially in
time dependent PDE case. In literature, this category of methodology is called
"reduced problem formulation" or "NAND (Nested Analysis and Design)."
Another approach is to put PDE and optimization solver in one code. In
literature, this method is called "allatonce approach" or "SAND (Simultaneous
Analysis and Design)." In this methodology, both state variables(i.e.
displacement or velocity) and control variables (i.e. external forces,
userdefined displacement, or angle of attack) are considered as optimization
variables. This implies no sensitivity analysis is required to compute
gradients.
One approach to solving time dependent PDE problems on parallel machines is to
divide the time domain into intervals and estimate the initial conditions by
solving the problem with a coarser (or computationally cheaper) time
integrator, then correct the difference at each boundary of time slices.
Several algorithms are invented and two representative methods are "Parareal
time domain decomposition" and "PITA (Parallel Implicit Timeintegration
Algorithm)."
In this talk, I will show an allatonce optimization algorithm which utilizes
SQP method and PITA. The algorithm shows two advantages. First, we can solve
all the separate adjoint equations by using PITA in order to compute Lagrange
multipliers since the Jacobian of constraints is nothing more than Hermitian
transpose of the original structure of PDE. Second, the PDE "solve" is
faster (because we are not solving the PDE). The adjustment to get the solution
to be continuous is made only when the equality PDE constraint violation is too
big. After some discussion of algorithm, computational results on flapping beam
will be presented.
4.19.11  Wade Spurlock (wadespur@stanford.edu)
Implementation of a global reducedorder model for the Presta method
The objective of this work is to continue the development of a fast, accurate,
predictive reducedorder model (ROM). The Presta (POD Residual Minimization
with Tensor Approximations) method is a nonlinear reducedorder modeling
approach that has generated fast, accurate results on a wide range of problems;
however its effectiveness has only been demonstrated in cases where the online
input is the same as the training input.
This work investigates the predictive ability of global Presta model. This
global model is constructed by combining snapshots from all training inputs
before POD computations and ROM construction; the resulting ROM is subsequently
used everywhere in the input space. Although this global approach often leads
to unstable PODGalerkin ROMs, this approach is likely to generate good results
for the Presta method. The reason for this is that the Presta model is
constructed using optimal approximations that lead to a priori convergence.
This talk presents results for a global Presta ROM tested on the
onedimensional inviscid Burgers' equation. Two input parameters  the
inlet boundary condition and a constant in the exponential forcing function 
are considered. The global Presta ROM generates a prediction at a nontraining
input with only 1.43% error and a speedup factor of 9.4. These results
validate the feasibility of the global Presta ROM approach. The implication of
this work is to extend the Presta method to fastturnaround design problems,
allowing multiple training inputs offline followed by rapid online simulations
for many different configurations.
3.04.11  Kranthi Kode (kranthi@stanford.edu)
Onthefly computation of reducedorder bases by iterative update of dominant singular subspaces
2.22.11  Xianyi Zeng (rhodusz@stanford.edu)
Improved limiter functions in MUSCL schemes on 1D nonuniform grids
For solving 1D conservation laws using finite volume methods, liner
reconstruction of data on each cell is performed to achieve second order
accuracy in space. In addition, limiter functions are introduced for stability
in the sense of total variation diminishing for such second order methods.
However, traditional 1D limiter functions are developed assuming uniform
meshes, and the mathematical conditions for second order accuracy and stability
of these limiters assume uniform cell sizes as well. Numerical tests show that
when the mesh is irregular, such limiters fail to provide second order
accuracy, though stability is still observed. The mathematics behind such
observations are studied, and conditions for second order accuracy and TVD
stability on nonuniform meshes are derived. In addition, the symmetry property
of limiter functions is also investigated, and the mathematical condition is
generalized to account for the irregular meshes. Some of the wellknown limiter
functions are modified to satisfy the order condition, the stability condition
and the symmetry condition; the achievement of desired order of accuracy are
confirmed by numerical simulations applying these limiters to solve 1D Euler
equations on highly irregular meshes.
12.08.10  David Amsallem (amsallem@stanford.edu)
Realtime aeroelastic predictions over a large parametric domain using
reducedorder models
The recently developed concept of parametric adaptation of
reducedorder bases using a database and interpolation algorithms on
manifolds is extended to linearized reducedorder operators. Two steps
are involved: (1) the transformation of the reducedoperators in
consistent bases, and (2) the interpolation of the transformed
operators on a suitable manifold, resulting for the first time in a
fully online method. It is illustrated with applications to
parametric studies of structural and aeroelastic systems in subsonic
to supersonic regimes. Its ability to detect mode crossing and veering
is highlighted.
11.17.10  Meir Messingher Lang (meirmsn@stanford.edu)
Modeling and simulation of a flapping wing airplane and a high altitude long endurance aircraft
In conventional aircraft, the generation of lift and thrust forces are delegated separately to the wings and the engine subsystems, respectively. In contrast, flapping wing airplanes generate their driving forces through the motion of their wings. This commonly involves large deformations that make the consideration of geometric nonlinearities important. This type of aircraft is attractive for smallscale low Reynolds number flight; this scenario arises in surveillance/reconnaissance environments where larger aircraft are impractical due to their size. In this project, a flapping wing toy model has been developed to perform a series of numerical simulations. The main goal is to study the effects of scale on the performance of this type of aircraft.
In a conceptually different yet physically related area, aircraft designed for High Altitude Long Endurance (HALE) flights are currently being studied and constructed. These aircraft may play an important role in future military surveillance/reconnaissance missions, atmospheric research, and telecommunication relay services. To make them as efficient as possible during long flights, these aircraft feature wings with high aspect ratios and low weight. As a result, these wings are inherently flexible and prone to large deformations during flight; if uncontrolled, this may lead to catastrophe. Hence, geometric nonlinearities must be considered. In this project, a HALE airplane model is developed from publicly available information in an effort to perform a series of numerical simulations. The main objective is to develop and demonstrate the capability of FRG's suite of codes to handle this type of problem.
This talk will present results for both projects, illustrate which simulations have been successful, and highlight the approaches taken to make them possible.
11.08.10  Jincong He, guest speaker (jincong@stanford.edu)
Enhanced linearized reducedorder models for subsurface flow simulation
Many reservoir simulation applications, such as optimization of well settings and history matching, require a large number of runs. Using traditional simulators for such problems is often very expensive computationally. In this thesis we extend and enhance the trajectory piecewise linearization (TPWL) method for accurate and efficient reducedorder modeling for such applications. In this approach, the reservoir equations are linearized around previously simulated training runs and the highdimension state space is projected into a lowdimension space using proper orthogonal decomposition (POD). With linearization and reduction, simulations that require hours to run can now be completed within a few seconds. TPWL does require overhead computations that correspond to the time required for a few fullorder simulations.
In this talk, both the accuracy and stability of the TPWL method are considered. A local resolution treatment, in which we preserve high resolution in well blocks and other important flow regions, is proposed to improve the accuracy of the method. Stability analysis shows that, particularly for cases with density differences between phases, the TPWL method can be unstable. The stability of the method can be quantified in terms of the spectral radius of an amplification matrix appearing in the TPWL model equation. Two different stabilizing methods are proposed. The first method seeks to minimize the spectral radius of the amplification matrix by choosing the optimized number of reduced variables. The second method stabilizes cases with density differences using a basis constructed from the same case without density differences. Both methods are tested on two reservoir problems of practical size with significant density differences. Results demonstrate that both methods are able to provide stabilized and accurate solutions for these cases.
The stabilized TPWL method is then implemented as a surrogate model within a generalized pattern search (GPS) optimization procedure. Results from the two production optimization problems shows that TPWL can be used as an efficient proxy for production optimizations.
10.27.10  Xianyi Zeng (rhodusz@stanford.edu)
A new framework for enhancing accuracy for ghost methods in solving FSI problems
Ghost fluid method has been applied in simulations of fluidfluid interactions successfully. It removes the requirements for body fitted meshes and mesh motions; instead, GFM uses ghost values in a ghost domain to capture the interface conditions implicitly. For fluidfluid interaction, the status variables on both sides are usually the same, providng a natural way to defining physically significant ghost values which aid in achieving high order accuracy. However for FSI, the status variables are different for solid and fluid, hence there is no natural way to obtain ghost values for fluid from the structure side; typically, all the ghost values for fluid must come from the fluid side itself.
Using the simple constant extrapolation to populate ghost values for fluid seems to be first order accurate in space, but it turns out that the local order of accuracy at interface is only zeroth order. Thus in order to achieve high order of accuracy, the ghost values must be defined very carefully.
In this presentation, a new framework for analyzing the accuracy of GFM applied to FSI problem is proposed, which divides the GFM procedure into two stages: identifying surrogate interfaces with proper surrogate interface values, and populating ghost values using operator matching. The analysis is based on simple wave equations as model equations for FSI, and both 1D cases and 2D cases are considered. Numerical results show that the framework works for various numerical schemes which are explicit in time, for solving wave equations with known moving interface/boundary conditions in both 1D and 2D, and the desired second order and even third order accuracy are obtained.
10.13.10  Sebastien Brogniez (brogniez@stanford.edu)
An insight into DEM for the advection diffusion equation
The Discontinuous Enrichment Method (DEM) has first been derived for the Helmholz equation with success. The use of particular solutions of the homogeneous equation as local enrichment functions improves greatly the accuracy of the numerical solution. Lately, DEM has been applied to the advectiondiffusion equation in the High Peclet regime with promising results. However, even though numerical results have shown the efficiency of the method, no theoretical proofs of wellposedness of the problem have been derived so far. This talk will tackle this gap between theory and practice.
9.15.10  Alex Main (alexmain@stanford.edu)
The ghost fluid method for the poor for underwater implosions with implicit time stepping
The Ghost Fluid Method for the Poor (GFMP) has proven to be an effective method for a wide variety of two phase flow problems. It fails, however, in cases with large density discontinuities. The Ghost Fluid Method for Poor with Exact Riemann Solver (GFMPERS), developed here by Arthur Rallu, is an extension of GFMP which performs well on underwater implosion problems involving large density jumps. The GFMPERS has been implemented previously in AEROF using an explicit time stepping scheme. In addition to reviewing the GFMPERS method, we will explain how it was extended to allow for implicit time integration in AEROF, discussing both first and second order accurate time integrators. Finally we will compare the computational cost of the explicit scheme with that of the implicit scheme on a set of test problems.
9.01.10  Dalei Wang (daleiw@stanford.edu)
Spacetime DGM for wave equations in 2D
This presentation discusses and ongoing project about solving wave equations in 2D using a method which is discontinuous in both time and space. Unlike semidiscrete methods, SpaceTime DGM does not break the PDE into a system of ODE's. The presentation will cover the theoretical background for the method and discuss practical issues which arose in the implmentation. Simulation results regarding stability, convergence, etc. will also be discussed.
8.18.10  Jon Gretarsson (jontg@stanford.edu)
Finding the fluidstructure interface
I'll be talking about the geometric challenge that arises from identifying fluidstructure edges. In an embedded framework, every edge that lies between any two adjacent fluid fluid nodes needs to be checked to see whether it crosses the structure interface. We have in our toolbox a robust routine for determining whether an edge intersects a triangle, but the computation is somewhat expensive; therefore the naive approach of applying this routine to every possible combination of fluid edge and surface triangle is impractical. In order to address this, we provide several spatiallyaware optimizations in order to quickly eliminate edgetriangle pairs which do not intersect.
8.04.10  Kevin Wang (icmewang@stanford.edu)
Algorithms for interface treatment and load computation in embedded boundary methods for fluid and fluidstructure interaction problems
Embedded boundary methods for CFD (Computational Fluid Dynamics) simplify a number of issues ranging from meshing the fluid domain, to designing and implementing Eulerianbased algorithms for fluidstructure applications featuring very large structural motions and/or deformations of the embedded surfaces. Unfortunately, such methods also complicate other issues such as the treatment of the wall boundary conditions in general, and fluidstructure transmission conditions in particular. This talk focuses on this aspect of the problem in the context of compressible flows, the finite volume method for the fluid, and the finite element method for the structure. First, we present a numerical method for treating simultaneously the fluid pressure and velocity conditions on static and dynamic embedded surfaces that is based on the exact solution of local, onedimensional, fluidstructure Riemann problems. Next, we describe two different conservative approaches for computing the flowinduced forces and moments on rigid and flexible structures. The performance of all three numerical algorithms is assessed for threedimensional fluid and fluidstructure interaction problems associated with aeronautical and underwater implosion applications.
7.21.10  Adam Larat (adam.larat@stanford.edu)
Very high order residual distribution schemes for steady hyperbolic equations
Residual Distribution Schemes (RDS) are a class of Numerical Methods for
Hyperbolic Equations that appeared in the early 80s. After 30 years of
research, a very rich framework has been developed. It can nowadays be truly
considered as an alternative to the traditional Finite Volume (FV) and Finite
Element (FE) Methods, even though it is still far from being fully developed.
During this presentation, a particular focus will be given to a generalization
of the LaxFriedrichs scheme in the RDS framework, this method being mainly
developed at INRIA Bordeaux SudOuest with the collaboration of Remi Abgrall
and Mario Ricchiuto. This particular scheme comes with proofs of consistency,
(k+1)th order accuracy, conservation, monotonicity preserving property for
steady scalar hyperbolic equations on Pk unstructured meshes. On the contrary
to FV schemes, the treatment of the equation by the RDS framework is
multidimensional while the stencil remains compact, even at very high order.
This provides some fast converging methods that can be easily parallelized.
When compared to the increasingly popular Discontinuous Galerkin (DG) method,
RDS use a conformal representation of the solution on the mesh, hence enabling
a considerable decrease of degrees of freedom, especially in 3D. After a very
theoretical enlightenment of these schemes and of the former words, the talk
will be illustrated by the generalization of the method to increasingly complex
test cases: hybrid meshes, 2D and 3D steady Euler problems, steady viscous
simulations. Pro and cons of the method will be practically emphasize and the
presentation will end by a quick summary of the further perspectives.
7.07.10  David Powell (dpowell1@stanford.edu)
Adaptation of the discontinuous enrichment method for multiscale composite applications
The multiscale nature of composite materials provides a number of interesting modeling challenges. While the behavior and interaction of the individual constituents of the material are typically very important for understanding the mechanical response, the large separation of scales between the macro and micro levels precludes using a model with enough detail to directly include this microstructure. This presentation will explore the potential use of the discontinuous enrichment method (DEM) as an approach to modeling composite materials where enrichment functions are used to introduce the microscale behavior without requiring any refinements to the mesh. Unlike traditional implementation of DEM, here the enrichment functions are generated numerically using a microscale simulation. Inspiration from a number of multiscale modeling techniques, in particular asymptotic expansion homogenization, is used to create these enrichment functions.
This presentation represents a work in progress.
6.23.10  Jon Gretarsson (jontg@stanford.edu)
Practical animation of compressible flow for shock waves and related phenomena
We propose a practical approach to integrating shock wave dynamics into traditional smoke simulations. Previous methods either simplify away the compressible component of the flow and are unable to capture shock fronts or use a prohibitively expensive explicit method that limits the time step of the simulation long after the relevant shock waves and rarefactions have left the domain. Instead, we employ a semiimplicit formulation of Euler's equations, which allows us to take time steps on the order of the fluid velocity (ignoring the more stringent acoustic wave speed restrictions) and avoids the expensive characteristic decomposition typically required of compressible flow solvers. We also propose an extension to Euler's equations to model combustion of fuel in explosions. The flow is twoway coupled with rigid and deformable solid bodies, treating the solidfluid interface effects implicitly in a projection step by enforcing a velocity boundary condition on the fluid and integrating pressure forces along the solid surface. As we handle the acoustic fluid effects implicitly, we can artificially drive the sound speed c of the fluid to infinity without going unstable or driving the time step to zero. This permits the fluid to transition from compressible flow to the far more tractable incompressible flow regime once the interesting compressible flow phenomena (such as shocks) have left the domain of interest, and allows the use of stateoftheart smoke simulation techniques.
5.24.10  Kevin Carlberg (carlberg@stanford.edu)
Efficient nonlinear model reduction via a leastsquares PetrovGalerkin projection and compressive tensor approximations
A PetrovGalerkin projection method is proposed for reducing the dimension of a discrete nonlinear static or dynamic computational model in view of enabling its processing in realtime. The right reducedorder basis is chosen to be invariant and is constructed using the Proper Orthogonal Decomposition method. The left reducedorder basis is selected to minimize the 2norm of the residual arising at each Newton iteration. Thus, this basis is iterationdependent, which enables it to capture nonlinearities, and leads to the globally convergent GaussNewton method. To avoid the significant computational cost of assembling the reducedorder operators, the residual and the action of the Jacobian on the right reducedorder basis are each approximated by the product of an invariant, largescale matrix, and an iterationdependent, smaller one. The invariant matrix is computed from a data compression procedure that meets proposed consistency requirements. The iterationdependent matrix is computed to enable the leastsquares reconstruction of some entries of the approximated quantities. The results obtained for the solution of a turbulent flow problem and nonlinear structural dynamics problems highlight the merit of the proposed consistency requirements. They also demonstrate the potential of this method to significantly reduce the computational cost associated with highdimensional models while retaining their accuracy.
5.10.10  Edmond Chiu (edmondc@stanford.edu)
On the change in predicted flutter characteristics due to presence and modeling of internal fluids
The analysis of the sloshing effect of an internal fluid on a flutter envelope has received little attention, at least in the open literature. This problem can affect many aircraft, especially high performance fighter jets with wingstore configurations. In the talk, I will discuss the modeling aspect of this problem and present some computational results that quantify the effects of fuel slosh on the flutter characteristics of a test wingstore configuration. In particular, the talk will feature the hydroelastic added mass model and a more trivial frozen mass model that are used to represent the internal fluid. Implications of the computational results obtained from flutter calculations performed using both internal fluid models combined with a validated CFDbased linearized method for solving fluidstructure interaction problems will also be discussed.
4.26.10  Radek Tezaur (rtezaur@stanford.edu)
Reconstruction methods for fast frequency sweep analysis
The solution of problems of the form
Zx = (K\omega^2 M +i\omega D)x = f(\omega),
that arise, for example, in structural dynamics and acoustics,
is considered. The response x is sought over a range of frequencies \omega.
The straightforward solution of such problems for fine increments of
\omega is computationally prohibitive, particularly when Z is a
largescale matrix. An alternative solution approach based on
Pad\'e approximation or a Galerkin projection to a subspace spanned
by the solution and its derivatives at a subset of frequencies are
investigated. Numerical results comparing the methods on applications
of interest are presented.
4.19.10  Irina Kalashnikova (irinak@stanford.edu)
A stable Galerkin reduced order model (ROM) for compressible flow
Simulation of unsteady 3D fluid flow remains, and will continue to remain for some time, an expensive endeavor. For this reason, efforts have been made to develop Reduced Order Models (ROMs): models that capture the essential dynamics of the full CFD simulations, but at a much lower computational cost. The use of ROMs in a predictive setting raises some fundamental questions about these models' the essential numerical properties, namely stability, consistency, and convergence. General results for any of the three properties are lacking for Proper Orthogonal Decomposision (POD)/Galerkin models of compressible fluid flow. This leads to practical limitations; for example, a ROM might be stable for a given number modes but unstable for other choices of basis size.
This talk focuses on numerical stability of linear ROMs for compressible flow constructed using the POD/Galerkin method for model reduction. The model reduction technique differs from the approach taken in many applications in that the Galerkin projection step is applied to the continuous system of PDEs, rather than the semidiscrete representation of the governing equations is projected onto the modes. It is demonstrated that the inner product used to define the Galerkin projection is intimately tied to stability of the resulting model. For the linearized Euler equations, a symmetry transform leads to a stable formulation for the inner product. Preservation of stability for the discrete implementation of the Galerkin projection is made possible using a weak enforcement of the boundary conditions, and piecewisesmooth finite element bases. Stability of the reduced order model for the fluid is demonstrated on several model problems, where a suitable approximation basis is generated using POD of a transient computational fluid dynamics simulation.
3.15.10  Xianyi Zeng (rhodusz@stanford.edu)
A highorder ghost method for solving moving boundary condition problem
In solving the fluid part of fluidstructure interaction problems with large interface deformations, methods involves mesh motion, like ALE, become infeasible due to the distortion of fluid mesh. A fixed fluid mesh is more favorable in such situations, which means that the mesh is nonconforming with the interface. Many methods (embedded, immersed, or ghost, etc.) have been developed to treat the interface. However, they are almost at best first order accurate at the interface, and many of them have zeroth order of accuracy.
In this talk, a new approach to achieve high order accuracy at interface is presented. This approach is based on ghost fluid method, whose original version usually has zeroth order of accuracy at the boundary. 1D wave equation under forced motion (moving boundary condition) is used as the model problem, and numerical examples will be shown that the new approach can actually achieve second, or even third order of accuracy for this simple problem.
3.8.10  Sebastien Brogniez (brogniez@stanford.edu)
Stability analysis of a staggered algorithm for conjugate heat transfer
This analysis is motivated by the interest in numerical procedures for coupling separate computations of thermal diffusion in a structure and a fluid. A single fully coupled code which would solve the entire fluidstructure domain is practically infeasible for complex geometries. A more practical approach is to use existing codes which accurately and efficiently solve each individual
part of the problem.
The objective of this talk is to present a method for studying the numerical stability for such a procedure as well as the following results. A stable and efficient procedure for solving coupled steadystate solution will also be derived from this analysis.
2.22.10  David Powell (dpowell1@stanford.edu)
Multiscale modeling of ballistic fabric and other fibrous materials
Highstrength fabrics such as Kevlar and Zylon are being widely used for body armor and for the protection of sensitive components of commercial and military structures. Since experimental testing on these materials is often expensive and time consuming, a computational model amenable to largescale numerical simulations on massively parallel processors provides an ideal approach to reducing expensive laboratory tests. To this end, a novel multiscale approach has been developed for modeling ballistic fabric that extracts information from the microscale fibril material properties in order to build the macroscale sheet. This approach incorporates the stochastic nature of the material due to the random variations in the yarn caused by the weaving process. Representing these variations is critical to capturing the heterogeneous damage and failure of the material, as confirmed by experimental tests. A parallel implementation of the proposed multiscale method has been developed with particular attention to issues involving contact mechanics. This presentation will focus on all of these issues and report on massively parallel largescale computations for large sheet applications, with both performance and validation results. Current and future work will be discussed including applications of this approach to other fibrous materials, such as collagen fibers in biological tissue.
1.30.10  David Amsallem (amsallem@stanford.edu)
A realtime method for interpolating on manifolds linear parametric timeinvariant reducedorder models
A new method for interpolating between timeinvariant parametric ROMs is presented. The method takes reducedorder linear operators precomputed for various values of the parameters as inputs and returns reduced operators at new values of the parameters. As such, the quantities involved do not depend on the size of the underlying fullorder model and the method has realtime capabilities. Furthermore, the proposed procedure is not codeintrusive as the reducedorder operators are quantities that are very often outputted. The present method relies on a heuristic rationale that results in an optimization problem on a matrix manifold. Numerical experiments will be presented and compared against other methods proposed in the literature, highlighting the advantage and potential of the method for several type of systems (mechanical, aeroelastic) and model reduction methods (moment matching method, POD).
11.18.09  Dalei Wang (daleiw@stanford.edu)
Spacetime DGM for wave equations
11.5.09  Irina Kalashnikova (irinak@stanford.edu)
The Discontinuous enrichment method (DEM) for advectiondominated fluid problems
10.21.09  Meir Messingher Lang (meirmsn@stanford.edu)
Flapping wing airplanes and HALE aircrafts
10.7.09  Harsh Menon (hmenon@stanford.edu)
Unified discretization of a class of nonlinear aeroelastic problems
9.21.09  Jon Gretarsson (jontg@stanford.edu)
Fluidstructure interactions with semiimplicit compressible flow
9.9.09  Sebastien Brogniez (brogniez@stanford.edu)
Stability and accuracy considerations for designing a time integration scheme on moving meshes
9.2.09  Paolo Massimi (pmassimi@stanford.edu)
A Discontinuous enrichment method for the solution of plate vibration problems
8.26.09  Julien Cortial (jcortial@stanford.edu)
Timeparallel solution of structural dynamics problems
The timeparallel framework for constructing parallel implicit timeintegration schemes (PITA)
has two major applications: (a) the exploitation of a number of processors that is far larger than the
maximum number of processors that spaceparallelism alone can efficiently utilize for reducing the CPU time
associated with the solution of partial differential equations, and (b) the parallelization of timedependent
applications for which spaceparallelism is unfeasible because of the very few degrees of freedom they involve.
To this effect, the fundamental principles of PITA are derived from classical domain decomposition techniques.
The timedomain is divided in timeslices whose boundary define a coarse timegrid. Seed values of the
timedependent solution are introduced on this coarse timegrid in order to initiate embarrassingly parallel
timeintegrations on the timeslices. A Newtonlike procedure is also designed to iteratively reduce the
solution jumps on the coarse timegrid. To avoid artificial resonance and numerical instability, a projector
is constructed for efficiently propagating a relevant component of the jump correction on the original (fine)
timegrid and accelerate convergence.
When the problem is timereversible, forward and backwardintime integrations can be carried out simultaneously
for an almost doubled parallel potential. In this talk, the application of these ideas to structural dynamics is
presented with experimental results and performance assessment.
8.19.09  Ulrich Hetmaniuk, guest speaker
Special finite element shape functions based on component mode synthesis
Abstract here
8.12.09  Kevin Carlberg (carlberg@stanford.edu)
Model reductionbased solution methods for realtime nonlinear simulation and repeated analyses
Model reduction is a powerful tool that can be used to dramatically
decrease the cost of numerical simulations in one of two contexts: the
realtime context and the repeated analyses context. However,
satisfying the computational time and accuracy requirements demanded by
problems in these categories is not a trivial task. To this end, one
model reductionbased solution method for each category is presented
that effectively meets the disparate demands of its respective context.
For realtime problems, extremely fast simulation time is required,
while moderate errors in the solution can be tolerated. In this talk, a
novel Gappy Proper Orthogonal Decomposition (POD) solution method is
presented for executing nonlinear simulations in near realtime. The
method is based on computing only a few (greedilychosen) rows of the
residual and Jacobian matrix at each Newton iteration.
Repeated analyses problems, such as those arising in design
optimization, are characterized by the need to solve consecutive linear
systems of equations of the form A^{i}x^{i}=b^{i}, where A^{i}
are nearby matrices. These problems typically have relatively stringent
accuracy requirements that typical model reduction techniques can fail
to meet. To this end, a novel PODbased iterative method is presented
for repeated analyses problems. The algorithm employs a POD basis to
accelerate convergence of solving each linear system.
8.5.09  Kevin Wang (icmewang@stanford.edu)
Computation of flowinduced forces on embedded meshes
Non bodyconforming CFD grids in which wet surfaces of various obstacles are embedded are gaining
popularity in many scientific and engineering applications. Despite many attractive advantages, these
methods tend to complicate issues such as enforcing the fluidstructure transmission conditions in general,
and the computation of the flowinduced loads on the wet surfaces of obstacles in particular.
In this talk, both of these issues are addressed. More specifically,
two load transfer algorithms are presented for fluid and
fluidstructure solvers based on embedded methods. In both algorithms,
the fluid Dirichlet boundary
condition ?—? or for fluidstructure interaction problems, the
kinematic transmission condition ?—?
is enforced via the analytical solution of an appropriate
onedimensional fluidstructure Riemann problem. As for load transfer,
the first algorithm, highlighted by conservation of energy,
approximates the structure wet surface by a selected set of control
volume faces, while the second algorithm reconstructs the structure wet
surface within the fluid mesh in order to obtain better accuracy. The
numerical properties of these methods are
theoretically analyzed. Their practical significances are also
highlighted for several threedimensional fluid and fluidstructure
interaction problems associated with underwater implosion and
aeronautical applications.
7.29.09  David Amsallem (amsallem@stanford.edu)
Algorithms for interpolation on matrix manifolds and application to model reduction
A new methodology is presented for interpolating parameterized
quantities belonging to matrix manifolds. It relies on identifying the
correct manifold for the given application, constructing the
appropriate logarithm mapping to move the interpolation data to a
tangent space to this manifold where a standard multivariate
interpolation algorithm can be applied, and constructing the
appropriate exponential mapping to bring back the computed result to
the manifold of interest. The proposed methodology is illustrated with
its application to realtime computational engineering using databases
of reducedorder models.
7.22.09  David Powell (dpowell1@stanford.edu)
Multiscale modeling and largescale transient simulation of ballistic fabric undergoing impact
Ballistic fabrics such as Kevlar and Zylon are finding new uses not
only as shielding for personnel but also in commercial and military
aircraft protecting flightcritical components in the event of a
highspeed ballistic impact. Since experimental tests on these
materials are often expensive and time consuming, a computational model
amenable to largescale numerical simulations on massively parallel
processors would provide an ideal alternative. To this effect, a novel
multiscale approach has been developed for modeling ballistic fabric
that extracts information from the microscale fibril material
properties to build the macroscale sheet. This approach incorporates
the stochastic nature of the material due to the random variations in
the yarn caused by the weaving process. Representing these variations
is critical to capturing the heterogeneous damage and failure of the
material as confirmed by experimental tests. A parallel implementation
of the proposed multiscale method has been developed with particular
attention to the contact problem. This talk will focus on all of these
issues and report on massively parallel largescale computations for
large sheet applications with both performance and validation results.
